1 Linking pattern and process in cultural landscapes. An empirical study based on spatially explicit indicators Thomas Wrbka, 1,* Karl-Heinz Erb, 2 Niels B. Schulz, 2 Johannes Peterseil, 1 Christoph Hahn, 1 Helmut Haberl 2 1 Institute of Ecology and Conservation Biology, University of Vienna Department of Conservation Biology, Vegetation Ecology and Landscape Ecology Althanstraße 14, 1090 Vienna, Austria 2 Institute of Social Ecology, Vienna. Alpen-Adria Universität Klagenfurt Graz - WienSchottenfeldgasse 29, 1070 Vienna, Austria Corresponding author: Key words: Landscape diversity; Landscape indicators; Human appropriation of net primary production (HANPP); Hemeroby; Land use intensity. Land Use Policy Special Issue Land Use and Sustainability Indicators Guest editors: H. Haberl, M. Wackernagel, T. Wrbka Published as: Wrbka, Thomas, Karl-Heinz Erb, Niels B. Schulz, Johannes Peterseil, Christoph Hahn, Helmut Haberl, Linking pattern and process in cultural landscapes. An empirical study based on spatially explicit indicators. Land Use Policy 21(3), doi:
2 2 Abstract Landscapes can be seen as the contingent and historically variable outcome of an interplay between socio-economic and biophysical forces. Landscape ecologists use a wealth of indicators based on landscape patterns to evaluate landscape diversity, landscape structure, the naturalness of landscapes, and land-use intensity. Indicators such as the human appropriation of net primary production (HANPP) are increasingly used to evaluate the changes in ecosystem processes induced by land use. Based on a study region in the central part of Lower Austria (Niederösterreich), this paper explores interrelations between patterns and processes in landscapes by empirically analysing (1) whether landform affects spatial patterns of HANPP, (2) whether HANPP is correlated with indicators of landscape naturalness, and (3) whether HANPP is correlated with landscape diversity and landscape patterns. According to our results, landform influences patterns of HANPP but cannot explain them entirely. We found strong monotonous correlations between HANPP and urbanity, an indicator for the prevalence of human-dominated land-cover classes, as well as between HANPP and the hemerobiotic state, an indicator for landscape naturalness. We found significant unimodal (hump-shaped) correlations between HANPP and the landscape diversity. Landscape pattern indicators showed weak but significant relations to HANPP. Introduction Human activities have become a dominant factor shaping most cultivated landscapes of the Earth (Goudie and Viles, 1997). Human, animal and machine labour expended in using the land can create outstanding cultural landscapes with high aesthetic, cultural and ecological value such as the paddy-field rice terraces of south-east Asia (Droste et al., 1995), but may as well result in land degradation as is the case in some karst regions in the Mediterranean (McNeill, 1992; Thirgood, 1981). The distribution of landforms such as steep slopes, fertile plains, inundated valleys in a landscape sets the frame for land use by determining factors such as accessibility, water and nutrient availability (Risser, 1990), but may over long periods of time also be changed through land use. On the other hand, land use serves distinct socioeconomic purposes: Land may supply materials and energy through hunting, agriculture or forestry, it may host infrastructure, or it may be needed to absorb wastes and emissions (Haberl et al., 2004). Landscapes can be seen as the contingent and historically variable outcome of this interplay between socio-economic and biophysical forces. During the evolution of cultural landscapes throughout the world, humans have developed adaptive land use techniques and created specific patterns of fields, farmsteads, remnant woodlots and the like that depended on both natural and socio-economic conditions (Grigg, 1974). In European agricultural landscapes the long history of land transformation has led to regionally distinct regular patterns of geometrically arranged landscape elements, reflecting the historical and cultural background of the prevailing land use system of a region (Bell, 1999). The spatial distribution of ecotopes, the so-called landscape structure, has therefore often been regarded as a mosaic of frozen processes; i.e. landscape structure assumedly mirrors the processes which had been going on in a landscape. This perception has even become a central paradigm in modern landscape ecology (Forman and Godron, 1986; Forman, 1995). While many ecosystem processes are difficult to observe directly, landscape
3 3 structure can be derived from mapping as well as from remote-sensing data; therefore, landscape structure was often not only used to evaluate the ecological value of landscapes, but also to judge ecological aspects of the sustainability of land use patterns (Odum and Turner, 1989; Wrbka et al., 1999b,c). In recent years, however, indicators such as the human appropriation of net primary production (HANPP) have become available that allow researchers to measure specific aspects of ecosystem processes caused by land use on the landscape scale (Haberl et al., 2001), making it possible to relate landscape patterns and the underlying processes, as the human influence in landscapes, and thus test hypotheses such as the frozen process paradigm. This paper seeks to improve our understanding of the relationship between landscape patterns and society-nature interaction by analysing the dependency between landforms, land-use induced changes in ecosystems, and landscape patterns. We analysed the following questions: (1) To what extent is land use intensity (measured as HANPP) determined or at least limited by landforms? (2) What is the relation between HANPP and landscape patterns? Study Area We studied the central part of the Austrian province Lower Austria (Niederösterreich) around Lower Austria s capital St. Pölten (Figure 1). The study area consists of the territory of three political districts and includes 3 cities, St. Pölten (50,000 inhabitants), Krems (23,000 inhabitants) and Tulln (12,000 inhabitants). The smallest spatial object of our analysis was the political municipality which is the lowest level for which official land-use statistics are available. The average size of the municipalities in the study region is 31.8 km² but ranges from 0.4 to km 2. Figure 1. Location of the study region. Situated in the eastern part of Austria: from west to east and south to north.
