Geoscience Reference
In-Depth Information
of 100 m. There are 44 classes in five major groups distinguished. With the usual
GIS methods, the structural information was intersected with CLC2000 in ArcGIS
(ESRI, Inc., Redlands, CA) appointing a thematical meaning to the individual land-
scape element. Because of differences in the Minimum Mappable Unit (MMU) and
minimum width (MMU segmentation - 0.25 ha, minimum width - 25 m; MMU
CLC2000 - 25 ha, minimum width - 100 m) more detailed structural than thematic
information is available. The combination of all input data allowed the building of a
consistent database of landscape-structure data.
Calculation of the landscape-structure indices was done with V-Late 1.1, Vector-
based Landscape Analysis Tools Extension (Lang & Tiede, 2003). V-Late is able
to calculate landscape metrics on three different levels - the patch level (metrics
of every single landscape element), the class level (averages of the metrics of all
landscape elements belonging to one land cover) and the landscape level (average
over all elements within one landscape no matter which thematic information they
possess). We selected five landscape metrics: Patch Size (MPS_ha), Patch Edge
(MPE), Shape Index (MSI), Perimeter-Area-Ratio (MPAR), and Fractal Dimension
(MFRACT).
Focus was given to those land-cover classes which are most interesting in agri-
cultural landscapes and which were present in at least nine selected regions with
a high percentage in area: non-irrigated arable land, pastures, complex cultivation
patterns, agricultural land with natural vegetation, and natural grasslands.
Statistical analysis was performed with R2.6.0 (R Development Core Team,
2006) and Statgraphics 5.0 with the main focus on extracting differences of land-
scape metrics regarding land-cover classes among the regions. Standard statistical
procedures were applied testing for normal distribution and variance homogeneity.
Mann-Whitney U test, discriminant and factor analysis were performed to detect
differences in terms of class-level landscape metrics among the regions. In par-
ticular, the factor analysis with varimax rotation was carried out to determine the
contribution of the individual landscape metrics to the variation in the data at class
level (Cumming & Vernier, 2002; Schindler, Poirazidis, &Wrbka, 2008). Quadratic
discriminant function analysis (QDA) was performed to test whether the function
with the variables classified according to the existing groups, and thus whether the
multivariate set of landscape metrics allows an accurate separation of the data into
SRRF-regions.
12.3 Results
In the different regions, the area of land-cover classes varied enormously
(Table 12.1). In all regions, agricultural and forest classes are the dominating land
cover. The share of anthropogenic, seminatural and natural cover is rather low.
Significant differences of single landscape-structure indices among the regions
were detected by box-and-whisker plots and the Mann-Whitney U test. Not every
landscape metric depicted differences where others did, also some variation between
the land-cover classes were exposed. Interestingly, the structure of natural grassland
Search WWH ::
Custom Search