Remotesensing 14 04933
Remotesensing 14 04933
Remotesensing 14 04933
Technical Note
Use of Remote Sensing Techniques to Estimate Plant Diversity
within Ecological Networks: A Worked Example
Francesco Liccari 1,2, * , Maurizia Sigura 2 and Giovanni Bacaro 1
1 Department of Life Sciences, University of Trieste, Via L. Giorgieri 10, 34127 Trieste, Italy
2 Department of Agricultural, Food, Environmental and Animal Sciences, University of Udine, Via delle
Scienze 206, 33100 Udine, Italy
* Correspondence: francesco.liccari@phd.units.it
Abstract: As there is an urgent need to protect rapidly declining global diversity, it is important to
identify methods to quickly estimate the diversity and heterogeneity of a region and effectively im-
plement monitoring and conservation plans. The combination of remotely sensed and field-collected
data, under the paradigm of the Spectral Variation Hypothesis (SVH), represents one of the most
promising approaches to boost large-scale and reliable biodiversity monitoring practices. Here, the
potential of SVH to capture information on plant diversity at a fine scale in an ecological network
(EN) embedded in a complex landscape has been tested using two new and promising method-
ological approaches: the first estimates α and β spectral diversity and the latter ecosystem spectral
heterogeneity expressed as Rao’s Quadratic heterogeneity measure (Rao’s Q). Both approaches are
available thanks to two brand-new R packages: “biodivMapR” and “rasterdiv”. Our aims were to
investigate if spectral diversity and heterogeneity provide reliable information to assess and monitor
over time floristic diversity maintained in an EN selected as an example and located in northeast
Italy. We analyzed and compared spectral and taxonomic α and β diversities and spectral and
landscape heterogeneity, based on field-based plant data collection and remotely sensed data from
Citation: Liccari, F.; Sigura, M.;
Sentinel-2A, using different statistical approaches. We observed a positive relationship between
Bacaro, G. Use of Remote Sensing
taxonomic and spectral diversity and also between spectral heterogeneity, landscape heterogeneity,
Techniques to Estimate Plant
Diversity within Ecological
and the amount of alien species in relation to the native ones, reaching a value of R2 = 0.36 and
Networks: A Worked Example. R2 = 0.43, respectively. Our results confirmed the effectiveness of estimating and mapping α and β
Remote Sens. 2022, 14, 4933. https:// spectral diversity and ecosystem spectral heterogeneity using remotely sensed images. Moreover, we
doi.org/10.3390/rs14194933 highlighted that spectral diversity values become more effective to identify biodiversity-rich areas,
representing the most important diversity hotspots to be preserved. Finally, the spectral heterogeneity
Academic Editors: Justin F Moat and
index in anthropogenic landscapes could be a powerful method to identify those areas most at risk of
Javier J Cancela
biological invasion.
Received: 29 June 2022
Accepted: 28 September 2022 Keywords: biodiversity patterns; free and open-source algorithms; multispectral; satellite images;
Published: 2 October 2022 spectral diversity maps; spectral heterogeneity maps; vegetation plots
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations. 1. Introduction
As there is an urgent need to protect rapidly declining global diversity [1], it is impor-
tant to identify methods to quickly estimate the diversity and heterogeneity of a region and
effectively implement monitoring and conservation plans. It is well known that biodiversity
Copyright: © 2022 by the authors.
assessment through field surveys has a very high cost both in terms of time and money.
Licensee MDPI, Basel, Switzerland.
Economic limitations often cause the inability to implement monitoring programs based on
This article is an open access article
distributed under the terms and
large-scale fieldwork [2]. Biodiversity monitoring programs must be planned on a sound
conditions of the Creative Commons
basis to obtain quality information and three aspects are considered particularly relevant,
Attribution (CC BY) license (https:// i.e., sampling design, sample size, and type of statistical analysis [3,4]. These requirements
creativecommons.org/licenses/by/ make it complex to obtain statistically valid monitoring data for better understanding and
4.0/). modeling of biodiversity over space and time [5]. In contrast to traditional field-based
samplings, Earth Observation and Remote sensing (RS/EO), based on airborne and satel-
lite systems, provides data and methods that, complementing field-based techniques, are
particularly useful for biodiversity monitoring, as it allows the observation of regions with
a high spatial and temporal resolution, thus enabling the production of maps for modeling
and monitoring diversity from local to global scales [6–8]. Operational methods for detect-
ing biodiversity patterns and ecosystem heterogeneity using remote-sensing data require
minimal supervision and do not rely on extensive ground-based data collection, as they are
non-expensive and ready-to-use methods [9]. From this point of view, the development of
free and open-source algorithms to measure and monitor biodiversity and/or ecosystem
heterogeneity from space provide robust, reproducible, and standardized estimates of
ecosystem functioning and services [10].
The combination of remotely sensed and field-collected data represents one of the most
promising approaches to boost large scale and reliable biodiversity monitoring practices [2].
To date, much research has considered the relationships between remotely sensed and
field-sampled data (e.g., [11–13]) under the paradigm of the Spectral Variation Hypothesis
(SVH), proposed for the first time by Palmer et al. [11] and further developed by Rocchini
et al. [14,15]. This concept hypothesizes that the variability of the spectral response of a
remotely sensed image could be used as a proxy to assess plant biodiversity. The ability
of SVH to detect plant diversity was tested on several ecosystems covering large areas
(e.g., [16–18]), but few studies (e.g., [19]) have investigated SVH application at a greater
level of detail over small, complex, and heterogeneous areas.
Typically, in these studies, diversity is assessed in term of α and β components [20,21],
accounting for taxonomic diversity in ground-based data as well as for spectral diversity in
remotely sensed data. Specifically, α diversity represents local diversity or diversity within
a community and β diversity represents compositional variation among communities.
Furthermore, another often neglected component of ecological diversity is represented by
ecosystem heterogeneity that is linked to a range of ecological processes and functions
that, in addition to species diversity patterns and change [8], includes metapopulation
dynamics [22], population connectivity [23] or gene flow [24].
Here, we decided to test the potential of SVH to capture information on plant diversity
at fine scale in a complex landscape, computing both α and β components and ecosystem
heterogeneity via remote sensing. The two new and promising methodological approaches
for estimating α and β spectral diversity and ecosystem heterogeneity have been tested
using the R packages “biodivMapR” [9,16] and “rasterdiv” [25,26], respectively. Specifically,
we investigated whether spectral diversity and heterogeneity can be used as proxies for
taxonomic diversity and landscape heterogeneity. In more detail, (i) we examined whether
the α and β components of spectral diversity can be compared with α and β taxonomic
diversity and (ii) whether spectral heterogeneity (measured as pixel reflectance variation)
is related to landscape heterogeneity and plant diversity in a complex landscape, where
natural and anthropogenic elements interact. The objective was to understand if spectral
data can be used to assess and/or monitor plant diversity and its dynamics in an Ecological
Network (EN) or more generally in natural environments over time.
We tested the reliability of the two methodological approaches on an existing EN
located in Friuli Venezia Giulia region (northeastern Italy), which was developed at the
local scale in the context of the Regional Environmental Landscape Plan [27]. The consid-
ered EN was modeled as a composite multi-species ecological network where the nodes
(natural habitats), corridors (links between natural habitats) capture favorable conditions
for biodiversity in an agricultural landscape matrix. Most of these natural habitats are
wetlands which are vulnerable ecosystems under climate change, extremely important for
the maintenance of biodiversity and among the most exploited and impacted by human
activity especially in Europe [28–30]. These environments are usually characterized by
marked vegetation zonation, associated with the environmental gradients, determined
primarily by hydrology [31], which host numerous species, including rare and endemic
ones.
Remote Sens. 2022, 14, x FOR PEER REVIEW 3 of 16
Remote Sens. 2022, 14, 4933 primarily by hydrology [31], which host numerous species, including rare and endemic
3 of 16
ones.