4 4 The region was selected to contain Austria s major geo-ecological land units. The Hercynian uplands (1), the Eastern lowlands (2), the Lower northern Alps (3) and finally the Danube floodplain (4) are represented in the study region (see Figure 2a). With a total area of 2864 km² and its share of major land units the study area seemed to be a good representative for the ecological variety of eastern Austria as well as the variety of prevailing land-use systems. The region is characterised by subcontinental climatic conditions with an average annual temperature of 8.8 C and a mean annual sum of precipitation from 600 to 850 mm. The altitude ranges from 115 m to 1,301 m above sea level. The subsoil is dominated by quaternary and tertiary sediments in the eastern pre-alpine lowlands and basins, by granite and gneiss bedrock in the Hercynian Uplands and by limestone and sandstone in the lower northern Alps. While the land-cover map (see Figure 2b) reveals a rather fine-grained mixture of land-cover classes throughout the region, it also indicates that the four land units differ with respect to their predominant land use system. We find mixed agriculture (cropland and grasslands) with small to medium-sized woodlots in the Hercynian uplands. The alpine part of the study region is characterised by a mixture of forests and permanent grassland. The Eastern lowlands are dominated by arable land, settlements, and permanent cultures (e.g., vineyards or orchards), whereas the river corridor of the Danube floodplain is still covered by deciduous riparian forests. Forests cover is about 39.1 % of the study region, cropland and permanent cultures 36.2 %, grasslands 16.8 % and settlements 3.4 %.
5 5 Figure 2. Situation of the study region: a) Map of the geo-ecological land units, and b) Land cover map of the study area (derived from Landsat TM-5 satellite images using a combination of a segmentation and supervised classification procedure).
6 6 Data and Methods Conceptual considerations The conceptual framework of our analysis is depicted in Figure 3. Solid lines represent relations explicitly analysed in this study whereas dashed lines stand for important influences that could not be analysed in this paper. Our basic assumption is that landscape patterns such as naturalness, landscape structure or landscape diversity are a result of ecosystem processes which in turn are influenced not only by natural factors such as climate and landforms, but also by socio-economic metabolism (see Haberl et al., 2004; Krausmann et al., 2003). Landforms: elevation, slope, roughness Climate Biogeochemical cycles HANPP: land-use induced changes in ecosystem processes (energetics) Socio-economic metabolism: material and energy flows Landscape patterns: naturalness, landscape structure and heterogeneity Regional landscape ecosystem Influences explicitly considered in this study Influences not explicitly considered in this study Figure 3. Conceptual overview of this study. While socio-economic metabolism can, in principle, be analysed for administrative units it is difficult, if not impossible, to assess for small spatial units as represented for example by pixels in a GIS grid. What can be analysed in a spatially explicit way, however, are the changes in ecosystem processes that result from socio-economic metabolism. As is discussed in other papers in this issue (Haberl et al., 2004; Krausmann et al., 2004), HANPP is an indicator of the extent to which ecosystem processes are altered by human activities. HANPP takes into account potential and actual ecosystem processes and reflects important components of socio-economic metabolism, above all the domestic extraction of biomass (Haberl et al., 2001; 2004; Krausmann, 2001). As indicators of landforms we used simple measures that can be derived from digital terrain models such as elevation, slope and relief roughness. In order to describe landscape patterns we used indicators that are commonly used in landscape ecology to evaluate the naturalness, structure and diversity of landscapes.
