Ecological Indicators 2 (2002) 3–15

Applicability of landscape metrics for the monitoring of landscape

change: issues of scale, resolution and interpretability
A. Lausch∗ , F. Herzog
UFZ Centre for Environmental Research, P.O. Box 2, D-04301 Leipzig, Germany

Abstract
In most parts of the world, land-use/land cover can be considered an interface between natural conditions and anthropogenic
influence. Indicators are being sought which reflect landscape conditions, pressures and related societal responses. Landscape
metrics, which are based on the number, size, shape and arrangement of patches of different land-use/land cover types, are
used-together with areal statistics-to quantify landscape structure and composition.
The applicability of landscape metrics for landscape monitoring has been investigated in a 700 km2 test region in eastern
Germany, where open cast coal mining has caused far reaching land-use changes in the course of this century. Time series of
maps (1912–2020) have been elaborated from various data sources (topographic maps, aerial photography, satellite images,
prospective planning material). Landscape metrics have been calculated for the entire test region and for ecologically defined
subregions at the landscape, class and patch level.
The results are presented and methodological issues are addressed, namely the impact of scale, spatial and temporal
resolution on the interpretability of landscape metrics. Critical issues are:
• the application of remote sensing methods, which is a pre-requisite for the area-wide monitoring of land-use change;
• standardised data processing techniques, which are vital for the spatial and temporal comparability of results;
• the selection of a manageable set of indicators which embraces the structural properties of landscapes;
• the choice of appropriate spatial units which allow for an integration of landscape indicators (which tend to relate to
cross-border phenomena) and socio-economic indicators (which are usually available for administrative entities or areas).
These issues are discussed in relation to the application of landscape indices in environmental monitoring.
© 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Environmental monitoring; Remote sensing; GIS; Land-use; Land cover; Landscape pattern

1. Introduction purposes and for the establishment of settlements.

Subsequent technical developments in the 19th and
Land-use is one of the main factors through 20th centuries, together with the growing human
which man influences the environment. Historically, population, have increased man’s need and ability
the most important land-use change imposed by to shape the environment according to his require-
man was the clearing of forest land for agricultural ments. The control of the hydrological regime of
landscapes (melioration, irrigation), industrialised
∗ Corresponding author. Tel.: +49-341-235-20-98; agricultural production methods and large-scale in-
fax: +49-341-235-25-11. frastructure works interfere more and more with our
E-mail address: lausch@alok.ufz.de (A. Lausch). landscapes.

1470-160X/02/$ – see front matter © 2002 Elsevier Science Ltd. All rights reserved.
PII: S 1 4 7 0 - 1 6 0 X ( 0 2 ) 0 0 0 5 3 - 5
4 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

The importance of land-use as an environmental Indicators which address landscape pattern and
parameter is reflected by the attention it receives in which are based on landscape geometry may prove
the actual discussion on environmental index develop- to be helpful in this context. There has been a con-
ment. The Organisation for Economic Co-operation siderable research effort in this field in the last two
and Development proposes a core set of 50 envi- decades, enabled by the rapid development of remote
ronmental indicators (OECD, 1998). Of these, eight sensing and geographic information systems (GIS).
relate to land cover/land-use (irrigated areas; area Correlations between landscape metrics and various
of forests; land-use changes; land cover conver- landscape functions are sought (Fig. 1). In the par-
sion; land-use and conservation; biodiversity, wildlife ticular field of landscape monitoring, the application
habitats, landscape; road infrastructure; habitat frag- of landscape metrics has been tested in a number of
mentation). However, values are presented only for studies representing a wide range of test areas and
indicators which can be derived from broad categories methods of data acquisition and treatment (Table 1).
of land-use statistics (OECD, 1999). More advanced There are considerable variations in the size of the
landscape indicators such as indicators for cultural test areas, the spatial and temporal resolutions, the
landscapes, agriculture and wildlife habitats, agricul- number of different land-use/land cover types (LT),
tural landscapes are discussed and their importance and the kind of raw data used. The most frequently
is stressed, but no operational indicators have been applied landscape indices belong to the broad cate-
presented to date (OECD, 2001). This also applies gory of edge and shape metrics. They quantify the
to a recent publication of the Statistical Office of the occurrence of ecotones, and are often related to patch
European Communities (EUROSTAT, 1998), where area, the fractal dimension, or the discrepancy be-
three levels of landscape indicators are mentioned: (1) tween actual and isodiametric shapes. Diversity mea-
statistical data on land cover and use (basically areal sures are usually derived from information theory and
statistics); (2) trends in land cover (relating mostly to often involve the use of Shannon’s diversity index.
landscape pattern) and (3) landscape elements with The number and size of patches (patch area) are also
a strong impact on the user’s perception. Whereas often measured, whereas metrics for landscape con-
values are available for the conventional level 1 indi- figuration (contagion indices) were seldom applied.
cators, level 2 and 3 indicators are yet to be developed The study presented here was carried out in a land-
by national initiatives. scape which has been subject to particularly profound

Fig. 1. The application of landscape metrics.