2. Materials
2. Materials and
and Methods
Methods
2.1. Study Site
2.1.
This study
This studywas
wascarried
carriedoutout
inin thethe lowlands
lowlands of the
of the Friuli
Friuli Venezia
Venezia Giulia
Giulia regionregion (NE
(NE Italy;
Italy; centroid
centroid coordinates: 45◦ 480 13.4
coordinates: 00 N–13◦ 080 11.000 E; Figure Figure
45°48′13.4″N–13°08′11.0″E; 1). The 1). Thearea
study study
has area has an
an extent of
298 kmof2 including a vast agricultural area bordered
extent 298 km2 including a vast agricultural areaby two riverby
bordered systems (Stella
two river and Corno).
systems (Stella
The
and landscape
Corno). Theis characterized by a mixed mosaic
landscape is characterized by a of intensively
mixed mosaicand extensivelyand
of intensively cultivated
exten-
areas,
sively settlements, semi-natural,
cultivated areas, settlements,and natural habitats,
semi-natural, including
and natural eight Natura
habitats, 2000 Special
including eight
Area
Naturaof Conservation (Habitats
2000 Special Area Directive 92/43/EEC)
of Conservation and nine92/43/EEC)
(Habitats Directive regional protected
and ninesites
re-
(biotopes), mainlysites
gional protected connecting wetland
(biotopes), mainly habitats.
connecting wetland habitats.
Figure 1.
Figure 1. Location of the study area (red square) in relation to the European continent (top right).
Representation of
Representation of the
the Ecological
Ecological Network
Network model
model and
and location
location of
of sampling
sampling units
units within
within the
the study
study
area (main figure). Colored lines and patches are corridors and nodes of the network, representing
area (main figure). Colored lines and patches are corridors and nodes of the network, representing
different habitat types and species-specific networks. EUNIS Habitat Codes are as follows: D4.11
different habitat types and species-specific networks. EUNIS Habitat Codes are as follows: D4.11
Schoenus nigricans fens; D5.24 Fen Cladium mariscus beds; E1.55 Eastern sub-Mediterranean dry grass-
Schoenus nigricans fens; D5.24 Fen Cladium mariscus beds; E1.55 Eastern sub-Mediterranean dry
land; E2.2 Low and medium altitude hay meadows; E3.51 Molinia caerulea meadows and related com-
grassland; E2.2 LowIllyrian
munities; G1.A1A and medium altitude haybetulus
Quercus-Carpinus meadows;
forests; Molinia
E3.51G1.11 caeruleaSalix
Riverine meadows and related
woodland; G1.22
communities; G1.A1A Illyrian Quercus-Carpinus betulus forests; G1.11 Riverine Salix woodland;
Southeast European Fraxinus-Quercus-Alnus forests; G1.41 Alnus swamp woods not on acid peat. G1.22
Southeast European Fraxinus-Quercus-Alnus forests; G1.41 Alnus swamp woods not on acid peat.
The geology of the area consists mainly of Quaternary sand, silt, and silt-clay sediments
The geology of the area consists mainly of Quaternary sand, silt, and silt-clay sediments
formed by glacial fluvial transport during the Pleistocene and by alluvial deposition during
formed by glacial fluvial transport during the Pleistocene and by alluvial deposition during
the Holocene. The area is characterized by an average annual temperature of about 13 °C and
the Holocene. The area is characterized by an average annual temperature of about 13 ◦ C
an average annual precipitation between 1100 and 1400 mm.
and an average annual precipitation between 1100 and 1400 mm.
The natural habitats of the area are woods, meadows, and fens and have been classi-
The natural habitats of the area are woods, meadows, and fens and have been classified
fied following EUNIS habitat classification [32,33]. Specifically part of the forests are dom-
following EUNIS habitat classification [32,33]. Specifically part of the forests are dominated
inated by Carpinus betulus and Quercus robur (EUNIS habitat codes G1.A1A) while the wet
by Carpinus betulus and Quercus robur (EUNIS habitat codes G1.A1A) while the wet forests
forests by Alnus glutinosa, Fraxinus spp., and Salix spp. (EUNIS habitat codes F3.23, F9.2,
by Alnus glutinosa, Fraxinus spp., and Salix spp. (EUNIS habitat codes F3.23, F9.2, G1.11,
G1.11, G1.22, G1.41), dry meadows are characterized by Arrhenatherum elatius, Brachypo-
G1.22, G1.41), dry meadows are characterized by Arrhenatherum elatius, Brachypodium
dium rupestre, Bromopsis erecta, Carex spp., Chrysopogon gryllus, Festuca rubra, Filipendula
rupestre, Bromopsis erecta, Carex spp., Chrysopogon gryllus, Festuca rubra, Filipendula vulgaris,
vulgaris, Lolium spp., Lotus spp., Trifolium spp. (EUNIS habitat codes E1.55, E2.2) while wet
Lolium spp., Lotus spp., Trifolium spp. (EUNIS habitat codes E1.55, E2.2) while wet meadows
meadows
by by Carex
Carex spp., spp.,
Molinia Molinia
spp., spp., and Filipendula
and Filipendula ulmaria
ulmaria (EUNIS (EUNIS
habitat habitat
codes E3.4, codes
E3.51)E3.4,
and
fens by Armeria helodes (endemic), Cladium mariscus, Equisetum palustre, Frangula alnus,
Remote Sens. 2022, 14, 4933 4 of 16
Lysimachia vulgaris, Molinia caerulea, Potentilla erecta, Salix cinerea, Scirpoides holoschoenus,
Schoenus nigricans, and Senecio fontanicola (endemic; EUNIS habitat codes D4.11, D5.24).
we chose to use multiple images and then averaging them across the sampling period
(May–September 2019) to compare the different performance. The resulting Sentinel 2
multi-spectral images were filtered to remove non-vegetated, shaded, and cloudy pixels.
The applied thresholds were as follows: (1) Normalized Difference Vegetation Index (NDVI)
higher than 0 to exclude non-vegetated pixels, (2) NIR > 1500 to remove shaded areas as
these are characterized by low overall reflectance, and (3) blue < 500 to ignore cloudy pixels
as residuals from atmospheric corrections may lead to increased reflectance in the blue
domain [9].
After the filtering, a Principal Component Analysis (PCA) was performed on a random
subset (21% ca) of the image to ensure computational efficiency. The result of this PCA
was then applied to order the whole image. Subsequently, a second filtering based on PCs
thresholding was applied automatically discarding the pixels showing values beyond the
mean PC value ± 3 standard deviations for any of the first five components and the mask
was updated accordingly. Finally, the PCA preprocessing including random pixel selection
was applied a second time with the updated mask. Based on PCA results, relevant features
for biodiversity mapping were selected considering the PCA outputs.
Spectral species mapping is based on k-means clustering of the components selected
from the PCA. First, a subset of pixel was extracted from PCA, equivalent to the number
of pixels randomly selected when computing PCA. This pool of pixels was then split into
n partitions, and k-means clustering was performed on each partition, in order to define
the adequate number of clusters. The clustering was then applied to the whole PCA image
in order to assign a cluster to each pixel based on closest centroid, for each repetition.
Therefore, the number of partition cluster maps result from this independent clustering.
Each cluster map was then processed until the computation of α- and β-diversity metrics,
and these diversity metrics were finally averaged. Spectral species mapping was performed
setting number of clusters parameter to 50. This value was suggested by Féret and de
Boissieu [9] for tropical forests, but it was also indicated that the number of clusters should
be set according to the level of heterogeneity of the landscape under study, in our case quite
high.
The α and β spectral diversity maps were produced through the computation of three
indexes (i.e., species richness and H’ for α diversity and BC for β diversity), based on the
distribution of clusters in the spectral species map for a window size set to 6 × 6 pixels
over the whole image. Testing different window sizes on the study area, we observed
that the smaller the window, the more accurate the estimate, but small windows may not
contain a sufficient number of pixels. For this reason, we decided to use a window of
6 × 6 pixels as a compromise. Finally, spectral diversity index values were extracted for
the sampled plots (plots containing fewer than three pixels were discarded) in order to
compare field inventories with diversity indices estimated by the “biodivMapR” package.