7 7 Data on land use and land cover The land cover map of the study region (Figure 2) is part of a land cover map of Austria derived through automatic segmentation and classification of Landsat TM5 images with a pixel size of 30x30m (Wrbka et al., 1998). This map distinguishes 20 land cover classes (see Figure 2). The segmentation procedure placed high emphasis on spectral homogeneity of segments and used a region growing method. The resulting segments were classified using a decision tree classification based on the six reflective Landsat TM5 bands. The decision rules were formulated using expert knowledge. These rules were refined in an iterative process based on information from training samples. Further methodological details can be found in Wrbka et al. (1998). The segments resemble the parcels which are the units of land use. The automatic image interpretation was optimized to get information on the landscape structure and configuration. Data on land use within these land cover classes were inferred from agricultural and forestry statistics as well as the Austrian forest inventory. For example, crops assumed to be growing on the various cropland classes were inferred from agricultural statistics that are available on the level of municipalities. Agricultural yields on cropland and meadows were also taken from agricultural statistics and were available as averages for each political district. Spatial reference units The optimal spatial scale level for analysing patterns and processes in agricultural landscapes is controversially discussed and depends largely on natural preconditions as well as on the intensity of human impacts. Banko et al. (2003) suggested the use of predefined, ecologically meaningful landscape types as reference units for national and international reporting of environmental indicators (OECD, 2001). We adopted this approach by delineating land units that were roughly homogenous with respect to their biophysical conditions, i.e. the four geoecological land units described above (Fig. 2). One of them, however, the Danube floodplain was excluded from further analysis because it covered a too small area. As reference unit for statistical analysis a grid mesh of 1x1 km size was used, in accordance with the Austrian national grid. The single grid cell which is the basic analysis units is termed as landscape cell in the following text. Every landscape cell was assigned to one of the four geo-ecological land units. Alternatively we performed our analyses on the basis of the municipalities. Each municipality was assigned to one of the four geo-ecological land units to which it had the major share. We found this approach useful because land use statistics, derived from census data, are aggregated and officially reported at this level. It can be expected that municipality-based results might eventually better link with policies and their assessment, but they presumably blur the picture of the bio-physical aspects of socio-economic processes in the region because they seldom reflect geo-ecological borders in the region. Thus we see it as a complementary procedure, enabling us to explore the strength and weaknesses of both spatial reference units which also reflect the different spatial scales. All variables, land form indicators, landscape structure indicators and HANPP, were calculated on the basis of these sample units, both the landscape cells (L) and the municipalities. The landscape cells allowed to perform proper statistics with equally sized sample units.
8 8 Indicators We distinguished three main groups of indicators: (a) landform indicators, (b) landscape structure indicators and (c) HANPP. A detailed overview of indicators (a) and (b) used in this study is presented in Table 1. Table 1. Variables used in the analyses. All variables were calculated on two different sample units (a) the landscape cell and (b) the municipality. Name of Variable Remarks Landform indicators Altitude* Slope* Curvature* Roughness* Average altitude of a certain surface area (m a.s.l.). The rate of maximum change in altitudinal (z) value from a certain surface area ( ). Concavity or convexity of the land surface describing if there are merely ridges and hills or valleys. Standard deviation of curvature describes the variation of the land surface, if a surface is merely flat or hilly. Landscape structure indicators a. Heterogeneity Landscape richness** Landscape diversity** b. Pattern Patch Size** Matrix distance** Shape Index** Maximum Incircle** Minimum-bounding-rectangle** Elongation** c. Landscape fragmentation Influence by traffic infrastructure*** Remoteness*** Meshsize*** Settlement distance** d. Landscape naturalness Urbanity** Hemeroby** Number of land cover categories per area unit. Probability of the land cover category in a reference unit. Calculated as Simpson Diversity Index. Average size of the segments (landscape element). Euclidian distance between elements belonging to the matrix. A measure for the shape compactness. The maximum radius of a circle which can be drawn within the boundaries of the element (core areas). Relation between the area of the element and the area of the bounding rectangle (complexity of the element). Relation between the shortest and the longest axis of the element. Weighted mean distance of a certain point in the landscape to the next road using a minimum operator. Weighted mean distance of a certain point in the landscape to the next road using a maximum operator. Average area within the spatial reference unit which is encircled by traffic lines. The average distance between settlement patches. The urbanity index: an indicator of the extent to which landscapes are dominated by strongly human-altered systems. Describes gradients of human influence on landscape. * Calculated on the base of a 250m resolution digital terrain model (DTM) ** Calculated on the base of the land cover map of the study area. *** Calculated on the base of the Austrian Road Map (Austrian map 1: BEV)
