Lapas

Download as pdf or txt
Download as pdf or txt
You are on page 1of 163

 

The Utility of Digital Aerial Surveys in Censusing  

Dipteryx panamensis, the Key Food and Nesting Tree of the  

Endangered Great Green Macaw (Ara ambigua) in Costa Rica 

by 

Sara L. M. Chun 

University Program in Ecology 
Duke University 
 

Date:_______________________ 
Approved: 
 
___________________________ 
Norman L. Christensen, Chair 
 
___________________________ 
Gary S. Hartshorn 
 
___________________________ 
Dean L. Urban 
 
___________________________ 
Robin L. Chazdon 
 

Dissertation submitted in partial fulfillment of 
the requirements for the degree of 
Doctor of Philosophy in the University Program in Ecology 
in the Graduate School 
of Duke University 
 
2008 
 

 
 

ABSTRACT 

The Utility of Digital Aerial Surveys in Censusing  

Dipteryx panamensis, the Key Food and Nesting Tree of the  

Endangered Great Green Macaw (Ara ambigua) in Costa Rica 

by 

Sara L. M. Chun 

University Program in Ecology 
Duke University 
 

Date:_______________________ 
Approved: 
 
___________________________ 
Norman L. Christensen, Chair 
 
___________________________ 
Gary S. Hartshorn 
 
___________________________ 
Dean L. Urban 
 
___________________________ 
Robin L. Chazdon 
 

An abstract of a dissertation submitted in partial 
fulfillment of the requirements for the degree of 
Doctor of Philosophy in the University Program in Ecology 
in the Graduate School 
of Duke University 
 
2008 

 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Copyright by 
Sara L. M. Chun 
2008 
 

 
 

Abstract

Remote sensing technologies offer an innovative way to study tropical forests 

and complement research from forest inventory plots.  From a logistic standpoint, 

remote sensing mitigates some of the physical, political, and economic challenges that 

complicate terrestrial studies of tropical forests.  From a biological standpoint, remotely 

sensed data can elucidate ecological phenomena and dynamics of tropical forests that 

may be challenging to observe on the ground.  I conducted a digital aerial survey during 

the peak blooming period for Dipteryx panamensis.  D. panamensis is a canopy‐emergent, 

keystone tree species within the lowland Atlantic forests of Costa Rica.  The aerial 

survey occurred over Maquenque National Wildlife Refuge, which is situated in the 

proposed San Juan‐La Selva Biological Corridor.  Maquenque was decreed in 2005 to 

protect the last breeding habitat for the endangered Great Green Macaw (Ara ambigua) in 

Costa Rica.  The Great Green Macaw depends on D. panamensis as its primary food and 

nesting resource.  However, because of Costa Rica’s complex deforestation history, the 

Great Green Macaw remains imperiled due to habitat fragmentation, degradation, and 

loss.  Only 25 to 35 breeding pairs and 250 individuals likely persist in the country.  My 

research had three main objectives.  First, I evaluated a classification technique that 

capitalized on the unique spectral signature of blooming D. panamensis in order to detect 

this species using a pattern recognition approach.  Second, I determined the landscape 

i v
 

density and distribution of this tree species using results from the aerial survey and 

classification.  I also noted environmental factors predicting potential D. panamensis 

habitat using a classification and regression tree (CART) model, and I subsequently 

calculated how much potential habitat exists in Costa Rica.  Third, I identified 

environmental conditions that indicate potential breeding habitat for the Great Green 

Macaw using a CART model, and I calculated how much potential breeding habitat 

exists in Costa Rica.  Results of the classification work indicate that D. panamensis can be 

identified based on its unique spectral signature.  In particular, the IHS (intensity, hue, 

saturation) color space was effective at resolving this tree.  Results of the density 

analysis suggest that canopy‐emergent D. panamensis trees occur at densities as high as 2 

trees/hectare.  This work demonstrates that D. panamensis exists in denser patches than 

previously determined from forest inventory plots.  Environmental conditions that will 

support high densities of D. panamensis habitat occur between 45 and 125 meters in 

elevation and on soil with an acid and clay profile.  Nearly 240,000 hectares could 

support high density patches of D. panamensis in Costa Rica.  Potential Great Green 

Macaw breeding habitat is defined by the density of D. panamensis trees within its 550‐

hectare breeding territory.  Approximately 67,000 hectares of Great Green Macaw 

breeding territory exist in Costa Rica.  Ultimately, the conservation of both D. panamensis 

and the Great Green Macaw in Costa Rica may require a multinational partnership 

between Costa Rica and Nicaragua.

 v
 

Dedication

Dedicated to the memories of my grandmothers, Popo Chun and Popo Lee, two 

women whose own journeys spirited me to where I am today.

v
 i
 

Contents

Abstract .........................................................................................................................................iv

List of Tables ................................................................................................................................xii

List of Figures ............................................................................................................................ xiii

Acknowledgments ......................................................................................................................xv

Chapter 1.  Introduction............................................................................................................... 1

1.1 A land of pura vida .......................................................................................................... 1

1.2 A bit about Costa Rica’s biodiversity ............................................................................ 5

1.3 A pock on the landscape:  Costa Rica’s history of deforestation ............................... 6

1.4 Costa Rica’s system of protected areas........................................................................ 10

1.5 New conservation directions ........................................................................................ 11

1.5.1 Background on Maquenque National Wildlife Refuge and the proposed San 
Juan‐La Selva Biological Corridor .......................................................................... 12

1.5.2 Current land use and human dimensions within the San Juan‐La Selva 
Biological Corridor.................................................................................................... 12

1.5.3 San Juan‐La Selva Biological Corridor design....................................................... 14

1.6 Research opportunities within the San Juan‐La Selva Biological Corridor............ 16

1.7 Synopses of dissertation chapters ................................................................................ 16

Chapter 2.  An evaluation of digital aerial photography as a means to survey Dipteryx 
panamensis, a canopy‐emergent tree species, during its peak blooming in 
Costa Rica................................................................................................................ 19

2.1 Introduction..................................................................................................................... 19

2.1.1 Remote sensing studies of tropical forests compared to forest  
inventory plots .......................................................................................................... 19

vii
 
 

2.1.2 Overview of remote sensing platforms used to identify individual tree crowns 
in the tropics .............................................................................................................. 22

2.1.3 Objectives.................................................................................................................... 26

2.2 Materials and methods .................................................................................................. 27

2.2.1 Study site:  San Juan‐La Selva Biological Corridor............................................... 27

2.2.2 Aerial survey technique............................................................................................ 29

2.2.3 Digital photographs .................................................................................................. 32

2.2.4 Image enhancement:  contrast stretch .................................................................... 34

2.2.5 Image enhancement:  conversion to intensity, hue, and  
saturation color space ............................................................................................... 37

2.2.6 Supervised classification .......................................................................................... 37

2.2.7 Post‐classification image processing:  recode........................................................ 41

2.2.8 D. panamensis clump density determination.......................................................... 42

2.2.9 Classification error assessment................................................................................ 43

2.3 Results .............................................................................................................................. 46

2.3.1 Comparison of image enhancement transformations .......................................... 46

2.3.2 Signature separability ............................................................................................... 46

2.3.3 Classification error assessment................................................................................ 49

2.3.4 High density images ................................................................................................. 50

2.4 Discussion........................................................................................................................ 50

Chapter 3.  Density distribution, habitat description, and fragmentation of a population 
of Dipteryx panamensis in northern Costa Rica................................................... 54

3.1 Introduction..................................................................................................................... 54

3.1.1 Objectives.................................................................................................................... 55
viii
 
 

3.2 Background ..................................................................................................................... 56

3.2.1 Overview of biogeographic studies in the tropics ................................................ 56

3.2.1.1 Limitations of terrestrial forest inventory plots............................................. 56

3.2.1.2 The importance of documenting the distribution of keystone species ...... 57

3.2.2 Background on study species:  D. panamensis ....................................................... 58

3.2.2.1 Conservation status and keystone species designation................................ 58

3.2.2.2 Phenological patterns ........................................................................................ 59

3.2.2.3 Interactions with other organisms................................................................... 61

3.2.2.4 Threats and conservation measures ................................................................ 62

3.2.2.5 Habitat requirements, distribution, and abundance..................................... 62

3.2.2.6 Pollen dispersal patterns for D. panamensis.................................................... 63

3.3 Materials and methods .................................................................................................. 65

3.3.1 Study Site:  San Juan‐La Selva Biological Corridor .............................................. 65

3.3.2 Landscape distribution of D. panamensis................................................................ 65

3.3.2.1 Digital aerial surveys......................................................................................... 65

3.2.2.2 D. panamensis density determinations:  semi‐automated supervised 
classification and manual density approach................................................. 65

3.2.2.3 Creation of a D. panamensis density map using the IDW interpolation 
method................................................................................................................ 68

3.3.3 Density trends over the study landscape and forested areas.............................. 71

3.3.4 Environmental factors predicting high density locations of D. panamensis:  a 
classification tree approach ..................................................................................... 71

3.3.5 Delineation of potential D. panamensis habitat ...................................................... 77

3.3.6 Fragmentation analysis of moderate density D. panamensis patches ................. 77
 ix
 

3.4 Results and discussion................................................................................................... 80

3.4.1 Landscape distribution of D. panamensis................................................................ 80

3.4.2 Density trends over the study landscape and forested areas.............................. 80

3.4.3 Environmental factors that predict locations of high D. panamensis density.... 86

3.4.4 Delineation of potential D. panamensis habitat ...................................................... 86

3.4.5 Degree of fragmentation of moderate density D. panamensis patches ............... 88

3.5 Conclusions ..................................................................................................................... 93

Chapter 4.  Determination and description of potential nesting habitat for the Great 
Green Macaw (Ara ambigua) in northern Costa Rica using a classification and 
regression tree model ............................................................................................ 97

4.1 Introduction..................................................................................................................... 97

4.1.1 Objectives.................................................................................................................... 99

4.2 Background ..................................................................................................................... 99

4.2.1 General description and ecology of the Great Green Macaw in Costa Rica ..... 99

4.3 Materials and methods ................................................................................................ 103

4.3.1 Determination of Great Green Macaw nesting habitat ...................................... 103

4.3.2 Existing Great Green Macaw nesting data .......................................................... 104

4.3.3 Development of Great Green Macaw nesting habitat model:  a classification 
tree approach ........................................................................................................... 105

4.4 Results and discussion................................................................................................. 108

4.4.1 Classification tree model ........................................................................................ 108

4.4.2 Environmental conditions predicting Great Green Macaw habitat ................. 108

4.4.3 Location of potential Great Green Macaw nesting habitat................................ 110

4.5 Conclusions ................................................................................................................... 112


 x
 

Chapter 5.  General conclusions.............................................................................................. 117

Appendix A:  Newspaper article published in La Nación describing “suspicious” 
overflights ........................................................................................................ 120

Appendix B:  Results of the signature separability matrix ................................................. 122

Appendix C:  Recent ruling that bans the use, exploitation, or extraction of  
D. panamensis ................................................................................................... 128

References .................................................................................................................................. 130

Biography................................................................................................................................... 143

 xi
 

List of Tables

Table 1:  Number of photos sampled for the classification error assessment .................... 44

Table 2:  Results of the classification error assessment.......................................................... 50

Table 3:  Response and predictor variables used in CART model to predict potential D. 
panamensis habitat ....................................................................................................... 75

Table 4:  Summary of response variable bins.......................................................................... 76

Table 5:  Description of bins for fragmentation analysis ....................................................... 78

Table 6:  Percentage of the total and forested landscape represented by each D. 
panamensis density bin .............................................................................................. 82

Table 7:  Summary of FRAGSTATS metrics............................................................................ 89

Table 8:  Response and predictor variables used in the CART model to predict potential 
Great Green Macaw habitat ...................................................................................... 107

xii
 
 

List of Figures

Figure 1:  Location of Costa Rica within Central America and protected areas in Costa 
Rica................................................................................................................................ 2

Figure 2:  Location of Maquenque National Wildlife Refuge and proposed San Juan‐La 
Selva Biological Corridor ........................................................................................... 4

Figure 3:  Digital camera platform and airplane used in aerial surveys ............................. 31

Figure 4:  Location of aerial surveys over the proposed San Juan‐La Selva Biological 
Corridor ...................................................................................................................... 33

Figure 5:  Sequence of image enhancement steps................................................................... 47

Figure 6:  Sample signature separability plot.......................................................................... 48

Figure 7:  Sample images with high densities of D. panamensis trees.................................. 51

Figure 8:  Land use classification within proposed San Juan‐La Selva  
Biological Corridor.................................................................................................... 72

Figure 9:  Density distribution of D. panamensis ..................................................................... 81

Figure 10:  Percentage of the total landscape represented in various D. panamensis 
density bins .............................................................................................................. 83

Figure 11:  Percentage of forested landscape represented in various D. panamensis 
density bins .............................................................................................................. 84

Figure 12:  Potential habitat that can support D. panamensis > 0.4 trees/hectare................ 87

Figure 13:  Distribution of 13 subpopulation patches with moderate D. panamensis 
density....................................................................................................................... 90

Figure 14:  Connected moderate density D. panamensis patches.......................................... 92

Figure 15:  Location of Indio‐Maíz Biological Reserve, Nicaragua in relation to potential 
D. panamensis habitat in Costa Rica....................................................................... 95

Figure 16:  Potential Great Green Macaw nesting habitat within Costa Rica .................. 111

xiii
 
 

Figure 17:  Location of Indio‐Maíz Biological Reserve, Nicaragua in relation to potential 
Great Green Macaw nesting habitat in Costa Rica ........................................... 115

xiv
 
 

Acknowledgments

Like most doctoral students, I never truly believed that I would arrive at this 

stage of the dissertation process.  To me, these final words of thanks represented some 

imaginary endpoint that would never be realized.  But here I am writing my 

acknowledgements.  I’m reminded of the plucky bunny hero in my favorite children’s 

book, “The Velveteen Rabbit” and his quest to be real and not just a stuffed animal that 

sat on a dusty shelf.  Well, I feel that all of the blood, sweat, and tears that I shed along 

my doctoral journey have finally transformed my thesis from the abstract into the real.  

And this realness is palpable now. 

Of course, many people assisted in the alchemy of converting my ideas into 

reality.  First and foremost, I need to recognize my two co‐advisors, Dr. Norm 

Christensen and Dr. Gary Hartshorn.  Your personalities and mentorship styles couldn’t 

have been more complementary.  Norm, thank you for your sage advice and affable 

demeanor throughout these years.  Though I have relished all of our conversations, I 

have particularly enjoyed listening to all of your musings on life.  And I never cease to 

be impressed by the depth and breadth of your ecological knowledge.  Gary, thank you 

for taking me into the field and teaching me about the Neotropics.  I will never forget 

driving the bumpy roads through the San Juan‐La Selva Corridor, hanging out in Boca 

Tapada, and sharing a hike on your favorite La Selva trail.  You made me appreciate 

Costa Rica as much as you do!  I must also acknowledge my two other committee 
xv
 
 

members, Dr. Robin Chazdon and Dr. Dean Urban.  Robin, I cherish the moments I 

overlapped with you at La Selva and learned from you.  You thought about my project 

even from afar and always had an interesting take on it.  I also enjoyed getting to know 

your family, and I admire how integral they are in your professional life.  Dean, thank 

you for sharing your knowledge on landscape ecology and getting me up to speed on R.  

I appreciate you welcoming me into your lab group meetings early on.  And your wry 

sense of humor always makes my day! 

Because of the international nature of my project, I have many people to 

acknowledge both near and far.  My pie‐in‐the‐sky idea of flying over the rain forest 

began with an innocent conversation with Dr. George Powell after I chatted with him 

about the Great Green Macaw and almendro tree back in the summer of 2004.  It’s 

probably a good thing I didn’t have a sense of all the places this project could go 

wrong—or that I would one day be mistaken as a drug trafficker!  Jorge Carranza 

Sánchez with Mexico’s Comisión Nacional de Áreas Naturales Protegidas (CONANP) 

took my idea and made it happen.  I owe him a huge debt of gratitude for traveling to 

Costa Rica and assisting me with the aerial photography.  Then, the Costa Rican pilots of 

AeroBell gave my project wings—literally.  I thank Olivier Chassot and Guisselle Monge 

Árias with the Great Green Macaw Research and Conservation Project.  Not only have 

they given me access to data, but they have almost singlehandedly brought the vision of 

Maquenque National Wildlife Refuge to life.  At the La Selva Biological Station, Drs. 

xvi
 
 

David and Deborah Clark are tropical ecologists extraordinaire.  In particular, David has 

shared his data on the big trees of La Selva and helped me think about my project.  Dr. 

Deedra McClearn gave me insight into the almendro’s flowering patterns.  I thank 

Marcia Snyder, former GIS manager at La Selva, for her geospatial assistance.  I also 

thank the many local scientists and staff at La Selva who helped with my travel and 

research logistics.  Finally, I am grateful to John and Chris Thomas for opening their 

home in Alajuela and providing me with fellowship, family, and food during my many 

trips to Costa Rica! 

At Duke University, many people have helped me as I wrangled with oodles of 

digital images and grappled with GIS.  I want to especially acknowledge Pete Harrell and 

Dr. Clinton Jenkins.  Both were hugely generous with their time and helped me surmount 

many roadblocks when I got stuck.  The following also assisted at various points during 

my analyses:  Dr. Pat Halpin, Joe Sexton, Ron Sutherland, and Dr. Jennifer Swenson.  The 

Nicholas School IT kept my computer running smoothly.  And Nancy Morgans and Meg 

Stephens both helped me navigate the inner workings of the Nicholas School. 

I have had the privilege of traversing the doctoral path in the company of many 

good friends.  We mutually supported each other and reminded one another that there 

is a life outside of school.  I feel especially blessed by my female friends who shared this 

academic journey with me.  I would like to say a heartfelt thanks to Ariana Sutton‐Grier, 

Dawoon Jung, Erica Tsai, Julie DeMeester, Melissa Kenney, Michelle Hersh, and Regina 

xvii
 
 

Peón Díaz Barriga.  I also would like to thank the other halves—Carl Salk, Jackson Fox, 

Nat Grier, and Paul Gong.  Thank you to my friend Rodney Vargas for all the times we 

lunched at Guglhupf and cheered on the Blue Devils.  To Jeff Clark, I enjoyed your real‐

world take on environmental issues.  Finally, I offer a sincere thank‐you to my adopted 

Chinese family, the Kaos (Rom, Jeannie, Andrew, and Bianca).  A chance meeting during 

my first day in Durham led to this lifelong friendship.  I have enjoyed all the afternoons 

making dumplings, and it’s been a delight to watch Bianca and Andrew grow up to such 

wonderful young adults. 

Now I come to my real family.  What can I say?  You have all been with me 

through every twist and turn in my life and allowed me the latitude to become who I 

am.  Thank you.  Mom and Dad, you accompanied me as my Duke adventure started.  

Wasn’t that cross‐country trip fun?  I’ll never forget it.  To my Mom, Ann Chun, I thank 

you for dreaming big dreams for me yet always reminding me to stay grounded.  To my 

Dad, Robert Chun, I thank you for all the little ways you show you care—like my car, 

the VIA magazines, and all the NY Times puzzles!  To my younger brother, Matthew, 

somehow we both found ourselves sharing this PhD experience together.  I think our 

parents have often wondered what they’d done wrong to have two kids suffering 

through their PhDs!  Well, of all the times we asked each other, “Why?” I know we both 

were secretly cheering each other on.  Thanks! 

xviii
 
 

Finally, I owe my deepest gratitude to my best friend, Jerry Hsu.  Although we 

were both together and apart, you walked beside me during every step of this journey 

and always inspired me to do my best.  You raised my spirits when I was down, buoyed 

my confidence when it wavered, and provided never‐ending entertainment when I 

needed a laugh.  Grand adventures now await us.  I know it!  

* * * 

I thank the following funding sources that made my doctoral research possible:  

Duke University Graduate School, Duke University Center for Latin American & 

Caribbean Studies, Organization for Tropical Studies, Sigma Xi Scientific Research 

Society, Kuzmier‐Lee‐Nikitine Fund, and the Lazar Scholarship through the Nicholas 

School’s Student International Discussion Group.  

xix
 
 

Chapter 1. Introduction

1.1 A land of pura vida

In 2007, more than 1.9 million international tourists (Fallas 2007) made their way 

to the tiny country of Costa Rica, which is nestled within the Central American land 

bridge between North and South America (Figure 1).  Many of these tourists came 

searching for adventure or sanctuary amongst the country’s dramatic volcanoes, lush 

forests, and picturesque beaches.  Whether these visitors identified themselves as 

ecotourists1 or not, most were engaged in some form of nature‐based tourism.  More 

than half of all nonresident visitors explored a national park or protected zone during 

their trip (ICT 2006).  And these visitors came to the right place, to a country whose 

unofficial motto of pura vida or “pure life” encapsulates the spirit of a people renowned 

for their conservation ethic and value of the natural world.   

There is ample evidence for why Costa Rica is heralded as an exemplar of how to 

conduct tropical conservation in a developing country.  Though only 51,100 km2 

(roughly the size of the state of West Virginia), Costa Rica boasts over 150 protected 

areas such as national parks, biological preserves, and wildlife refuges that cover nearly 

25% of its national territory (Sanchez‐Azofeifa et al. 2003, WRI 2003).  Maquenque  

                                                      

1 While many definitions of ecotourism abound, The International Ecotourism Society (2008) has defined it as 
“responsible travel to natural areas that conserves the environment and improves the wellbeing of local 
people”. 
 1
 

Figure 1:  Location of Costa Rica within Central America and protected areas in Costa 
Rica 

 2
 

National Wildlife Refuge (MNWR), which is embedded within the proposed San Juan‐

La Selva (SJLS) Biological Corridor (Figure 2), is one of the newest areas to join that list 

of protected areas.  Decreed in 2005, though not yet operational, MNWR may soon draw 

some of Costa Rica’s 1.9 million visitors.  If lucky, these tourists may catch a glimpse of 

the dazzling Great Green Macaw (Ara ambigua 2), which feeds on and nests in Dipteryx 

panamensis 3, a widespread tree of the lowland Atlantic forest. 

Yet the story of MNWR and the rest of Costa Rica’s protected areas is a complex 

one.  Although this country is a model for tropical conservation and has pioneered 

innovative strategies like debt‐for‐nature swaps, bio‐prospecting, and payments for 

ecosystem services, Costa Rica had one of the highest deforestation rates during the 20th 

century among Latin American countries.  While this pattern of land clearing 

dramatically altered the natural landscape, it also galvanized a cadre of local biologists 

such as Mario Boza and Alvaro Ulgado (Boza 1993).  And this spirit of pura vida 

continues to pave the way for new conservation efforts like MNWR and the SJLS 

Corridor that will be discussed later. 

                                                      

2 Ara ambiguus is also an accepted scientific name for the Great Green Macaw.  Please see David and 
Gosselin (2002). 
3 Based on a recent nomenclatural ruling, this tree should be referenced as Dipteryx oleifera (Stevens et al. 

2001).  However, published studies on this tree in Costa Rica continue to use Dipteryx panamensis.  Hence, I 
am following that precedent in this document. 
 3
 

Figure 2:  Location of Maquenque National Wildlife Refuge and proposed San Juan‐
La Selva Biological Corridor 

 4
 

1.2 A bit about Costa Rica’s biodiversity

Costa Rica exhibits exceptional biodiversity and a diverse array of ecosystems.  