We compared these values by linear regression, correlation analysis, and using the R
package “Metrics” [44] that computes evaluation metrics (i.e., RMSE and bias) that are
commonly used in supervised machine learning to compare actual and predicted values.
to the transformation of the pixel value in 8-bit, a larger number of pixels was necessary
to generate variability in the index calculation. In addition, we also calculated the Rao’s
Q on an NDVI time series (i.e., one image per month for the year 2019; QNDVI ) rescaled
to 8-bit radiometric resolution, with a moving window of 9 × 9 pixels and alpha set to
1, 5, and infinite. Here again, Rao’s Q values were extracted for all the sampled plots
and the relationships between spectral heterogeneity and field data were estimated using
Generalized Additive Models (GAM) and transformation-based Redundancy Analysis
(tb-RDA). We hypothesized that in a highly heterogeneous landscape, such as the one
under study, relationships were to be sought between spectral heterogeneity, landscape
heterogeneity, and the amount of native and alien species. For this reason, in the GAMs we
considered Rao’s Q values deriving from both NDVI time series (QNDVI ) and multispectral
single image (Qmulti ) as response variables and ratio of alien to native species richness
(RatioAN) and Shannon index by land use category (ShannonLU) calculated in an area of
250 m around the plots as predictive variables.
Six GAMs were considered, the response variables were alternatively QNDVI 1, QNDVI 5,
QNDVI Inf, Qmulti 1, Qmulti 5, and Qmulti Inf, and the predictive variables were always Shan-
nonLU as smooth term and RatioAN as linear term. A Gaussian distribution was used for
the error and a cubic regression spline was selected for the smooth term by means of the
“mgcv” R package [46]
Regarding β diversity, the contribution of previous variables (Qmulti , RatioAN, and
ShannonLU) with the addition of three other variables (i.e., native species richness, N.Nat;
focal species richness, N.Foc; and habitat type) to the observed community composition
was considered using tb-RDA [47]. QNDVI and alien species richness were not considered
as they were highly collinear with Qmulti and RatioAN, respectively. The tb-RDA was
based on Hellinger [48] pre-transformed species composition matrix. Species abundances
were log transformed before Hellinger transformation was done. These transformations
were made possible as tb-RDA supports the use of many data transformations to perform
ordination, offering much more flexibility for the analysis of community data [49]. Variance
partitioning was then computed in order to assess which group of variables (habitat, land
use, and Rao’s Q) contributed more to explain the variability in the community composition.
3. Results
3.1. Comparison of α and β Spectral Diversity vs. Measured Taxonomic Diversity
The resulting α and β spectral diversity maps, obtained from the “biodivMapR”
package, are shown in Figures 2 and S1.
The comparison between the observed values of α and β taxonomic diversity cal-
culated from sampled plots and the ones remotely estimated via the “biodivMapR” are
reported below.
Considering the results for the single image (3 June 2019, Figure S1), the regression
between observed taxonomic and estimated spectral species richness (Figure S2a) gave an
RMSE = 16.68 and a bias = 15.15 that indicate a high underestimation of the number of
species. We also computed the Pearson correlation between the observed species richness
and the estimated one obtaining a value of 0.16 (p = 0.03). This result was partly expected
considering the study of Féret and Asner [16] where they observed an underestimation that
could be explained by the limited number of spectral species compared to the maximum
taxonomic diversity. However, as stated by Féret and Asner [16], it can be easily corrected
with a linear factor derived from the relationship obtained between field data and estimation
(for example, in our case by 0.4, Figure S2b).
Remote
Remote Sens. 2022, 14,
Sens. 2022, 14, 4933
x FOR PEER REVIEW 7 of 16
Figure 2.
Figure 2. Spectral
Spectral α
α diversity
diversity map,
map, expressed
expressed asas Shannon
Shannon index
index (H’),
(H’), of
of the
the study
study area
area (a)
(a) and
and of
of the
the
EN nodes (b). Spectral β diversity map, expressed as Bray–Curtis dissimilarity index (BC), produced
EN nodes (b). Spectral β diversity map, expressed as Bray–Curtis dissimilarity index (BC), produced
by the projection of the n × n dimensional space of the dissimilarity matrix into an n × 3 dimensional
by the projection of the n × n dimensional space of the dissimilarity matrix into an n × 3 dimensional
space (PCoAs), of the study area (c) and of the EN nodes (d). Similar colors, resulting from the PCoA
space (PCoAs),
ordination, of the study
represent similararea (c) and
spectral of the
plant EN nodes (d). Similar colors, resulting from the PCoA
communities.
ordination, represent similar spectral plant communities.
The comparison between the observed values of α and β taxonomic diversity calcu-
The regression between observed taxonomic and estimated spectral H0 (Figure S3)
lated from sampled plots and the ones remotely estimated via the “biodivMapR” are re-
yielded an RMSE = 0.39 in H’ units and a bias = −0.01 that indicated a slight overestimation
ported below.
of H’. We also computed the Pearson correlation between the observed H’ and the estimated
Considering the results for the single image (3 June 2019, Figure S1), the regression
H’ obtaining a value of 0.53 (p < 0.001). The regression between observed taxonomic and
between observed taxonomic and estimated spectral species richness (Figure S2a) gave an
estimated spectral BC (Figure S4) yielded an RMSE = 0.17 in BC units and a bias = 0.06
RMSE = 16.68 and a bias = 15.15 that indicate a high underestimation of the number of
that indicated a slight underestimation of BC dissimilarity. Mantel correlation between the
species. We also computed the Pearson correlation between the observed species richness
observed BC and the estimated one yielded a value of 0.48 (p < 0.001).
and the estimated one obtaining a value of 0.16 (p = 0.03). This result was partly expected
Considering the results for the averaged multiple images (from May 2019 to September
considering
2019, Figure 2),thethe
study of Féretbetween
regression and Asner [16] where
observed they and
taxonomic observed an underestimation
estimated spectral species
that could be explained by the limited number of spectral species
richness gave an RMSE = 14.53 and a bias = 14.66 that indicate a high underestimation compared to the maxi-of
mum taxonomic diversity. However, as stated by Féret and Asner
the number of species. We also computed the Pearson correlation between the observed [16], it can be easily
corrected
species with aand
richness linear
thefactor derived
estimated one from the relationship
obtaining a value of 0.18obtained between
(p = 0.01). In thisfield
case,data
the
and estimation
species richness(for example,
estimation in our case
produced by 0.4, better
a slightly Figureresult
S2b). than that of the single image.
The regression
The regression between
between observed
observed taxonomic
taxonomic and and estimated
estimated spectral
spectral H H′0 (Figure
(Figure S3)
S5)
yielded an RMSE = 0.39 in H’ units and a bias = −0.01 that indicated a slight
yielded an RMSE = 0.36 in H’ units and a bias = 0.06 that indicated a slight underestimation overestimation
of H’.
of H’.WeWealsoalsocomputed
computed thethe Pearson
Pearson correlation
correlation between
between the observed
the observed H’ and H’theandestimated
the esti-
mated
H’ H’ obtaining
obtaining a valuea ofvalue
0.60of(p0.53 (p < 0.001).
< 0.001). The regression
The regression betweenbetween observed
observed taxonomic
taxonomic and
and estimated
estimated spectral
spectral BC (Figure
BC (Figure S6) S4) yielded
yielded an an
RMSERMSE = 0.17
= 0.15 inin
BC BCunits
unitsand
andaabiasbias== 0.05
0.06
that indicated
that indicatedaaslight
slightunderestimation
underestimationofof BCBC dissimilarity.
dissimilarity. Mantel
Mantel correlation
correlation between
between the
the observed BC and the estimated one yielded a value
observed BC and the estimated one yielded a value of 0.52 (p < 0.001). of 0.48 (p < 0.001).