 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 5

Table 1 Large-scale lignite mining started in the 1920s and

Range of characteristics of 13 investigations (published in various culminated in the 1980s. By 1990, surface mining had
papers) which applied landscape metrics for landscape monitoring;
occurred on almost 40% of the area and about 90
frequency of application of categories of metrics (Herzog et al.,
2001) km2 were awaiting recultivation (RPW, 1998). Leipzig
South had been transformed into an industrial region
Parameter Range
with numerous environmental problems caused by the
Area of test regions (km2 ) 3.2–9678 mining itself, by the artificial modifications of the
Source of data Aerial region’s hydrological regime (lowering of the ground-
photography/topographic
maps—satellite images
water table, dislocations of riverbeds) and by the asso-
Type of data (spatial resolution) Vector-Raster (2–200 m) ciated carbo-chemical and brown coal industry. About
Number of land-use/land 2–17 60 settlements have been destroyed since 1928 and
cover classes about 23,000 inhabitants re-located (Berkner, 1995;
Temporal resolution (a) 8–30 Kabisch, 1997). After the German re-unification in the
Landscape metrics employed
Patch area 58% of studies
1990s, mining was drastically reduced and the indus-
Edge and shape 75% trial plants closed. Today, sanitation, reclamation and
Diversity 75% the regeneration of the landscape’s regulatory func-
Configuration 8% tions are the main issues. Novel landscape structures
establish, often spontaneously, and there are large ar-
eas with a high potential for nature protection (e.g.
modifications by means of surface lignite mining. This Durka and Altmoos, 1997).
type of landscape is well suited for developing and
testing methods of landscape monitoring because the 2.2. Data acquisition
landscape changes occur very rapidly. Preliminary re-
sults have been presented by Steinhardt et al. (1999) For Leipzig South, Spot-XS images from 1990,
who proposed the ‘Hemeroby Index’ as a human im- 1994 and 1996 were analysed with a hierarchical clas-
pact indicator. In this paper methodological questions sification procedure based on maximum-likelihood,
related to the selection of a set of landscape metrics resulting in eleven LT for each image. In addition, the
for landscape monitoring are addressed. landscape development concept of the regional plan-
ning authority (RPW, 1996) for 2020 was digitised
and included in the analysis in order to allow for the
2. Materials and methods evaluation of future developments. The spatial resolu-
tion of multispectral SPOT-XS of 20 m/pixel allowed
2.1. Leipzig South only visual recognition of the traffic network but no
proper classification. Because the values of landscape
The Leipzig lowland bay is a flat basin, which was metrics depend to a large extent on linear landscape
filled up in the Tertiary period with sediments. When elements (Lausch and Menz, 1999), the traffic net-
the basin lowered itself during the Tertiary period, the work was extracted from the digital biotope map of
groundwater level rose and the area became swampy. Saxony, which is based on CIR aerial photography
This resulted in the formation of several lignite seams (Frietsch, 1997). It was transformed to raster format
of 5–20 m in diameter. Today, the landscape consists (10 m/pixel) and merged with the classified satellite
of even ice-age deposits, interspersed by large, flat images, which had beforehand been transformed to
valleys. Topography is essentially flat with a steady the raster cell size of 10 m/pixel as well.
rise from the southern edge of the city of Leipzig Leipzig South was subdivided into 50 landscape
(110 m above sea level) to Borna town (160 m above units at the meso scale (2–36 km2 ) based on a clas-
sea level) (Eissmann, 1975). Until the early 20th cen- sification of the Saxony Academy of Science (Haase
tury, Leipzig South was an agricultural region with and Mannsfeld, 1987). Three major landscape types
comparatively intensive arable production and grass- are present: (1) valleys of large rivers and riverplains,
land in the river plains (Herzog and Heinrich, 1997). (2) plains on loose rocks and (3) mosaics in technical
6 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

Fig. 2. The test areas ‘Leipzig South’ and ‘Espenhain’ in western Saxony, Germany, and their respective subdivision into landscape units.