9 9 As landform indicators we used basic variables such as elevation, slope, curvature and relief roughness as landform indicators (Nichols et al., 1998; O Brian, 1998). Altitude was considered as average altitude of a certain surface area (m a.s.l.). Slope was defined as the rate of maximum change of the altitudinal values for a certain surface area ( ). The curvature describes whether the surface area at a certain spot is upwardly convex or concave shaped and reflects if the surface is merely characterised by ridges and hills, valleys or plains. The roughness of the relief is defined as the standard deviation of curvature. It describes if the surface area is merely flat or hilly. Higher values reflect a high variation of the surface whereas low values reflect a plain and smooth surface. All land form indicators were derived from a digital terrain model (DTM) with a spatial resolution of 250 m which is used to describe the macro and meso-relief. Averages and standard deviation were computed for each sample unit, both the landscape cells and the municipalities, using the ARC/Info Grid module and the ArcView 3.3 Spatial Analyst extension. We used HANPP (Vitousek et al., 1986) as an indicator for changes in the availability of trophic energy in ecosystems induced by land use (Haberl et al., 2004; Krausmann et al., 2004). HANPP was calculated based on the land-use and land-cover data described above using methods explained in more detail elsewhere (Haberl et al., 2001). Basically, we used information on potential vegetation, climate and elevation to calculate potential productivity; harvest indices to extrapolate the productivity of croplands and meadows from agricultural statistics; and agricultural and forestry statistics to calculate harvest on cropland, grasslands and in forests. HANPP% expresses HANPP as the percentage of the potential vegetation s NPP appropriated by humans. In calculating HANPP we only considered the aboveground compartment. Indicators of landscape patterns were derived from the land cover map of the study area. The segments of the land cover map resulting from the automatic satellite image interpretation were treated as landscape elements, which are the basic spatial units at the landscape scale (Forman and Godron, 1986; Forman, 1995). Each segment was assigned to one of the 20 land cover classes and the various indices were calculated on the patch (for the individual segments) as well as on the class level (for the land cover category within the spatial reference unit). In a second step, average values of the measures on the patch level that describe patch shape and configuration were computed for the respective sample unit: the landscape cell and the municipality. We selected a comparatively small set of variables from the suite of landscape-metrics suggested by various authors (e.g., Elkie et al., 1999; Haines-Young and Chopping, 1996; Hargis et al., 1997; 1998; McGarigal and Marks, 1994) according to our research questions. We considered four types of indicators: (a) Landscape heterogeneity indicators indicators for landscape heterogeneity based on information theory, (b) landscape pattern indicators indicators for the size and shape of landscape elements, (c) landscape fragmentation indicators indicators for the fragmentation and dissection of landscapes and (d) landscape naturalness indicators. We analysed two indicators of landscape heterogeneity. Landscape richness (LR) is defined as the number of land-cover categories per unit of area (km²).
10 10 N LR * A N is the total number of land cover categories within the sample unit and A is the total area of the sample unit (m², e.g. 1 km² for the landscape cell or the area of the municipality). Landscape diversity (LD) was calculated as the Simpson Diversity Index (McGarigal and Marks, 1994), defined by the formula LD 1 m Pi i 1 ² where P i is the probability of the occurrence of the land cover category i within the spatial reference unit and m is the number of land cover categories within the spatial reference unit. The Simpson Diversity Index is similar to the Dominance Index introduced by O Neill et al. (1988) and reflects to what extent a land cover category dominates the land mosaic within the sample unit. The value of LD increases as more land cover categories with similar proportions build up the land use mosaic and decreases as the land use mosaic is dominated by just a few land cover categories. We analysed six indicators for the size and shape of landscape elements. Patch Size (PS) is defined as the average size of the segments within the spatial reference unit (ha). This indicator is used to describe the grain size of the land mosaic within the sample unit. Matrix Distance (MD) is defined as the Euclidian distance (in m) between the segments which belong to the matrix. The matrix was identified as the prevailing land cover category within a 1 km radius of each 30x30m pixel of the land cover map. The average for each sample unit was calculated. The value increases as the matrix becomes more fragmented and decreases as the matrix becomes more connected. The Shape Index (SI) is a measure for the shape compactness as SI=1 when the shape is circular and increases without limit as the segment becomes more irregular. SI is defined as SIi Pi 2* * Ai and MSI m i 1 SIi NP where P i is the perimeter of the segment I, A i is the area of the segment I and m and NP is the total number of segments within the sample unit. The Maximum In-Circle (MXIN) is defined as the maximum radius of a circle which can be drawn within the boundaries of the segment. It is a measure for the size of the core area of a habitat which is an important quality for the appearance of interior species (Forman and Godron, 1986; Forman, 1995). The average was calculated for each sample unit. The Minimum Bounding Rectangle Fill (MBRF) is used as an indicator for the complexity of the segment. It is defined as the relation between the segment area and the area of the bounding rectangle which best fits the segment. The value increases as the segment becomes more complex and decreases towards 1 with decreasing complexity. A rectangular field shows the value 1. The Patch Elongation (ELON) is defined as the relation between the shortest and the longest axis of the segment. The value increases as the segment becomes more elongated.