Considering
The spectral the results formap,
β diversity the averaged
expressed multiple
as BC images (from May
dissimilarity index, 2019 to Septem-
produced by
ber projection
the 2019, Figure of 2),
thethen ×regression
n dimensionalbetweenspaceobserved taxonomic and
of the dissimilarity estimated
matrix into anspectral
n×3
species richness
dimensional gave
space an RMSE
(PCoAs, = 14.53
Figure andpresents
2c,d), a bias = a14.66
goodthat indicate
estimate of anatural
high underesti-
habitats,
mation of the number of species. We also computed the Pearson correlation between the
mated H’ obtaining a value of 0.60 (p < 0.001). The regression between observed taxonomic
and estimated spectral BC (Figure S6) yielded an RMSE = 0.15 in BC units and a bias = 0.05
that indicated a slight underestimation of BC dissimilarity. Mantel correlation between
the observed BC and the estimated one yielded a value of 0.52 (p < 0.001).
Remote Sens. 2022, 14, 4933 The spectral β diversity map, expressed as BC dissimilarity index, produced by 8 ofthe
16
projection of the n × n dimensional space of the dissimilarity matrix into an n × 3 dimen-
sional space (PCoAs, Figure 2c,d), presents a good estimate of natural habitats, showing
a distribution
showing along the
a distribution positive
along PCoA1 PCoA1
the positive axis foraxis
79%for
of 79%
the pixels
of the contained in the in
pixels contained nodes
the
of the EN.
nodes of the EN.
3.2.Spectral
3.2. SpectralHeterogeneity
Heterogeneityvs.
vs.Landscape
LandscapeHeterogeneity
Heterogeneityand
andTaxonomic
TaxonomicPlant
PlantDiversity
Diversity
Theresulting
The resultingQQNDVI Qmulti
NDVIInf and Q multiInf
Inf spectral
spectral heterogeneity maps, obtained
obtained from
fromthe
the
“rasterdiv”package,
“rasterdiv” package,are
areshown
shownininFigure
Figure3.3.
Figure3.3.Rao’s
Figure Rao’sQQindex,
index,calculated
calculatedfrom
fromthe theNDVI
NDVItimetimeseries
seriescovering
coveringthe theyear
year2019
2019(Q(QNDVI) with
NDVI ) with
theweight
the weightfor forthe
the distance
distance matrix
matrix setset to infinite,
to infinite, for study
for the the study
area area (a). Rao’s
(a). Rao’s Q index,
Q index, calculated
calculated from
from the 10 bands of the Sentinel 2 image of 3 June 2019 (Q multi) with the weight for the distance
the 10 bands of the Sentinel 2 image of 3 June 2019 (Qmulti ) with the weight for the distance matrix set
matrix set to infinite, for the study area (b).
to infinite, for the study area (b).
Weobserved
We observedsignificant
significantrelationships
relationships inin all
all GAMs
GAMs between
between spectral
spectral heterogeneity,
heterogeneity,
landuse
land usediversity,
diversity,and
andalien
aliennative
nativespecies
speciesrichness
richnessratio,
ratio,except
exceptfor
forthe
theterm
termRatioAN
RatioANinin
Q multi1 GAM (Table 1, Figure S7). The adjusted R2 2 increased as the weight for the distance
Qmulti 1 GAM (Table 1, Figure S7). The adjusted R increased as the weight for the distance
matrixwas
matrix washigher
higherfor
forboth
bothQQ NDVI and Qmulti GAMs. Comparing the models with the same
NDVI and Qmulti GAMs. Comparing the models with the same
distance weight, those with Q
distance weight, those with Qmulti always had
multi always had aa higher
higher goodness-of-fit
goodness-of-fit (Table
(Table1).
1).The
Thebest
best
model (R2 = 0.43) was the one using the Qmulti with the highest distance weight. The linear
term RatioAN was always positive related to Rao’s Q in all models, while the smooth
term ShannonLU was more positively related to Rao’s Q the greater the distance weight
considered (Figure S7).
Remote Sens. 2022, 14, 4933 9 of 16
Table 1. Summary of generalized additive models (GAMs) for spectral heterogeneity (Rao’s Q
index calculated from NDVI timeseries (QNDVI ) and from a multispectral image (Qmulti ) with three
different weights for the distance matrix (i.e., 1, 5, infinite) vs. alien native species richness ratio
(RatioAN, linear term) and Shannon index calculated on land uses (ShannonLU, smooth term).
Est. ± SE = estimate ± standard error; Edf = effective degrees of freedom.
The tb-RDA ordination explained 36.20% of the variance, the first three axes accounting
for 11.30%, 10.15%, and 5.41% of the total explained variance, respectively (Figure 4). The
first seven axes out of eighteen exceeded the threshold of statistical significance (p < 0.05).
The first axis was correlated with native (N.Nat) and focal (N.Foc) species richness and
ratio of alien to native species richness (RatioAN) and the second with land use diversity
and spectral heterogeneity (ShannonLU and Qmulti Inf, Figures 4 and S8). The axes from
three to seven mainly described the difference between plant community composition in
different habitats dictated by the presence of hygrophilous or xerophilous, woodland or
grassland species. Forest habitats were mainly distributed along the second axis based
on a gradient of spectral heterogeneity (RDA2 −0.54), land use diversity (RDA2 −0.44),
and ratio of alien to native species richness (RDA2 −0.38) while fens and meadows were
distributed on the first and third axes along a gradient of focal (RDA1 −0.40; RDA3 −0.38)
and native species richness (RDA1 −0.37; RDA3 0.36). These gradients showed that higher
values of Rao’s Q spectral heterogeneity were more related to wet habitats, and with those
habitats where the land use diversity of the surrounding landscape was higher, features
that are also known to often promote plant invasion.
Remote Sens.
Remote Sens.2022,
2022,14,
14,x4933
FOR PEER REVIEW 10 of 16 10 of 1
Figure4.4.tb-RDA
Figure tb-RDA ordination
ordination based
based on Hellinger
on Hellinger pre-transformed
pre-transformed species species composition
composition matrix, wit
matrix, with
sitegrouped
site grouped perper habitat
habitat and and displaying
displaying the following
the following variables:variables: focal
focal species species
richness richness
(N.Foc), (N.Foc), na
native
tive species
species richnessrichness
(N.Nat),(N.Nat), Rao’s calculated
Rao’s Q index, Q index, calculated from the
from the 10 bands 10Sentinel
of the bands of the Sentinel
2 image of 3 June2 image o
3 June
2019 2019
with the with
weight the
forweight for the
the distance distance
matrix set tomatrix
infinite set to infinite
(Qmulti INF), and multiINF),
(Qratio and
of alien to ratio
nativeof alien t
native species richness (RatioAN), Shannon index on land use
species richness (RatioAN), Shannon index on land use diversity (ShannonLU). diversity (ShannonLU).
The
Thevariance partitioning
variance on species
partitioning composition
on species data highlighted
composition that habitatthat
data highlighted was habitat
the wa
main explanatory factor, accounting for 29% of the total variation (Figure S9). Interestingly,
the main explanatory factor, accounting for 29% of the total variation (Figure S9). Inter
spectral heterogeneity and land use diversity contributed only to 8% of total variation, with
estingly, spectral heterogeneity and land use diversity contributed only to 8% of total var
the latter almost completely negligible in explaining plant community variations.
iation, with the latter almost completely negligible in explaining plant community varia
4.tions.