Biological inventories have identified over 200 species of mammals, 845 species of birds, 

160 species of amphibians, 218 species of reptiles, and 10,000 species of vascular plants 

comprising 4% of the total number of the world’s plant species (Boza 1993).  Costa Rica 

also exhibits a high degree of endemism.  For example, Hartshorn and Hammel (1994) 

estimate that 10% of the plants found in Costa Rica’s Atlantic lowlands are endemic to 

the country.  Scientists attribute this disproportionate amount of biodiversity to the fact 

that Costa Rica is located along the Mesoamerican land bridge that links the continents 

of North and South America and has facilitated the dispersal of plants and animals over 

time (Wallace 1997).  Costa Rica also has many diverse and unique ecosystems.  In fact, 

this country can be classified into 23 different life zones (Powell et al. 2000).  The life 

zone system uses gradients of mean temperature, annual precipitation, and 

evapotranspiration to predict plant communities (Holdridge 1947, Holdridge et al. 1971).  

Costa Rica is also included in the biodiversity hotspot of Mesoamerica due to this 

region’s high concentration of endemic species yet exceptional loss of habitat overall 

(Myers et al. 2000). 

 5
 

1.3 A pock on the landscape: Costa Rica’s history of


deforestation

It may seem odd that a country that has protected nearly 25% of its national 

territory had one of the highest tropical deforestation rates during the 20th century.  Yet 

it is a testament to how far Costa Rica has come in terms of balancing economic growth 

with natural resource preservation.  The arrival of the Spanish in the 1500s marked the 

beginning of sporadic forest clearing in Costa Rica.  This trend accelerated until the mid 

1900s as thousands of hectares of forests were cleared for agriculture and cattle pastures 

(Pfaff and Sanchez‐Azofeifa 2004).  Many researchers attribute the rise in forest clearing 

following World War II to various demographic, social, and political factors at both 

national and international levels that put pressure on the country’s natural resources.  

First, Costa Rica’s population expanded four‐fold in the two generations following 

World War II from less than 800,000 to over 3 million people (Rosero‐Bixby and Palloni 

1998).  By the 1960s, Costa Rica had one of the world’s fastest population growth rates—

a 4% natural increase/year (Rosero‐Bixby and Palloni 1998).  The annual population 

growth rate has slowed considerably and is projected to be 1.5% between 2005 and 2010; 

the total population is about 4.5 million people (United Nations Economic and Social 

Affairs 2006).  Demographers and ecologists have acknowledged a causal link between 

pressures from population growth and increased natural resource use (Rosero‐Bixby 

and Palloni 1998).  The peak in population growth during the 1960s stemmed from 

 6
 

Costa Rica’s national policies that prioritized demographic growth and agricultural 

production (Pfaff and Sanchez‐Azofeifa 2004).  This ultimately led to forest clearing.   

Costa Rica’s presence on the international commodities market during the 20th 

century also affected spatiotemporal patterns of deforestation and influenced which 

crops were planted in the wake of forest clearing.  In the early 1900s, lands that could 

grow coffee were cleared first (Pfaff and Sanchez‐Azofeifa 2004).  These areas were 

concentrated in the central volcanic region of the country.  Depending on the shifting 

prices of various commodities throughout the 20th century, forest was also removed to 

support crops like banana, sugar cane, pineapple, and cattle.  The latter resulted in what 

Myers (1981) termed the “hamburger connection”, as foreign appetite for cheap beef 

resulted in the widespread clearing of tropical forests in Central and South America in 

order to raise cattle.  

So how have these demographic and economic factors affected land‐use patterns, 

and how much forest was lost during the peak of Costa Rica’s land clearing?  Ultimately, 

there have been numerous studies tracking Costa Rica’s deforestation trends, and new 

remote sensing technologies have afforded a more precise documentation of land‐use 

patterns.  Joyce (2006) has prepared a comprehensive resource that summarizes land‐use 

change in Costa Rica from 1966 to 2006.  Here, I will just highlight some major trends 

regarding deforestation rates.  Based on aerial photos and Landsat Multispectral Scanner 

data collected from five individual years between 1940 and 1983, Sader and Joyce (1988) 

 7
 

note that the total primary forest remaining in 1940, 1950, 1961, 1977, and 1983 was 67%, 

56%, 45%, 32%, and 17%, respectively.  Furthermore, deforestation occurred primarily in 

tropical dry and moist life zones during the early period.  Between 1977 and 1983, the 

annual forest clearing rate was 8.8% in tropical wet forest, and over 78% of the primary 

forest in this life zone had been lost.  During this same period, tropical moist forest was 

lost at an average annual rate of 16.4% and almost no primary forest in this life zone 

remained by 1983.  Many scientists concur that Costa Rica’s deforestation rate peaked in 

the 1970s.  In fact, Costa Rica had the fifth highest deforestation rate in the world from 

1976 to 1980 as calculated by the United Nations Food and Agricultural Organization 

(Sanchez‐Azofeifa et al. 2001).  The annual deforestation rate remained high between 

1986 and 1991 as forests were cleared annually at 4.2%/year (Sanchez‐Azofeifa et al. 

2001).  By the end of this period, almost all forest within the tropical moist forest and 

premontane moist forest life zones had been eliminated.   

In addition, Costa Rica’s social and political history may also be inextricably 

linked to its deforestation record.  Costa Rica managed to avoid most of the strife, civil 

wars, and economic turmoil that plagued many of its Mesoamerican neighbors like 

Nicaragua.  In fact, Costa Rica suffered only one short civil war in 1948 after which the 

army was disbanded; to this day, Costa Rica has no standing army (Campbell 2002).  

This enlightened view of the military enabled the country to focus its spending on social 

programs that benefit its populace.  Costa Rica’s universal healthcare system is one of 

 8
 

the best in Latin America and comparable to industrialized nations (Rosero‐Bixby and 

Palloni 1998).  For example, life expectancy is on par with the United States at 79.7 and 

75.0 years for females and males, respectively.  In addition, this country has a well‐

developed educational system and boasts one of the most educated citizenry with a 98% 

literacy rate for youth between 15 and 24 years of age according to WRI (2003).  Such 

progressive national policies have resulted in a populace accustomed to a standard of 

living akin to developed countries, putting pressure on Costa Rica’s natural resources.   

However, these economic and social successes have put this country in the 

position to support and nurture national conservation efforts.  The 1970s were the 

inception of Costa Rica’s national parks system as a cadre of conservation‐minded 

people decided to do something about Costa Rica’s rampant loss of primary forest.  

Deforestation has slowed in the last two decades.  Coupled with reforestation efforts 

largely begun after 1979 as well as widespread secondary forest regeneration, some of 

the damage done to Costa Rica’s forests has been mitigated (Butterfield 1994, Chazdon 

2003).  Foreign tourists are now flocking to this country that offers the political stability 

and amenities not present in many Latin American countries.  In addition, new 

economies like ecotourism offer a way to bridge economic development with 

sustainable natural resource use. 

 9
 

1.4 Costa Rica’s system of protected areas

Costa Rica’s nascent conservation movement was borne out of a desire to protect 

areas of natural beauty in the wake of high deforestation in the 1960s.  Boza (1993) 

acknowledges that the first national parks were created in areas of stunning scenic 

beauty, on historic sites of national significance, or in areas of demonstrated 

conservation importance.  Thus, the first national parks were created in the early 1970s 

and included Poás, Cahuita, and Santa Rosa National Parks.  At this time, the early 

conservationists had just five guards and a vehicle, but little experience and no 

administrative support (Boza 1993).  From these humble beginnings, Costa Rica now 

boasts over 150 protected areas that include not only national parks but wildlife refuges 

and national monuments that attract some of the 1.9 million annual visitors.  These 

protected areas are now managed under the National System of Conservation Areas 

known as SINAC, which is part of the government’s Ministry of Environment and 

Energy (MINAE).  Even though over 25% of the country’s national territory is 

designated as some type of protected area, only 12 of Costa Rica’s 23 life zones are 

adequately represented with more than 5,000 ha under conservation (Powell et al. 2000).  

Those 12 life zones account for 98% of the total area protected in the country.  The 

remaining 11 life zones are represented in just 2% of the total area protected in the 

country, or not represented at all.  The majority of the protected areas are concentrated 

along higher elevation regions running along Costa Rica’s volcanic backbone.  Life zones 

1  0
 

in lowland areas are less well represented as these areas were settled by humans and 

were more susceptible to forest clearing.  For example, only 9% of the tropical wet forest 

life zone is protected (Powell et al. 2000).  Not only do Powell et al. (2000) advocate the 

expansion of existing protected areas in order to target underrepresented life zones, but 

they also champion the establishment of corridors in order to facilitate the movement of 

organisms with large habitat needs and foster genetic flow.  Corridors have long been 

proposed by conservation biologists as a means to link species, habitats, or processes 

that operate at multiple spatial and temporal scales (Noss 1991).  Despite Costa Rica’s 

impressive conservation achievements, it is apparent that more can still be done. 

1.5 New conservation directions

With all that Costa Rica has achieved, it is not resting on its laurels.  Decreed in 

2005, Maquenque National Wildlife Refuge (MNWR) represents the newest addition to 

Costa Rica’s network of protected areas.  Because MNWR is located within lowland 

Atlantic rain forest, it fills some of the gaps missing in the system of protected zones.  In 

addition, it represents the core area within a larger biological corridor effort known as 

the San Juan‐La Selva (SJLS) Biological Corridor that would link protected areas in 

southern Nicaragua to Costa Rica’s Central Volcanic Range reserve.  MNWR and the 

greater SJLS Corridor are the setting for the research contained in this dissertation.  

1  1
 

1.5.1 Background on Maquenque National Wildlife Refuge and the


proposed San Juan-La Selva Biological Corridor

The SJLS Corridor originated in March 2001 when the SJLS Executive Committee 

was formed to implement the vision of the Mesoamerican Biological Corridor (MBC) 

project within Costa Rica.  The MBC is a transnational effort among Guatemala, Belize, 

El Salvador, Honduras, Nicaragua, Costa Rica, and Panama, as well as the five southern 

states of Mexico (Kaiser 2001).  Established in 1997, the purpose of the MBC project is to 

unite protected areas throughout the Central American isthmus by either expanding 

extant protected areas or restoring intervening lands.  The SJLS Corridor represents one 

span of the MBC that will link five protected areas in southern Nicaragua (including the 

264,000 ha Indio‐Maíz Biological Reserve) with Costa Rica’s Central Volcanic Range 

reserve (Chassot et al. 2004).  It comprises 246,608 ha of fragmented lands and is located 

in northern Costa Rica within the provinces of Heredia and Alajuela in the Huetar Norte 

region of the country.  The SJLS Corridor takes its name after the San Juan River that 

forms the border between Costa Rica and Nicaragua and the La Selva Biological Station, 

a field station operated by the Organization for Tropical Studies in Costa Rica. 

1.5.2 Current land use and human dimensions within the San Juan-La
Selva Biological Corridor

An estimated 45,000 people live in the SJLS corridor’s several towns and smaller 

population centers (Chassot et al. 2004).  Approximately, 30,000 people inhabit 

communities adjacent to the corridor such as Puerto Viejo, La Virgen, and Horquetas de 

1  2
 

Sarapiquí.  According to official census data from 2000, about 30 inhabitants per square 

kilometer reside within the proposed corridor region (INEC 2001).  The private 

environmental research firm CEDARENA (Centro de Derecho Ambiental y de los 

Recursos Naturales) conducted an exhaustive land tenure survey to identify property 

owners and delineate inholdings within the Huertar Norte region.  They ultimately 

identified over 700 private farms with a registered title within MNWR.  As the refuge is 

developed, compensation and relocation of land owners will be a significant challenge. 

Land use within the proposed corridor is varied and includes primary forest, 

secondary forest, deforested areas, and agricultural zones.  Much of the non‐forested 

areas comprise a matrix of abandoned cattle pastures, tree plantations, and various 

agricultural crops.  This region’s land conversion can be attributed to government 

incentives that encouraged the expansion of the agricultural frontier and the clearance of 

forest for cattle pasture (Butterfield 1994).  With the advent of numerous forestry laws as 

well as a payments for ecosystem services program, much of this previously cleared 

land is either regenerating naturally or being reforested.  Landowners who maintain 

their land in ways that promote the following services—mitigation of greenhouse gas 

emissions; water use for domestic purposes, irrigation, and hydroelectricity; biodiversity 

conservation; or ecotourist and recreational purposes—are allowed to apply to the 

payments for ecosystem services program for financial remuneration (Snider et al. 2003).   

1  3
 

1.5.3 San Juan-La Selva Biological Corridor design

The SJLS Executive Committee has outlined a strategy to consolidate the matrix 

of public and private lands and implement the SJLS Corridor (Chassot et al. 2004).  

Lands will be purchased, and the property rights transferred to SINAC.  The initial 

design for the SJLS Corridor integrates three land‐use levels with increasing levels of 

biodiversity protection and ecosystem integrity within the core (Figure 2).  These levels 

include 1) Core Protected Area, 2) Corridor Nuclei, and 3) Corridor Matrix.  The goal of 

the SJLS corridor initiative is to protect 100% of the core protected area and nuclei and 

50% of the matrix (Chassot et al. 2004).  A more detailed description of each region 

follows.   

MNWR forms the heart of the SJLS Corridor, as it constitutes the Core Protected 

Area.  It was officially decreed in 2005.  Totaling approximately 59,700 ha or 24% of the 

corridor, MNWR is situated just south of the Nicaraguan border.  Its boundaries are 

roughly circumscribed by three rivers, the Río San Juan to the north and the Río Toro 

and Río Sarapiquí to the southeast.  The Río San Carlos flows through the reserve and 

divides it into two sections.  MNWR comprises intact lowland Atlantic forest dominated 

by a high density of the tree, D. panamensis.  Lowland Atlantic forest is an ecosystem 

type that is poorly represented in Costa Rica’s suite of protected areas.  In addition to 

safeguarding a unique habitat assemblage, one of the primary purposes of MNWR is to 

protect the dwindling population of the Great Green Macaw, which depends on D. 

1  4
 

panamensis as a food resource and nesting substrate.  MNWR forms the core of the 

proposed SJLS Corridor by uniting other wildlands and protected areas throughout 

Costa Rica and Nicaragua such as Barra del Colorado Wildlife Refuge to the east, La 

Selva Biological Station and Braulio Carrillo National Park to the south, and the Indio‐

Maíz Biological Reserve to the north in Nicaragua. 

The Corridor Nuclei consist of three blocks of habitat constituting about 41,150 

ha or about 17% of the SJLS Corridor (Chassot et al. 2004).  Tirícias‐Crucitas, the largest 

of the nuclear patches, totals 27,273 ha and would buffer protected areas in adjoining 

Nicaragua.  The Cerros Astilleros‐Lomas de Sardinal nuclear area encompasses 9,043 ha, 

while the Cerros Arrepentidos nuclear area comprises 3,838 ha.  While these nuclei have 

been identified, none has officially been protected.  Nor have any of the blocks been 

prioritized for acquisition. 

The remainder of the SJLS Corridor that does not comprise MNWR or the 

Corridor Nuclei is designated as Corridor Matrix.   Currently, the matrix is the most 

fragmented and degraded portion of the corridor and includes the highest degree of 

human habitation and varied land uses.  This region presents the highest opportunities 

for restoration or natural forest regeneration. 

1  5
 

1.6 Research opportunities within the San Juan-La Selva


Biological Corridor

The development of MNWR and the SJLS Corridor adds an exciting chapter to 

Costa Rica’s complex yet extraordinary conservation story.  Borne as a reaction to the 

troubling patterns of deforestation during the 20th century, Costa Rica’s conservation 

movement now has coalesced to consider the visionary SJLS Corridor project, which has 

ecological ramifications beyond the country’s borders.  This ongoing conservation 

project has provided numerous opportunities for scientific research, and I am grateful 

for the opportunity to participate in and contribute to the ongoing SJLS corridor efforts.  

It has been an exciting process to observe!  While the doctoral research contained in this 

dissertation addresses broad ecological questions, it is my hope that this research may 

also assist the ongoing conservation efforts in the proposed SJLS Corridor. 

1.7 Synopses of dissertation chapters

Chapter 2 is entitled “An evaluation of digital aerial photography as a means to 

survey Dipteryx panamensis, a canopy‐emergent tree species, during its peak blooming in 

Costa Rica”.  Here, I examine the utility of digital aerial surveys that coincided with the 

unique phenological period for D. panamensis to analyze this tree species’ distribution 

and density across the SJLS Corridor.  The primary purpose of this research was to 

develop a semi‐automated classification approach that keyed in on the unique spectral 

signature of blooming D. panamensis trees.  This tree species is a keystone species and an 

1  6
 

important nesting and food resource for the Great Green Macaw, the flagship organism 

around which MNWR was created.  Ultimately, this chapter highlights advances in 

remote sensing technologies.  In addition, it justifies the use of digital aerial 

photography as a cost‐effective means of surveying tropical forests. 

In Chapter 3, which is titled “Density distribution, habitat description, and 

fragmentation of a population of Dipteryx panamensis in northern Costa Rica”, I utilize 

the density estimates distilled from the aerial photography methods described in 

Chapter 2 to create an interpolated surface showing the distribution of D. panamensis 

throughout the SJLS Corridor.  In addition, I identify environmental factors that predict 

hotspots of high D. panamensis density by using a classification and regression tree 

(CART) model.  Finally, I examine whether patches of D. panamensis are disconnected 

from each other based on measured pollen dispersal distances.  

In the final chapter entitled “Determination and description of potential nesting 

habitat for the Great Green Macaw (Ara ambigua) in northern Costa Rica using a 

classification and regression tree model”, I examine what environmental factors predict 

optimal Great Green Macaw nesting habitat.  Because the Great Green Macaw relies on 

mature D. panamensis trees as a nesting substrate, I use the interpolated surface showing 

the distribution and density of D. panamensis to guide this analysis.  In addition, I 

quantify how much potential nesting habitat exists within the vicinity of the SJLS 

1  7
 

Corridor.  Finally, I note if there are any nesting habitat patches that should be 

prioritized and targeted in the ongoing conservation efforts. 

1  8
 

Chapter 2. An evaluation of digital aerial photography


as a means to survey Dipteryx panamensis, a canopy-
emergent tree species, during its peak blooming in
Costa Rica

2.1 Introduction

2.1.1 Remote sensing studies of tropical forests compared to forest


inventory plots

Tropical forests are widely considered one of the most diverse and productive 

terrestrial ecosystems.  Though they cover less then 7% of Earth’s terrestrial surface 

(they are restricted between 23.5° north and south latitudes), they may harbor nearly 

half of the world’s biodiversity (Reid and Miller 1989).  Incredible accounts of species 

richness have been reported for tropical forests worldwide.  For example, Gentry (1988) 

found that one hectare (ha) of rain forest near Iquitos, Perú contained 283 species of trees 

> 10 cm dbh (diameter at breast height), while Whitmore et al. (1985) counted 233 

vascular plant species on just 0.01 ha of lowland tropical rain forest in Costa Rica.  

Compare this to the fact that Europe, north of the Alps and west of Russia, has only 50 

indigenous tree species while eastern North America has but 171 (Whitmore 1998).  

Tropical regions also exhibit high degrees of endemism.  Gentry (1992) noted that the 

greatest numbers of endemic species occur in Amazonia, Central America (including 

subtropical Mexico), coastal Brazil, and the northern Andes.  In addition, tropical 

ecosystems are considered to be crucial sinks for CO2 and may play an important role in 

1  9
 

sequestering carbon and mitigating the effects of global climate change.  Yet tropical 

regions are threatened from a variety of anthropogenic forces such as deforestation, 

agricultural intensification, pasture expansion, and urbanization (Lambin et al. 2003).   

Achard et al. (2002) determined that the pace of tropical deforestation and forest 

degradation from 1990 to 1997 was much higher than previously thought and could 

have huge consequences on the ability of these systems to act as carbon sinks.   

Sophisticated remote sensing technologies now enable tropical ecologists to track 

these threats from afar, garner a large spatial perspective of tropical forests, and explore 

these amazing ecosystems that so beguile scientist and layman alike.  Until recently, 

tropical ecologists have relied on a global network of terrestrial forest inventory plots to 

study tropical forests.  While forest inventory plots have provided invaluable, long‐term 

data on ecological, biological, and physical factors such as tropical biodiversity, forest 

dynamics, and nutrient cycling (Condit 1995, Sheil 1995, Clark and Clark 1999), they 

have limitations.  Clark et al. (2004b) note that many forest inventory plots are limited in 

size, sparsely distributed around the globe, or biased in their placement.  In addition, 

forest inventory plots are relatively small; the mean size of the 17 forest dynamics plots 

run by the Smithsonian’s Center for Tropical Forest Science is 32.3 ha (CTFS 2008).  

Thus, they tend to poorly sample the large trees generally defined as those > 60 cm 

diameter above buttress (Clark et al. 2004b).  These large, canopy‐emergent trees have 

crowns that soar many meters high making them difficult to study from the ground due 

2  0
 

to obscuration by the intermediate foliage (Clark et al. 2004b, Trichon and Julien 2006).  

Canopy studies using cranes, walkways, or tree‐climbing equipment were initiated 

largely in the 1980s and enabled tropical ecologists to access the tree crown (Parker et al. 

1992, Nadkarni 1994).  However, these canopy studies only provide a localized glimpse 

of ecological processes and afford limited detail at landscape scales.   

Tropical ecologists have long recognized the potential of remote sensing 

technologies to complement research in forest inventory plots and enhance the study of 

tropical forests in general.  Remote sensing refers to the acquisition of information on an 

object, area, or phenomenon by a device that is not in contact with the object, area, or 

phenomenon under study (Lillesand et al. 2004).  From a logistic standpoint, remote 

sensing mitigates some of the physical, political, and economic challenges that 

complicate terrestrial studies of tropical forests.  Tropical ecologists can access remote 

landscapes via a “bird’s‐eye view” of the landscape.  These areas could be otherwise 

challenging to reach because of physical barriers, or dangerous to enter due to political 

unrest.  Economically, the use of computer‐assisted interpretation of remotely sensed 

data may reduce time and costs of traditional field methods (Trichon 2001).    

From a biological standpoint, remotely sensed data can elucidate ecological 

phenomena and dynamics of tropical forests.  Trichon (2001) asserts that aerial photos 

may be used to identify particular trees of interest such as commercially important, rare, 

endemic, or keystone species.  Tree crown architecture and phenology can be studied 

2  1
 

(Trichon 2001) as well as forest dynamics such as tree mortality, survivorship, and 

growth (Clark et al. 2004b).  Finally, probably the greatest opportunity of remote sensing 

is the ability to study tropical forests across large spatial or temporal extents. 

New remote sensing platforms and the availability of multispectral and 

hyperspectral data are bringing us closer to conducting automated, individual tree 

species recognition.  As noted above, forest inventory plots tend to undersample very 

large, canopy‐emergent trees.  The advances in individual tree species recognition and 

finer sensors may elucidate this important structural guild over much larger study areas 

(Clark et al. 2004b).   

2.1.2 Overview of remote sensing platforms used to identify


individual tree crowns in the tropics

Turner et al. (2003) describe the multitude of remote sensing platforms and 

sensors used for biological conservation studies.  Here, I will discuss those that have 

been specifically used to examine tropical forests and highlight some that may enable 

the study of individual tree crowns.  In addition, I will detail the applications of high‐

resolution, digital aerial surveys as they are the remote sensing method used for the 

work detailed in this chapter.   

Previously, most studies in the tropics relied on remote sensing data coming 

from spaceborne, satellite platforms such as Landsat Thematic Mapper sensors with a 

spatial resolution of 30 m.  This platform produces medium spatial resolution imagery 

2  2
 

that is relatively inexpensive and has been useful in efforts to identify broad land‐

use/land‐cover classifications as well as to detect hotspots of deforested or fragmented 

zones (Skole and Tucker 1993, Sanchez‐Azofeifa et al. 2003, Van Laake and Sanchez‐

Azofeifa 2004).  These sensors can resolve aspects of tropical forest structure such as 

differences between young secondary forest and primary forest.  However, due to their 

coarse spatial resolution, these sensors are unable to differentiate older secondary 

growth or selectively logged forest from primary forest (Steininger 2000).  In addition, 

medium spatial resolution imagery such as Landsat Thematic Mapper cannot detect 

features such as individual tree crowns. 