11 11 We analysed four indicators for the fragmentation and dissection of landscapes which were based on a national road map (1:500,000 scale) and the land cover map. The Influence by Traffic Infrastructure (SPDMI, Wrbka et al., 2001) is defined as the weighted average distance of a certain point in the landscape to the next traffic line using a minimum operator when combining the different traffic layers. An expert based estimation of traffic intensity was used as weight. The distance is calculated as the true surface distance using the ARC/Info GRID function pathdistance. The value of SPDMI decreases with increasing influence by traffic infrastructure. Wilderness areas show high SPDMI values. The average value within the sample unit is calculated. Remoteness (SPDMA; Wrbka et al., 2001) is defined as the weighted average distance (see Influence by Traffic Infrastructure) of a certain point in the landscape to the next traffic line using a maximum operator. Distant areas are up-weighted and areas close to main traffic zones are down-weighted. The weighting and calculation procedure is similar to SPDMI. The value of SPDMA increases as the landscape gets more remote and decreases with increasing anthropogenic influence. The average value within the sample unit is calculated. Mesh size (SNET) is defined as the average area (in km²) which is encircled by traffic infrastructure. The average value within the sample unit is calculated. Settlement Distance is defined as connectedness of settlement areas and calculated as the average Euclidian distance between medium to highly sealed land cover categories using the ARC/Info GRID function pointdistance. The average value within the sample unit is calculated. As landscape naturalness indicators we used the urbanity index and hemeroby. The urbanity index (O Neill et al., 1988) is often used in landscape ecology as an indicator of the extent to which landscapes are dominated by strongly human-altered systems. It is defined as U A Urbanity log10 F W B where U denotes urban area, A agricultural area (cropland, agriculturally used grasslands), F forest areas, W water and wetland areas and B natural or semi-natural biotopes ( natural areas ). The concept of hemeroby was introduced by Jalas (1955) to describe gradients of human influence on landscape and flora. It was originally based on data of the share of life-forms (e.g., trees, shrubs, grasses) and introduced species of vascular plants. It was extended by central European ecologists by integrating parameters that describe human impacts on ecosystems such as land-use types, plant communities and soils (Blume and Sukopp, 1976; Bornkamm, 1980; Sukopp 1972, 1976). Data on hemeroby are given on a ordinal scale ranging from level 1 ( ahemerob; i.e., without actual human impact) to level 7 ( metahemerob; i.e., artificial landscape elements that do not resemble the originally prevalent biocoenoses). To assess the hemerobiotic state of each land cover type a hemeroby value was assigned based on expert judgement and the experience of an extensive field survey (Wrbka et al., 1999b,c). An area weighted average value of the hemerobiotic state was calculated for each sample unit. A similar method was recently used by Steinhardt et al. (1999).
12 12 Statistical analyses Following the conceptual model used in this study (see Figure 3) we first analysed the influence of land forms on land use intensity as assessed by HANPP%, using ordinary least square regressions techniques (Azola and Harrell, 2001). Variables were selected in a stepwise forward selection process. Linear and non-linear functions (restricted cubic splines piecewise polynomials of order higher than linear with linear restrictions to the upper and lower tail of the function; see Harrell, 2001) were fitted to the data and tested for their explanatory value. The resulting models were tested for an increase of r² using ANOVA (Ftest, p<0.05). Model validation was performed by a bootstrap method (resampling with replacement) with B=100 repetitions to get corrected r² values. Additionally we performed linear and non linear Spearman rho² rank correlation analyses (Azola and Harrell, 2001). The relationship between HANPP% and the various landscape structure indicators were analysed in a second step. HANPP%, as an indicator describing human-induced changes in processes in landscape ecosystems, was used as explanatory variable. The landscape structure indicators listed above were used as response variables. All landscape structure indicators were analysed individually. Again we performed an ordinary least square regression analysis. Linear as well as non-linear functions (e.g., restricted cubic splines) were fitted to the data. Model validation was performed using a bootstrap method with B=200 repetitions. In addition we performed Spearman rho² rank correlation analyses, examining linear as well as quadratic functions. First the landscape heterogeneity indicators (H) were analysed, second the landscape pattern indicators (P) as for example patch size and shape, third the landscape fragmentation indicators (F) and fourth the landscape naturalness indicators (N). The analysis was performed for two different levels of the spatial reference units: (a) the total study region (in the following text termed as Total ) and (b) the geo-ecological land units (in the following text identified by their names as given above). This design was used to assess the differences within the study region and between the different geo-ecological land units. The analysis was done using both sample units, landscape cells (n=2698) and municipalities (n=90). The analysis was performed using SPlus and the Hmisc and Design library (Azola and Harell, 2001). Results Dependence of HANPP on land form A highly significant result was obtained for nearly all spatial reference units (see Table 2). The different geo-ecological land units showed, for the landscape cells as well as for the municipalities, a slightly different degree of influence of landform on HANPP%. In general the relation between HANPP and land form variables was weaker for the landscape cell sample unit (corrected r² ranging from 0.17 to 0.38) than for the municipalities (corrected r² ranging from 0.68 to 0.83). Figure 4 presents the spatial distribution of some indicator values used in the analysis.