Discussion
This research aimed at investigating the use of SVH in a complex and anthropogenic
4. Discussion
landscape, using two new and promising methodological approaches for estimating α
and βThis
spectral diversity
research aimed andatecosystem heterogeneity.
investigating the use ofTheir
SVH outputs differed
in a complex andbut both
anthropogeni
gave important
landscape, information
using two new on plant diversity expendable
and promising for planning
methodological data collection
approaches and
for estimating α
monitoring campaigns for biodiversity conservation programs.
and β spectral diversity and ecosystem heterogeneity. Their outputs differed but both
The relationships between floristic and spectral α and β diversity indices provided
gave important information on plant diversity expendable for planning data collection
evidence of the potential, but also of the limits, of remote sensing data as proxies of plant
and monitoring
diversity campaigns
[5]. In particular, in ourforcase
biodiversity
study, in aconservation
heterogeneousprograms.
and anthropogenic land-
scape,The
where relationships
natural habitatbetween
patches floristic
(the nodesand spectral
of the EN) areαembedded
and β diversity indices provided
in an agricultural
evidence
matrix, ofrelationships
these the potential, butslightly
were also ofweaker
the limits, of remotewith
in comparison sensing
other data
studiesasinproxies
homo- of plan
diversity
geneous [5]. Inhabitats
natural particular, in our case
[17,18,50–54]. study, inFéret
For example, a heterogeneous
and Asner [16],and anthropogenic land
in homogeneous
natural
scape, forests,
where reported a weak relationship
natural habitat patches (the between
nodes observed
of the EN) taxonomic and estimated
are embedded in an agricul
spectral species richness that could be easily corrected with a linear factor
tural matrix, these relationships were slightly weaker in comparison with other derived from thestudies in
relationship obtained between field data and estimation, as we also noticed in our results.
homogeneous natural habitats [17,18,50–54]. For example, Féret and Asner [16], in homo
Moreover, Féret and Asner [16] in tropical forests reported an underestimation of both α
geneous natural forests, reported a weak relationship between observed taxonomic and
and β diversity indices but with high correlation rates (between 0.73 and 0.86 for α and
estimated
between spectral
0.61 and 0.75species
for β). richness that
In contrast, wecould be easily
observed corrected with
an overestimation of αadiversity
linear factor de
rived from the relationship obtained between field data and estimation, as we also noticed
in our results. Moreover, Féret and Asner [16] in tropical forests reported an underestima
tion of both α and β diversity indices but with high correlation rates (between 0.73 and
0.86 for α and between 0.61 and 0.75 for β). In contrast, we observed an overestimation o
Remote Sens. 2022, 14, 4933 11 of 16
(H’) and an underestimation of β diversity (BC) and both diversity indexes achieved lower
correlation values between floristic and spectral values (i.e., 0.53 and 0.48, respectively).
However, considering the results from the averaged multiple images, we observed an
underestimation of both indexes but with higher correlations between floristic and spectral
values (i.e., 0.60 and 0.52, respectively). Our lower correlation values could derive, on one
hand, from the lower relationship between spectral and taxonomic diversity observed by
Rossi et al. [55] in grassland ecosystems. On the other hand, they may be determined by the
plot dimension (10 m × 10 m) in comparison with that of the window used to calculate the
spectral species (60 m × 60 m): in a highly heterogeneous landscape, such as the one under
study, the signal can vary a lot moving away from the sampled plot due to plant community
and land use variations, driven by many factor such as habitat type, anthropic pressure,
landscape structure, and edge effects [56,57]. The effect of the grain size on the robustness
in the relationship between spectral diversity and taxonomic diversity has been examined
by different authors (e.g., [14,58]) and all of them concluded that the increase of the spatial
scale of analysis, from both field and remotely sensed data, increased correlation between
spectral heterogeneity and species richness. This issue, defined Modifiable Areal Unit
Problem (MAUP), is a well-known pattern in landscape ecology and has been exhaustively
discussed and analyzed by Jelinski and Wu [59]. This partly confirms that the performance
of the spectral variability as a proxy of species diversity is influenced by the location and
the extent of the reference region [60]. However, we found positive correlations between
taxonomic and spectral diversity that could be applied, for example, for guiding sampling
planning. Moreover, in our case, the relationship between taxonomic and spectral diversity
values became more accurate for high diversity values (Figures S3–S6), thus highlighting
that spectral diversity values become more reliable for biodiversity-rich areas that also
represent the most important diversity hotspots to be monitored and preserved.
In addition, the spectral maps in Figure 2 have given evidence of the real differences
between plots both in terms of α and β diversity. In fact, taking into consideration the
entire study area and the sorting of the β diversity values of the pixels on the three axes of
the PCoA (Figure 2c,d), it was possible to observe that the majority of the pixels linked to
the positive part of the first axis of the sorting, represented in violet, corresponded to the
forested nodes of the EN.
The results produced by “rasterdiv” package (Figure 3) highlighted the influence of
the surrounding landscape composition and fragmentation on the values expressed by the
Rao’s Q heterogeneity index. In fact, the areas with higher values of spectral heterogeneity
were not those that we would expect to be richer in biodiversity, but those that were
characterized by more anthropogenic impact (high values of land use diversity) and so
also to biological invasion (high values of alien to native species richness ratio). Remote
sensing data can provide information on complex systems, which depend on the original
radiometric and spectral resolution, giving different results and interpretations depending
on the composition of the study area and the type of existing vegetation. Using an ecological
parallelism, the spectral space defined by many bands is analogous to the Hutchinson’s
hypervolume, defined by a set of n independent axes corresponding to those variables
(abiotic and biotic) shaping species’ niches [61,62]. In this case, spectral space was expected
to be related to both species’ niches and their relative diversity [63], and this was the case,
albeit with a relationship opposite to that expected. That is, greater spectral difference was
found to be related to greater ratio of alien to native species richness rather than greater
native or focal species richness (Figure 4). However, this result should not be neglected
as it relates well to the use of remote sensing techniques for monitoring invasive alien
plants across vast areas [64]. Many studies have demonstrated the capability of remote
sensing approaches to detect invasive plant species and to map their distribution [65–70]
and certainly the use of the Rao’s Q heterogeneity index in anthropogenic landscapes could
be a powerful method to identify those areas potentially more prone to biological invasion.
GAMs (Table 1, Figure S7) showed that the greater the weighting between the spectral
distance of pixels, the greater the relationship between spectral heterogeneity, land use
Remote Sens. 2022, 14, 4933 12 of 16
diversity, and ratio of alien to native richness. It was interesting to observe that Qmulti
was better explained than QNDVI by the above-mentioned variables, probably because the
amount of spectral information contained in the 10 bands used in Qmulti was greater than
that contained in the two bands of QNDVI , although the latter was calculated over a longer
time frame. Spectral heterogeneity analysis suggests that the indexes can be interpreted in
the opposite way (e.g., −Qmulti or −QNDVI ) in our case study and could allow to identify a
method to detect core areas within nodes (i.e., patches of natural habitats) of an EN. The
core area is the inner part of a node that is less affected by the external impacts and to
the edge effects. These latter are important ecological processes that are closely related to
species habitat protection [71], community dynamics [72], and ecological restoration [73,74].
The analysis of heterogeneity using Rao’s Q can thus represent a new tool to be integrated
in the context of EN structure optimization.
Variance partitioning (Figure S9) pointed out that the variable contributing the most to
explaining differences among communities was habitat while the contribution of land use
diversity is completely negligible in this context. Instead, spectral heterogeneity contributes
to nearly one-third of the explained variation. This shows that the Rao’s Q in complex
areas, not dominated by a single habitat, is unable to account for variation among different
communities. However, this result also suggests, observing the forests in the ordination
plot (Figures 4 and S8), Rao’s Q potential to explain variation in the composition of the
community in environments dominated by forest habitats.
Our results highlighted the effectiveness of estimating and mapping α and β spectral
diversity and ecosystem spectral heterogeneity using remotely sensed images. This is
currently a key topic in ecology and could provide landscape managers with rapid and
effective tools to estimate and monitor global change [5]. Moreover, this study confirms
once again the robustness and importance of SVH for estimating and monitoring diversity
in different habitats [16,18,19,52–54]. In addition, we suggest experimenting with spectral
heterogeneity analysis in the field of landscape ecology (e.g., ENs structure analysis) as
well as the use of spectral diversity maps as fast approach in data-poor settings as starting
base.