and artificially generated landscape units (Fig. 2). The the 1973 map, where aerial photos (1:12,000 grey
latter (22 anthropogenic landscape units) came into scale) were used—and the data processed using a GIS
existence through surface mining. (Arc/Info 7.03). Linear landscape elements (i.e. roads,
For one of those technical landscape units and running water, tree rows, etc.) were integrated into
its surrounding countryside, the Espenhain quarry, the maps as polygons (lines buffered for the average
a detailed investigation of its historical development width of the corresponding land-use types), resulting
was conducted based on a time series of four digital in maps with a maximum of 29 different LT. The
maps (1912/1944/1973/1989). Land-use was deter- Espenhain test region was subdivided into 20 natural
mined from 1:25,000 topographic maps—except for landscape units, this time based only on the original
 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 7

topography and geology in 1912. They belong to metrics were computed with the raster version
two major landscape types: (1) River valleys (Broad (Leipzig South) and the vector version (Espenhain)
U-shaped, Narrow U-shaped, V-shaped) and Plains of FRAGSTATS (Vers. 2.0, McGarigal and Marks,
(Plains on sand and gravel, Plains on pushed glacial 1994) at the landscape and—except for diversity
sediments). These micro-scale landscape units range metrics—at class and patch levels. They were cal-
between 0.1 and 19.4 km2 , many of them cut by the culated for the entire test regions as well as for the
map’s limit (Fig. 2). Because the landscape units’ landscape units. Of the 46 metrics which are avail-
size influences the values of some metrics (Herzog able, 24 could be determined for the Espenhain test
et al., 2001), a size-matched sample of six pairs of area and 27 for Leipzig South. The others would have
River valley landscape units and Plain landscape units required additional information which was not avail-
was formed. The specifications of data acquisition able, could only be calculated on raster bases (in the
and processing for the two nested test areas (Leipzig case of Espenhain) or made no sense in the context
South, Espenhain) are summarised in Fig. 3. of landscape monitoring. Some of the indicators for
For each map-year (Leipzig South as well as Espen- Leipzig South were calculated at the patch level and
hain), areal statistics were calculated and landscape then aggregated to the landscape level.

Fig. 3. Data type and processing for Leipzig South and for the Espenhain test area.
8 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

Statistical analysis was conducted with STATIS- et al. (1995) was adapted, which is based on factor
TICA 5.1 StatSoft Inc. (1997). analysis (see also Herzog et al., 2001). For Leipzig
South, the classification-tree method was used (Fig. 4).
In the Espenhain test region, for each time year,
3. Results and discussion Spearman’s correlation coefficients were computed for
all 24 metrics in order to reduce redundancies. If the
3.1. Indicator selection coefficient between two metrics was 0.9 or more, one
of the two metrics was chosen to represent both of
Numerous landscape metrics have been proposed them. This selection was arbitrary except for diversity
(e.g. Forman and Godron, 1986; O’Neill et al., 1988; metrics, where Simpson-based indices were preferred
Turner and Gardner, 1991). When working with land- to Shannon-based metrics. The use of Shannon’s di-
scape metrics one is confronted with the question of versity metrics has been criticised in cases when rich-
selecting indicators relevant for the area and the prob- ness (i.e. the number of LT) is below 100 (Yue et al.,
lem under investigation. For example, if landscape 1998). The resulting sets of metrics differed slightly
fragmentation is to be examined, one will choose in- between map-years. A number of 16 metrics were
dicators which relate to patch size, nearest neighbour- retained which belonged to the set of at least one
hood, core area, etc. In the Leipzig study area, we map-year.
sought to identify those indicators which best reflect The metrics were then grouped into the four cate-
the landscape’s temporal change. Selection can be gories “patch area metrics”, “edge and shape metrics”,
based on expertise and experience (e.g. Herzog and “diversity metrics” and “configuration metrics”.
Lausch, 2001) or on statistical approaches (this paper). Within the first three categories, a factor analysis was
In the Espenhain test region, the procedure of Riiters conducted over all four map-years in order to identify

Fig. 4. Approaches for determining a set of key landscape metrics in the Espenhain test area and in Leipzig South.
 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 9