13 13 Figure 4. Illustration of the spatial distribution within the study region of exemplary indicators used in the analysis: a) altitude (m a.s.l.), b) roughness, c) landscape richness (number of land cover categories), d) landscape diversity (Simpson Diversity Index), e) HANPP% (Human Appropriation of Net Primary Production in percent of potential NPP), and f) the mean Hemerobiotic State. and process in cultural landscapes. An empirical study based on spatially explicit indicators. Land Use Policy 21(3), doi:
14 14 Table 2. Dependence of HANPP on land form. The analysis was performed for the whole study region (Total) as well as for the geo-ecological land units separately on the basis of two different sample units (a) the landscape cell (1x1km) and (b) the municipalities: n is the sample size, corr. r² is the corrected r² value resulting from model validation (bootstrap method), B is the number of repetitions used for the validation. All models are significant at a level of p < except n.s. where no significant model could be obtained. Model n Explanatory variables corr. r 2 B (a) Landscape cell Total 2698 altitude, slope, roughness Hercynian uplands 906 altitude, roughness Eastern lowlands 1095 altitude, slope, roughness Lower northern Alps 576 altitude, slope (b) Municipalities Total 90 roughness Hercynian uplands 32 altitude, roughness Eastern lowlands 38 n.s. Lower northern Alps 19 roughness For the whole study region altitude, slope and roughness turned out to be the best predicting variables to explain the variance of HANPP%, using landscape cells as sample units. The resulting regression model total explains 38% of the variance (corrected r² = 0.38, p> 0.001). Only non linear functions (restricted cubic splines with 4 knots for elevation and slope and a log10 function for the roughness of the relief) were used in the model. Generally HANPP% decreases with increasing altitude and slope (Figure 5). HANPP% decreases rapidly between 200m and 600m a.s.l., but there is greater variance and a weaker correlation in higher altitudes. The negative influence of slope on HANPP% is stronger for slight to moderate slopes and weaker in steep terrain. Roughness shows a nearly linear response function, but with high variation. The dependence of HANPP% on land form indicators differs markedly between the geoecological units. For the Hercynian uplands elevation and the roughness of the relief turned out to be the best predictors for HANPP% (corrected r²=0.34; p < 0.001). The weakest regression coefficient were found for the Lower northern Alps (corrected r²=0.17; p < 0.001) and the Eastern lowlands (corrected r²=0.20; p < 0.001). Different sets of variables had the best explanatory value for the different geo-ecological land units (see Table 2).
15 15 Figure 5. Influence of landform on HANPP. Scatterplots (a-c) and response functions (d-f) of the variables entered in the model and explaining most of the variation of HANPP% for the total study region (n=2698 landscape cells; ordinary least square regression, corrected r²=0.38, p<0.001): a) altitude, b) slope and c) roughness. The variables were fitted using linear and non linear functions (e.g. restricted cubic splines). Figure 4 shows the spatial distribution of selected indicators in the study region. A comparison of Figure 4a with Figure 4b reveals that roughness tends to be larger at higher altitudes with the exception of a part of the Hercynian uplands. This region is a plateau-like landscape with an average elevation of 650 m a.s.l. and low roughness values. This region is characterized by nutrient-poor sandy soils and an average annual temperature of 7.9 C, much less favourable geological and climate conditions for agriculture than in the other geoecological units. This explains why this region is less intensively used than the lowland areas, even if the relief would permit more intensive cultivation.