The observed relationship between spectral and floristic diversity, in a complex and
anthropogenic landscape, supports SVH as a method to quickly estimate α and β diversity
and heterogeneity. Moreover, it is suggested to explore their variation across regions to
effectively implement monitoring and conservation plans allowing the production of maps
for modeling and monitoring diversity from local to global scales [6–8].
5. Conclusions
The problem of quickly estimating the diversity and heterogeneity of a region to imple-
ment monitoring and conservation plans has a relevant impact on biodiversity management
and natural resources conservation. Biodiversity assessment is today mainly performed by
field survey, which is a time and cost-consuming method. This work explored the potential
of SHV for estimating and monitoring diversity in an anthropogenic landscape, composed
of different habitats. In this framework, two distinct methodological approaches for esti-
mating α and β spectral diversity and ecosystem heterogeneity have been tested using the
R packages “biodivMapR” and “rasterdiv”. These are based on different indices, but both
are promising in the measurement of relationship between taxonomic diversity and spectral
diversity. However, in heterogenous areas, they achieve lower correlation values between
floristic and spectral values compared to the results obtained in areas characterized by
extended and homogeneous natural landscapes, suggesting that spectral variability as a
proxy of species diversity could be influenced both by the location characteristics (e.g.,
high or low landscape diversity, land use fragmentation) and by the extent of the reference
regions. These aspects should be further explored. However, positive correlations between
taxonomic and spectral diversity have been found, highlighting that spectral diversity
values become more reliable for biodiversity-rich areas. The related thematic maps can
contribute to informing processes of landscape management and planning. In fact, a fast
Remote Sens. 2022, 14, 4933 13 of 16
method to discover biodiversity hotspots could support the detection of changes during
a given time frame related to land use intensification/extensification or allow to address
ground monitoring efforts on the potential biodiversity hot spots. Regarding the weak-
nesses of the two methods, in the case of “biodivMapR”, the possibility of working on time
series would have to be implemented and the estimation of spectral species would have
to be improved, which, as noted by the authors themselves, are always underestimated.
In the case of “rasterdiv”, it would be appropriate to give different weight to radiometric
variation depending on land use or discriminating between anthropogenic and natural
areas.
Remote sensing data can provide information on complex systems, which depend on
the original radiometric and spectral resolution, giving different results and interpretations
depending on the composition of the study area and the type of existing vegetation. We
found that this is true in particular in heterogeneous and anthropogenic landscapes where
natural habitat patches are embedded in a heterogeneous agricultural matrix. This young
research field needs further experimenting with both spectral diversity and spectral het-
erogeneity analysis to explore the potential of Sentinel-2A satellite data, available for free
on ESA’s Sentinel Scientific Data Hub (https://scihub.copernicus.eu, accessed on 12 May
2022), which represent a powerful basis to map and monitor natural ecosystems.
Data Availability Statement: Data presented in this study are available upon request to the corre-
sponding author.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. IPBES. Summary for Policymakers of the Global Assessment Report on Biodiversity and Ecosystem Services of the Intergovernmental
Science-Policy Platform on Biodiversity and Ecosystem Services; Díaz, S., Settele, J., Brondízio, E.S., Ngo, H.T., Guèze, M., Agard, J.,
Arneth, A., Balvanera, P., Brauman, K.A., Butchart, S.H.M., et al., Eds.; IPBES Secretariat: Bonn, Germany, 2019; 56p. [CrossRef]
2. Vihervaara, P.; Auvinen, A.P.; Mononen, L.; Törmä, M.; Ahlroth, P.; Anttila, S.; Böttcher, K.; Forsius, M.; Heino, J.; Heliola, J.; et al.
How essential biodiversity variables and remote sensing can help national biodiversity monitoring. Glob. Ecol. Conserv. 2017, 10,
43–59. [CrossRef]
3. Yoccoz, N.G.; Nichols, J.D.; Boulinier, T. Monitoring of biological diversity in space and time. Trends Ecol. Evol. 2001, 16, 446–453.
[CrossRef]
4. Maccherini, S.; Bacaro, G.; Tordoni, E.; Bertacchi, A.; Castagnini, P.; Foggi, B.; Gennai, M.; Mugnai, M.; Sarmati, S.; Angiolini,
C. Enough Is Enough? Searching for the Optimal Sample Size to Monitor European Habitats: A Case Study from Coastal Sand Dunes.
Diversity 2020, 12, 138. [CrossRef]
5. Rocchini, D.; Salvatori, N.; Beierkuhnlein, C.; Chiarucci, A.; de Boissieu, F.; Förster, M.; Garzon-Lopez, C.X.; Gillespie, T.W.;
Hauffe, H.C.; He, K.S.; et al. From local spectral species to global spectral communities: A benchmark for ecosystem diversity
estimate by remote sensing. Ecol. Inform. 2021, 61, 101195. [CrossRef]
6. Féret, J.B.; Rocchini, D.; He, K.S.; Nagendra, H.; Luque, S. Forest species mapping. In A Sourcebook of Methods and Procedures for
Monitoring Essential Biodiversity Variables in Tropical Forests with Remote Sensing; GOFC-GOLD, GEO BON, Ed.; Report Version
UNCBD COP-13 GOFC-GOLD Land Cover Project Office; Wageningen University: Wageningen, The Netherlands, 2017.
7. Rocchini, D.; Boyd, D.S.; Féret, J.B.; Foody, G.M.; He, K.S.; Lausch, A.; Nagendra, H.; Wegmann, M.; Pettorelli, N. Satellite remote
sensing to monitor species diversity: Potential and pitfalls. Remote Sens. Ecol. Conserv. 2016, 2, 25–36. [CrossRef]
8. Rocchini, D.; Luque, S.; Pettorelli, N.; Bastin, L.; Doktor, D.; Faedi, N.; Feilhauer, H.; Féret, J.B.; Foody, G.M.; Gavish, Y.; et al.
Measuring β-diversity by remote sensing: A challenge for biodiversity monitoring. Methods Ecol. Evol. 2018, 9, 1787–1798.
[CrossRef]
9. Féret, J.B.; de Boissieu, F. biodivMapR: An r package for α- and β-diversity mapping using remotely sensed images. Methods Ecol.
Evol. 2020, 11, 64–70. [CrossRef]
10. Rocchini, D.; Neteler, M. Let the four freedoms paradigm apply to ecology. Trends Ecol. Evol. 2012, 27, 310–311. [CrossRef]
11. Palmer, M.W.; Earls, P.G.; Hoagland, B.W.; White, P.S.; Wohlgemuth, T. Quantitative tools for perfecting species lists. Environmetrics
2002, 13, 121–137. [CrossRef]
12. Rocchini, D.; Hernández-Stefanoni, J.L.; He, K.S. Advancing species diversity estimate by remotely sensed proxies: A conceptual
review. Ecol. Inform. 2015, 25, 22–28. [CrossRef]
13. Lausch, A.; Heurich, M.; Magdon, P.; Rocchini, D.; Schulz, K.; Bumberger, J.; King, D.J. A Range of Earth Observation Techniques
for Assessing Plant Diversity. In Remote Sensing of Plant Biodiversity, 1st ed.; Cavender-Bares, J., Gamon, J., Townsend, P., Eds.;
Springer Nature: Cham, Switzerland, 2020; pp. 309–348.
14. Rocchini, D.; Chiarucci, A.; Loiselle, S.A. Testing the spectral variation hypothesis by using satellite multispectral images. Acta
Oecol. 2004, 26, 117–120. [CrossRef]
15. Rocchini, D.; Balkenhol, N.; Carter, G.A.; Foody, G.M.; Gillespie, T.W.; He, K.S.; Kark, S.; Levin, N.; Lucas, K.; Luoto, M.; et al.