the major factors and the metrics dominating them. Table 2

The factors which explained 10% or more of the total Results of factor analysis (principal components analysis, varimax
rotation) and factor loadings for individual landscape metrics,
variability were retained. For each factor, the index
cumulated over the four map-years of the Espenhain test region
with the highest sum of factor loadings over the four
map-years was chosen to represent this factor. In ad- Factor 1 Factor 2 Factor 3
dition to the seven metrics selected by factor analysis, Patch area metrics
the interspersion and juxtaposition index (IJI), which Eigenvalue 7.86 5.52 2.68
is the only index for landscape configuration available Percent of total variability 39.29 27.58 13.40
Cumulated factor loadings
for vector images in FRAGSTATS, is part of the final MPS 3.25 0.42 0.11
indicator set (Table 2). NP 0.29 3.73 −0.62
For Leipzig South, the approach was somewhat dif- LPI 0.29 −1.24 2.80
ferent. Whereas the factor analysis in the Espenhain PSSD 3.09 1.16 0.21
test region concerned the size-matched sample of PSCV 1.66 1.88 1.20
River valleys and Plains, the classification trees were Edge and shape
applied solely to the anthropogenic landscape units Eigenvalue 10.09 6.72 1.82
Percent of total variability 42.03 28.02 7.60
caused by surface mining. We wanted to isolate the in-
Cumulated factor loadings
dicators which showed the strongest reaction towards AWMPFD 3.45 0.88 0.49
the temporal land cover change in those landscape LSI 0.39 3.10 1.78
units between 1990 and 2020. The classification-tree MSI −2.37 1.68 0.25
method (Breiman et al., 1984) is a data mining tech- AWMSI 1.88 2.12 1.41
DLFD −0.74 2.24 1.90
nique which is well suited to examine the interaction
ED 3.17 1.06 0.09
of various indicators. In contrast to “traditional” multi-
variate analysis, the classification-tree method is based Diversity
Eigenvalue 8.77 4.55 0.80
on a hierarchy of predictions, which allow for a rank- Percent of total variability 54.81 28.42 5.02
ing of the importance of individual metrics for indicat- Cumulated factor loadings
ing differences between landscapes at different times. SIDI 2.94 −0.10 2.00
The total of 27 landscape metrics computed for PRD 0.42 3.71 −0.30
Leipzig South were subject to the analysis. They were PR 2.35 −2.30 1.57
SIEI 2.86 0.11 1.99
grouped in the same categories as for the Espenhain
case study and classified within those categories. The Configuration
IJI – – –
category Neighbourhood metrics was added which in
FRAGSTATS is available only for raster-based im- Bold: factors explaining 10% or more of the total variability and
ages. With the help of univariate splitting based on final selection of metrics. For details and metrics’ formulae see
McGarigal and Marks (1994). AWMPFD: area-weighted mean
discrimination analysis, the indicators were ranked in
patch fractal dimension, AWMSI: area-weighted mean shape in-
an step-wise procedure. Once the separation between dex, DLFD: double log fractal dimension, ED: edge density, IJI:
the map-years 1990/1994/1996/2020 was optimal, a interspersion and juxtaposition index, LPI: largest patch index,
model was considered complete. Those metrics for LSI: landscape shape index, MPS: mean patch size, MSI: mean
which an importance of 50 or more was computed are shape index, NP: number of patches, PR: patch richness, PRD:
patch richness density, PSCV: patch size coefficient of variation,
considered to represent the respective category. For
PSSD: patch size standard deviation, SIDI: Simpson’s diversity
them, the class-level metrics for the LT woods, forest, index, SIEI: Simpson’s evenness index.
water, farmland, mining were classified according to
the same procedure in order to identify the LT which
have the strongest impact (Table 3). patch area metrics, mean patch size (MPS) and patch
Indicators computed at the landscape level yield density (PD), are strongly determined by the LT
relatively general information averaged over the entire water—the lakes which will form in the open pits
landscape (unit) under investigation. Metrics com- until 2020. For edge and shape metrics, the mean
puted at the class level are helpful for the understand- shape index (MSI) of farmland and of forest patches
ing of landscape development. The two outstanding have the strongest impact. Neighbourhood metrics
10 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

Table 3
Indicator set resulting from the classification-tree method for landscape level metrics and for class level metrics for the LT mining, water,
forest, woods, farmland
Landscape level indicator Rank Importancea Class level indicator Rank Importancea

Patch area metrics

MPS 1 100 MPS Forest 2 82
MPS Water 3 81
PD 2 69 PD Water 1 100
PD Mining 4 65
PSCV 3 42
PSSD 4 12
LPI 5 9
Edge and shape
MPFD 1 100 MPFD Forest 3 84
MSI 2 59 MSI Farmland 1 100
MSI Forest 2 96
MSI Water 4 67
MSI Woods 5 57
DLFD 3 25
AWMPFD 4 21
ED 5 15
AWMSI 6 13
LSI 7 9
TE 8 3
Neighbourhood
MNN 1 100 MNN Mining 2 69
NNCV 2 64 NNCV Water 1 100
NNCV Woods 3 51
NNSD 3 35
Diversity
No ranking
Configuration
IJI – – IJI Mining 1 100
IJI Forest 2 96
IJI Woods 3 85
IJI Farmland 4 84
CONTAG – – –
Bold: metrics with importance ≥ 50. For details and metrics’ formulae see McGarigal and Marks (1994). AWMPFD: area weighted mean
patch fractal dimension, AWMSI: area weighted mean shape index, CONTAG: Contagion index, DLFD: double log fractal dimension, ED:
edge density, IJI: interspersion and juxtaposition index, LPI: largest patch index, LSI: landscape shape index, MNN: mean nearest-neighbour
distance, MPFD: mean patch fractal dimension, MPS: mean patch size, MSI: mean shape index, NNCV: Nearest-Neighbour Coefficient
of Variation, NNSD: nearest-neighbour standard deviation, PD: patch density, PSCV: patch size coefficient of variation, PSSD: patch size
standard deviation, TE: total edge.
a 0 = low, 100 = high.