16 16 The models based on municipalities as sample units show similar results, but in general the relation is markedly better. For the total study region 68% (corrected r²=0.68; p < 0.001) of the variation of HANPP% could be explained only by roughness. HANPP% decreases rapidly with an increasing variability of the surface area. The roughness was a key explanatory variable in all models based on the municipality as sample units. In contrast to the previous results for the landscape cells a clear relation between HANPP% and the landform indicators could be demonstrated also for the Lower northern Alps. Only for the Eastern lowlands no significant model could be obtained (see Table 2). Dependence of landscape patterns on HANPP We analysed four different groups of landscape structure indicators: (a) landscape heterogeneity indicators, (b) landscape pattern indicators, (c) landscape fragmentation indicators and (d) landscape naturalness indicators. Samples of the spatial distribution of some indicator values in the study region are given in Figure 4. Significant relations between landscape heterogeneity and HANPP could be obtained for the total study region, for all geo-ecological land units at the landscape cells level, and for some indicators at the municipality level. Landscape richness as well as landscape diversity calculated for the landscape cells showed an unimodal, hump-shaped response curve (Figure 6). Landscape diversity and richness are highest at intermediate levels of HANPP. We found significant relations between HANPP% and landscape richness (corrected r²=0.35; p < 0.001) as well as between HANPP% and landscape diversity (corrected r²=0.43; p < 0.001). This also holds for the different geo-ecological land units, but in this case the response curves are slightly skewed. For the Lower northern Alps we found a left-skewed relation between richness and HANPP, in the Eastern lowlands they were distorted to the right, whereas the Hercynian Uplands yielded a classical unimodal curve. Landscape diversity revealed an even clearer optimum curve than did landscape richness. Nevertheless for all spatial reference units a more or less clear unimodal response curve could be obtained which means that landscape diversity (LD) and richness (LR) are highest at intermediate HANPP levels. All models were significant at p < (see Figure 6). When using municipalities as sample units (Table 3) valid and significant models could be obtained only for landscape richness (LR) in the Hercynian uplands (corrected r²=0.49, p<0.001) and for landscape diversity (LD) in the Eastern lowlands (corrected r²=0.65, p<0.001). HANPP was only weakly, if at all, correlated to indicators of landscape pattern. On the basis of the landscape cells as sample units weak but significant correlation could be shown for patch size, matrix distance, shape index and the maximum in-circle (Figure 6). Generally the relations are better for the separate geo-ecological land units (corrected r² ranging from 0.30 to 0.44, p<0.001) than for the total study region (corrected r² ranging from 0.23 to 0.31, p<0.001). Matrix distance shows an unimodal response curve to HANPP (see Figure 6). For the municipalities as sample units nearly no valid model could be obtained.
17 17 Table 3. Dependence of landscape structure on HANPP. The analysis was performed for the whole study region (Total) as well as for the geo-ecological land units separately on the basis of two different sample units (a) the landscape cell (1x1km) and (b) the municipalities: n is the sample size, corr. r² is the corrected r² value resulting from model validation (bootstrap method with 200 repetitions). All models are significant at a level of p < except n.v. where no valid model could be obtained. Total Hercynian uplands Eastern lowlands Low. north. Alps corr. r 2 corr. r 2 corr. r 2 corr. r 2 (a) Landscape cell n=2698 n=906 n=1095 n=576 H Landscape richness H Landscape diversity P Patch Size P Matrix distance P Shape Index P Maximum Incircle P Minimum bounding rectangle P Elongation 0.02 n.v F Influence by traffic infrastructure F Remoteness F Meshsize F Settlement distance N Hemeroby N Urbanity (b) Municipalities n=90 n=32 n=38 n=20 H Landscape richness n. v. n. v. H Landscape diversity 0.19 n. v n. v. P Patch Size n. v. n. v. n. v. n. v. P Matrix distance n. v. n. v. n. v. n. v. P Shape Index n. v. n. v. n. v. n. v. P Maximum Incircle n. v. n. v. n. v. n. v. P Minimum bounding rectangle 0.17 n. v. n. v. n. v. P Elongation n. v. n. v. n. v. n. v. F Influence by traffic infrastructure 0.22 n. v. n. v. n. v. F Remoteness n. v. n. v. n. v. n. v. F Meshsize 0.28 n. v. n. v. n. v. F Settlement distance 0.35 n. v. n. v. n. v. N Hemeroby N Urbanity Landscape structure indicators: (H) landscape heterogeneity, (P) landscape pattern, (F) landscape fragmentation and (N) landscape naturalness.