Remotely sensed spectral heterogeneity as a proxy of species diversity: Recent advances and open challenges. Ecol. Inform. 2010,
5, 318–329. [CrossRef]
16. Féret, J.B.; Asner, G.P. Mapping tropical forest canopy diversity using high-fidelity imaging spectroscopy. Ecol. Appl. 2014, 24,
1289–1296. [CrossRef] [PubMed]
17. Heumann, B.W.; Hackett, R.A.; Monfils, A.K. Testing the spectral diversity hypothesis using spectroscopy data in a simulated
wetland community. Ecol. Inform. 2015, 25, 29–34. [CrossRef]
18. Torresani, M.; Rocchini, D.; Sonnenschein, R.; Zebisch, M.; Marcantonio, M.; Ricotta, C.; Tonon, G. Estimating tree species
diversity from space in an alpine conifer forest: The Rao’s Q diversity index meets the spectral variation hypothesis. Ecol. Inform.
2019, 52, 26–34. [CrossRef]
19. Marzialetti, F.; Cascone, S.; Frate, L.; Di Febbraro, M.; Acosta, A.T.R.; Carranza, M.L. Measuring Alpha and Beta Diversity by
Field and Remote-Sensing Data: A Challenge for Coastal Dunes Biodiversity Monitoring. Remote Sens. 2021, 13, 1928. [CrossRef]
20. Whittaker, R.H. Vegetation of the Siskiyou mountains, Oregon and California. Ecol. Monogr. 1960, 30, 279–338. [CrossRef]
21. Whittaker, R.H. Evolution and measurement of species diversity. Taxon 1972, 21, 213–215. [CrossRef]
22. Fahrig, L. Landscape heterogeneity and metapopulation dynamics. In Key Topics in Landscape Ecology; Wu, J., Hobbs, R., Eds.;
Cambridge University Press: Cambridge, UK, 2007; pp. 78–91.
23. Malanson, G.P.; Cramer, B.E. Landscape heterogeneity, connectivity, and critical landscapes for conservation. Divers. Distrib. 1999,
5, 27–39. [CrossRef]
Remote Sens. 2022, 14, 4933 15 of 16
24. Lozier, J.D.; Strange, J.P.; Koch, J.B. Landscape heterogeneity predicts gene flow in a widespread polymorphic bumble bee,
Bombus bifarius (Hymenoptera: Apidae). Conserv. Genet. 2013, 14, 1099–1110. [CrossRef]
25. Marcantonio, M.; Iannacito, M.; Thouverai, E.; Da Re, D.; Tattoni, C.; Bacaro, G.; Vicario, S.; Ricotta, C.; Rocchini, D. rasterdiv:
Diversity Indices for Numerical Matrices. 2021, R Package Version 0.2-3. Available online: https://CRAN.R-project.org/package=
rasterdiv (accessed on 12 October 2021).
26. Rocchini, D.; Thouverai, E.; Marcantonio, M.; Iannacito, M.; Da Re, D.; Torresani, M.; Bacaro, G.; Bazzichetto, M.; Bernardi, A.;
Foody, G.M.; et al. rasterdiv—An Information Theory tailored R package for measuring ecosystem heterogeneity from space: To
the origin and back. Methods Ecol. Evol. 2021, 12, 1093–1102. [CrossRef] [PubMed]
27. Sigura, M.; Boscutti, F.; Buccheri, M.; Dorigo, L.; Glerean, P.; Lapini, L. La Rel dei Paesaggi di Pianura, di Area Montana
e Urbanizzati. Piano Paesaggistico Regionale del Friuli-Venezia Giulia (Parte Strategica) E1 -Allegato alla Scheda di RER.
Regione Friuli-Venezia Giulia. 2017. Available online: http://www.regione.fvg.it/rafvg/cms/RAFVG/ambiente-territorio/
pianificazione-gestione-territorio/FOGLIA21/#id9 (accessed on 12 October 2021).
28. Jones, T.A.; Hughes, J.M.R. Wetland inventories and wetland loss studies: A European perspective. In Waterfowl and Wetland
Conservation in the 1990s: A Global Perspective, Proceedings of the IWRB Symposium, St Petersburg Beach, FL, USA, 12–19 November
1993; Moser, M., Prentice, R.C., van Vessem, J., Eds.; IWRB Spec. Publ. No. 26; IWRB: Slimbridge, UK, 1993; pp. 164–169.
29. European Commission. Life and Europe’s Wetlands. Restoring a Vital Ecosystem; European Commission: Luxembourg, 2007.
30. Jantke, K.; Schleupner, C.; Schneider, U.A. Gap analysis of European wetland species: Priority regions for expanding the Natura
2000 network. Biodivers. Conserv. 2011, 20, 581–605. [CrossRef]
31. Liccari, F.; Castello, M.; Poldini, L.; Altobelli, A.; Tordoni, E.; Sigura, M.; Bacaro, G. Do Habitats Show a Different Invasibility
Pattern by Alien Plant Species? A Test on a Wetland Protected Area. Diversity 2020, 12, 267. [CrossRef]
32. Davies, C.E.; Moss, D.; Hill, M.O. EUNIS Habitat Classification Revised 2004; Report to the European Topic Centre on Nature
Protection and Biodiversity; European Environment Agency: Copenhagen, Denmark, 2004.
33. Chytrý, M.; Tichý, L.; Hennekens, S.M.; Knollová, I.; Janssen, J.A.M.; Rodwell, J.S.; Peterka, T.; Marcenò, C.; Landucci, F.;
Danihelka, J.; et al. EUNIS habitat classification: Expert system, characteristic species combinations and distribution maps of
European habitats. Appl. Veg. Sci. 2020, 23, 648–675. [CrossRef]
34. Liccari, F.; Sigura, M.; Tordoni, E.; Boscutti, F.; Bacaro, G. Determining plant diversity within interconnected natural habitat
remnants (ecological network) in an agricultural landscape: A matter of sampling design? Diversity 2022, 14, 12. [CrossRef]
35. Liccari, F.; Boscutti, F.; Bacaro, G.; Sigura, M. Connectivity, landscape structure, and plant diversity across agricultural landscapes:
Novel insight into effective ecological network planning. J. Environ. Manag. 2022, 317, 115358. [CrossRef]
36. Bartolucci, F.; Peruzzi, L.; Galasso, G.; Albano, A.; Alessandrini, A.; Ardenghi, N.M.G.; Astuti, G.; Bacchetta, G.; Ballelli, S.; Banfi,
E.; et al. An updated checklist of the vascular flora native to Italy. Plant. Biosyst. 2018, 152, 179–303. [CrossRef]
37. Galasso, G.; Conti, F.; Peruzzi, L.; Ardenghi, N.M.G.; Banfi, E.; Celesti-Grapow, L.; Albano, A.; Alessandrini, A.; Bacchetta, G.;
Ballelli, S.; et al. An updated checklist of the vascular flora alien to Italy. Plant. Biosyst. 2018, 152, 556–592. [CrossRef]
38. Copernicus Open Access Hub. Sentinel-2 Data. Available online: https://scihub.copernicus.eu/ (accessed on 15 September 2021).
39. SNAP-ESA Sentinel Application Platform v7.0. Available online: http://step.esa.int (accessed on 15 September 2021).
40. Ustin, S.L.; Gamon, J.A. Remote sensing of plant functional types. New Phytol. 2010, 186, 795–816. [CrossRef]
41. Homolová, L.; Malenovský, Z.; Clevers, J.G.P.W.; García-Santos, G.; Schaepman, M.E. Review of optical-based remote sensing for
plant trait mapping. Ecol. Complex. 2013, 15, 1–16. [CrossRef]
42. Shannon, C.E. A mathematical theory of communication. Bell. Syst. Tech. J. 1948, 23, 379–423. [CrossRef]
43. Bray, J.R.; Curtis, J.T. An ordination of the upland forest communities of Southern Wisconsin. Ecol. Monogr. 1957, 27, 325–349.