are again dominated by water but also by the re- is mining. Due to the ongoing and projected recla-
maining open pits. Whereas the distance between mation activities, mining areas become smaller and
the first will decrease, the distance between open more distant from each other. This leads to an in-
pits will increase by 2020. At the landscape level, crease of the middle distance of the open-cast mining
mean nearest neighbour distance (MNN) increases by areas (indicator MNN Mining) of only some me-
about 50% between 1990 and 2020 (from approxi- ters to almost 200 m. To some extent, mining will
mately 50–75 m). The LT with the strongest change be replaced by water (the open pits will be filled
 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 11

up by rising ground water) and there will be water the dominating LT, in Espenhain as well as in Leipzig
bodies dispersed all over Leipzig South by 2020. South in general, but about one quarter of the Leipzig
This leads to a decrease of the variation of the dis- South farmland in 1990 is reclaimed dumps. Land
tance between lakes, reflected by a decrease of the cover changes between 1990 and 1996 mainly con-
Nearest-Neighbour Coefficient of Variation (NNCV) cerned the reduction of mining per ground, which is in-
of the LT water (NNCV Water), which dominates creasingly covered by pioneer plant species. By 2020,
NNCV at the landscape level. the planned reclamation activities will further reduce
The classification-tree method failed to rank diver- mining, but also spontaneous vegetation. Forest will
sity metrics because the differences between the nine become an important land cover type, together with
diversity metrics were too small over the period un- water due to the lakes which will form in the aban-
der investigation. Also, it was not possible to rank the doned open pits.
two configuration metrics at the landscape level. At To what extent are those obvious landscape changes
the class level, however, IJI for four out of the five LT reflected by landscape metrics? As an example, this
differ strongly between the map-years. The Contagion is investigated for two metrics which were part of the
Index (CONTAG) is not computed for individual LT. selected indicator sets for Espenhain as well as for
Leipzig South: mean patch size (MPS) and intersper-
3.2. Analysis and interpretation sion and juxtaposition index (IJI). MPS is computed
by dividing the total landscape (or class) area A by the
Areal statistics of the historical Espenhain analysis number of patches N (McGarigal and Marks, 1994):
show that, at the beginning of this century, Leipzig

A 1
South was a rural area with farmland being the dom- MPS = (1)
N 10, 000
inating land-use type (Fig. 5). Arable land prevailed
on the plains, permanent grassland in the river valleys The arrangement of patches and LT in the landscape
and floodplains. There was a considerable share of (landscape composition) is assessed via the intersper-
“ecological infrastructure” (tree rows, hedges, ripar- sion and juxtaposition index (IJI). It is calculated from
ian woods, waterbodies, etc.) but only relatively small the relationship between the length of each edge type
patches of forest. By the end of the 1980s, the situa- eik and total edge of the landscape E (at the landscape
tion had changed drastically. Mining and settlements level)/the length of the respective edge type involved
expanded at the expense of, mainly, arable land and eik (at the class level) divided by a term based on the
grassland. Farmland (arable land, grassland) remains number of LT m :

Fig. 5. Areal statistics of the Espenhain test area (1912–1989) and of Leipzig South (1990–2020) (Herzog and Lausch, 2001).
12 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

Fig. 6. Evolution of mean patch size (MPS) at the landscape level (a) and for selected LT at the class level (b) for the Espenhain test
region (1912–1989) and for Leipzig South (1990–2020) (Herzog and Lausch, 2001).

Landscape level : Class level :

  m   m m
− m i=1 k=i+1 [(eik /E) × ln(eik /E)]
− m k=1 [(eik / k=1 eik ) × ln(eik / k=1 eik )]
IJI = (100) IJI =
ln(1/2[m (m − 1)]) ln(m − 1)
(2) × (100) (3)

Fig. 7. Evolution of the Interspersion-juxtaposition Index (IJI) in the Espenhain test area (1912–1989) and in Leipzig South (1990–2020)
at the landscape level (a) and for selected LT (b).
 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 13