High resolution, spaceborne, multispectral sensors such as the commercial 

IKONOS, SPOT, and QuickBird satellites have provided new insight into tropical forest 

dynamics and may even enable the discrimination of individual tree crowns.  The most 

recent SPOT satellite was launched in May 2002 and provides multispectral and 

panchromatic imagery.  QuickBird is a commercial satellite that was launched in 2001 

and provides multispectral imagery at 2.4 and 2.8 meter resolutions, and panchromatic 

imagery at 60 and 70 centimeter resolutions.  IKONOS is another commercial satellite 

launched in 1999 that collects high‐resolution imagery at 1 and 4 meter resolution and 

offers both multispectral and panchromatic imagery.  Palace et al. (2008) and Asner et al. 

(2002) have had some success in estimating crown widths of individual trees and 

examining forest canopy structure within the Brazilian Amazon using IKONOS 

2  3
 

imagery.  However, they found it difficult to separate adjacent, intermingled tree 

crowns.  Clark et al. (2004b) were able to detect individual tree crowns from IKONOS 

images within the La Selva Biological Station and accurately measure crown area.  They 

also had moderate success differentiating secondary forests and selectively logged 

forests from primary forests.  Clark et al. (2004a) also were able discern individual, 

emergent tree crowns using QuickBird imagery over the La Selva Biological Station in 

Costa Rica.  Researchers in the temperate zone, specifically Canada, have had more 

success in conducting semi‐automated, individual tree crown recognition using a new 

generation of high resolution, multispectral imagery known as CASI (Compact Airborne 

Spectral Imager) (Leckie et al. 2003, Leckie et al. 2005).  However, tree diversity at their 

Canadian study site was much lower and comprised three dominant coniferous species, 

each with distinctive crown architecture.  Thus, the application of this work may be 

limited in tropical forests where the tree diversity is so much higher and crown structure 

more heterogeneous. 

While these multispectral sensors provide limited ability to distinguish 

individual tree crowns and identify species, Clark et al. (2005) were able to use airborne, 

hyperspectral sensor data from the HYDICE (HYperspectral Digital Imagery Collection 

Experiment) sensor to differentiate seven canopy‐emergent trees in Costa Rica at the 

leaf, pixel, and crown levels.  Hyperspectral sensors are passive sensors that acquire 

simultaneous images in many relatively narrow, contiguous and/or non‐contiguous 

2  4
 

spectral bands through the ultraviolet, visible, and infrared portions of the 

electromagnetic spectrum.  This study was the first published report on automated 

tropical forest tree classification using high resolution, hyperspectral imagery.  It is a 

significant advance in using remote sensing to identify individual tree crowns and 

species in the tropics. 

Despite the advances in these high‐tech multispectral and hyperspectral sensors 

and the progress in automated tree crown recognition, aerial photography still offers 

many advantages in identifying and studying individual tree crowns within tropical 

forests.  Tropical ecologists have long recognized the usefulness of large‐scale aerial 

photography in tropical species identification (Myers 1982).  Now, traditional film‐based 

cameras have given way to digital cameras that have resolutions in the megapixel range.  

Digital cameras can be mounted on helicopters or small planes thereby giving the 

tropical ecologist flexibility to time flights depending on the season in order to capture 

specific phenological events.  The flying altitude can be controlled, making the 

identification of individual tree crowns more assured compared to present airborne or 

spaceborne imagery.  In addition, with the widespread availability of digital cameras, 

aerial photography may be an affordable complement to more expensive airborne or 

spaceborne imagery.  In the tropics, where funding can be limited and conservation is 

frequently conducted on a “shoe‐string” budget, digital aerial photography may provide 

a cost‐effective, accessible, and large‐scale method to tropical forest conservation.   

2  5
 

Recent work by tropical ecologists in Central and South America have  

successfully showed that human photo‐interpreters had an 87% success rate in 

identifying tropical forest trees in French Guiana in large‐scale aerial photographs 

(Trichon and Julien 2006).  Jansen et al. (2008) timed aerial overflights to coincide with 

fruiting of several palm species on Barro Colorado Island, Panama in order to ascertain 

if fruit abundance based on photogrammetric analysis correlated with direct fruit 

counts.  Both of these studies, however, relied on manual photo interpretation of images.  

In addition, aerial photographs captured a small spatial extent of less than 25 ha.  Foody 

et al. (2005) stress that while visually interpreting aerial photographs may produce 

accurate maps, it is an inherently subjective method.  Instead, digital image classification 

techniques should be further investigated.  Few studies have explored image 

classification methods using digital aerial photographs capturing a large spatial extent to 

systematically identify individual tree crowns within a GIS framework. 

2.1.3 Objectives

In this study, I examine the feasibility, cost‐effectiveness, and practicality of 

censusing and discriminating Dipteryx panamensis using high resolution digital aerial 

surveys timed to its peak blooming.  D. panamensis is an ecologically and economically 

important tree species.  This work was conducted over the lowland Atlantic rain forests 

of Costa Rica.  My main objectives were to: 

2  6
 

1. Evaluate the feasibility of using digital aerial surveys timed with the 

unique phenological period for D. panamensis to analyze this tree species’ 

density across a large spatial extent; 

2. Develop an image enhancement, classification, and machine learning 

approach for the computer‐automated discrimination of this canopy‐

emergent tropical tree species; 

3. Identify high density areas where D. panamensis trees are distributed; 

4. Assess the advantages and disadvantages of this classification approach. 

2.2 Materials and methods

2.2.1 Study site: San Juan-La Selva Biological Corridor

The digital aerial surveys were conducted over a portion of the proposed San 

Juan‐La Selva Biological Corridor (SJLS) located within northeastern Costa Rica (Figure 

2).  The SJLS Corridor is situated in northern Costa Rica within the provinces of Heredia 

and Alajuela within the Huetar Norte region.  The SJLS Corridor spans approximately 

246,608 hectares and comprises a matrix of land uses including primary forest, 

secondary forest, agriculture, pasture, and human settlements.  The predominant forest 

type within the SJLS Corridor is lowland Atlantic rain forest dominated by the tree 

species D. panamensis.  The creation of the SJLS corridor is the vision of Costa Rica’s 

Ministry of the Environment as well as several, local non‐governmental organizations 

2  7
 

(NGOs).  The SJLS Corridor forms one link of the regional Mesoamerican Biological 

Corridor effort that aims to restore habitat to promote the movement of organisms with 

large habitat needs and link protected areas within the seven Central American 

countries and the four states of southern Mexico (Kaiser 2001).   In 2001, the SJLS 

Corridor executive committee (comprised of NGOs, government agencies, and research 

groups) was charged with implementing the SJLS Corridor effort within Costa Rica.  The 

executive committee has outlined a strategy to consolidate the matrix of public and 

private lands within the proposed corridor.  The design of the SJLS corridor integrates 

three land‐use levels with increasing levels of biodiversity protection and ecosystem 

integrity (Chassot et al. 2004).  These levels include the Corridor Matrix, the Corridor 

Nuclei, and the Core Protected Area (Figure 2).  Maquenque National Wildlife Refuge 

(MNWR) forms the Core Protected Area within the SJLS Corridor.  In 2005, MNWR was 

officially decreed by Costa Rica’s Ministry of the Environment and was allotted 

conservation funding and protection.  MNWR was selected as the Core Protected Area 

because it represents one of the last breeding habitats for the endangered Great Green 

Macaw (Ara ambigua) in Costa Rica.  Studies by Powell et al. (1999) indicate that the 

Great Green Macaw depends on D. panamensis for food and nesting.  The conservation 

implications of the Great Green Macaw will be addressed within Chapter 4 of this 

dissertation. 

2  8
 

2.2.2 Aerial survey technique

From July 4‐8, 2005, I conducted an extensive aerial survey over 140,178 hectares 

or 57% of the SJLS Corridor 1.  I chartered a Cessna 206 airplane from AeroBell, a local 

Costa Rican airplane company, to fly transects over the SJLS Corridor.  In general, flights 

took off at 7 a.m. and lasted until 4 p.m. Costa Rica time.  A total of 21 hours were flown.  

I timed the overflights to coincide with the peak blooming for D. panamensis, and I 

selected the flight dates based on published descriptions of D. panamensis flowering 

patterns and personal observations by tropical ecologists at La Selva Biological Station.  

Phenological studies conducted by Frankie et al. (1974) indicate that adult D. panamensis 

trees begin blooming about one month into the Caribbean rainy season in Costa Rica.  

Although the trigger for the blooming event varies annually, this event generally begins 

in late May and continues into early August.  After the onset of flowering, individual 

adult D. panamensis trees flower asynchronously over a six‐week period with peak 

blooming occurring in mid‐July (Perry and Starrett 1980).  Flowers are a bright purple.  

A more detailed description of the phenological patterns of D. panamensis follows in 

Chapter 3 of this dissertation.   

Jacaranda copaia is a less common tree species in the project area.  However, it can 

reach the canopy and produces purple flowers.  Thus, this tree could potentially be 

                                                      

 As an interesting side note, my aerial surveys did not go unnoticed by the locals living in northern Costa 
1

Rica.  For more information, please see Appendix A.  
2  9
 

confused for flowering D. panamensis in the digital images.  However, Wright and van 

Schaik (1994) note an offset in the flowering periods for J. copaia and D. panamensis in the 

tropical moist forest of Barro Colorado Island, Panama, and James et al. (1998) observe 

that J. copaia flowers in the late dry season in the Coto Brus Valley of Costa Rica.  

Because D. panamensis begins to flower at the start of the rainy season, I was confident 

that I was photographing this species and not J. copaia. 

I employed a Nikon D1X digital camera with an image sensor of 5.3 megapixels 

to photograph the SJLS Corridor and capture blooming, canopy‐emergent D. panamensis 

trees.  The Nikon camera was mounted outside of the Cessna airplane via a stable 

platform.  Figure 3 depicts the digital camera platform.  A test flight was performed to 

ascertain the highest elevation that could be flown and still resolve individual, blooming 

D. panamensis crowns.  Although the pilot had to adjust the flying altitude at times due 

to the presence of clouds, the Cessna maintained an average flying altitude of about 1371 

meters above sea level (4500 feet) during the overflights.  A cable connected the Nikon 

camera to a laptop computer located inside the Cessna cabin.  The digital photographs 

were stored on the laptop as they were taken.  In addition, a GPS unit captured 

continuous positional information of the Cessna’s flight path.  These positional data 

were also recorded onto the laptop that was running the program Garmin MapSource.  

The positional information was subsequently related to each digital photograph. 

3  0
 

Figure 3:  Digital camera platform and airplane used in aerial surveys 

3  1
 

The aerial survey required 21 flight hours to complete.  A total of 63 flight 

transects were flown. Figure 4 depicts the flight lines of the aerial survey.  Forty‐four of 

the flight transects were oriented east‐west.  These transects were generally located over 

the southern portion of the SJLS Corridor as well as MNWR.  Nineteen flight transects 

traversed the northwestern section of the SJLS corridor and were oriented north‐south.  

In general, adjacent flight transects were separated a distance of 1 kilometer.  

Photographs within the same flight transect were the most similar in terms of flying 

elevation, time of day flown, and sun angle.  Ultimately, 2,626 digital images were 

collected that spanned about 140,178 hectares of the 246,608‐hectare SJLS Corridor.  

Some transects were flown a second time because clouds obscured the first pass.  Thus, 

there were 2,550 images in the final image count without duplicates. 

2.2.3 Digital photographs

The Nikon camera’s shutter speed was adjusted based upon the flying altitude so that 

adjacent photos within transects overlapped approximately 20%.  In addition, 

photographs between transects overlapped about 20%.  The raw, unprocessed digital 

photographs were collected in the Nikon Electronic Format (.nef) which is the 

proprietary digital image format used by Nikon digital cameras.  All digital images 

contained a red, green, and blue (RGB) spectral band.  The images did not contain an 

infrared band.  The raw images were unreferenced and lacked a geographic coordinate 

system.  However, general positional information could be ascertained for each photo by 

3  2
 

Figure 4:  Location of aerial surveys over the proposed San Juan‐La Selva Biological 
Corridor 
3  3
 

matching the data from the onboard GPS unit to time/date information collected with 

each photograph.  Based on an average flying altitude of 4500 ft., the digital images 

contained an average field of view (FOV) of 120 hectares and a resolution of about 0.3 

meters/pixel. 

2.2.4 Image enhancement: contrast stretch

After the digital photographs were acquired, I converted the photographs to a 

format suitable for analysis in a remote sensing platform.  I used the Nikon Capture 4 

program to convert the digital photographs from the .nef format to .tiff format.  At this 

stage, I rotated the images so that the images were oriented with north at the top.  Next, 

I converted the .tiff images to .img files (Erdas Imagine proprietary file format) using the 

remote sensing program, Erdas Imagine 8.7.  Images were retained as unsigned 8‐bit. 

Because I undertook this remote sensing effort to census only one canopy‐

emergent tropical tree species, I enhanced my images in order to unpack and accentuate 

the spectral detail of the blooming, purple D. panamensis crowns.  The original, digital 

images contained red, green, and blue bands.  When displaying digital images on a 

computer monitor, the appropriate bands are assigned to a red, blue, and green channel.  

Here, every pixel within a composite, digital image can be represented within a 3‐

dimensional coordinate space, or color cube, where the three primary colors comprise 

the three axes (Lillesand et al. 2004).  In a visual monitor using an 8‐bit‐per‐pixel data 

encoding, each pixel can be depicted by a “digital number” that ranges from 0‐255.  The 

3  4
 

RGB color cube is defined by the brightness values of each of the three primary colors.  

As a technical term, brightness refers to the strength or intensity of a color. 

A digital image’s spectral histogram depicts the distribution of the number of 

pixels and the range of digital numbers (DN) for each channel.  In the histogram, the x‐

axis contains the DNs ranging from 0 to 255, while the y‐axis represents the number of 

pixels within the image containing a particular DN.  Pixels with a smaller DN will 

appear darker on‐screen, while those with a larger DN appear brighter.  During 

exploratory testing of the digital images, it was apparent that the histograms for each of 

the three color bands were tightly clumped, thus underutilizing the full range of DN 

values available for display and analysis.  In order to unpack the spectral data contained 

within the digital images, I performed a contrast stretch.  The purpose of a contrast 

stretch is to expand the narrow range of brightness values typically present in the 

spectral bands of an input image over a wider range of DN values.  The result is an 

output image that has been accentuated in order to enhance the contrast between 

features of interest making the image more interpretable to an image analyst (Lillesand 

et al. 2004).  Standard contrast stretches exist such as a linear stretch or a histogram 

equalization stretch.  However, the specific contrast stretch that was performed on all the 

digital images was a Percentage Look‐Up Table (LUT) Stretch.  This type of contrast 

stretch clips off a specified percentage of pixels at the low and high ends of the 

histogram and then allocates the remaining pixels linearly across the DN values, from 0‐

3  5
 

255.  The purpose of such a stretch is to trim away “noise” at the ends of the histogram 

that correspond to features that are not of interest to the image analyst. 

I performed a contrast stretch first on one candidate image within each of the 63 

flight transects and then applied these enhancements to images across the remainder of 

the same flight transect.  Images within the same flight transect were more similar in 

terms of scene conditions compared with images within different flight transects.  

Because flying elevation, flight time, and flight direction varied more across flight 

transects and affected the digital scene within photographs, I performed subsequent 

image enhancements and transformations based on the unit of the flight transect.  For 

each candidate image, I clipped 10% off the left end of the histogram and 2.5% off the 

right end of the histogram.  Finally, the contrast was adjusted across the image by 60%.  

These contrast enhancements were performed on the RGB‐format, digital images using 

the Contrast Tool in Erdas Imagine 8.7.  The Percentage LUT Stretch accentuated the 

contrast of the blooming D. panamensis trees within the digital images and unpacked the 

spectral data for this specific feature of interest.  Finally, I matched the remaining images 

within a particular flight transect to the candidate image’s histogram using the batch 

function.  This procedure known as a histogram matching, converts the histogram of one 

image to resemble the histogram of another (LeicaGeosystems 2003). 

3  6
 

2.2.5 Image enhancement: conversion to intensity, hue, and


saturation color space

Following the contrast stretch, I performed another image enhancement to all of 

the digital images.  At this step, the digital images were still within the original RGB 

color space.  I converted the contrast‐stretched, RGB images to an alternate color space 

known as intensity, hue, and saturation (IHS).  This color space uses intensity, hue, and 

saturation as the three positional parameters instead of red, green, and blue and is based 

on a color sphere rather than a cube (Jenson 2005).  Intensity is the overall brightness of a 

color, and values range between 0 and 1.  Hue is the principal wavelength of the pixel, or 

dominant color, and values extend between 0 and 360.  Saturation refers to the purity of 

the color relative to gray, and values range between 0 and 1.  Gillespie et al. (1986) assert 

that the three color bands, RGB, tend to be highly correlated amongst each other.  

Conversion of images in RGB color space to IHS color space serves to extract intensity, 

hue, and saturation values from each of the red, green, and blue bands and reduce band 

intercorrelation.  In addition, this technique has proven quite effective in differentiating 

green and senescent plant materials in digital imagery having only three bands and 

lacking a near infrared band such as the imagery used in my study (Laliberte et al. 2006).   

2.2.6 Supervised classification

The purpose of the classification step was to extract spectral patterns contained 

within the IHS band data of the digital images.  Computer processors are able to 

3  7
 

recognize and sort patterns present in the pixels of digital images and provide 

meaningful interpretation of features within the scene.  In this next step, I extracted 

spectral patterns contained within the IHS band data of the enhanced digital images.  I 

capitalized on the unique spectral signature of blooming D. panamensis trees in order to 

identify individuals across my digital images.  All pixels within the digital images were 

used for the supervised classification. 

I collected spectral signatures on the IHS‐converted image in preparation for a 

supervised classification.  Spectral signatures were contained in a single signature file.  I 

used the same candidate image selected for the contrast stretch step to collect the 

signatures.  Spectral signatures are sample pixels that represent a certain feature of 

interest that the classification algorithm can then use to classify other pixels in the image.  

Examples of common features could be water, urban areas, bare ground, or vegetation.  

Signatures are based on training samples collected within the image to be classified.  I 

collected training samples using the Region Growing Tool, which is a more conservative 

method than a user‐defined Area of Interest (AOI).  The Region Growing Tool defines 

training samples based on specified user parameters such as a maximum training 

sample size or a maximum distance that candidate pixels can be to each other spectrally.  

Although I was most interested in defining tight signatures representing blooming D. 

panamensis trees, I also collected signatures for other land features.  This enabled the 

supervised classification to allocate pixels for features besides blooming D. panamensis 

3  8
 

into other classes.  Some examples of additional classes included vegetation, water, 

agriculture, dark regions, bare ground, roads, and clouds.  I examined the distribution of 

classes using the signature mean plot paying particular attention to the D. panamensis 

classes.  In addition, I examined the separability of the D. panamensis signatures 

compared to the other signature classes by constructing a separability matrix.  A 

separability matrix compares if training classes are spectrally distinct from each other, 

and separability is based on the transformed divergence metric.  The maximum possible 

divergence value is 2,000, and values less then 1,500 indicate spectrally similar classes.   

I also examined the classification error rates by using a Contingency Matrix.  In 

all, a total of 64 signature files were collected to classify images in each flight transect.  

The number of signature files is slightly higher than the number of flight transects (63) 

due to the fact that some transects had to be flown at two different times due to the 

presence of clouds.  Thus, these transects were analyzed separately. 

I used the training sample for each candidate image to perform a supervised 

classification for the rest of the IHS‐transformed images within each transect.  This is a 

very unconventional classification method, as normally images are classified using their 

unique signature file.  However, I considered this method to be appropriate, expedient, 

and effective in my analyses for several reasons.  From a practicality standpoint, because 

I had several thousand digital images to analyze, I needed an efficient way to classify 

my images without collecting signature files for each image.  As mentioned earlier, I 

3  9
 

selected a candidate image from each flight image to conduct the initial image 

enhancements and then gather training signatures because images within transects were 

more similar to each other due to the flight elevation, time/day flown, and sun angle.  

Thus, I considered it appropriate to use a single signature file for a candidate image as a 

proxy for the rest of the images within a flight transect.  Finally, I was only interested in 

resolving blooming D. panamensis trees throughout all of the images.  Thus, I did not 

require such a precise classification level accuracy for signatures other than D. 

panamensis. 

I initially performed a few “test‐runs” of the supervised classification approach 

on the images.  Based on these test‐runs, it was apparent that the classification technique 

performed the best in so‐called ideal conditions where the image scene contained mostly 

forest as well as blooming D. panamensis.  Other image scenes that contained a lot of bare 

ground, various types of agriculture such as pineapple plantations, or non‐forested areas 

tended to confuse the classifier.  Based on this information, I applied the classification 

technique on 801 of the 2,550 images.  These images represented 31% of the total images 

flown and were the ones that contained the most contiguous forest/vegetation cover.   

Ultimately, I used a pixel‐by‐pixel classification approach using a maximum 

likelihood classifier.  A maximum likelihood classifier is based on the probability that a 

pixel belongs to a particular class.  The basic equation assumes that these probabilities 

are equal for all classes, and that the input bands have normal distributions 

4  0
 

(LeicaGeosystems 2003).  The maximum likelihood classifier is the oft‐cited classifier 

approach used in remote sensing of forest trees (McGraw et al. 1998). 

2.2.7 Post-classification image processing: recode

Following the supervised classification, I conducted a series of processing steps 

in order to prepare the images for analysis to determine how well the automated 

classification approach had identified blooming D. panamensis crowns.  

The classified images were all recoded to identify pixels classified as blooming D. 

panamensis trees.  All other pixels that had been aggregated into another signature file 

that was not designated as D. panamensis were recorded as background.  A clump 

function was then run to aggregate these contiguous pixels as a unit.  Finally, two sieve 

threshold functions were performed to remove pixel clumps smaller than 150 m2 and 

larger than 2100 m2.  I selected these lower and upper sieve thresholds to target and 

remove small and large clumps erroneously classified as D. panamensis.  Running the 

smaller sieve threshold removed the “salt‐and‐pepper” effect present in the images and 

eliminated small clumps that did not correspond to D. panamensis.  This sieve level did 

remove some clumps that corresponded to actual D. panamensis trees.  However, a 

sensitivity analysis revealed that the sieve level of 150 m2 minimized the error of non D. 

panamensis clumps being included as this species.  The upper sieve level of 2100 m2 was 

selected after analyzing a subsample of 35 images and determining the largest D. 

panamensis clump in these images as well as the smallest and largest clumps erroneously 

4  1
 

identified as D. panamensis. The large sieve level of 2100 m2 effectively removed large 

clumps that were not D. panamensis.  In my analyses, few D. panamensis trees actually 

achieve a crown area greater than 2100 m2.  Thus, few D. panamensis trees were removed 

due to this sieve function.  These sieve thresholds also corresponded to the biology of D. 

panamensis and represented lower and upper crown areas of canopy‐emergent D. 

panamensis trees.  

2.2.8 D. panamensis clump density determination

Following the sieve step, the number of clumps within each image was 

quantified.  In the majority of cases, a “clump” represents an individual D. panamensis 

tree.  However, it is possible that a “clump” could represent several, neighboring trees 

with adjoining crowns.  This is because the maximum likelihood classifier is unable to 

discern discrete objects.  Therefore, adjacent D. panamensis trees are perceived as a single 

“clump” in the final classification output.  There have been recent breakthroughs in 

computer assisted pattern recognition techniques that use segmentation in addition to 

traditional pixel‐based classification approaches.  This new approach, known as object‐

oriented classification, first segments a landscape into “objects” based on similarity 

criteria such as texture (Meinel and Neubert 2004).  Thus, a tree crown would be more 

easily separated from other landscape features and potentially separable from other 

adjacent tree crowns.  I briefly considered using an object‐oriented classification 

approach to classify my images, and I evaluated several programs including eCognition.  