[CrossRef]
44. Hamner, B.; Frasco, M. Metrics: Evaluation Metrics for Machine Learning. R package version 0.1.4. 2018. Available online:
https://CRAN.R-project.org/package=Metrics (accessed on 12 October 2021).
45. Rao, C.R. Diversity and dissimilarity coefficients: A unified approach. Theor. Popul. Biol. 1982, 21, 24–43. [CrossRef]
46. Wood, S.N. Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear
models. J. R. Stat. Soc. 2011, 73, 3–36. [CrossRef]
47. Legendre, P.; Gallagher, E. Ecologically meaningful transformations for ordination of species data. Oecologia 2001, 129, 271–280.
[CrossRef] [PubMed]
48. Legendre, P.; Legendre, L. Numerical Ecology, 3rd ed.; Elsevier: Amsterdam, The Netherlands, 2012.
49. Blanchet, F.G.; Legendre, P.; Bergeron, J.A.C.; He, F. Consensus RDA across dissimilarity coefficients for canonical ordination of
community composition data. Ecol. Monogr. 2014, 84, 491–511. [CrossRef]
50. Nagendra, H.; Rocchini, D.; Ghate, R.; Sharma, B.; Pareeth, S. Assessing plant diversity in a dry tropical forest: Comparing the
utility of Landsat and Ikonos satellite images. Remote Sens. 2010, 2, 478–496. [CrossRef]
51. Hall, K.; Reitalu, T.; Sykes, M.T.; Prentice, H.C. Spectral heterogeneity of QuickBird satellite data is related to fine-scale plant
species spatial turnover in semi-natural grasslands. Appl. Veg. Sci. 2012, 15, 145–157. [CrossRef]
52. Warren, S.D.; Alt, M.; Olson, K.D.; Irl, S.D.H.; Steinbauer, M.J.; Jentsch, A. The relationship between the spectral diversity of
satellite imagery habitat heterogeneity, and plant species richness. Ecol. Inform. 2014, 24, 160–168. [CrossRef]
53. Arekhi, M.; Yilmaz, O.Y.; Yilmaz, H.; Akyüz, Y.F. Can tree species diversity be assessed with Landsat data in a temperate forest?
Environ. Monit. Assess. 2017, 189, 586. [CrossRef]
Remote Sens. 2022, 14, 4933 16 of 16
54. Madonsela, S.; Cho, M.A.; Ramoelo, A.; Mutanga, O. Remote sensing of species diversity using Landsat 8 spectral variables.
ISPRS J. Photogramm. 2017, 133, 116–127. [CrossRef]
55. Rossi, C.; Kneubühler, M.; Schütz, M.; Schaepman, M.E.; Haller, R.M.; Risch, A.C. Spatial resolution, spectral metrics and biomass
are key aspects in estimating plant species richness from spectral diversity in species-rich grasslands. Remote Sens. Ecol. Conserv.
2021, 8, 297–314. [CrossRef]
56. Porensky, L.M.; Young, T.P. Edge-Effect Interactions in Fragmented and Patchy Landscapes. Conserv. Biol. 2013, 27, 509–519.
[CrossRef] [PubMed]
57. Amici, V.; Rocchini, D.; Filibeck, G.; Bacaro, G.; Santi, E.; Geri, F.; Landi, S.; Scoppola, A.; Chiarucci, A. Landscape structure effects
on forest plant diversity at local scale: Exploring the role of spatial extent. Ecol. Complex. 2015, 21, 44–52. [CrossRef]
58. Oldeland, J.; Wesuls, D.; Rocchini, D.; Schmidt, M.; Jurgens, N. Does using species abundance data improve estimates of species
diversity from remotely sensed spectral heterogeneity? Ecol. Indic. 2010, 10, 390–396. [CrossRef]
59. Jelinski, D.E.; Wu, J. The modifiable areal unit problem and implications for landscape ecology. Landsc. Ecol. 1996, 11, 129–140.
[CrossRef]
60. Schmidtlein, S.; Fassnacht, F.E. The spectral variability hypothesis does not hold across landscapes. Remote Sens. Environ. 2017,
192, 114–125. [CrossRef]
61. Hutchinson, G. Homage to Santa Rosalia or why are there so many kinds of animals? Am. Nat. 1959, 93, 145–159. [CrossRef]
62. Blonder, B. Hypervolume concepts in niche and trait based ecology. Ecography 2018, 41, 1441–1455. [CrossRef]
63. Thouverai, E.; Marcantonio, M.; Bacaro, G.; Da Re, D.; Iannacito, M.; Marchetto, E.; Ricotta, C.; Tattoni, C.; Vicario, S.; Rocchini,
D. Measuring diversity from space: A global view of the free and open source rasterdiv R package under a coding perspective.
Community Ecol. 2021, 22, 1–11. [CrossRef]
64. Rocchini, D.; Andreo, V.; Förster, M.; Garzon-Lopez, C.X.; Gutierrez, A.P.; Gillespie, T.W.; Hauffe, H.C.; He, K.S.; Kleinschmit,
B.; Mairota, P.; et al. Potential of remote sensing to predict species invasions: A modelling perspective. Prog. Phys. Geogr. Earth
Environ. 2015, 39, 283–309. [CrossRef]
65. Müllerová, J.; Brůna, J.; Dvořák, P.; Bartaloš, T.; Vítková, M. Does the data resolution/origin matter? satellite, airborne and Uav
imagery to tackle plant invasions. ISPRS—Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. 2016, 41B7, 903–908. [CrossRef]
66. Skowronek, S.; Asner, G.P.; Feilhauer, H. Performance of one-class classifiers for invasive species mapping using airborne imaging
spectroscopy. Ecol. Inform. 2017, 37, 66–76. [CrossRef]
67. Skowronek, S.; Ewald, M.; Isermann, M.; Kerchove, R.V.D.; Lenoir, J.; Aerts, R.; Warrie, J.; Hattab, T.; Honnay, O.; Schmidtlein,
S.; et al. Mapping an invasive bryophyte species using hyperspectral remote sensing data. Biol. Invasions 2017, 19, 239–254.
[CrossRef]
68. Vaz, A.S.; Alcaraz-Segura, D.; Campos, J.C.; Vicente, J.R.; Honrado, J.P. Managing plant invasions through the lens of remote
sensing: A review of progress and the way forward. Sci. Total Environ. 2018, 642, 1328–1339. [CrossRef]
69. Lopatin, J.; Dolos, K.; Kattenborn, T.; Fassnacht, F.E. How canopy shadow affects invasive plant species classification in high
spatial resolution remote sensing. Remote Sens. Ecol. Conserv. 2019, 5, 302–317. [CrossRef]
70. Ewald, M.; Skowronek, S.; Aerts, R.; Lenoir, J.; Feilhauer, H.; Van De Kerchove, R.; Honnay, O.; Somers, B.; Garzón-López,
C.X.; Rocchini, D.; et al. Assessing the impact of an invasive bryophyte on plant species richness using high resolution imaging
spectroscopy. Ecol. Indic. 2020, 110, 105882. [CrossRef]
71. Paton, P. The effect of edge on avian nest success: How strong is the evidence? Conserv. Biol. 1994, 8, 17–26. [CrossRef]
72. Fagan, W.; Cantrell, R.; Cosner, C. How habitat edges change species interaction. Am. Nat. 1999, 153, 165–182. [CrossRef]
73. Loveridge, A.; Hemson, G.; Davidson, Z.; Macdonald, D. African lions on the edge: Reserve boundaries as “attractive sinks”. Biol.
Conserv. Wild Felids 2010, 283, 283–304.
74. An, Y.; Liu, S.; Sun, Y.; Beazley, R. Construction and optimization of an ecological network based on morphological spatial pattern
analysis and circuit theory. Landsc. Ecol. 2021, 36, 2059–2076. [CrossRef]