IJI approaches 0 when adjacencies are unevenly dis- extent. This is not surprising because data models and
tributed; IJI = 100 if all patch types are equally ad- scales of the Espenhain and Leipzig South studies
jacent to all other patch types (McGarigal and Marks, were not the same and because different initial groups
1994). of indices were available before the selection. Still,
In Figs. 6 and 7 the evolution of MPS and IJI at the there are some similarities. MPS was retained in both
landscape level (Figs. 6a and 7a) and for selected LT studies to represent Patch area metrics. For edge and
(Figs. 6b and 7b) is shown for the Espenhain test area shape metrics, fractal indices (MPFD—mean patch
(1912–1989) and for Leipzig South (1990–2020). fractal dimension, AWMPFD—area weighted mean
The modifications of the indicators’ values can be patch fractal dimension) ranked particularly high in
explained by increase in surface mining from the both studies. And landscape configuration can be
1940s onwards, the resulting re-arrangement of traffic captured by IJI—if not at the landscape then at the
and ecological infrastructure (also influenced by agri- class level. This outcome is congruent with the results
cultural intensification) and the projected landscape of similar investigations—although, due to differing
which will be dominated by a series of lakes form- data sets and indicator computation, comparability is
ing in the former open pits (see Herzog and Lausch, limited. Still, patch characteristics such as average
2001; Herzog et al., 2001). From a methodological size and shape (measured by MPS, LSI—landscape
point of view, two observations are of interest. shape index in our case) were retained by Cain et al.
(1997) and Riiters et al. (1995). Landscape com-
1. Indicator values for Espenhain and Leipzig South
position (IJI, CONTAG—contagion index) and—to
are at different levels. MPS computed for Leipzig
some extent—the largest patch index (LPI) and patch
South was generally lower, IJI was slightly higher.
density (PD) are comparable to indicators for ‘im-
Although small patches <0.15 ha were removed
age texture’ (contagion index, see Li and Reynolds,
during the process of satellite image classification
1993) and ‘large-patch density-area scaling’ selected
and analysis—raster satellite images still contain
by Riiters et al. (1995).
far more small pixels than vector images which, to
The metrics ‘actual values’ however, were at dif-
some extent, are generalised in the process of map-
ferent levels (Figs. 6 and 7) for Espenhain and for
ping and interpretation. This leads to an overall
Leipzig South. This is basically due to the differ-
lower level of mean patch size and its distribution
ent data models used (vector/raster). The digitising of
appears to be more uniform due to comparatively
maps or aerial photographs, which results in vector
higher spatial differentiation of raster-based satel-
data, encompasses more than just making analogous
lite images (higher values of IJI).
data digitally available. In the digitising process, the
2. Class level indicators (Figs. 6b and 7b) are helpful
spatial information is interpreted and generalised (in
in understanding general trends appearing at the
the case of map based work, a similar interpretation
landscape level (Figs. 6a and 7a). For example,
was already done by the cartographer). Digital satel-
MPS Lignite pit, IJI Lignite pit, MPS Water and
lite images, on the other hand, are raster based. Dur-
IJI Water reflect the development of lignite pits
ing their interpretation, generalisation is usually less
later becoming water bodies.
and spatial information is conserved. Also, classifica-
tion errors of individual pixels can never be avoided
4. Conclusions completely and accuracy levels of more than 80–85%
can hardly be achieved for SPOT-XS images (Albertz,
The Leipzig mining region proved well suited for 1991; Hildebrandt, 1996).
testing the application of landscape metrics for land- This dependence upon the data model and process-
scape monitoring. The findings of Riiters et al. (1995); ing limits the comparability of the results of individual
Cain et al. (1997) were confirmed that relatively few studies and consists a major drawback for the applica-
metrics suffice to capture landscape pattern. Both tion of landscape metrics. If they were to be applied at
methods applied (factor analysis, tree-classification a larger scale and enter policy relevant sets of environ-
method) resulted in sets of a manageable size. The mental indicators such as OECD (1998), there would
actual indicators selected, however, differed to a large be a need for standardisation at this level.
14 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15

A particular problem of environmental statistics is differentiation of LT (number and type of LT