4  2
 

However, due to price considerations and software availability, I ultimately opted for 

the pixel‐by‐pixel classification approach in Erdas Imagine.  Indeed, the fact that some 

clumps represent multiple D. panamensis trees is trivial considering the number of D. 

panamensis stems across the 140,178‐ha study area.  In addition, from a macro‐biological 

standpoint, most organisms that depend on D. panamensis for resources would be 

unlikely to perceive two, comingled tree crowns as separate entities. 

D. panamensis density per image was then determined by dividing the number of 

D. panamensis clumps identified in the classified image by the image’s area.  The area for 

each image was calculated based on the field of view (FOV) represented in each image.  

The FOV calculation incorporated the focal length of the digital camera, the height above 

the landscape at which the image was taken, as well as the pixel dimensions of the 

image. 

2.2.9 Classification error assessment

I performed a post‐classification error assessment to determine various types of 

errors associated with the supervised classification. The purpose of an error assessment 

is to examine the different types of classification errors known as omission and 

commission.  For a thorough overview of assessing error in classifications of remotely 

sensed data, please see Congalton (1991).  Error of omission, also known as producer’s 

or exclusion error, reflects the probability of a pixel being correctly classified into the 

target category.  Error of commission, also known as user’s or inclusion error, reflects 

4  3
 

the probability that a pixel classified into a given category actually represents that 

category on the ground. An error assessment is most often represented by an error 

matrix.    

Because the supervised classification that I performed was unconventional, I 

performed a nontraditional error assessment based solely on the ability of the classifier 

to distinguish D. panamensis clumps.  I performed a manual error assessment on the 

classified images following the sieve step.  I divided the images into two sets based 

primarily on the direction in which the aerial surveys were flown.  The purpose of 

aggregating the images into two sets prior to conducting the error assessment was to 

organize images that were collected under similar flying conditions.  The two image sets 

were named MNWR (Maquenque National Wildlife Refuge) and NZ (Northern Zone) 

after the regions over which they were flown.  Table 1 summarizes the number of photos 

sampled for the error assessment for the images in the MNWR and NZ regions.   

Table 1:  Number of photos sampled for the classification error assessment 

Region  Number of photos  Total number of  Percentage of images 


sampled for error  classified images  sampled for error 
MNWR  25  633  3.9% 
NZ  12  168  7.0% 
Total  37  801  4.6% 
 

The error assessment itself consisted of a manual assessment of how well the 

classifier identified D. panamensis clumps.  First, I manually identified and counted the 

4  4
 

number of D. panamensis clumps within a candidate image from either the MNWR or 

NZ Region.  Next, I compared the D. panamensis clumps that I manually identified with 

the D. panamensis clumps the classifier identified.  Based on the manually identified D. 

panamensis clumps, I quantified the number of D. panamensis clumps correctly identified 

by the classifier that also corresponded to the D. panamensis clumps identified manually.  

In addition, I quantified the number of tree clumps the classifier missed as well as the 

number of non D. panamensis clumps that the classifier misidentified.  The former 

reflects the error of omission for the classifier while the latter reflects the error of 

commission for the classifier.  Because I was only interested in the classifier’s success in 

identifying D. panamensis and because I trained the original classifier to identify D. 

panamensis and not other landscape features, this error assessment did not include other 

landscape categories.  A traditional accuracy assessment would include errors of 

omission and commission for multiple landscape categories.  Hence, a complex error 

matrix could be generated.  Ultimately, my classified images reflect a binary world—D. 

panamensis or non D. panamensis. 

4  5
 

2.3 Results

2.3.1 Comparison of image enhancement transformations

The image enhancement transformations were an appropriate method to analyze 

digital photography that contained RGB bands but lacked an infrared band.  Figure 5 

shows the sequence of image enhancements and transformations for one image. 

2.3.2 Signature separability 

Sixty‐four separate signature files were collected in preparation for performing 

the supervised classification.  At least one candidate signature file was collected per 

flight transect, and the same signature file was used to classify images within a 

particular transect.  I examined the specific training samples collected for blooming D. 

panamensis within each signature file to ensure separability.  Because of the sheer 

number of signature files, I present the results of the separability analysis from one 

signature file.  Training classes collected for blooming D. panamensis crowns were 

distinct compared to other landscape features such as vegetation, bare ground, water, or 

shadow areas.  Figure 6 is a graph of the mean plot for each of the training samples in 

this candidate signature file.  The graph shows the value for each of the three bands in 

the IHS color space for each training class.  In general, D. panamensis training samples 

were the most distinct within the hue band of the IHS color space.  The mean value for 

this band was 47.4 and ranged from 24.2 to 60.9 indicating that this region of the hue 

band correlated with the spectral signature for blooming D. panamensis.  
4  6
 

Figure 5:  Sequence of image enhancement steps 

A) Original RGB image that has been contrast stretched; B)  Image that has been 
converted to IHS; C) Output from supervised classification; D) Image that has been 
recoded for just D. panamensis; E) Image that has been sieved to remove small and 
large tree clumps; F) Original RGB image for comparison. 

4  7
 

  

Figure 6:  Sample signature separability plot 

4  8
 

Results from the signature separability matrix are shown in Appendix B.  All of 

the training classes gathered for this candidate signature are shown, and multiple 

training classes were collected to capture the spectral signature of blooming D. 

panamensis crowns.  What is notable is that all of the D. panamensis training classes are 

separate and distinct from the other training classes, as no value is less then 1,500.  This 

would indicate that training samples are spectrally similar.  Some of the D. panamensis 

training classes are spectrally similar to each other as evidenced by values less than 

1,500, which is acceptable considering that these classes are sampling the same 

environmental feature. 

2.3.3 Classification error assessment

An error assessment was performed to determine the errors of omission and 

errors of commission over the entire project area.  Errors were also determined 

separately for the MNWR and NZ regions.  Together, the MNWR and NZ regions form 

the entire project area.   The overall error of commission was 22.2% while the overall 

error of omission was 11.9%.  The error of commission was acceptable based on other 

published studies (Lunetta et al. 2004, Foody et al. 2005).  Table 2 summarizes the errors 

based on the classification error assessment. 

4  9
 

Table 2:  Results of the classification error assessment 

Region  Percent Error of Omission  Percent Error of Commission 


MNWR  6.7%  27.1% 
NZ  22.7%  11.9% 
Overall 
(adjusted to account 
11.9%  22.2% 
for # of photos in 
each region) 
 

2.3.4 High density images

The automated classification technique was successful in classifying high density 

areas of D. panamensis.  The classifier worked especially well when applied to image 

scenes containing the most intact, contiguous forest.  Examples of high density regions 

are show in Figure 7.  These images display densities of 2.0 trees/ha and 1.83 trees/ha. 

2.4 Discussion

This study underscores that digital aerial photography still has a place amongst 

high‐tech airborne and spaceborne remote sensing technologies.  This technique can be 

used across a large spatial extent to investigate canopy‐emergent trees.  Canopy‐

emergent trees are one of the least studied tropical tree guilds due to their soaring 

crowns that often make them challenging to study from the ground (Clark et al. 2004b).  

In addition, high resolution digital photography can be analyzed en masse in a GIS 

platform by using a semi‐automated classification approach to identify individual tree 

crowns.  While an infrared band may increase the spectral separability of features,  

5  0
 

Figure 7:  Sample images with high densities of D. panamensis trees 

Blooming, canopy‐emergent D. panamensis trees are identifiable from their purple 
crowns.  Image A (97.8 ha) has a density of 1.83 trees/ha.  Image B (95.2 ha) has a 
density of 2.0 trees/ha. 
5  1
 

alternative color spaces such as IHS and other image enhancements may help resolve 

standard RGB digital images.  Timing the overflights during a particular phenological 

period for the focal species—such as a peak blooming event—may also boost the success 

rate of the classifier.  D. panamensis is a unique species, however.  Fortuitously, this 

study species has a distinct peak blooming event when other canopy‐emergent tree 

species have only green foliage, making it a prime candidate to calibrate a classification 

technique.  However, this technique may need to be adjusted for species without a clear 

phenological signal.  

Although this classification approach represents a semi‐automated approach to 

identifying and analyzing the spatial distribution of a particular tropical tree species, 

this method does have some limitations.  While a classification algorithm was developed 

to extract the spectral signature of blooming D. panamensis trees, the classification 

approach worked in ideal conditions when the image scene contained contiguous forest 

cover.  The classifier had a difficult time differentiating spectral differences of blooming 

D. panamensis with bare ground and water present.  Thus, only a limited number of the 

total images could be analyzed using the classifier. 

Despite some of these limitations, digital aerial photography may offer some 

advantages in terms of time and money compared to spaceborne sensors.  At this point, 

image scenes captured from high resolution spaceborne sensors such as SPOT, 

Quickbird, or IKONOS may be prohibitively expensive for large‐scale surveys such as 

5  2
 

this project that extended over 100,000 ha.  If the digital camera is in hand, the main 

expense for an aerial survey will be the rental of the airplane or helicopter.  In this study, 

the cost of renting the plane was less than $5,000.  

Ultimately, digital aerial surveys may complement more expensive remote 

sensing imagery.  This approach may be particularly handy in augmenting the spatial 

extent of field plots, overcoming physical barriers, and sampling sites that may be 

dangerous or difficult to enter on the ground.  In the field of tropical ecology and 

conservation, practitioners often look for creative ways to conduct research in light of 

limited funding.  Digital aerial photography may be one way to surmount such 

challenges.  

5  3
 

Chapter 3. Density distribution, habitat description, and


fragmentation of a population of Dipteryx panamensis in
northern Costa Rica

3.1 Introduction

Dipteryx panamensis is a large tree that occurs in lowland tropical moist and 

tropical wet forests along the Atlantic coast from Nicaragua to Colombia (Bonaccorso et 

al. 1980, Flores 1992).  Along with several other tree species, D. panamensis comprises an 

important component of the emergent canopy layer.  Canopy‐emergent trees are defined 

as having crowns that extend above the mean canopy layer (Clark and Clark 1987).  

Tropical ecologists have struggled to accurately document the distribution and 

population densities of tropical emergent trees using forest inventory plots.  Although 

forest inventory plots have long been used to examine biogeographic patterns of floristic 

species richness, density, and distribution in the tropics (Condit 1995), small field plots 

may poorly sample large trees.  These are generally defined as > 60 cm diameter at 

breast height (Clark et al. 2004b).  Because large, canopy‐emergent tree species like D. 

panamensis are frequently distributed at densities of less than one adult/ha (Clark and 

Clark 1987), the size or placement of forest inventory plots may skew their density or 

distribution values at a landscape level.  However, remote sensing techniques present 

new opportunities to examine the distribution of canopy‐emergent trees across a larger 

spatial extent and present a more accurate representation of biogeographic patterns.  

5  4
 

3.1.1 Objectives

I sought to document the landscape distribution and density of D. panamensis in 

the lowland Atlantic forests of Costa Rica.  This tree plays an important role as a 

keystone species in this region and is particularly important as a nesting and food 

resource for the endangered Great Green Macaw (Ara ambigua).  Specifically, I used 

digital aerial surveys to census this tree species across 140,178 ha and augment the 

coverage of traditional terrestrial plots.  In addition, I used the results of this work to 

examine the connectivity of subpopulations based on pollen dispersal distances.  In 

particular, I address the following specific questions in this chapter: 

1. What is the distribution of D. panamensis across the study landscape? 

2. What are the density trends for this species across the study landscape 

and over forested areas? 

3. What environmental factors predict locations of high D. panamensis 

density? 

4. How much potential habitat exists within the study area? 

5. Are subpopulations of D. panamensis fragmented or connected based 

upon measured pollen dispersal distances? 

5  5
 

3.2 Background

3.2.1 Overview of biogeographic studies in the tropics

The renowned English naturalist, Alfred Russel Wallace who lived from 1823‐

1913 is often considered the father of biogeography.  Wallace (1876) is famous for his 

studies on the distributions of animals and plants of the Malay Archipelago which are 

documented in numerous books, most notably in the two‐volume work “The 

Geographical Distribution of Animals”.  Biogeography, the discipline that Wallace 

pioneered, is the scientific field that examines the spatial, distributional, and abundance 

patterns of biological diversity, both past and present (Lomolino et al. 2005).  Tropical 

ecologists have long been curious about patterns of species distribution.  For example, a 

central question driving ongoing studies in the tropics has been what abiotic and biotic 

factors promote the latitudinal gradient in species diversity (Pianka 1966, Stevens 1989).  

In other words, why do so many species’ distributions intersect at the tropics and how 

do these species coexist?   

3.2.1.1 Limitations of terrestrial forest inventory plots 

Examining the spatial distribution and abundance of any organism is 

constrained by the ability to census a large enough geographic area to garner an accurate 

signal.  Early tropical ecologists sought to explore patterns of tree species richness 

through small plot‐based studies.  For example, Black et al. (1950) conducted an early 

5  6
 

examination of floristic diversity and population density of trees in Amazonian forests 

within three 1‐ha plots.  However, because of the rarity of many tropical tree species, 

small plot studies may distort or all together miss uncommon species thereby belying 

their true spatial abundance and distribution.  Even in their pioneering study, Black et al. 

(1950) cautioned against drawing conclusions on species diversity and abundance based 

on a small sample area.  A network of 50‐hectare plots throughout the tropics has sought 

to augment the spatial area of study as well as compare forests throughout different 

regions of the tropics (Condit 1995).  These plots afford detailed study of floristic 

richness, composition, and distribution as every individual ≥ 1 cm in diameter is 

mapped and identified.  While these plots are larger, Clark et al. (2004b) note that 

tropical inventory plots are often subjectively located, too small to capture processes of 

interest, or rarely sited using a statistical design. 

3.2.1.2 The importance of documenting the distribution of keystone species 

While tropical ecologists are often interested in documenting the spatial 

distribution and abundance of rare tree species as these species may contribute to high 

levels of both alpha and beta diversity (Pitman et al. 1999), attention should be given to 

the distribution and biogeographic patterns of more common tropical tree species at 

intermediate to large scales.  In a study of 21 forest plots distributed across 40,000 ha of 

the Peruvian Amazon, Pitman et al. (1999) determined that while the rarest species 

occurred at a density of 1 stem/36.375 ha, 88% of tree species occurred at densities <1 

5  7
 

individual/ha and at least half occurred somewhere at densities > 1.5 stems/ha.  This 

work suggests that most tree species in tropical regions may be habitat generalists that 

occur at low densities but large absolute population size.  Unlike rarer species, these 

ubiquitous tree species may actually be the linchpin in an ecosystem and an important 

keystone resource (Terborgh 1986, Mills et al. 1993, Nason et al. 1998).  Much like a 

keystone in an arch, a keystone species plays a disproportionate role in an ecosystem’s 

functioning, and the loss or removal of a keystone species could cause a dramatic shift in 

that ecosystem.  Documenting the biogeographic ranges and habitat preferences of 

common, keystone trees can also assist conservation endeavors aimed at protecting 

these species and others that depend on them. 

3.2.2 Background on study species: D. panamensis

3.2.2.1 Conservation status and keystone species designation 

Because of D. panamensis’ keystone status and importance in an ongoing 

conservation project, I documented the distribution and density of flowering, canopy‐

emergent individuals across 140,178 ha.  This work occurred over Costa Rica’s lowland 

Atlantic forest and employed digital aerial surveys.  While relatively ubiquitous, this 

species is currently listed as a Convention on International Trade in Endangered Species 

of Wild Fauna and Flora (CITES) Appendix III species (CITES 2008).  Appendix III 

species are those that are regulated within the jurisdiction of a particular country to 

5  8
 

prevent or restrict exploitation.  The country has also identified the need for cooperation 

by other countries in order to control trade of this species. 

In the lowland Atlantic forests of Costa Rica, D. panamensis and about a half 

dozen other tree species form an important component of the emergent canopy layer.  

Clark and Clark (1987) assert that the demography of tropical emergent species like D. 

panamensis remains poorly known due to the low density of post‐seedling individuals 

and the fact that most adults occur at densities less than 1 tree/ha.  Adult D. panamensis 

trees may reach 50 m in height and close to 1.5 m diameter at breast height (Flores 1992); 

reproductive maturity is likely attained at about 30 m in height (Perry and Starrett 1980).  

Based on tree ring counts and 14C dating, individuals may live close to 300 years with a 

theoretical maximum age of 645 years (Fichtler et al. 2003).   

Like the fig tree (Nason et al. 1998), D. panamensis is considered another keystone 

tree resource in the tropics.  The fruits of D. panamensis mature at the onset of the dry 

season, a time when the fewest tree species present edible, ripe fruits (Frankie et al. 1974, 

Bonaccorso et al. 1980).  In a given year, a large adult D. panamensis tree may produce an 

estimated 10,000 fruits (Bonaccorso et al. 1980).  However, individual trees may oscillate 

between years of high and low fruit production. 

3.2.2.2 Phenological patterns 

The phenological pattern of D. panamensis has been well studied.  D. panamensis 

trees flower annually, and the timing of the first flowers occurs immediately after the 

5  9
 

months with highest light and lowest rainfall (OʹBrien et al. 2008).  Thus, flowering 

generally commences around May prior to the onset of the rainy season in northeastern 

Costa Rica.  Individuals of D. panamensis bloom asynchronously with overlapping 6‐

week flowering periods (Perry and Starrett 1980).  Flowering generally peaks in July and 

declines by August.  However, O’Brien (2008) observed some D. panamensis trees still 

flowering into October.  Flowers, which are arranged along lateral or terminal panicles, 

generally last for one day before falling to the ground.  The presentation of relatively 

few flowers each day by an individual tree and the asynchronous overlap of blooming 

individuals seem to encourage inter‐tree pollinator movement and favor out‐crossing 

(Perry and Starrett 1980).  While it has been long thought that this species is an obligate 

out‐crosser (Perry and Starrett 1980), recent genetic work by Hanson et al. (2008) reveals 

instances of self‐pollination.   

Immature fruits first appear in early July, and mature fruits are present on the 

trees from November to March (Bonaccorso et al. 1980).  The fruits are elliptical drupes 

that comprise a thin exocarp, a fleshy mesocarp, and a hard endocarp that protects a 

single seed (De Steven 1988).  As fruit production ceases, D. panamensis trees exhibit 

moderate deciduousness.  Individuals shed their leaves synchronously during the driest 

months from February to May (OʹBrien et al. 2008).  A new flush of leaves soon emerges. 

6  0
 

3.2.2.3 Interactions with other organisms 

In keeping with D. panamensis’ designation as a keystone tree resource, this 

species participates in a complex web of interactions with other organisms.  Not only 

does it require other species for critical services like pollination and seed dispersal, but it 

also supports a host of other species.  The prime pollinators include at least 13 species of 

medium‐to‐large bees (Perry and Starrett 1980).  Bonaccorso et al. (1980) observed at 

least 16 species of mammals that will eat the fruit.  Seed commensals that only consume 

the fleshy mesocarp, do not penetrate the endocarp, and leave the seed intact include 

primates, bats, marsupials, rodents, and carnivores.  Some of these species may even 

disperse seeds away from the parent tree.  Fruit bats, particularly in the genus Artibeus 

are notable dispersers of D. panamensis fruit and will often carry the drupes considerable 

distances until they reach a feeding roost often in a palm.  Seed predators that damage 

the stony endocarp and frequently consume the seed include many rodents such as red‐

tailed squirrel (Sciurus granatensis), agouti (Dasyprocta punctata), paca (Agouti paca), and 

spiny rat (Proechimys semispinosus) (Bonaccorso et al. 1980).  Hundreds of migratory or 

resident birds may frequent D. panamensis trees to forage.  A notable species includes the 

endangered Great Green Macaw (Ara ambigua).  Powell et al. (1999) noted that D. 

panamensis accounted for 85% of Great Green Macaw feeding observations during the 

month of January.  In addition, 86% of Great Green Macaw nests occur in the hollows of 

this tree (Chassot et al. 2002, Monge et al. 2003). 

6  1
 

3.2.2.4 Threats and conservation measures 

Although D. panamensis is relatively abundant within the lowland Atlantic 

forests of Costa Rica, it is threatened primarily from logging since its wood is used to 

make plywood.  However, due to the density, weight, and crystalline deposit content of 

D. panamensis wood, this species was extremely difficult to fell with axes and manual 

saws (Flores 1992).  Until recently, solitary individuals stood in pastures cleared of more 

pliable tree species.  Only with the advent of steel chain saws could this tree be targeted.  

Because of this species’ keystone status within the lowland Atlantic forest ecosystem as 

well as its significance to the Great Green Macaw, Costa Rica’s Ministry of the 

Environment enacted a series of forestry laws that restrict logging of D. panamensis by 

50%.  However, enforcement still remains a challenge.  In 2005, a new national wildlife 

refuge known as Maquenque was decreed (refer to material in Chapter 1).  Situated in 

the larger San Juan‐La Selva (SJLS) Biological Corridor, the primary purpose of this 

59,700‐ha refuge is to protect breeding and nesting habitat of the Great Green Macaw.  

This mandate effectively affords protection to D. panamensis populations. 

3.2.2.5 Habitat requirements, distribution, and abundance 

Although aspects of D. panamensis’ phenology are well‐studied, specific details 

about its habitat requirements, distribution, and abundance across a larger spatial extent 

are still needed.  Information about those factors could assist the ongoing conservation 

efforts while augmenting basic biological data on the species.  The following broad 
6  2
 

habitat conditions were described by Flores (1992).  D. panamensis is generally found on 

humid plateaus with alluvial or sandy soils and sometimes acid and clay soils at 

elevations between 20‐1300 meters above sea level.  What is known about the densities 

of D. panamensis seedlings, saplings, and adult trees come from relatively small plot 

studies.  As noted earlier, plot size and the delineation of plot boundaries may skew 

density values.  The few estimates of adult D. panamensis densities within intact forest 

come from plots established at the La Selva Biological Station and include trees that are 

being followed for a long‐term demographic project.  Reported density measurements 

range from 0.45 trees/ha (85 ha plot) to 0.97 trees/ha (65‐ha plot) (Hanson et al. 2006, 

Hanson et al. 2008).  However, these density estimates are difficult to interpret as the 

plots substantially overlap and are relatively small making them susceptible to skew.  

Nonetheless, Clark and Clark (1987) assert that tropical emergent species like D. 

panamensis are frequently distributed at densities of less than 1 adult/ha challenging our 

abilities to study their demographics, abundances, or distribution across large spatial 

scales.  However, remote sensing techniques present new opportunities to examine this 

species, especially large adults, from an aerial vantage point.  

3.2.2.6 Pollen dispersal patterns for D. panamensis 

A recent study by Hanson et al. (2008) examined the effects of varying levels of 

habitat fragmentation on the genetic structure, genetic diversity, and pollen dispersal 

distances of adult and progeny populations of D. panamensis within four different 

6  3
 

habitats—continuous forest, forest fragments, pastures adjacent to fragments, and 

isolated pastures.  Using genetic techniques to analyze parentage, Hanson et al. (2008) 

determined that D. panamensis progeny located in isolated pastures were the results of 

the longest pollination dispersal distances compared to the other three habitats.  Pollen 

traveled a mean of 557 m with the furthest distance being 2.3 km for these offspring in 

isolated pasture.  This is compared to the shortest pollen dispersal distance among 

continuous forest trees of 240 m.  While pollen tended to travel farther amongst trees in 

pastures, these trees exhibited lower outcrossing rates compared to trees in continuous 

forest or forests fragments.  This suggests that long distance pollen movement is 

possible for these pasture trees, but the frequency is a lot lower due to higher rates of 

self‐pollination.  Thus, these tree populations could be at risk of long‐term genetic 

erosion due to the fragmented landscape.  