18 18 Figure 6. Influence of HANPP on landscape structure indicators. Scatterplots (a-d) and response functions (e-h) for some exemplary indicators used in the analysis: a) Landscape richness (corrected r²=0.35), b) Maximum incircle (corr. r²=0.31), c) Matrix distance (corr. r²=0.25) and d) Meshsize (corr. r²=0.24). All variables were analysed separately using ordinary least square regression techniques. Non linear functions (e.g. cubic or restricted cubic splines) were used. All models are significant at p<0.001.
19 19 With respect to landscape fragmentation indicators, significant relations were rare. Only for mesh size (Figure 6) and settlement distance a very weak relation could be obtained. Higher land use intensity is correlated with a denser traffic infrastructure, but nevertheless the relation is very weak (corrected r²=0.24, p<0.001). The models for the three geo-ecological land units show no relation at all (see Table 3). For the municipalities no valid model at all could be obtained. Hemeroby and urbanity show a high correlation to HANPP% (Figure 7). The corrected r² values ranges between 0.57 and 0.91 for the landscape cells and between 0.38 and 0.95 for the municipalities (Table 3). The relation between HANPP and the landscape naturalness indicators is nearly linear showing a decreasing naturalness with increasing HANPP levels. But the variation in the naturalness values at high HANPP levels is higher than for low HANPP values. The models for the separate geo-ecological land units show some differences, but in general the trend is similar. Even for the municipalities as sample units good correlations could be obtained: corrected r² values ranged from 0.86 to 0.94 with the exception of the Lower northern Alps (p<0.001). Figure 7. Influence of HANPP on landscape naturalness indicators. Scatterplots (a-b) and response functions (cd) for the indicators used in the analysis: a) Hemeroby Index (corrected r²=0.84), b) Urbanity Index (corr. r²=0.87). All variables were analysed separately using ordinary least square regression techniques. Non linear functions (e.g. cubic or restricted cubic splines) were used. All models are significant at p<0.001.
20 20 Discussion The influence of land form on the intensity of land use Only 38% of the spatial variance of HANPP could be explained by factors describing the geoecological conditions in the study region. Climatic variables are only indirectly considered; for example, through the gradients in temperature und precipitation with elevation. Geological and soil condition could not be taken into consideration due to the lack of detailed data. Figure 4 shows the spatial distribution of selected indicators used in the survey for the study region. The different geo-ecological land units exhibit clear differences that can be explained by differences in their biophysical conditions. But also within one geo-ecological land unit we find different patterns, for example in the Eastern lowlands north and south of the Danube, even though these areas are very similar with respect to geological and climatic preconditions: While the northern part of the Eastern lowlands is dominated by a diverse mosaic of crop farming and permanent cultures (mostly viticulture; Figure 2b) the southern part is dominated by large, rectangular-shaped fields for crop and fodder crop farming. These differences can not be explained just by biophysical variables, but must be seen as a result of different farming traditions and thus of socio-economic conditions. Cultural landscapes have, in contrast to natural and semi-natural landscapes, special characteristics. The disturbance regime as well as the major material and energy fluxes in these transformed landscapes are controlled to a large extent by humans (Fischer-Kowalski et al., 1997; Nassauer, 1995). This is done by the different land use practices applied for meadows, arable land or forests. Decisions about land use are made according to the local agro-ecological characteristics which are nested in a hierarchy of social, economical and technical constraints (Burel and Baudry, 1995; Deffontaines et al., 1995; Nassauer, 1995). Cultural landscapes can thus only be understood by analysing the interplay between biophysical and socio-economic patterns and processes (Nassauer, 1995). Within this conceptual framework we interpret our findings as follows: Spatial patterns of land use intensity as measured by HANPP are influenced by a number of factors. Even in industrialized countries landforms explain HANPP patterns to some extent, but of course other factors can also play a significant role. We would expect that, by considering a set of biophysical factors such as climate, geology, etc. as well as social and economic factors such as agricultural policies (Krausmann et al., 2003), energy policy (Haberl et al., 2003), farming traditions, etc., one could explain these pattern to a larger extent. Landscape structure and intensity of land use does the process and pattern paradigm hold? Odum and Turner (1989) found that the landscape elements of the Georgia landscape in the early 1930s had a higher fractal dimension than the elements of the same region in the 1980s. During the same period of time the use of fertilizers, pesticides and other agrochemicals increased dramatically. This illustrates that the growing human impact on the land may result in a landscape with decreasing geometrical complexity. The same can be said for agricultural landscapes in north-eastern Germany, where Stachow and Piorr (1995) detected a negative correlation between indices describing the complexity of patch shapes of landscape elements on one hand and the hemerobiotic state on the other hand. Wrbka et al. (1998) showed for several Austrian cultural landscapes that there is a significant influence of the hemerobiotic