the spatial unit they refer to. Whereas socio-economic considered) (Herzog and Lausch, 2001).
indicators are usually available for administrative 3. A few key landscape metrics must be selected. On
entities or areas, many environmental phenomena of- the one hand, more case studies using a statistical
ten manifest themselves regardless of administrative approach, as described here, should be conducted.
boundaries. Relating environmental indices to regions If one or two metrics are regularly retained, their
delimited according to ecological criteria (landscape suitability is obvious. At the same time, however,
units as in Leipzig South, catchments, landscape expert knowledge may be used to identify metrics
types, etc.) would increase their sensitivity and in- which can be used for particular purposes. These
terpretability. For example, in the European Union will often be class (not landscape) level metrics.
this approach has been adopted in the Water Frame- For example, again with respect to habitat func-
work Directive (2000/60/EC). Water management is tion, the MPS, Core Area and MNN of forest
supposed to be organised at the level of river basins patches may be important. MPS and the configu-
according to water quality standards specified for 25 ration (IJI) of farmland may be used to assess the
individual limnofaunal regions within Europe. This interspersion of agricultural land with other LT. At
is one example where an “eco-region” approach is a higher level of abstraction, human impact indi-
preferred over the usual “nation state” approach. Its cators should also be considered (e.g. Yue et al.,
major drawback will be that not only socio-economic 1998, 1990; Steinhardt et al., 1999).
indicators must be made available at the level of river
The potential of landscape metrics to yield informa-
basins but that also administrative structures needed
tion on landscape pattern should be further examined
for the implementation of measures must be created
in relation to national and international initiatives on
at this level, which must co-ordinate their actions
the development of environmental land-use/land cover
with the existing administrative bodies.
related indicators (Denisov et al., 1997; EUROSTAT,
Whether they be related to eco-regions or admin-
1998; OECD, 1998). Further research and technical
istrative regions, landscape metrics would need to be
development is required in the field of remote sensing
harmonised. With respect to database, data processing
and indicator selection. At the same time, concerted
and indicator selection, the following considerations
pilot projects on the applicability of landscape metrics
apply.
for landscape monitoring should be initiated in differ-
ent regions.
1. Remote sensing data must be available with a
sufficient spatial resolution to capture linear land-
scape features not only visually but also to ac- Acknowledgements
tually extract and classify them. Although they
are not important in terms of areal statistics, they Part of this research was funded by the Ger-
are indispensable for the evaluation of landscape man National Space Agency (DLR) under the grant
pattern. With respect to the landscapes’ habitat FKZ50EE9512.
function, for example, some are barriers (e.g. traf-
fic infrastructure, rivers) and others are corridors
and specific habitats (e.g. hedges, rivers). Spatial References
resolution (multispectral), therefore, should be less
Albertz, J., 1991. Grundlagen der Interpretation von Luft-und
than 5 m. Satellitenbildern: Eine Einführung in die Fernerkundung. Wiss.
2. Data processing must be harmonised. Basic re- Buchges., Darmstadt.
quirements concern the size of raster-cells, the Berkner, A. 1995. Die beeinträchtigen Oberflächengewässer des
definition of patches (whether diagonal adjacent Südraumes Leipzig mit besonderer Berücksichtigung der Pleiße.
In: Zukunft Südraum Leipzig. Sonderheft Hochschule für
pixels of the same LT are considered one patch
Technik, Wirtschaft und Kultur in Leipzig: 10–21.
or two), the definition of a minimum patch size Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J., 1984.
(elimination of patches below a certain size in Classification and Regression Trees. Wadsworth & Brooks/Cole
order to reduce the “salt and pepper” effect), the Advanced Books & Software, Monterey.
 A. Lausch, F. Herzog / Ecological Indicators 2 (2002) 3–15 15