While this study by Hanson et al. (2008) elucidates the effect of habitat loss on the 

genetic structure of D. panamensis, it focused on the southern extent of the proposed SJLS 

Corridor and was conducted at a relatively small spatial scale.  The study sites were 

separated by a mean distance of 18.4 km, and the largest study site was 65 ha.  Results 

from the work by Hanson et al. (2008) could be used to scale up to a landscape level in 

order to study the separation of D. panamensis stands based on pollen dispersal 

distances. 

6  4
 

3.3 Materials and methods

3.3.1 Study Site: San Juan-La Selva Biological Corridor

Patterns of D. panamensis distribution, density, and fragmentation were 

discerned at a landscape level across 140,178 hectares within the proposed San Juan‐La 

Selva Biological Corridor (SJLS) located in northeastern Costa Rica (Figure 2) (refer to 

material in Chapter 1).  

3.3.2 Landscape distribution of D. panamensis

3.3.2.1 Digital aerial surveys 

The landscape distribution of canopy‐emergent D. panamensis trees was based on 

a digital aerial survey from July 4‐8, 2005.  The survey was timed to coincide with this 

species’ peak blooming.  For a detailed description of the aerial survey methods as well 

as the post‐flight image processing, please refer to Chapter 2.  

3.2.2.2 D. panamensis density determinations:  semi‐automated supervised 
classification and manual density approach   

After the digital images were captured, the images were adjusted and classified 

using a supervised approach (Chapter 2).  The purpose of these steps was to spectrally 

enhance the images in preparation for a supervised classification.  The supervised 

classification then identified blooming D. panamensis trees in an automated fashion.  Of 

the original 2,550 digital images that were gathered during the aerial survey, 801 were 

6  5
 

analyzed using the automated, supervised classification approach.  Thus, only about 

31% of the digital images collected were analyzed using the classification approach.  The 

classifier performed well on images such as these that contained contiguous forest or 

vegetation cover.  Based on the error assessment (Chapter 2) that identified the errors of 

omission and commission for the classification within the two survey areas termed 

MNWR (Maquenque National Wildlife Refuge) and NZ (Northern Zone), I 

systematically adjusted the number of tree clumps identified within each image scene by 

a uniform percentage.  Although the error within all images was not uniform, this 

approach addressed the average error throughout.  The adjustment value was based on 

the manual error assessment and represented:   

Number of D. panamensis clumps manually identified within each image


Number of D. panamensis clumps identified by the classifier
 

Here, the numerator of this equation represents the actual number of blooming 

D. panamensis within each image scene, while the denominator reflects the number of 

blooming D. panamensis as determined by the classifier.  In general, the classifier tended 

to overestimate the number of D. panamensis clumps within each scene.  All images 

within the MNWR region were adjusted by 79%.  However, the NZ images did not 

require adjustment.  There is a reason for the difference in adjustment values between 

the two sites.  Images within the NZ were captured on the same day and mostly 

possessed scenes of contiguous, uninterrupted forests.  These were ideal conditions for 

the application of the classifier.  Images within MNWR were captured over several days 
6  6
 

and possessed more variable scene conditions like agriculture, pastures, towns, or water.  

The classifier was less successful over these scenes and tended to overestimate the 

number of D. panamensis tree clumps within the MNWR study area.  Although I applied 

a systematic adjustment for these images, this does not completely remove all error from 

the images.  However, the degree of error remaining within these images regarding the 

number of D. panamensis tree clumps is acceptable compared to the number of trees that 

are being surveyed over the entire 140,178‐hectare study area. 

The classified images likely contained a higher density of D. panamensis due to 

the preponderance of intact forest within the scene.  Thus, I selected a spatially stratified 

subset of the remaining images to perform a manual count of D. panamensis within each 

scene.  These images had not been analyzed via the supervised classification.  A total of 

503 images were selected for the manual count representing about 20% of the total 

digital images collected.  Because these images contained less contiguous forest and 

more human altered landscapes, the manual count was not exceptionally time intensive 

and provided a more accurate assessment of D. panamensis trees than the supervised 

classification approach.  However, the supervised classification was appropriate for 

image scenes with contiguous vegetation and a lot of blooming D. panamensis as the 

classifier provided a more automated, systematic, and complete approach.  A manual 

effort for these particular images would have been prohibitively time consuming. 

6  7
 

Following either the classifier or manual approach, D. panamensis density within 

each image was determined by dividing the total number of D. panamensis tree clumps 

within each image by the image’s area.  The area for each image was calculated based on 

the field of view (FOV) represented in each image.  The FOV calculation incorporated 

the focal length of the digital camera, the height above the landscape that the image was 

taken, and the pixel dimensions of the image.  The average FOV for all classifier and 

manual images was 113.45 ha (SD 24.01).  Because the locations of all images had been 

recorded with an onboard GPS unit during the original aerial survey, the density value 

for each image could be spatially depicted within a GIS framework. 

3.2.2.3 Creation of a D. panamensis density map using the IDW interpolation method 

In order to depict the spatial variation in D. panamensis density across the study 

area, I created an interpolated surface within a GIS framework of all of the D. panamensis 

densities garnered from individual images.  Interpolation is a technique that enables the 

creation of a continuous raster surface from sampled data.  Because it is often impossible 

to perform a complete census of an area due to financial or time constraints, 

interpolation is a method by which intervening, non‐sampled locations are estimated 

from sampled values (Theobald 2003).   

Numerous interpolation methods exist such as spline, kriging, natural neighbor, 

and inverse distance weighting (IDW).  Selecting the appropriate technique depends on 

the type of data to be displayed, data accuracy, and the spatial distribution of the data 

6  8
 

points.  Interpolators like IDW and spline are deterministic because they assign values to 

unsampled locations based on the surrounding measured values and mathematical 

formulas that determine the smoothness of the resulting surface.  Kriging is considered a 

geostatistical method and the resulting interpolation is based on a statistical model that 

includes autocorrelation.  In addition, interpolators can be exact or inexact.  Exact 

interpolators preserve the original input value at the sample location.  Exact 

interpolators are suitable for data that can be measured reliably and contain a lot of 

certainty.  With inexact interpolators, however, input values may vary from their 

original value in order to achieve a better fit in the resulting interpolation. 

Because the D. panamensis density values covered the project area in an evenly 

spaced fashion, I utilized the IDW interpolation method using the Geostatistical Analyst 

tool in ArcGIS 9.1 to create a continuous map of D. panamensis density.  The IDW 

method was appropriate because as an exact interpolator, it preserved the original 

density input values.  The IDW interpolation also explicitly assumes that closer things 

are more similar to one another than things that are farther apart.  The IDW method 

estimates new cell values in the output raster surface by averaging the values of the 

sample data points in the vicinity of each cell (ESRI 2005).  The closer an input point is to 

the cell being estimated, the more weight it has in determining the value of that cell.  

Thus, the weight of a measured point diminishes with distance—hence the term 

“inverse distance weighting”.  The power parameter in the IDW method affects how fast 

6  9
 

the weights decrease.  If the power parameter equals 0, there is no decrease with 

distance.  As the power parameter rises above 0, the weights of distant points decrease 

rapidly.  The Geostatistical Analyst tool in ArcGIS 9.1 has a way to optimize the power 

parameter.  In general, the power parameter should be calibrated depending on the 

ecological phenomena being modeled.  For example, a larger power parameter should 

be used to model phenomena that have a more localized effect.   

A total of 1,304 D. panamensis density inputs were used for the interpolation.  

These density inputs were distilled from the same number of images.  The image sample 

represents 51% of the total number of images gathered and covers approximately 

147,856 ha.  This area exceeds the size of the area originally flown (140,178 ha) as some 

images overlap in coverage.  An optimized power function of 1.24 was used in the IDW 

interpolation model.  The sample window was an ellipsoid shape and included at least 4 

neighboring input points.  A cross‐validation was performed on the data set in order to 

determine how well the model performed.  The cross‐validation step systematically 

omits a point and compares its predicted value to its true value.  The deviation from the 

predicted to the true value is expressed by the root mean square error (RMSE) given 

below: 

RMSE = (∑ x − x 0 )2 /n
 

The RMSE is the square root of the average squared difference between the 

predicted value (x) and the “true” value (x0) where n is the number of points.  The RMSE 

7  0
 

for the interpolation model was 0.18.  The optimized power parameter of 1.24 produced 

the lowest RMSE compared to power parameters of other values. 

Based on the interpolation, I produced a density map showing the distribution of 

D. panamensis across the study area.  The map displayed the densities based on 

increments of 0.1. 

3.3.3 Density trends over the study landscape and forested areas

I used the interpolated density map to analyze density trends across the study 

landscape.  I was specifically interested in locations where high densities of D. 

panamensis were predicted.  In particular, I examined the distribution of D. panamensis 

densities across the total landscape as well as over only forested area.  The forested areas 

were derived from a land‐use classification of a satellite image over the northeastern 

part of Costa Rica.  The satellite classification shows land use from 1997‐2000 (Figure 8).  

This satellite classification was conducted by the Earth Observation Systems Lab, 

University of Alberta, Canada.  The satellite classification’s extent is cropped and does 

not cover the entire area of the aerial survey.  Thus, some of the analyses had to be 

abridged. 

3.3.4 Environmental factors predicting high density locations of D.


panamensis: a classification tree approach

I used the D. panamensis density map as well as other GIS data to identify 

environmental conditions that have the potential to support regions of high D.

7  1
 

Figure 8:  Land use classification within proposed San Juan‐La Selva Biological 
Corridor 

7  2
 

panamensis density.  For this analysis, I used a classification and regression tree (CART) 

model.  CART models are one of many methods that can be used to predict species‐

habitat distributions and identify potential habitat (Guisan and Zimmermann 2000).  

However, CART models offer several advantages over traditional methods such as 

linear discriminant analysis or cluster analysis due to their flexibility.  De’ath and 

Fabricius (2000) note that CART models are uniquely equipped to deal with ecological 

data that are often complex, unbalanced, or missing values.  They are a nonparametric 

approach to distinguishing differences among groups (Breiman et al. 1984, Vayssieres et 

al. 2000).  In addition, because CART models are graphical in nature, each conditional 

path can easily be executed in a GIS platform in order to display habitat distributions 

(Urban et al. 2002). 

The response variable in a CART model can either be categorical or continuous.  

If it is the former, a classification tree is produced; if it is the latter, a regression tree is 

obtained. The explanatory variables may be categorical, continuous, or a mix of both.  

CART models produce a decision tree by recursively partitioning the data.  At each split, 

data are aggregated into two mutually exclusive groups, each of which is as 

homogeneous as possible.  The subgroups are then partitioned further until the data 

cannot be split anymore or a stopping criterion is met.  The objective is to partition the 

response variable into homogeneous groups but also keep the tree reasonably small 

(Deʹath and Fabricius 2000).  Because CART models attempt to partition all the data, 

7  3
 

they can often over‐fit the data.  Large decision trees may actually be meaningless.  

Thus, these decision trees are then pruned back to the desired size in order to avoid 

over‐fitting the data.   

Decision trees are presented graphically with a root node at the top, which 

represents the undivided data, and branches and leaves toward the bottom.  Terminal 

leaves represent homogeneous data that no longer can be partitioned.  The pathway to 

each terminal leaf in the decision tree defines a set of environmental conditions under 

which the splitting rule associated with that leaf will apply (Moore et al. 1991).  It is this 

set of conditions that can be entered into a GIS platform and used to graphically display 

potential habitat. 

In order to identify potential habitat that can support high density D. panamensis 

stands, I utilized a CART model.  The response variable was D. panamensis density, 

which was binned into different density categories making the model output a 

classification tree.  Environmental predictor variables were a mixture of categorical and 

continuous.  Predictor variables were selected based on an a priori knowledge of 

environmental factors that could be important in determining D. panamensis occurrences 

and potential habitat.  These were gleaned from the literature and included information 

on soil, elevation, and rivers.  Descriptions of the specific response and predictor 

variables that were considered in the model are summarized in Table 3. 

7  4
 

Table 3:  Response and predictor variables used in CART model to predict potential 
D. panamensis habitat 

Name  Description  Variable Type 


Response variable:     
DENSITY  Values of D. panamensis  Categorical 
density binned in 4 
increments 
Predictor variable:     
SOILOR  Soil order  Categorical 
SOILSUB  Soil suborder  Categorical 
SOILGG  Soil great group  Categorical 
SOILTOPO  Soil topography  Categorical 
ELEV  Elevation of terrain  Continuous 
RIV  Distance to nearest river  Continuous 
 

Data for the response and predictor variables were contained within GIS rasters 

or shapefiles that were standardized to the same spatial extent and geographic 

projection.  GIS data were projected to WGS 1984 UTM Zone 16N.   All GIS analyses 

were conducted using ArcGIS 9.1.  Data for the response variable, DENSITY, consisted 

of 1,265 observations representing a particular D. panamensis density bin.  Hence, the 

response variable was categorical.  These density bins were distilled from the original D. 

panamensis density values originating from the digital images covering the SJLS 

Corridor.  Density values were obtained using both an automated pattern recognition 

technique as well as a manual assessment of blooming D. panamensis crowns (Section 

3.2.2).  Each density value was geographically referenced to the location in which the 

original photograph was taken and all points were contained in a GIS shapefile.  The 

density bins are summarized in Table 4. 

7  5
 

Table 4:  Summary of response variable bins 

Bin Name  D. panamensis density range  Number of observations 


Bin #1  0.00 trees/hectare  85 
Bin #2  0.01‐0.20 trees/hectare  404 
Bin #3  0.21‐0.40 trees/hectare  251 
Bin #4  >0.41 trees/hectare  525 
Total Observations    1,265 
 

The predictor variables were obtained by using the D. panamensis shapefile and 

intersecting the point locations with the raster files for each predictor variable in order to 

extract the underlying data values.  GIS data for the environmental predictor variables 

originated from the Organization for Tropical Studies.   

The CART analysis was performed in the R Statistical Program using the “tree” 

package (R Development Core Team 2008).  I tested several classification trees to 

determine which predictor variables were relevant in identifying potential habitat that 

could support high D. panamensis density.  Predictor variables that did not produce a 

logical split in the classification tree were removed from consideration.  The “tree” 

settings were customized for the analysis.  For example, the minimum required number 

of observations per node was changed from 20 to 10.  Classification trees were run 

utilizing the cross‐validation tool and with it disabled.  The final classification tree was 

run using the following environmental variables:  SOILSUB and ELEV.  This 

classification tree required no pruning.   

7  6
 

3.3.5 Delineation of potential D. panamensis habitat

I was specifically interested in the environmental conditions that could support 

D. panamensis populations at densities > 0.4 trees/ha.  Thus, I used the parameters from 

the classification tree in order to produce a map depicting potential D. panamensis 

habitat. 

3.3.6 Fragmentation analysis of moderate density D. panamensis


patches

I utilized the D. panamensis density map developed in Section 3.2.3 to analyze 

how habitat patches of this species were connected throughout the landscape.  Hanson 

et al. (2008)  observed that D. panamensis progeny located in isolated pastures were the 

results of the longest pollination dispersal distances, with the farthest distance being 2.3 

km.  Thus, I used this distance as a threshold.  Distances below this value would indicate 

that two discrete patches were connected based on pollen flow between them.  Distances 

above this value would denote that two patches were disconnected.  I delineated D. 

panamensis patches by categorizing the density map into four different density bins.  

These density bins roughly corresponded to habitat quality.  Regions without any D. 

panamensis trees had densities between 0 and 0.1 trees/ha.  Low density regions had 

densities between 0.1 and 0.2 trees/ha.  Moderate density areas exhibited densities 

between 0.2 and 0.5 trees/ha.  Finally, high density regions possessed densities > 0.5 

trees/ha.  These bins are detailed in Table 5.  The purpose of this analysis was to gauge 

7  7
 

one aspect of habitat fragmentation based on pollen dispersal.  Examining seed dispersal 

patterns could have provided a glimpse of seedling recruitment, the effectiveness of 

animal seed dispersers, or another aspect of habitat fragmentation.  However, only data 

on pollen movement were available.      

Table 5:  Description of bins for fragmentation analysis 

Name  Density bin  % of Area 


(D. panamensis tree clumps/ha) 
No D.  
0‐0.1  32% 
panamensis 
Low density  0.1‐0.2  15% 
Moderate  
0.2‐0.5  32% 
density 
High density  >0.5  21% 
 

I used the program FRAGSTATS to analyze how moderate density patches of D. 

panamensis were connected throughout the landscape based on the pollen dispersal 

distance of 2.3 km.  FRAGSTATS was developed by McGarigal et al. (2002), and it is 

designed to compute a wide variety of landscape metrics at patch, class, and landscape 

scales.  In particular, I examined the connectivity of moderate density patches to other 

moderate or high density patches.  Only patches that were over 20 ha and were 

contained entirely within the bounds of the study area were included.  I also determined 

the mean patch size and the mean Euclidean distance of each moderate density patch to 

its nearest neighbor.  The latter metric provided a measure of habitat patch isolation 

across the landscape.  

7  8
 

I also computed the connectance index between these moderate density D. 

panamensis patches.  In FRAGSTATS, connectance measures the number of functional 

joinings between patches of the same class that are within a specified distance of each 

other.  It assesses whether each patch is connected to every other patch in the landscape 

based on the distance threshold.  Here, I used the value of 2.3 km based on the longest 

pollen dispersal distance measured by Hanson et al. (2008).   

The fragmentation analysis described above was primarily a heuristic exercise 

and employed several simplifying assumptions about the landscape and the ecology of 

pollen and gene flow.  First, this analysis examined whether direct pollination events 

could occur among trees located in a discrete moderate‐density patch to trees located in 

another moderate‐density patch.  I utilized the distance threshold of 2.3 km, which 

represents the maximum pollination dispersal distance between D. panamensis trees 

measured by (Hanson et al. 2008).  The analysis ignores the potential contribution of 

isolated trees within the intervening areas between patches that are located in either low 

or no‐density regions.  Thus, I am not claiming that moderate‐density patches within the 

landscape are genetically isolated from each other, nor am I maintaining that gene flow 

does not occur between these patches.  I am merely examining whether direct 

pollination events could occur.  In addition, I am not specifically examining gene flow as 

my analysis is not occurring at the individual tree level—rather the patch level.  

7  9
 

Ultimately, the purpose of this analysis is to apply Hanson et al.’s (2008) results based on 

a localized study and scale up to a landscape level. 

3.4 Results and discussion

3.4.1 Landscape distribution of D. panamensis

Figure 9 shows the distribution of D. panamensis density across the 140,178‐ha 

study area.  Densities are binned into increments of 0.1 tree clumps/ha.  The 

northwestern part of the study area contains the highest densities with, canopy‐

emergent trees reaching 2 tree clumps/ha.  This high‐density region borders Nicaragua. 

3.4.2 Density trends over the study landscape and forested areas

Table 6 summarizes the percentage of the total landscape surveyed that is 

represented in each D. panamensis density bin.  Table 6 also shows the distribution of D. 

panamensis density over only forested areas.  The satellite classification that was used to 

identify forested areas does not cover the entire extent of the density map.  In particular, 

the satellite classification does not extend into the extreme northern part of the study 

area where D. panamensis densities are the highest.  Hence, approximately 3,388 ha of the 

original D. panamensis density map are not included in this analysis which is why these 

values are missing.  Figure 10 represents the percentage of the total landscape 

represented in each D. panamansis density bin.  Figure 11 represents the percentage of 

the forested landscape represented in each D. panamensis density bin.

8  0
 

Figure 9:  Density distribution of D. panamensis 

8  1
 

Table 6:  Percentage of the total and forested landscape represented by each D. 
panamensis density bin 

Density Bin Range  Area (ha)  Percentage  Area (ha)  Percentage 


(trees/hectare)  (over total  (over total  (over forest  (over forest 
landscape)  landscape)  cover)  cover) 
0‐0.1  44095.05  31.45%  18870.84  21.72% 
0.1‐0.2  21574.89  15.39%  13539.24  15.59% 
0.2‐0.3  16135.74  11.51%  10550.79  12.15% 
0.3‐0.4  14922.72  10.64%  10261.08  11.81% 
0.4‐0.5  13531.32  9.65%  10139.22  11.67% 
0.5‐0.6  12820.05  9.14%  10661.13  12.27% 
0.6‐0.7  8317.89  5.93%  7143.57  8.22% 
0.7‐0.8  4600.44  3.28%  3956.04  4.55% 
0.8‐0.9  1617.48  1.15%  1192.95  1.37% 
0.9‐1.0  676.35  0.48%  384.48  0.44% 
1.0‐1.1  503.46  0.36%  156.33  0.18% 
1.1‐1.2  284.58  0.20%  11.7     0.01% 
1.2‐1.3  204.84  0.15%  Data Missing  Data Missing 
1.3‐1.4  200.61  0.14%  Data Missing  Data Missing 
1.4‐1.5  100.89  0.07%  Data Missing  Data Missing 
1.5‐1.6  119.88  0.09%  Data Missing  Data Missing 
1.6‐1.7  170.91  0.12%  Data Missing  Data Missing 
1.7‐1.8  215.37  0.15%  Data Missing  Data Missing 
1.8‐1.9  82.98  0.06%  Data Missing  Data Missing 
1.9‐2.0  18.63  0.01%  Data Missing  Data Missing 
Total  140,194.08  100%  86,867.37  100% 
 

8  2
 

Figure 10:  Percentage of the total landscape represented in various D. panamensis 
density bins 

8  3
 

Figure 11:  Percentage of forested landscape represented in various D. panamensis 
density bins 

8  4
 

  This research highlights that canopy‐emergent D. panamensis trees can occur at 

densities as high as 2 trees/ha.  Nearly 20 ha were identified that support this density 

level.  In addition, nearly 600 ha contain densities over 1.5 trees/ha, and 1,900 ha contain 

densities over 1.0 tree/ha.  The aerial survey could only detect individual trees that 

protruded from the canopy and those that were in flower.  Thus, the total density of this 

species would be higher if other life‐history stages like understory trees or non‐

flowering individuals were included. 

Density patterns within the La Selva Biological Station are also worth noting.  La 

Selva, which is operated by the Organization for Tropical Studies, comprises about 1,600 

ha of tropical wet forests.  Only 413 ha of La Selva (25%) contained densities between 

0.1‐0.2, and 96 ha of La Selva (6%) contained densities between 0.2‐0.3 ha.  The 

remainder of the field station had densities less then 0.1 tree clumps/ha. 

While it is often thought that canopy‐emergent tree species like D. panamensis 

exist at densities less then 1 adult‐tree/ha (Clark and Clark 1987), this work clearly 

shows that this species can occur at much higher densities.  However, the distribution of 

this species throughout the entire study area averages between 0.3 and 0.4 trees/ha.  

Thus, over a large spatial area, the relative densities are more moderate.  Nonetheless, 

the results of this work may assist in ongoing conservation endeavors such as MNWR 

and the SJLS Corridor aimed at protecting this keystone tree and other species within 

the lowland Atlantic rain forest of Costa Rica.   

8  5
 

3.4.3 Environmental factors that predict locations of high D.


panamensis density

Results from the classification tree model indicated the environmental conditions 

predicted to support D. panamensis stands with a density of 0.4 tree clumps/ha or 

greater.  These environmental conditions occur between elevations of 45‐125 meters and 

on the Humult, Aquent, and Tropept‐Aquept soil suborders.   

3.4.4 Delineation of potential D. panamensis habitat

Figure 12 shows the distribution of potential habitat predicted to support D. 

panamensis at densities of 0.4 tree clumps/ha and above.  The map highlights the greater 

SJLS Corridor vicinity.  Potential habitat covers about 236,568 hectares and is the result 

of the classification tree model.  Although the potential habitat includes the Humult, 

Aquent and Tropept‐Aquept soil suborders, 96% of the potential habitat occurs on the 

Humult suborder.  Humult soils are soils with high organic mater content (humus) that 

are generally found on high ground (tierra firme).  They belong to the Ultisol soil order.  

Soils belonging to this order are normally found on old land surfaces and are formed 

from the processes of clay mineral weathering (Brady and Weil 2002).  They are low in 

bases and have subsurface horizons of illuvial clay accumulations.   

Ultimately, the classification tree model predicts that D. panamensis will occur 

within a much narrower elevation range than a previous habitat description by Flores 

(1992).  She stated that the elevation distribution of this species is between 20 and 1300 

8  6
 

Figure 12:  Potential habitat that can support D. panamensis > 0.4 trees/hectare 

8  7
 

meters.  This species is much more restricted to lowland habitat then previously 

described as it is predicted to occur between 45 and 125 meters in elevation.  The soil 

conditions predicted to support densities of this species >0.4 tree clumps/ha generally fit 

prior published reports.  Flores (1992) stated that this species is found on alluvial or 

sandy soils and sometimes in acid and clay soils.  The Humult soil suborder fits this soil 

profile because it is derived from clay mineral weathering, is low in bases, and has a 

high organic content.   

Elevation emerged as an important environmental predictor variable in the 

CART model because it likely serves as a proxy for other edaphic conditions that can 

support D. panamensis.  For instance, soils between 45 and 125 meters in elevation may 

exhibit appropriate soil texture or drainage conditions for this species. 

Based on the habitat analysis, 236,568 ha have the potential to support this 

species at densities over 0.4 trees/ha.  This area represents less then 5% of the country.  

Thus, despite this species’ designation as a common, generalist tree, conservation 

measures that prohibit its removal should be renewed and strengthened.  In addition, 

potential habitat should be protected and/or restored. 

3.4.5 Degree of fragmentation of moderate density D. panamensis


patches

Thirteen patches with moderate D. panamensis densities were analyzed for 

whether direct pollination events could occur between trees within each discrete patch.  

8  8
 

These 13 patches are indicated in Figure 13.  These patches were derived from the D. 

panamensis density map by applying a threshold to identify density regions between 0.2 

and 0.5 tree‐clumps/ha.  I considered patches connected if 2.3 km or less separated them.  

Patches were fragmented if they were separated by over 2.3 km.  Basic metrics were also 

collected for these 13 patches including patch size and Euclidean distance to the nearest 

patch.  Results for these metrics are summarized in Table 7.  The mean patch size was 

263 ha while the mean Euclidean distance was 1.52 km. 

Table 7:  Summary of FRAGSTATS metrics 

Metric  Mean  Lowest  Highest 


Patch Size  263 ha  32 ha  1843 ha 
Euclidean Distance  1.52 km  0.16 km  6.15 km 
 

The connectance index for these 13 patches was 12.1%.  The connectance index is 

a measure of the functional joinings between patches where each pair of patches is either 

connected or not based on a specific distance threshold.  It is reported as a percentage of 

the maximum possible connections given the number of patches.  The distance threshold 

for this landscape was 2.3 km based on the longest pollen dispersal distance measured 

by Hanson et al. (2008).  This connectance index indicates that of the maximum possible 

connections between the 13 patches, only 12.1% of the potential connections are less than 

2.3 km. 

8  9
 

Figure 13:  Distribution of 13 subpopulation patches with moderate D. panamensis 
density 

9  0
 

Figure 14 illustrates the connectivity and fragmentation of the 13 moderate 

density patches based on a pollen dispersal distance of 2.3 km.  Connected patches are 

shaded the same color.  Based on this analysis, there are four separate subpopulations.  

It is unlikely that direct pollination events can occur between subpopulations of different 

colors.  Seven of the patches (shaded dark blue) are connected to a larger habitat patch.  

One subpopulation (shaded light green) is located within the La Selva Biological Station.  

Based on the extent of this study area, direct pollination events are unlikely between this 

patch and the other 12 moderate density patches. 

The purpose of this analysis was to gauge one aspect of fragmentation based on 

pollen dispersal distances across patches of D. panamensis.  This study does not suggest 

that the four subpopulations are genetically isolated from each other as there are 

intervening D. panamensis trees in the low or no density regions.  However, what this 

work implies is that there are certain moderate density patches that have direct pollen 

flow between patches and are hence more connected.  These patches may be more 

disconnected to other moderate density patches that occur over 2.3 km away.  Thus, 

trees within a linked subpopulation may be more genetically similar to each other 

compared to trees in patches farther afield.  Based on the scope of this work, the trees 

located within the subpopulation at the La Selva Biological Station may be genetically 

more similar to each other than trees in the other subpopulations.  Pollen from these 

trees may not be flowing to other clusters of moderate density patches. 

9  1
 

Figure 14:  Connected moderate density D. panamensis patches 

9  2
 

3.5 Conclusions

This study indicates that canopy‐emergent D. panamensis trees have the potential 

to exist at densities approaching 2 adult trees/ha.  This density is much higher than 

previous measurements which ranged from 0.45 trees/ha to 0.97 trees/ha (Hanson et al. 

2006, Hanson et al. 2008).  Elevation and soil suborder are also important conditions that 

predict appropriate habitat that can support D. panamensis.  In particular, elevation may 

serve as a proxy for other important edaphic conditions such as drainage or soil texture.  

In addition, this work examines the landscape distribution of an important keystone 

species across 140,178 ha.  Because this area lies within a proposed biological corridor, 

results from this work may inform the ongoing conservation endeavors.   

The FOV of the digital images that were used to determine the density estimates 

averaged 113 ha.  The FOV is analogous to the size of a very large terrestrial forest 

inventory plot.  However, these sampling units that comprised the overall density map 

were much larger than most forest inventory plots in the tropics.  As Clark et al. (2004b) 

stress, remote sensing provides a way to increase the size of areas sampled and improve 

the statistical sampling design.  Because the crowns of adult D. panamensis often extend 

above the canopy, the aerial survey method also permitted the study of a species that 

can be difficult to track from the ground.   

Because D. panamensis was censused during its peak blooming, the results 

suggest that there is a sizable reproductive population.  Although a copious crown of 

9  3
 

flowers may not necessarily translate into an abundant fruit crop, this work does 

indicate that dense areas may experience high reproductive output.  Many animals, 

including the endangered Great Green Macaw, consume D. panamensis fruits.  Thus, a 

healthy, reproductive population is crucial to sustain the organisms dependent on this 

keystone resource.  However, the results of the fragmentation analysis indicate that the 

landscape contains some D. panamensis subpopulations where pollen may not be moving 

directly.  This could affect the long‐term genetic diversity of certain regions.  An 

interesting complement to this work would be to study patterns of D. panamensis gene 

flow through seed dispersal.  Patterns of seed dispersal could provide another glimpse 

of habitat fragmentation based on movement of seeds by mammalian seed dispersers as 

well as recruitment of seedlings.  This work could elucidate new habitat opportunities 

for the Great Green Macaw.   

Finally, this work suggests that similarly dense regions of D. panamensis exist 

across Costa Rica’s border.  The highest densities of this species were located in the 

northern part of the aerial survey right at the border with Nicaragua’s Indio‐Maíz 

Biological Reserve (Figure 15).  This region likely has similar, high densities of D. 

panamensis but distributed over a much larger geographic area.  Indio‐Maíz Biological 

Reserve is one of Nicaragua’s largest protected areas covering 264,000 ha.  Despite the 

fact that many local inhabitants still extract natural resources here due to the country’s 

tumultuous history, this reserve is one of the largest remaining stands of primary rain 

9  4
 

Figure 15:  Location of Indio‐Maíz Biological Reserve, Nicaragua in relation to 
potential D. panamensis habitat in Costa Rica 

9  5
 

forest in all of Central America (Nygren 2004).  Thus, the Indio‐Maíz Biological Reserve 

is an important piece in the larger landscape.  Subsequent research on D. panamensis 

should occur in the Indio‐Maíz Biological Reserve in order to extend the biogeographic 

study of this species.  In addition, future conservation efforts should promote 

multinational partnerships that protect biodiversity spanning both countries.

9  6
 

Chapter 4. Determination and description of potential


nesting habitat for the Great Green Macaw (Ara
ambigua) in northern Costa Rica using a classification
and regression tree model

4.1 Introduction

The Great Green Macaw (Ara ambigua) is the second largest New World macaw 

and one of only two macaw species native to Costa Rica.  Because this species is 

threatened by habitat loss and degradation, nest‐robbing and trapping for the 

international bird trade1, as well as persecution via hunting (Guedes 2004), the Great 

Green Macaw is listed as endangered on the International Union for Conservation of 

Nature’s (IUCN) Red List (EN A2cd+3cd) (BirdLife International 2005).  In addition, the 

Convention on International Trade in Endangered Species of Wild Fauna and Flora has 

designated the Great Green Macaw as an Appendix I species (CITES 2008).  Appendix I 

species are threatened with extinction, and all international trade in commercial 

specimens is prohibited.   

In northeastern Costa Rica, Powell et al. (1999) estimate that only 35 Great Green 

Macaw breeding pairs and 200 individuals remain.  A follow‐up to this study suggests 

that the overall population may have expanded slightly to 250 individuals (Vargas M. 

2008).  While habitat fragmentation and degradation are affecting the persistence of this 

                                                      

1 Great Green Macaw nestlings have fetched $300 apiece in Costa Rica (Bjork and Powell 1999). 
9  7
 

species in Costa Rica, scientists contend that the loss of breeding habitat and removal of 

its nesting trees are having the greatest impact on this species (Powell et al. 1999).  Like 

most other macaw species, the Great Green Macaw is a secondary cavity nester.  

Breeding pairs rely on large hollows in mature trees to lay eggs and raise their young.  

In particular, Great Green Macaw breeding pairs have an affinity for one particular tree 

species.  A census of Great Green Macaw nests that occurred between 1994 and 2003 

identified that nearly 90% were located in large Dipteryx panamensis trees (Powell et al. 

1999, Chassot et al. 2002).  Not only is this tree species important as a nesting substrate 

for the Great Green Macaw, but its fruits are an important food resource throughout the 

breeding season.  Although D. panamensis is a relatively common, canopy‐emergent 

species within the lowland Atlantic forest of Costa Rica, it has been logged extensively 

in the past. 

The creation of Maquenque National Wildlife Refuge (MNWR) in 2005 

represents a positive first step in the protection of potential nesting and breeding habitat 

for the Great Green Macaw in northern Costa Rica (Figure 2).  This region was targeted 

because it contains several large nesting trees used by Great Green Macaw breeding 

pairs.  In addition, MNWR protects foraging habitat that may be critical during the 

Great Green Macaw’s breeding season.  MNWR is also embedded within the larger San 

Juan‐La Selva (SJLS) Biological Corridor effort that aims to connect protected areas in 

southern Nicaragua to those in central Costa Rica (Chapter 1). 

9  8
 

4.1.1 Objectives

To complement this conservation project, I conducted research to further 

elucidate the ecology of the Great Green Macaw’s nesting and breeding habitat.  I 

utilized the results of an aerial survey that I performed in 2005 documenting the 

distribution and density of D. panamensis trees across 140,178 ha in northern Costa Rica 

(Chapters 2 and 3).  I synthesized the biogeographic distribution of D. panamensis with 

the locations of known nesting trees used by the Great Green Macaw in order to assess 

the quantity and quality of nesting habitat for this species.  In particular, I addressed the 

following questions:  

1. What environmental factors predict optimal Great Green Macaw nesting 

habitat? 

2. Based on these environmental factors, how much potential habitat exists 

within northern Costa Rica? 

3. What areas should be targeted as conservation priorities? 

4.2 Background

4.2.1 General description and ecology of the Great Green Macaw in


Costa Rica

The Great Green Macaw belongs to the family Psittacidae.  It is one of 17 species 

of macaws, of which one is extinct, another is extinct in the wild, and seven other species 

are endangered (Texas A&M University 2006).  In general, macaws are characterized by 

9  9
 

a long tail, large beak, and bare facial patch (Abramson 1999).  Adult Great Green 

Macaws are mostly yellowish‐green with a scarlet forehead.  This species approaches 31 

inches in length and can weigh up to 1.3 kg (Stiles and Skutch 1989).  The Great Green 

Macaw occurs as two sub‐species.  The nominate race occurs in humid, lowland forests 

of the Atlantic/Caribbean lowlands of eastern Honduras, Nicaragua, and Costa Rica, 

both slopes of Panama, and in northwestern Colombia; the race guayaquilensis occurs on 

the Pacific slopes of Western Ecuador (Snyder et al. 2000, BirdLife International 2005).  

According to Snyder et al. (2000), the Central American populations are restricted to four 

isolated subpopulations that are confined to lowland Atlantic forest.  The subpopulation 

that is the focus of this chapter occurs from southeastern Nicaragua to northeastern 

Costa Rica.  (Hereinafter, I will use population to refer to the Costa Rican portion of the 

Nicaraguan/Costa Rican subpopulation.)   

Until recently, very little was known about the Great Green Macaw’s food, 

nesting, and habitat requirements.  As of 1989, its nest was still undescribed (Stiles and 

Skutch 1989).  One of the first extensive examinations of ecological, behavioral, and 

demographic dynamics of Costa Rica’s Great Green Macaw population was initiated in 

1994 by the Great Green Macaw Research and Conservation Project, which is now 

affiliated with the Centro Científico Tropical (CCT) in San José, Costa Rica.  This 

research by Powell et al. (1999) comprised radio‐telemetry efforts to determine Great 

Green Macaw home ranges, monitoring and data collection on all known nest sites, and 

100
 
 

phenological studies of tree species that constituted its diet.  During the study period, 

Powell et al. (1999) estimated that Costa Rica’s population of Great Green Macaw 

averaged around 200 individuals with no more than 35 reproductively active pairs.  A 

follow‐up to this study was conducted in 2008, and it indicated that the overall 

population may have increased to 250 Great Green Macaw individuals (Vargas M. 2008).  

In northern Costa Rica, the breeding range has shrunk by 90% and is restricted to about 

1,120 km2.  This area abuts southern Nicaragua and the San Juan River to the north and 

is bordered by the Sarapiquí and San Carlos Rivers to the east and west, respectively.  In 

2005, part of this region was decreed MNWR.   

The Great Green Macaw is dependent on D. panamensis trees for food and 

nesting.  As secondary cavity nesters, Great Green Macaw breeding pairs select large D. 

panamensis trees with hollows in order to construct nests.  This tree is subject to 

numerous threats that may cascade and affect the Great Green Macaw’s long‐term 

viability (Chapter 3).  The scarcity of natural nest sites may be one of the major factors 

limiting the number of adult macaws—including the Great Green—that can reproduce 

during a breeding season (Munn 1992).  Cavities take a long time to form since the 

process is initiated by decomposition when water or moisture enters the wood, drains 

down the inside of the trunk or branch, and empties out of the tree through fissures and 

cracks (Mawson and Long 1994).  While macaws will clean and maintain cavities, they 

do not form them on their own, hence the designation as secondary cavity nesters.  

101
 
 

Because macaws may use the same nesting tree year after year, the removal of a single 

nesting tree represents a reduction in reproductive output over the life of the tree, since 

hundreds of macaw chicks may be reared from a single tree during its lifetime 

(Brightsmith 2005).  In addition, the fact that most wild macaw species have a naturally 

low reproductive rate only compounds these anthropogenic threats and underscores the 

importance of identifying and protecting their nesting and breeding habitat (Nycander 

et al. 1999).   

Between 1994 and 2002, 66 active Great Green Macaw nests located in northern 

Costa Rica were tracked by CCT scientists.  Of these, 57 (86%) were located in large D. 

panamensis trees (Powell et al. 1999, Chassot et al. 2002).  The remaining nests were 

located in large trees of the following species:  Carapa nicaraguensis, Goethalsia meiantha, 

and Vochysia ferruginea (Powell et al. 1999).  In northern Costa Rica, nesting occurs 

between December and June with egg‐laying occurring in late January.  Nests contain 

eggs in January or February and nestlings between February and April.  Most chicks 

fledge by late April.   

The Great Green Macaw closely tracks fruit production of a variety of tree 

species and may forage on fruit from over 30 species as they move between Costa Rica’s 

lowlands and higher volcanic regions.  Great Green Macaws consume D. panamensis 

fruit from about September to April (Chapter 3).  During peak D. panamensis fruit 

production in January, this resource may account for 85% of the Great Green Macaw’s 

102
 
 

diet (Powell et al. 1999).  January is also a crucial time in the Great Green Macaw’s 

reproductive cycle as this month coincides with its egg‐laying period and precedes 

hatching and nestling development.  Around June, the birds disperse from lowland, 

Atlantic forests to higher elevations of Costa Rica’s volcanic ranges and drier forests in 

Nicaragua.  This seasonal migration correlates with the cessation of fruiting for D. 

panamensis, and the birds are likely searching for alternative food sources at higher 

elevations.  Between April and August when D. panamensis fruit are scarce, Great Green 

Macaws will switch to fruit from Sacoglottis trichogyna.  Powell et al. (1999) reported that 

D. panamensis and S. trichogyna accounted from between 65% and 95% of foraging trees 

between October and July. A study of Great Green Macaws in Ecuador also noted this 

population’s seasonal movements coupled with changes in food availability (Berg et al. 

2007).   

4.3 Materials and methods

4.3.1 Determination of Great Green Macaw nesting habitat

Snyder et al. (2000) conjectured that it is the loss of the large nesting trees, 

specifically D. panamensis in Costa Rica that may jeopardize the long‐term viability of 

this species.  Results from work that I conducted regarding the distribution of D. 

panamensis density across 140,178 ha in northern Costa Rica indicate that this tree species 

still exists in relatively high densities (Chapter 3).  Reproductive, adult D. panamensis can 

103
 
 

reach densities as high as 2 trees/ha.  Nearly 20% of the land surveyed exhibits densities 

of 0.5 trees/ha and higher, and over 50% exhibits densities of 0.2 trees/ha and higher.  

However, an analysis of landscape connectivity and fragmentation indicated that 

moderate density patches of D. panamensis exhibit a very small average patch size.  

Direct pollen flow may also be lacking between these patches as they are separated by 

distances greater then the maximum pollen dispersal distance measured for this species 

(Chapter 3). 

Because the quantity and quality of nesting habitat is critical for the Great Green 

Macaw, I investigated whether there are environmental factors that predict optimal 

Great Green Macaw nesting habitat, how much potential habitat exists within northern 

Costa Rica, and if there are any habitat patches that should be prioritized for 

conservation.  To address these points, I developed a habitat model to identify critical 

habitat variables that predict optimal Great Green Macaw nesting habitat.  I then used 

these variables as well as existing GIS base layers in order to create a map of potential 

GGM nesting habitat.  This work occurred in the vicinity of MNWR within the proposed 

SJLS Corridor. 

4.3.2 Existing Great Green Macaw nesting data

I used data documenting the locations of existing Great Green Macaw nest sites 

and information regarding habitat requirements during the breeding season.  These data 

were collected as part of a larger effort by the CCT to detail the Great Green Macaw’s 

104
 
 

ecology (Powell et al. 1999).  Based on telemetry studies, CCT scientists identified that 

breeding pairs utilize a home range of 550 ha on average.  This territory does not overlap 

with the breeding habitat of other pairs.  In addition, CCT researchers identified active 

Great Green Macaw nests as well as formerly active nests that had been cut.  Nests were 

located based on extensive field surveys and interviews with local residents, and the 

nest locations were then recorded using a GPS.  These data are maintained as a GIS 

point file, which I used in my analyses.  CCT scientists identified 66 trees that the Great 

Green Macaw had used from 1994 to 2003 as nesting sites.  Nest sites were not always 

active in consecutive years because nesting pairs often skip a nest site that had been 

active the previous breeding season.  Of the 66 sites identified, 57 (86%) were located in 

large D. panamensis trees (Chassot et al. 2002).  Because of data availability, I utilized the 

locations of 51 nest sites that had been identified from 1994‐2003.  Of these 51 nest sites, 

10 trees had been cut by the end of the survey period.  In some cases, the nests had been 

cut deliberately even after the tree had been protected and demarcated as a nesting tree.  

In these instances, it is likely that the Great Green Macaw chicks were being targeted.  In 

other cases, some nest trees had been removed unwittingly. 

4.3.3 Development of Great Green Macaw nesting habitat model: a


classification tree approach

I used environmental variables that are important components in delineating 

Great Green Macaw nesting habitat.  These environmental variables were contained in 

105
 
 

GIS raster data that covered the project area.  All GIS data were projected to WGS 1984 

UTM Zone 16N, and raster data had the same grid cell size of 30 m.   

I modeled potential Great Green Macaw nesting habitat by using a Classification 

and Regression Tree (CART) approach (Chapter 3).  The CART analysis was performed 

in the R Statistical Program using the “tree” package (R Development Core Team 2008).  

For background on CART models, please see Chapter 3.  I considered the locations of the 

Great Green Macaw nest sites as presences.  I intersected the presence points with the 

raster surfaces for each environmental variable in order to extract the underlying data 

values and create a spreadsheet with the resulting data values.  Because five of the 

original 51 nest points fell outside of the extent of the environmental layers, I removed 

these points from my analysis.  Thus, the 46 nests represented presences.   

In order to create a complementary portrait of the landscape as a whole, I 

generated 50 points and randomly arrayed them spatially across the project area.  The 

random points were at least 1 km away from each other in distance, and they were 

restricted from a 550 ha buffer around each of the nest points, which corresponded to 

the breeding territory of a Great Green Macaw pair.  Ultimately, one of the random 

points fell outside the extent of the environmental layers resulting in 49 points.  These 49 

points reflect average background conditions across the project area.  Similar to the 

presence points, I intersected the background points with the raster surfaces for each 

environmental variable in order to extract the underlying data values and create a 

106
 
 

spreadsheet with the resulting background values.  The presence and background points 

represented the response variable in the CART model.  Because the response variable 

was categorical, the model was run as a classification tree.  I evaluated several 

classification tree models that included various combinations of the environmental 

predictor variables.  Predictor variables that did not produce a logical split in the 

classification tree were removed from consideration.  A description of the response and 

predictor variables appears in Table 8. 

Table 8:  Response and predictor variables used in the CART model to predict 
potential Great Green Macaw habitat 

Name  Description  Variable Type 


Response variable:     
This file represented either point locations of 
known Great Green Macaw nest locations or 
NEST_PRESENCE  Categorical 
random points representing habitat 
“background”. 
Predictor variable:     
Raster surface depicting D. panamensis density 
DENSITY  Continuous 
distribution throughout the project area. 
Mean D. panamensis density within a 550 ha 
buffer.  D. panamensis density was characterized 
DENS_550HA 
in this way because Great Green Macaw breeding  Continuous 
 
pairs utilize 550 ha of non‐overlapping territory 
during the breeding season to forage. 
Mean forest cover within a 100 ha region ranging 
FOREST_COV  Continuous 
from no forest cover (0) to high forest cover (1). 
Land use type:  forest cover, deforested, no 
LANDUSE  Categorical 
forest, or water. 
This raster surface represented the distance to the 
RIVERS_DIST  Continuous 
nearest river. 
This raster surface represented the distance to the 
ROADS_DIST  Continuous 
nearest road. 
ELEV  Elevation of the terrain.   Continuous 

107
 
 

4.4 Results and discussion

4.4.1 Classification tree model

The final classification tree included two environmental predictor variables 

(DENS_550HA and ELEV) and was calibrated with 11 terminal nodes.  This 

classification tree was pruned based on “misclassification” and produced a final 

classification tree containing three terminal nodes.  A confusion matrix was constructed 

to determine the classification success in assigning group membership of the “presence” 

samples.  The classification success for this group was 91%.  Thus, only 4 of the 46 

known nest “presence” points did not occur in areas satisfied by environmental 

conditions predicted by the CART model. 