Cain, D.H., Riiters, K., Orvis, K., 1997. A multi-scale analysis of Industrie-und Bergbauregionen-Eine Chance für den Südraum
landscape statistics. Landsc. Ecol. 122, 199–212. Leipzig? Ring B.G.I. (Ed.), Teubner Verlagsgesellschaft,
Denisov, N.B., Mnatsakanian, R.A., Semichaevsky, A.V., 1997. Stuttgart, Leipzig, pp. 113–137.
Environmental Reporting in Central and Eastern Europe: Lausch, A., Menz, G., 1999. Bedeutung der Integration linearer
A Review of Selected Publications and Frameworks. Elemente in Fernerkundungsdaten zur Berechnung von Land-
UNEP/DEIA/TR.97-6, GA/205031-97/1 and CEU/50-97.1. schaftsstrukturmaßen. PFG 3, 185–194.
Durka, W., Altmoos, M., 1997. Naturschutz in der Bergbaufol- Li, H., Reynolds, J.F., 1993. A new contagion index to quantify
gelandschaft als Teil einer nachhaltigen Landschaftsent- spatial patterns of landscapes. Landsc. Ecol. 8 (3), 155–162.
wicklung. In: Ring B.G.I. (Ed.), Nachhaltige Entwicklung in McGarigal, K., Marks, B., 1994. Fragstats—Spatial Pattern
Industrie-und Bergbauregion–Eine Chance für den Südraum Analysis Program for Quantifying Landscape Structure. Forest
Leipzig? Teubner Verlagsgesellschaft, Stuttgart, Leipzig, pp. Science Department, Oregon State University, Corvallis.
52–72. OECD, 1998. Environmental Indicators: Towards Sustainable
Eissmann, L., 1975. Das Quartiär der Leipziger Tieflandsbucht Development. Organisation for Economic Co-operation and
und angrenzender Gebiete um Saale und Elbe. Schriftenreihe Development, Paris.
für geologische Wissenschaften, Heft 2. OECD, 1999. OECD Environmental Data—Compendium 1999
EUROSTAT, 1998. European landscapes: Farmers maintain more edition. Organisation for Economic Co-operation and Develop-
than half of the territory. Statistics in Focus—Agriculture. ment, Paris.
(EUROSTAT online document at http://europa.eu.int/en/comm/ OECD, 2001. Environmental Indicators for Agriculture, vol. 3.
eurostat/serven/pdf/sb-land.pdf). Methods and Results. Organisation for Economic Co-operation
Forman, R.T.T., Godron, M., 1986. Landscape Ecology, Wiley, and Development, Paris.
New York. O’Neill, R.V., Krummel, J.R., Gardner, R.H., Sugihara, G.,
Frietsch, G., 1997. Ergebnisse der CIR-Biotoptypen-und Landnut- Jackson, B., DeAngelis, D.L., Milne, B.T., Turner, M.G.,
zungskartierung und ihre Anwendungsmöglichkeiten in der Zygmunt, B., Christensen, S.W., Dale, V.H., Graham, R.L.,
Naturschutzpraxis–Einführungsvortrag. Sächsische Akademie 1988. Indices of landscape pattern. Landsc. Ecol. 1 (3), 153–
für Natur und Umwelt 3/97, pp. 7–11. 162.
Haase, G., Mannsfeld, K., 1987. Konzeption und Beispielarbeiten Riiters, K.H., O’Neill, R.V., Hunsacker, C.T., Wickham, J.D.,
für eine Naturraumtypen-Karte der DDR im mittleren Maßstab Yankee, D.H., Timmins, S.P., Jones, K.B., Jackson, B.L., 1995.
1: 500000/1: 200000 (Projekt NTK). In Naturraumerkundung A factor analysis of landscape pattern and structure metrics.
und Landnutzung, Heinzmann G. (Ed.), Beiträge zur Geo- Landsc. Ecol. 10 (1), 23–39.
graphie, Akademie Verlag, Bd. 34, pp. 143–191. RPW, 1996. Regionalplanung in Westsachsen. Regionaler Planu-
Herzog, F., Lausch, A., 2001. Supplementing land-use statistics ngsverband Westsachsen Grimma, Karte Entwicklungskonzept
with landscape metrics: some methodological considerations. Landschaft, Stand Januar 1995, Entwurf vom 09.08.1996,
Environmental Monitoring and Assessment 72, 37–50. Maßstab 1: 75 000.
Herzog, F., Lausch, A., Müller, E., Thulke, H.H., Steinhardt, RPW, 1998. Braunkohlenplanung in Westsachsen. Regionaler
U., Lehmann, S., 2001. Landscape metrics for the assessment Planungsverband Westsachsen, Leipzig.
of landscape destruction and rehabilitation. Environ. Manage. StatSoft, Inc. 1997. STATISTICA für Windows. Tulsa.
27 (1), 91–107. Steinhardt, U., Herzog, F., Lausch, A., Müller, E., Lehmann,
Herzog, F., Heinrich, K., 1997. Die Landwirtschaft im Südraum S., 1999. The hemeroby index for landscape monitoring and
Leipzig–Nachhaltig geschädigt. In: Ring, I., Teubner, B.G. evaluation, 16. In: Pykh, Y.A., Hyatt D.E., Lenz., R.J.M., (Eds.),
(Eds.), Nachhaltige Entwicklung in Industrie-und Bergbaure- Environmental Indices—Systems Analysis Approach, EOLSS
gion—Eine Chance für den Südraum Leipzig? Verlagsge- Publ. Oxford, pp. 237–254.
sellschaft, Stuttgart, Leipzig, pp. 191–220. Turner, M.G., Gardner, R.H. (Eds.), 1991. Quantitative Methods
Hildebrandt, G., 1996. Fernerkundung und Luftbildmessung. in Landscape Ecology. Springer-Verlag, New York.
Herbert Wichmann Verlag, Heidelberg. Yue, T.X., Haber, W., Herzog, F., Cheng, T., Zhang, H.Q., Wu,
Kabisch, S., 1997. Siedlungsstrukturelle Einschnitte infolge Q.H., 1998. Models for DLU strategy and their applications.
des Braunkohlenbergbaus. In: Nachhaltige Entwicklung in Ekológia (Bratislava) 17 (Suppl. 1), 118–128.