4.4.2 Environmental conditions predicting Great Green Macaw habitat

Results from the classification tree model indicate that Great Green Macaw pairs 

nest in habitat where adult, reproductive D. panamensis trees occur in densities of 0.2 

trees/ha or greater within a 550 ha buffer around the nesting tree and where the ground 

elevation is less then 132 meters.  D. panamensis density within the breeding territory is 

the most important environmental condition predicting potential Great Green Macaw 

nesting habitat.  If this condition is satisfied, then the overall quality of the surrounding 

forest is insignificant.  Thus, if the mean density of adult, reproductive D. panamensis 

trees meets or exceeds 0.2 trees/ha, Great Green Macaw pairs may nest in habitat with a 

108
 
 

range of forest cover.  Results from the classification tree indicate that there is no 

relationship between the distance of nesting trees to rivers or roads. 

The results of this study underscore the importance of large, reproductive D. 

panamensis trees in delineating potential Great Green Macaw nesting habitat.  Breeding 

habitat occurs in lowland areas (<132 m) where D. panamensis trees grow at densities that 

meet or exceed 0.2 trees/ha within the 550‐ha breeding territory.  D. panamensis trees are 

a critical resource for this bird during all phases of the breeding season—from the time it 

selects a nesting tree in late December to when juveniles fledge in June.  Nearly 90% of 

all Great Green Macaw nests identified in northern Costa Rica are located within 

hollowed cavities of large D. panamensis trees.  Results of the habitat analysis suggest 

that Great Green Macaw pairs select nesting trees that are surrounded by relatively 

dense stands of reproductive D. panamensis trees.  D. panamensis fruit likely sustain the 

adults, chicks, nestlings, and fledglings over the course of the breeding season.  

Fortuitously, the Great Green Macaw breeding season coincides with the peak 

production of D. panamensis fruit.  During the nest‐building period in late December, 

adults are likely foraging for food in the vicinity of the nest tree.  Once the eggs are laid, 

the female remains on the nest while the male forages within the breeding territory; 

when nestling are present, both parents will forage for food (Powell et al. 1999).  Because 

breeding Great Green Macaw pairs will utilize a 550‐ha swathe of forest around the nest 

109
 
 

tree in order to forage for food, the quality and quantity of D. panamensis trees in the 

vicinity are crucial.   

4.4.3 Location of potential Great Green Macaw nesting habitat

Figure 16 shows the distribution of potential Great Green Macaw nesting habitat 

within the lowland Atlantic rain forest of northern Costa Rica.  Approximately 67,271 ha 

of potential Great Green Macaw nesting habitat exists within the northern reaches of the 

study area.  Assuming that Great Green Macaw breeding pairs require 550 ha of non‐

overlapping habitat during the breeding habitat, northern Costa Rica could support 

about 120 breeding macaw pairs.  It is estimated that only 35 breeding pairs currently 

utilize this area (Powell et al. 1999).  The “holes” in the map, where non‐habitat occurs 

within predicted nesting habitat, are regions that generally occur above the elevation 

threshold predicted by the CART model.  Many of these regions do meet the criteria for 

D. panamensis density within the breeding territory.  While these regions would most 

likely be able to support Great Green Macaw nests, these regions were excluded from 

predicted habitat since the CART model was influenced by the locations of the known 

nest trees.  These trees generally occurred in lowland areas. 

The habitat analysis did not identify forest cover as a primary indicator of Great 

Green Macaw habitat.  This study suggests that as long as adequate D. panamensis stands 

exist within the breeding territory, Great Green Macaw may be willing to nest within a 

110
 
 

Figure 16:  Potential Great Green Macaw nesting habitat within Costa Rica 

Yellow nest sites were excluded from the CART model since they fell outside of the 
study area.  Green nest sites did not occur in areas predicted by the CART model to be 
potential Great Green Macaw nesting habitat.  The area circled in yellow was identified 
as a conservation priority as it contains the highest densities of D. panamensis.

111
 
 

mosaic of landscapes such as pastures, forest fragments, or secondary forests.  Of course, 

relatively dense stands of D. panamensis are likely correlated with forest cover, and of the 

67,271 ha of potential Great Green Macaw nesting habitat identified, nearly 76% occurs 

in areas with forest cover greater than 50%.  However, the possibility that Great Green 

Macaw may nest in areas of varying forest cover may represent an opportunity for those 

working to expand Costa Rica’s Great Green Macaw population.  In fact, Bjork and 

Powell (1999) noted that several Great Green Macaw nesting trees were located in cattle 

pastures.  

4.5 Conclusions

The landscape within the lowland Atlantic forests of northeastern Costa Rica 

represents a patchwork of primary forest surrounded by a network of pastures, 

agricultural zones, and regenerating forests.  This network of multiple land uses is a 

reminder of a time when Costa Rica had one of the highest deforestation rates, 

averaging around 3.7% from the early 1970s to the early 1990s (Sanchez‐Azofeifa et al. 

2003).  However, many ecologists suggest that tropical forest restoration may undo some 

of the damage done by prior anthropogenic degradation (Lamb et al. 2005, Chazdon 

2008).  Targeted restoration of D. panamensis trees within abandoned pasture may 

jumpstart the recovery of land that is currently not viable for Great Green Macaw 

nesting or foraging.  Such actions may ensure the long‐term presence and viability of 

this species within Costa Rica’s forests.  
112
 
 

However, while targeted habitat restoration and forest regeneration may 

represent one solution to maintaining Costa Rica’s Great Green Macaw population, 

habitat protection must be the first step, and existing nesting trees should be 

safeguarded.  In fact, a recent decision by Costa Rica’s 4th Chamber of the Supreme 

Court on September 16, 2008 states that MINAE must abstain from the continuation or 

initiation of the use, exploitation, or extraction of D. panamensis (Fallas 2007) (Appendix 

C).  This is a very positive development. 

Understanding what environmental conditions predict potential Great Green 

Macaw habitat is also essential in developing or fine‐tuning any conservation program 

for this species.  MNWR was established in 2005, and one of its primary purposes is to 

protect breeding habitat for the Great Green Macaw in Costa Rica.  Approximately 

43,700 ha of land identified as potential Great Green Macaw breeding habitat lies within 

the boundaries of MNWR. This covers nearly 75% of the new wildlife refuge.  However, 

an important region of potential breeding habitat that is currently unprotected is located 

in the northwestern part of the study area (Figure 16).  My selection of this patch as a 

conservation priority is based on the following criteria.  Not only does this area meet the 

minimum requirements of D. panamensis density within a 550‐ha breeding territory, but 

this zone possesses some of the highest measured densities of adult D. panamensis trees 

on record of nearly 2 adult trees/ha (Chapter 3).  Thus, I am basing my recommendation 

on the extent of D. panamensis density in this region.  In addition, I do not believe that 

113
 
 

surveys have been conducted in this region for Great Green Macaw nests.  This region 

would benefit from extensive ground studies of this bird species.  Future efforts to 

preserve Great Green Macaw breeding habitat need to prioritize this habitat patch.  At 

the very least, the logging moratorium against removing large D. panamensis trees 

should be strictly enforced in this area. 

Despite emerging information about the Great Green Macaw’s breeding habitat 

and current conservation efforts to preserve habitat, Costa Rica’s population of Great 

Green Macaw may never reach the potential of 120 breeding pairs predicted here.  

Habitat fragmentation and other anthropogenic threats may simply be too widespread.  

Although this research indicates that there is more than enough foraging habitat during 

the breeding season because D. panamensis densities are high, the number of suitable 

nesting trees may be limited.  Even with adequate food during the breeding season as 

represented by high densities of reproductive D. panamensis trees, it is the scarcity of 

natural nest sites that may be limiting the number of adult Great Green Macaws that 

reproduce annually (Munn 1992).   

While Costa Rica’s Great Green Macaw may remain on the brink, a note of hope 

lies across the San Juan River and over the border in Nicaragua.  The Indio‐Maíz 

Biological Reserve is located just across Costa Rica’s northeastern border (Figure 17).  It 

comprises nearly 264,000 ha of lowland Atlantic forest, the same type of forest where 

Great Green Macaw pairs breed in Costa Rica.  Because of Nicaragua’s recent political 

114
 
 

Figure 17:  Location of Indio‐Maíz Biological Reserve, Nicaragua in relation to 
potential Great Green Macaw nesting habitat in Costa Rica 

115
 
 

history, much of the Indio‐Maíz Biological Reserve remains intact and unlogged.  While 

extensive field surveys of Great Green Macaw in Nicaragua remain scarce, it is 

estimated that the Nicaraguan population is nearly ten times the size of the population 

south of the border and that the Costa Rican population is merely a subpopulation of the 

much larger Nicaraguan population (Powell et al. 1999).  Most likely, the Indio‐Maíz 

Biological Reserve contains extensive forest swathes with extremely high densities of D. 

panamensis trees only seen in a few patches in northern Costa Rica.  Likely, many areas 

exceed the minimum Great Green Macaw nesting requirement of 0.2 trees/ha within the 

breeding territory.  Thus, to maintain the Great Green Macaw within Costa Rica, 

ongoing efforts to protect the habitat in Nicaragua need to be funded.  In addition, the 

level of research that has been conducted in Costa Rica should be extended north.  For 

example, known nesting trees should be identified and protected and the breeding 

population should be more accurately censused.  If appropriate GIS and remote sensing 

data exist, a similar study could be initiated to identify potential Great Green Macaw 

nesting habitat in the Indio‐Maíz Biological Reserve based on the parameters from the 

present work. 

116
 
 

Chapter 5. General conclusions

This dissertation provides new insights into the utility of digital aerial surveys in 

censusing an ecologically and economically important tropical tree species across a large 

spatial extent.  I demonstrate that digital aerial surveys can be timed to capture the 

unique blooming pattern for the canopy‐emergent tree, Dipteryx panamensis.  In 

particular, I developed a semi‐automated, pattern recognition technique that capitalized 

on the distinct spectral signature of blooming D. panamensis trees.  Specifically, the 

alternate color space, IHS (intensity, hue, and saturation), enhanced the images and 

enabled D. panamensis to be identifiable by the computer classifier.  The aerial survey 

occurred in Costa Rica over a portion of two new protected areas—the proposed San 

Juan‐La Selva Biological Corridor and Maquenque National Wildlife Refuge.  It covered 

over 140,000 ha.  Results of this work suggest that digital aerial surveys may provide a 

cost effective and efficient approach to studying tropical landscapes that may be 

dangerous or challenging to enter on the ground.  This approach is especially effective in 

studying the distribution of large, canopy‐emergent tree species, like D. panamensis, 

which are often difficult to study from the ground.  In addition, digital aerial surveys 

may augment the size of traditional forest inventory plots.  Due to biases in placement, 

terrestrial plots often misrepresent the abundance or density of large, infrequently 

distributed trees. 

117
 
 

Besides testing a novel remote sensing technique, this dissertation also examined 

the landscape distribution and conservation implications of D. panamensis.  I created a 

raster surface depicting the density of D. panamensis across the entire aerial survey extent 

using the output from the digital photographs.  This work provided basic, ecological 

information regarding D. panamensis, and indicated that reproductive, canopy‐emergent 

trees can exist in densities as high as 2 adult trees/ha.  An analysis of the environmental 

conditions that can support D. panamensis showed that this tree species is restricted to 

regions between 45 and 125 meters in elevation on soil with an acid and clay profile. 

Finally, the distributional work on D. panamensis provided the basis for an 

analysis on the environmental factors that predict nesting habitat for the endangered 

Great Green Macaw (Ara ambigua).  Because D. panamensis is a critical food and nesting 

resource for the Great Green Macaw, this dissertation highlighted important 

conservation implications and priorities for both D. panamensis and the Great Green 

Macaw.  Results from this dissertation underscored that Great Green Macaw breeding 

pairs nest in habitat where adult, reproductive D. panamensis trees occur in densities of 

0.2 trees/ha or greater within a 550 ha buffer around the nesting tree. 

Ultimately, the future of the Great Green Macaw may rely on the conservation of 

the Indio‐Maíz Biological Reserve in Nicaragua.  This protected area is one of the largest 

remaining stands of primary rain forest in Central America.  It is likely that D. 

panamensis occurs in high densities in Nicaragua similar to or greater than those 

118
 
 

identified in Costa Rica.  In addition, Costa Rica’s population of Great Green Macaw, 

which numbers only around 250 individuals, is likely a mere subpopulation of the larger 

population that exists over the border in Nicaragua.  Subsequent research and 

conservation efforts of both D. panamensis and the Great Green Macaw in Costa Rica 

should include a bi‐national partnership with Nicaragua.  Such a visionary collaboration 

may ensure the survival of the Great Green Macaw for many more generations! 

119
 
 

Appendix A: Newspaper article published in La Nación

describing “suspicious” overflights

120
 
 

I had the dubious distinction of having my aerial survey work written up in the July 7, 
2005 issue of Costa Rica’s national newspaper, La Nación (Hernández 2005).  While the 
aerial surveys were conducted for research purposes, the last sentence of this article 
suggests an alternative purpose of the overflights! 

121
 
 

Appendix B: Results of the signature separability matrix

122
 
 

Results of the Signature Separability matrix* 

Training Sample
Black_1 Black_2 Black_3 Vegetation_1 Vegetation_2 Vegetation_3 Vegetation_4
Names

Black_1 0
Black_2 977 0
Black_3 1443 290 0
Vegetation_1 2000 2000 2000 0
Vegetation_2 2000 2000 2000 1073 0
Vegetation_3 2000 2000 2000 2000 2000 0
Vegetation_4 2000 2000 2000 2000 2000 1960 0
Vegetation_5 2000 2000 2000 2000 2000 719 1644
Vegetation_6 2000 2000 2000 2000 2000 593 1989
Vegetation_7 2000 2000 2000 2000 2000 1785 425
Vegetation_8 2000 2000 2000 2000 2000 2000 2000
Vegetation_9 2000 2000 2000 2000 2000 1752 1547
Dipteryx
panamensis_1 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_2 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_3 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_4 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_5 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_6 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_7 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_8 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_9 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_10 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_11 2000 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_12 2000 2000 2000 2000 2000 2000 2000
Water_1 2000 2000 2000 2000 1999 2000 2000
Water_2 2000 2000 2000 2000 2000 2000 2000
Bare_Ground_1 2000 2000 2000 2000 2000 2000 2000
Bare_Ground_2 2000 2000 2000 2000 2000 2000 2000
Road_1 2000 2000 2000 2000 2000 2000 2000
Road_2 2000 2000 2000 2000 2000 2000 2000
Pasture_1 2000 2000 2000 2000 2000 1130 1993
Pasture_2 2000 2000 2000 2000 2000 2000 2000
Pasture_3 2000 2000 2000 2000 2000 1921 2000
Bare_Ground_3 2000 2000 2000 2000 2000 2000 2000
Road_3 2000 2000 2000 2000 2000 2000 2000
*Numbers after the name indicate that multiple training samples were collected to 
describe a training class.  Training samples are distinct if values are greater than 1,500. 

123
 
 

 Results of the Signature Separability matrix, cont. 

Names of
Dipteryx
Training Vegetation_5 Vegetation_6 Vegetation_7 Vegetation_8 Vegetation_9
panamensis_1
Classes
Black_1
Black_2
Black_3
Vegetation_1
Vegetation_2
Vegetation_3
Vegetation_4
Vegetation_5 0
Vegetation_6 1408 0
Vegetation_7 1005 1938
Vegetation_8 2000 2000 2000 0
Vegetation_9 1035 1956 943 2000 0
Dipteryx
panamensis_1 2000 2000 2000 2000 2000 0
Dipteryx
panamensis_2 2000 2000 2000 2000 2000 664
Dipteryx
panamensis_3 2000 2000 2000 2000 2000 1870
Dipteryx
panamensis_4 2000 2000 2000 2000 2000 1835
Dipteryx
panamensis_5 2000 2000 2000 2000 2000 640
Dipteryx
panamensis_6 2000 2000 2000 2000 2000 1158
Dipteryx
panamensis_7 2000 2000 2000 2000 2000 1851
Dipteryx
panamensis_8 2000 2000 2000 2000 2000 1832
Dipteryx
panamensis_9 2000 2000 2000 2000 2000 1883
Dipteryx
panamensis_10 2000 2000 2000 2000 2000 1998
Dipteryx
panamensis_11 2000 2000 2000 2000 2000 2000
Dipteryx
panamensis_12 2000 2000 2000 2000 2000 1982
Water_1 2000 2000 2000 2000 2000 2000
Water_2 2000 2000 2000 2000 2000 2000
Bare_Ground_1 2000 2000 2000 2000 2000 2000
Bare_Ground_2 2000 2000 2000 2000 2000 2000
Road_1 2000 2000 2000 2000 2000 2000
Road_2 2000 2000 2000 2000 2000 2000
Pasture_1 1727 1694 1945 2000 1981 2000
Pasture_2 2000 2000 2000 2000 2000 2000
Pasture_3 1999 1761 2000 2000 2000 2000
Bare_Ground_3 2000 2000 2000 2000 2000 2000
Road_3 2000 2000 2000 2000 2000 2000

124
 
 

Results of the Signature Separability matrix, cont. 

Dipteryx Dipteryx Dipteryx Dipteryx Dipteryx Dipteryx


Names of Training
panamensis panamensis panamensis panamensis panamensis panamensis
Classes
2 3 4 5 6 7
Black_1
Black_2
Black_3
Vegetation_1
Vegetation_2
Vegetation_3
Vegetation_4
Vegetation_5
Vegetation_6
Vegetation_7
Vegetation_8
Vegetation_9
Dipteryx
panamensis_1
Dipteryx
panamensis_2 0
Dipteryx
panamensis_3 1672 0
Dipteryx
panamensis_4 1358 943 0
Dipteryx
panamensis_5 1286 1817 1854 0
Dipteryx
panamensis_6 1671 1652 1813 232 0
Dipteryx
panamensis_7 1568 597 960 1875 1837 0
Dipteryx
panamensis_8 1516 1995 1907 1977 1995 1993
Dipteryx
panamensis_9 1147 1801 869 1977 1990 1505
Dipteryx
panamensis_10 1993 1997 1774 2000 2000 1998
Dipteryx
panamensis_11 2000 2000 2000 2000 1999 2000
Dipteryx
panamensis_12 1676 1997 1955 1998 2000 1975
Water_1 2000 2000 2000 2000 2000 2000
Water_2 2000 2000 2000 2000 2000 2000
Bare_Ground_1 2000 2000 2000 2000 2000 2000
Bare_Ground_2 2000 2000 2000 2000 2000 2000
Road_1 2000 2000 2000 2000 2000 2000
Road_2 2000 2000 2000 2000 2000 2000
Pasture_1 2000 2000 2000 2000 2000 2000
Pasture_2 2000 2000 2000 1997 1982 2000
Pasture_3 2000 2000 2000 2000 2000 2000
Bare_Ground_3 2000 1999 2000 1995 1966 2000
Road_3 2000 2000 2000 2000 2000 2000

125
 
 

Results of the Signature Separability matrix, cont. 

Names of Dipteryx Dipteryx Dipteryx Dipteryx Dipteryx


Training panamensis panamensis panamensis panamensis panamensis Water_1 Water_2
Classes 8 9 10 11 12
Black_1
Black_2
Black_3
Vegetation_1
Vegetation_2
Vegetation_3
Vegetation_4
Vegetation_5
Vegetation_6
Vegetation_7
Vegetation_8
Vegetation_9
Dipteryx
panamensis_1
Dipteryx
panamensis_2
Dipteryx
panamensis_3
Dipteryx
panamensis_4
Dipteryx
panamensis_5
Dipteryx
panamensis_6
Dipteryx
panamensis_7
Dipteryx
panamensis_8 0
Dipteryx
panamensis_9 1700 0
Dipteryx
panamensis_10 1882 1903 0
Dipteryx
panamensis_11 2000 2000 2000 0
Dipteryx
panamensis_12 1724 1457 2000 2000 0
Water_1 2000 2000 2000 2000 2000 0
Water_2 2000 2000 2000 2000 2000 2000 0
Bare_Ground_1 2000 2000 2000 2000 2000 2000 1999
Bare_Ground_2 2000 2000 2000 2000 2000 2000 2000
Road_1 2000 2000 2000 2000 2000 2000 2000
Road_2 2000 2000 2000 2000 2000 2000 2000
Pasture_1 2000 2000 2000 2000 2000 2000 2000
Pasture_2 2000 2000 2000 2000 2000 2000 2000
Pasture_3 2000 2000 2000 2000 2000 2000 2000
Bare_Ground_3 2000 2000 2000 2000 2000 2000 2000
Road_3 2000 2000 2000 2000 2000 2000 2000

126
 
 

Results of the Signature Separability matrix, cont. 

Names of Bare_ Bare_ Bare_


Pasture_ Pasture_ Pasture_
Training Ground Ground Road_1 Road_2 Ground Road_3
1 2 3
Classes 1 2 3
Black_1
Black_2
Black_3
Vegetation_1
Vegetation_2
Vegetation_3
Vegetation_4
Vegetation_5
Vegetation_6
Vegetation_7
Vegetation_8
Vegetation_9
Dipteryx
panamensis_1
Dipteryx
panamensis_2
Dipteryx
panamensis_3
Dipteryx
panamensis_4
Dipteryx
panamensis_5
Dipteryx
panamensis_6
Dipteryx
panamensis_7
Dipteryx
panamensis_8
Dipteryx
panamensis_9
Dipteryx
panamensis_10
Dipteryx
panamensis_11
Dipteryx
panamensis_12
Water_1
Water_2
Bare_Ground_1 0
Bare_Ground_2 2000 0
Road_1 2000 2000 0
Road_2 2000 1994 2000 0
Pasture_1 2000 2000 2000 2000 0
Pasture_2 2000 2000 1867 2000 2000 0
Pasture_3 2000 2000 2000 2000 1991 2000 0
Bare_Ground_3 2000 1999 1992 2000 2000 940 2000 0
Road_3 2000 1996 2000 1273 2000 2000 2000 2000 0

127
 
 

Appendix C: Recent ruling that bans the use,

exploitation, or extraction of D. panamensis

128
 
 

This article describes the recent ruling on D. panamensis and was published in Costa 
Rica’s national newspaper, La Nación (Ávalos R. 2008). 

129
 
 

References 

Abramson, J. 1999. Anatomy and physiology. Pages 39‐71 in J. Abramson, B. L. Speer, 
and J. B. Thomsen, eds. The Large Macaws:  Their Care, Breeding and Conservation. 
Raintree Publications, Fort Bragg, CA. 

Achard, F., H. D. Eva, H. J. Stibig, P. Mayaux, J. Gallego, T. Richards, and J. P. 
Malingreau. 2002. Determination of deforestation rates of the worldʹs humid 
tropical forests. Science 297: 999‐1002. 

Asner, G. P., M. Palace, M. Keller, R. Pereira, J. N. M. Silva, and J. C. Zweede. 2002. 
Estimating canopy structure in an Amazon forest from laser range finder and 
IKONOS satellite observations. Biotropica 34: 483‐492. 

Ávalos R., A. 2008. Sala IV prohíbe explotación del almendro amarillo. La Nación, San 
José, Costa Rica. 

Berg, K. S., J. Socola, and R. R. Angel. 2007. Great Green Macaws and the annual cycle of 
their food plants in Ecuador. Journal of Field Ornithology 78: 1‐10. 

BirdLife International. 2005. Ara ambiguus in IUCN, ed. 2007 IUCN Red List of Threatened 
Species. Accessed on 14 August 2008, http://www.iucnredlist.org. 

Bjork, R., and G. V. N. Powell. 1999. Buffonʹs macaw: some observations on the Costa 
Rican population, its lowland forest habitat and conservation. Pages 387‐393 in J. 
Abramson, B. L. Speer, and J. B. Thomsen, eds. The Large Macaws:  Their Care, 
Breeding and Conservation. Raintree Publications, Fort Bragg, CA. 

Black, G. A., T. Dobzhansky, and C. Pavan. 1950. Some attempts to estimate species 
diversity and population density of trees in Amazonian forests. Botanical Gazette 
111: 413‐425. 

Bonaccorso, F. J., W. E. Glanz, and C. M. Sandford. 1980. Feeding assemblages of 
mammals at fruiting Dipteryx panamensis (Papilionaceae) trees in Panama: seed 
predation, dispersal, and parasitism. Revista De Biologia Tropical 28: 61‐72. 
130
 
 

Boza, M. A. 1993. Conservation in action: past, present, and future of the national park 
system of Costa Rica. Conservation Biology 7: 239‐247. 

Brady, N. C., and R. R. Weil. 2002. The Nature and Properties of Soils. Prentice Hall, New 
Jersey. 

Breiman, L., J. Friedman, C. J. Stone, and R. A. Olshen. 1984. Classification and Regression 
Trees. CRC Press, Boca Raton, Florida. 

Brightsmith, D. J. 2005. Parrot nesting in southeastern Peru: seasonal patterns and 
keystone trees. Wilson Bulletin 117: 296‐305. 

Butterfield, R. P. 1994. Forestry in Costa Rica:  status, research priorities, and the role of 
La Selva Biological Station. Pages 317‐328 in L. A. McDade, K. S. Bawa, H. A. 
Hespenheide, and G. S. Hartshorn, eds. La Selva : ecology and natural history of a 
neotropical rain forest. University of Chicago Press, Chicago. 

Campbell, L. M. 2002. Conservation narratives in Costa Rica: conflict and co‐existence. 
Development and Change 33: 29‐56. 

Center for Tropical Forest Science (CTFS). 2008. Accessed August 22, 2008, 
<http://www.ctfs.si.edu/doc/plots/Site%20Summary%20Statistics.html>. 

Chassot, O., G. Monge, U. Aleman, G. V. N. Powell, and S. Palminteri. 2002. Lapa verde, 
estable pero inviable. Ambientales 24: 18‐23. 

Chassot, O., G. Monge, G. V. N. Powell, P. Wright, and S. Palminteri. 2004. Final San 
Juan‐La Selva, Costa Rica Biological Corridor. Tropical Science Center, San José, 
Costa Rica. 

Chazdon, R. L. 2003. Tropical forest recovery: legacies of human impact and natural 
disturbances. Perspectives in Plant Ecology Evolution and Systematics 6: 51‐71. 

131
 
 

Chazdon, R. L. 2008. Beyond deforestation: restoring forests and ecosystem services on 
degraded lands. Science 320: 1458‐1460. 

CITES. 2008. Convention on International Trade in Endangered Species of Wild Fauna 
and Flora, Appendices I, II, and III. Pages 1‐47. 

Clark, D. A., and D. B. Clark. 1999. Assessing the growth of tropical rain forest trees: 
Issues for forest modeling and management. Ecological Applications 9: 981‐997. 

Clark, D. B., and D. A. Clark. 1987. Population ecology and microhabitat distribution of 
Dipteryx panamensis, a Neotropical rain forest emergent tree. Biotropica 19: 236‐
244. 

Clark, D. B., C. S. Castro, L. D. A. Alvarado, and J. M. Read. 2004a. Quantifying 
mortality of tropical rain forest trees using high‐spatial‐resolution satellite data. 
Ecology Letters 7: 52‐59. 

Clark, D. B., J. M. Read, M. L. Clark, A. M. Cruz, M. F. Dotti, and D. A. Clark. 2004b. 
Application of 1‐M and 4‐M resolution satellite data to ecological studies of 
tropical rain forests. Ecological Applications 14: 61‐74. 

Clark, M. L., D. A. Roberts, and D. B. Clark. 2005. Hyperspectral discrimination of 
tropical rain forest tree species at leaf to crown scales. Remote Sensing of 
Environment 96: 375‐398. 

Condit, R. 1995. Research in large, long‐term tropical forest plots. Trends in Ecology & 
Evolution 10: 18‐22. 

Congalton, R. G. 1991. A review of assessing the accuracy of classifications of remotely 
sensed data. Remote Sensing of Environment 37: 35‐46. 

David, N., and M. Gosselin. 2002. The grammatical gender of avian genera. Bulletin of the 
British Ornithologistsʹ Club 122: 257‐282. 

132
 
 

Deʹath, G., and K. E. Fabricius. 2000. Classification and regression trees:  a powerful yet 
simple technique for ecological data analysis. Ecology 81: 3178‐3192. 

De Steven, D. 1988. Light gaps and long‐term seedling performance of a Neotropical 
canopy tree (Dipteryx panamensis, Leguminosae). Journal of Tropical Ecology 4: 407‐
411. 

ESRI. 2005. ArcGIS 9.1 Desktop Help. Accessed 10 September 2008, 
<http://webhelp.esri.com/arcgisdesktop/9.1/index.cfm?TopicName=welcome>. 

Fallas, H. 2007. País cierra el año con llegada de 1,9 milliones de turistas. La Nación, San 
José, Costa Rica. 

Fichtler, E., D. A. Clark, and M. Worbes. 2003. Age and long‐term growth of trees in an 
old‐growth tropical rain forest, based on analyses of tree rings and 14C. Biotropica 
35: 306‐317. 

Flores, E. M. 1992. Dipteryx panamensis. Pages 1‐22. Árboles y semillas del Neotrópico/Trees 
and seeds from the Neotropics. Museo Nacional de Costa Rica, Herbario Nacional 
de Costa Rica, San José, Costa Rica. 

Foody, G. M., P. M. Atkinson, P. W. Gething, N. A. Ravenhill, and C. K. Kelly. 2005. 
Identification of specific tree species in ancient semi‐natural woodland from 
digital aerial sensor imagery. Ecological Applications 15: 1233‐1244. 

Frankie, G. W., H. G. Baker, and P. A. Opler. 1974. Comparative phenological studies of 
trees in tropical wet and dry forests in the lowlands of Costa Rica. Journal of 
Ecology 62: 881‐919. 

Gentry, A. H. 1988. Tree species richness of upper Amazonian forests. Proceedings of the 
National Academy of Sciences of the United States of America 85: 156‐159. 

Gentry, A. H. 1992. Tropical forest biodiversity:  distributional patterns and their 
conservational significance. Oikos 63: 19‐28. 

133
 
 

Gillespie, A. R., A. B. Kahle, and R. E. Walker. 1986. Color enhancement of highly 
correlated images.  1.  Decorrelation and HSI contrast stretches. Remote Sensing of 
Environment 20: 209‐235. 

Guedes, N. M. R. 2004. Management and conservation of the large macaws in the wild. 
Ornitologia Neotropical 15: 279‐283. 

Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat distribution models in 
ecology. Ecological Modelling 135: 147‐186. 

Hanson, T., S. Brunsfeld, and B. Finegan. 2006. Variation in seedling density and seed 
predation indicators for the emergent tree Dipteryx panamensis in continuous and 
fragmented rain forest. Biotropica 38: 770‐774. 

Hanson, T. R., S. J. Brunsfeld, B. Finegan, and L. P. Waits. 2008. Pollen dispersal and 
genetic structure of the tropical tree Dipteryx panamensis in a fragmented Costa 
Rican landscape. Molecular Ecology 17: 2060‐2073. 

Hartshorn, G. S., and B. E. Hammel. 1994. Vegetation types and floristic patterns. Pages 
73‐89 in L. A. McDade, K. S. Bawa, H. A. Hespenheide, and G. S. Hartshorn, eds. 
La Selva:  Ecology and Natural History of a Neotropical Rain Forest. The University of 
Chicago Press, Chicago. 

Hernández, C. 2005. Vecinos reportan otra vuelo sospechoso en la zona norte. La Nación, 
San José, Costa Rica. 

Holdridge, L. R. 1947. Determination of world plant formations from simple climatic 
data. Science 105: 367‐368. 

Holdridge, L. R., W. C. Grenke, W. H. Hatheway, T. Liang, and J. A. Tosi, Jr. 1971. Forest 
Environments in Tropical Life Zones:  A Pilot Study. Pergamon, New York. 

Instituto Costarricense de Turismo (ICT). 2006. Informe de encuestas de turismo 2006:  
turistas no residentes que salieron por Aeropuerto Juan Santamaría, IV trimestre 
2006. Accessed 29 August 2008, 
134
 
 

<http://www.visitcostarica.com/ict/backoffice/treeDoc/files/IV_Trimestre_2006%2
0‐%20No%20Residentes%20‐%20AIJS.pdf>. 

Instituto Nacional de Estadística y Censos (INEC). 2001. IX Censo nacional de población 
y de vivienda. INEC, San José, Costa Rica. 

International Ecotourism Society. 2008. The International Ecotourism Society:  
definitions and principles. Accessed 29 August 2008, 
<http://www.ecotourism.org/>. 

James, T., S. Vege, P. Aldrich, and J. L. Hamrick. 1998. Mating systems of three tropical 
dry forest tree species. Biotropica 30: 587‐594. 

Jansen, P. A., S. A. Bohlman, C. X. Garzon‐Lopez, H. Olff, H. C. Muller‐Landau, and S. J. 
Wright. 2008. Large‐scale spatial variation in palm fruit abundance across a 
tropical moist forest estimated from high‐resolution aerial photographs. 
Ecography 31: 33‐42. 

Jenson, J. R. 2005. Introductory Digital Image Processing: A Remote Sensing Perspective. 
Prentice Hall, Upper Saddle River, New Jersey. 

Joyce, A. T. 2006. Land Use Change in Costa Rica as Influenced by Social, Economic, Political, 
and Environmental Factors:  1966‐2006. Litografía e Imprenta LIL, S.A., San José, 
Costa Rica. 

Kaiser, J. 2001. Bold corridor project confronts political reality. Science 293: 2196‐2199. 

Laliberte, A. S., A. Rango, and E. L. Fredrickson. 2006. Separating green and senescent 
vegetation in very high resolution photography using an intensity‐hue‐
saturation transformation and object based classification. American Society for 
Photogrammetry and Remote Sensing, Reno, Nevada. 

Lamb, D., P. D. Erskine, and J. A. Parrotta. 2005. Restoration of degraded tropical forest 
landscapes. Science 310: 1628‐1632. 

135
 
 

Lambin, E. F., H. J. Geist, and E. Lepers. 2003. Dynamics of land‐use and land‐cover 
change in tropical regions. Annual Review of Environment and Resources 28: 205‐
241. 

Leckie, D. G., F. A. Gougeon, N. Walsworth, and D. Paradine. 2003. Stand delineation 
and composition estimation using semi‐automated individual tree crown 
analysis. Remote Sensing of Environment 85: 355‐369. 

Leckie, D. G., F. A. Gougeon, S. Tims, T. Nelson, C. N. Burnett, and D. Paradine. 2005. 
Automated tree recognition in old growth conifer stands with high resolution 
digital imagery. Remote Sensing of Environment 94: 311‐326. 

LeicaGeosystems. 2003. ERDAS Field Guide, Atlanta, Georgia. 

Lillesand, T. M., R. W. Kiefer, and J. W. Chipman. 2004. Remote Sensing and Image 
Interpretation. John Wiley & Sons, New York. 

Lomolino, M. V., B. R. Riddle, and J. H. Brown. 2005. Biogeography. Sinauer Associates, 
Sunderland, MA. 

Lunetta, R. S., D. M. Johnson, J. G. Lyon, and J. Crotwell. 2004. Impacts of imagery 
temporal frequency on land‐cover change detection monitoring. Remote Sensing of 
Environment 89: 444‐454. 

Mawson, P. R., and J. L. Long. 1994. Size and age parameters of nest trees used by four 
species of parrot and one species of cockatoo in southwest Australia. Emu 94: 
149‐155. 

McGarigal, K. S., S. A. Cushman, M. C. Neel, and E. Ene. 2002. FRAGSTATS: Spatial 
Pattern Analysis Program for Categorical Maps. Computer software program 
produced by the authors at the University of Massachusetts, Amherst. 
http://www.umass.edu/landeco/research/fragstats/fragstats.html. 

McGraw, J. B., T. A. Warner, T. L. Key, and W. R. Lamar. 1998. High spatial resolution 
remote sensing of forest trees. Trends in Ecology & Evolution 13: 300‐301. 
136
 
 

Meinel, G., and M. Neubert. 2004. A comparison of segmentation programs for high 
resolution remote sensing data. Pages 1097‐1102. XX ISPRS Congress, Commission 
IV, Istanbul, Turkey. 

Mills, L. S., M. E. Soulé, and D. F. Doak. 1993. The keystone‐species concept in ecology 
and conservation. Bioscience 43: 219‐224. 

Monge, G., O. Chassot, G. V. N. Powell, S. Palminteri, U. Aleman, and P. Wright. 2003. 
Ecología de la lapa verde (Ara ambigua) en Costa Rica. Zeledonia 7: 4‐12. 

Moore, D. M., B. G. Lees, and S. M. Davey. 1991. A new method for predicting 
vegetation distributions using decision tree analysis in a geographic information 
system. Environmental Management 15: 59‐71. 

Munn, C. A. 1992. Macaw biology and ecotourism, or when a bird in the bush is worth 
two in the hand. Pages 47‐72 in S. R. Beissinger and N. Snyder, eds. New World 
Parrots in Crisis:  Solutions fro Conservation Biology. Smithsonian Institution Press, 
Washington, D.C. 

Myers, B. J. 1982. Large‐scale color aerial photographs—a useful tool for tropical 
biologists. Biotropica 14: 156‐156. 

Myers, N. 1981. The hamburger connection: how Central Americas Forests Become 
North Americaʹs Hamburgers. Ambio 10: 3‐8. 

Myers, N., R. A. Mittermeier, C. G. Mittermeier, G. A. B. da Fonseca, and J. Kent. 2000. 
Biodiversity hotspots for conservation priorities. Nature 403: 853‐858. 

Nadkarni, N. M. 1994. Diversity of species and interactions in the upper tree canopy of 
forest ecosystems. American Zoologist 34: 70‐78. 

Nason, J. D., E. A. Herre, and J. L. Hamrick. 1998. The breeding structure of a tropical 
keystone plant resource. Nature 391: 685‐687. 

137
 
 

Noss, R. F. 1991. Landscape connectivity:  different functions at different scales. Pages 
27‐39 in W. E. Hudson, ed. Landscape Linkages and Biodiversity. Island Press, 
Washington, D.C. 

Nycander, E., D. H. Blanco, K. M. Holle, A. del Campo, C. A. Munn, J. L. Moscoso, and 
R. D. G. 1999. Manu and Tambopata: nesting success and techniques for 
increasing reproduction in wild macaws in Southeastern Peru. Pages 423‐443 in J. 
Abramson, B. L. Speer, and J. B. Thomsen, eds. The Large Macaws:  Their Care, 
Breeding and Conservation. Raintree Publications, Fort Bragg, CA. 

Nygren, A. 2004. Contested lands and incompatible images: the political ecology of 
struggles over resources in Nicaraguaʹs Indio‐Maíz reserve. Society & Natural 
Resources 17: 189‐205. 

OʹBrien, J. J., S. F. Oberbauer, D. B. Clark, and D. A. Clark. 2008. Phenology and stem 
diameter increment seasonality in a Costa Rican wet tropical forest. Biotropica 40: 
151‐159. 

Palace, M., M. Keller, G. P. Asner, S. Hagen, and B. Braswell. 2008. Amazon forest 
structure from IKONOS satellite data and the automated characterization of 
forest canopy properties. Biotropica 40: 141‐150. 

Parker, G. G., A. P. Smith, and K. P. Hogan. 1992. Access to the upper forest canopy with 
a large tower crane. Bioscience 42: 664‐670. 

Perry, D. R., and A. Starrett. 1980. The pollination ecology and blooming strategy of a 
Neotropical emergent tree, Dipteryx panamensis. Biotropica 12: 307‐313. 

Pfaff, A. S. P., and G. A. Sanchez‐Azofeifa. 2004. Deforestation pressure and biological 
reserve planning: a conceptual approach and an illustrative application for Costa 
Rica. Resource and Energy Economics 26: 237‐254. 

Pianka, E. R. 1966. Latitudinal gradients in species diversity: a review of concepts. 
American Naturalist 100: 33‐46. 

138
 
 

Pitman, N. C. A., J. Terborgh, M. R. Silman, and P. Nuñez. 1999. Tree species 
distributions in an upper Amazonian forest. Ecology 80: 2651‐2661. 

Powell, G. V. N., P. Wright, U. Aleman, C. Guindon, S. Palminteri, and R. Bjork. 1999. 
Research Findings and Conservation Recommendations for the Great Green 
Macaw (Ara ambigua) in Costa Rica. Centro Científico Tropical, San José, Costa 
Rica. 

Powell, G. V. N., J. Barborak, and M. Rodriguez. 2000. Assessing representativeness of 
protected natural areas in Costa Rica for conserving biodiversity: a preliminary 
gap analysis. Biological Conservation 93: 35‐41. 

R Development Core Team. 2008. R:  A Language and Environment for Statistical 
Computing, Vienna, Austria. 

Reid, W. V., and K. Miller. 1989. Keeping Options Alive:  The Scientific Basis for Conserving 
Biodiversity. World Resources Institute, Washington, D.C. 

Rosero‐Bixby, L., and A. Palloni. 1998. Population and deforestation in Costa Rica. 
Population and Environment 20: 149‐185. 

Sader, S. A., and A. T. Joyce. 1988. Deforestation Rates and Trends in Costa‐Rica, 1940 to 
1983. Biotropica 20: 11‐19. 

Sanchez‐Azofeifa, G., R. C. Harriss, and D. L. Skole. 2001. Deforestation in Costa Rica: a 
quantitative analysis using remote sensing imagery. Biotropica 33: 378‐384. 

Sanchez‐Azofeifa, G. A., G. C. Daily, A. S. P. Pfaff, and C. Busch. 2003. Integrity and 
isolation of Costa Rica’s national parks and biological reserves: examining the 
dynamics of land‐cover change. Biological Conservation 109: 123‐135. 

Sheil, D. 1995. A critique of permanent plot methods and analysis with examples from 
Budongo Forest, Uganda. Forest Ecology and Management 77: 11‐34. 

139
 
 

Skole, D., and C. Tucker. 1993. Tropical deforestation and habitat fragmentation in the 
Amazon:  satellite data from 1978 to 1988. Science 260: 1905‐1910. 

Snider, A. G., S. K. Pattanayak, E. O. Sills, and J. L. Schuler. 2003. Policy innovations for 
private forest management and conservation in Costa Rica. Journal of Forestry 101: 
18‐23. 

Snyder, N., P. McGowan, J. Gilardi, and A. Grajal, eds. 2000. Parrots. Status Survey and 
Conservation Action Plan 2000‐2004. IUCN, Gland, Switzerland and Cambridge, 
UK. 

Steininger, M. K. 2000. Satellite estimation of tropical secondary forest above‐ground 
biomass: data from Brazil and Bolivia. International Journal of Remote Sensing 21: 
1139‐1157. 

Stevens, G. C. 1989. The latitudinal gradient in geographical range:  how so many 
species coexist in the tropics. American Naturalist 133: 240‐256. 

Stevens, W. D., C. Ulloa Ulloa, A. Pool, and O. M. Montiel, eds. 2001. Flora de Nicaragua. 
Missouri Botanical Garden Press, St. Louis, Missouri, USA. 

Stiles, F. G., and A. F. Skutch. 1989. A guide to the birds of Costa Rica. Comstock, Ithaca. 

Terborgh, J. 1986. Keystone plant resources in the tropical forest. Pages 330‐344 in M. E. 
Soulé, ed. Conservation Biology:  The Science of Scarcity and Diversity. Sinauer, 
Sunderland, MA. 

Texas A&M University. 2006. Disappearing nest egg: researcher studying declining 
numbers of macaws. ScienceDaily. Accessed 1 September 2008, 
<http://www.sciencedaily.com/releases/2006/10/061023192542.htm>. 

Theobald, D. M. 2003. GIS Concepts and ArcGIS Methods. Conservation Planning 
Technologies, Fort Collins, Colorado. 

140
 
 

Trichon, V. 2001. Crown typology and the identification of rain forest trees on large‐scale 
aerial photographs. Plant Ecology 153: 301‐312. 

Trichon, V., and M. P. Julien. 2006. Tree species identification on large‐scale aerial 
photographs in a tropical rain forest, French Guiana ‐ application for 
management and conservation. Forest Ecology and Management 225: 51‐61. 

Turner, W., S. Spector, N. Gardiner, M. Fladeland, E. Sterling, and M. Steininger. 2003. 
Remote sensing for biodiversity science and conservation. Trends in Ecology & 
Evolution 18: 306‐314. 

United Nations Economic and Social Affairs. 2006. World Population Prospects:  The 
2006 Revision and World Urbanization Prospects. Accessed 07 September 2008, 
<http://esa.un.org/unpp>. 

Urban, D., S. Goslee, K. Pierce, and T. Lookingbill. 2002. Extending community ecology 
to landscapes. Ecoscience 9: 200‐212. 

Van Laake, P. E., and G. A. Sanchez‐Azofeifa. 2004. Focus on deforestation: zooming in 
on hot spots in highly fragmented ecosystems in Costa Rica. Agriculture 
Ecosystems & Environment 102: 3‐15. 

Vargas M., A. 2008. Repunta población de lapas verdes en el país. La Nación, San José, 
Costa Rica. 

Vayssieres, M. P., R. E. Plant, and B. H. Allen‐Diaz. 2000. Classification trees:  an 
alternative non‐parametric approach for predicting species distributions. Journal 
of Vegetation Science 11: 679‐694. 

Wallace, A. R. 1876. The Geographical Distribution of Animals. Harper & Brothers, New 
York. 

Wallace, D. R. 1997. Central American landscapes. Pages 72‐96 in A. G. Coates, ed. 
Central America:  A Natural and Cultural History. Yale University Press, New 
Haven. 
141
 
 

Whitmore, T. C., R. Peralta, and K. Brown. 1985. Total species count in a Costa Rican 
tropical rain forest. Journal of Tropical Ecology 1: 375‐378. 

Whitmore, T. C. 1998. An Introduction to Tropical Rain Forests. Oxford University Press, 
New York. 

World Resources Institute (WRI). 2003. Population, health, and human well‐being–Costa 
Rica. Accessed 28 August 2008, <http://earthtrends.wri.org/>. 

Wright, S. J., and C. P. van Schaik. 1994. Light and the phenology of tropical trees. 
American Naturalist 143: 192‐199. 
 
 

142
 
 

Biography

Sara L. M. Chun was born in 1976.  She grew up in Mountain View, California.  

In 1998, she graduated from Harvard College with a bachelor’s degree in Biology.  

During her college years, Sara made several research trips to the tropical countries of 

Costa Rica, Malaysia, and Indonesia.  These trips piqued Sara’s curiosity of tropical 

systems.  After college, Sara spent three years working as a restoration ecologist at H. T. 

Harvey & Associates, an environmental consulting firm based in San Jose, California.  

Some of her job duties included designing and monitoring riparian and wetland 

restoration sites, writing environmental reports such as Mitigation and Monitoring 

Plans, and drafting scopes and budgets for project proposals.  In the fall of 2001, Sara 

began the doctoral program in the University Program in Ecology at Duke University.  

Sara is committed to balancing academic research with teaching and professional 

development.  As a doctoral student at Duke University, Sara participated in the 

Preparing Future Faculty (PFF) program.  The PFF program matches doctoral students 

with a faculty mentor at a local teaching college so that participants can observe and 

experience faculty responsibilities at a variety of academic institutions.  In the fall of 

2007, Sara taught her own graduate course entitled, “Ecosystem Science and 

Management” to 23 master’s students.  This course was conducted through the Duke 

Environmental Leadership program, which is an online executive master’s program.   

143
 
 

Sara’s doctoral research was supported by grants from the Duke University 

Graduate School, Duke University Center for Latin American & Caribbean Studies, 

Nicholas School of the Environment, Organization for Tropical Studies, and Sigma Xi 

Scientific Research Society. 

144
 

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy