Geometric and Landcover Signatures of Local Authorities in Peloponnesus GEORGE CH. MILIARESIS Department of Geology University of Patras Geology Department, University of Patras, Rion 265-04 GREECE gmiliar@upatras.gr http://miliaresis.tripod.com Abstract: - The quantification of knowledge related to the terrain and the landcover of local authorities in Southern Greece (Peloponnesus) is performed from a digital elevation model and the CORINE landcover database. Each local authority is parametrically represented by a set of attributes related to its relief (geometric) signature. Local authorities are classified on the basis of K-means cluster analysis in an attempt to see how they are organized into groups and cluster derived geometric signatures are defined. Finally each cluster is parametrically represented on the basis of the occurrence of the Corine landcover classes included and thus, landcover signatures are derived. Geometric and landcover signatures might be used in an attempt to determine the local authority vulnerability and assist the planning at a local authority level. Key-Words: - Spatial objects, Parametric representation, Classification, Digital elevation model, Corine 1 Introduction Nowadays, digital elevation models (DEMs) represents the earth s relief at regional to moderate scale [7]. Segmentation techniques have been developed to automate the interpretation of terrain features from DEMs [4]. Various parameters were devised in an attempt to characterize the landscape, and its sensitivity to natural hazards like landslides [8]. Additionally, a comprehensive land cover database is available, for the 25 EC Member States and other European countries, named Corine [1]. Corine provides quantitative data on land cover, consistent and comparable across Europe at an original scale of 1: 100000 using 44 Corince landcover classes (CLCs) of a 3-level nomenclature [1]. Vulnerability is termed the sensitivity of a region to the influence of unfavorable and dangerous natural events or phenomena [10]. Vulnerability is function of the relief (geomorphology), landcover and geology (lithology and tectonics). Local authorities in Greece establish an artificial terrain partition framework that in many cases contradicts to the geomorphologic and environmental organization of both terrain and landcover. On the hand local authorities are the first level objects related to the economic development. The quantification of knowledge related to the landscape organization and the environment of local authorities from both digital elevation models and the Corine landcover database is vital for the planning and economic development at both a country and a European level. The aim of this research effort is to parametrically represent each local authority with attributes related to relief as well as to landcover. These data will allow deriving the unique local authority signatures evident. These signatures could assist the planning and the economic development by providing an object based representation and modeling scheme of Greece, as well as the assessment of hazards related to the upcoming global climatic change. 2 Methodology A terrain partition framework is required first, allowing the segmentation of the landscape to elementary objects [3]. These objects are the local authorities. Then, each object is parametrically represented on the basis of its both 3-dimensional arrangement [6] and landcover. Finally, the objects are mapped according to a terrain classification scheme [5]. 2.1 Study area The study area (Fig.1) corresponds to Peloponnesus in Southern Greece, bounded by longitude 21 o to 23.5 o E and latitude 36.5 o to 38.5 o N. 2.2 Data The Hellenic Military Geographical Service topographic maps were used for the production of the DEM and the digitization of the borders of local
authorities (Fig. 1). The map scale was 1:50.000, with 20 m contour interval, georeferenced to EGSA87, the Greek Geodetic System [9]. A DEM with spacing 100 m (Fig. 1) was derived by using digital contour lines and selected elements from spot heights, drainage, ridges, lakes and shorelines. a certain landcover type in Peloponnesus) of the 44 Corine classes is given in the frequency histogram of Fig. 3. Fig.3. Occurrence of the CLCs. The CLCs in the study area with occurrence greater than 0.01% are given in Fig.4. Fig.1. The DEM of the study area and the local authorities border. The Corine landcover raster map [1] of Peloponnesus was reprojected to EGSA87 [9] and resampled to a 100 m grid by nearest neighbor (Fig.2). Fig.2. Corine landcover map of Peloponnesus. The occurrence (percentage of the area covered by Fig. 4. CLCs with occurrence >0.01%. 2.3 Terrain parametric representation Local authorities form polygons (Fig.1). Each polygon is actually a 3-d spatial object as it can be seen in Fig. 1. The parametric representation of 3-d spatial objects is achieved [5] by the logarithm of their size (lna), mean elevation (H), roughness (sd), local relief (LR), mean gradient (G), and the hypsometric integral (HI). More specifically: LnA is indication of the area extent. H is correlated to the annual rainfall height and vegetation type. LR equals to the elevation range and it is correlated to the terrain diversity within the object SD is the standard deviation of elevation and correlated to the height variability. G expresses mean slope per object and it is correlated to the intensity of the erosion/deposition processes. HI is used as an indicator of the relative amount of land (from the base of the mountain to its top) that
was removed by the erosion process. HI reflects the stage of landscape development (an indicator of the cycle of erosion). More specifically, for HI above 0.6 the area is in youthful stage, for HI in the range 0.35 to 0.6 the area is in the equilibrium (maturity erosion cycle) phase and HI below 0.35 characterises a transitory monadnock phase in landscape development [12]. In general, high values of HI indicate that most of the topography is high relative to the mean, such as a smooth surface cut by deeply incised streams. From the mathematical point [11] of view HI is computed by the equation (1). HI Hmean H max = H max H min (1) The cross-correlation between the attributes for the local authorities is computed (Table 1). Table 1. Correlation matrix. lna H LR R G HI lna 1 H 0.4 1 LR 0.57 0.7 1 sd 0.48 0.64 0.95 1 G 0.4 0.75 0.77 0.73 1 HI -0.28 0.38-0.01 0.04 0.23 1 2.4 Geometric signature Cluster analysis is a multivariate procedure which is commonly used for regional classification. It is based on some measurement of distance among objects, (for example Euclidean distance), which is calculated in a c-dimensional space, where c represents the number of attributes used in the clustering process [2]. The centroid method was employed that requires a priori definition of the number of clusters. The results for 5 clusters are shown in Table 2. Table 2. Number of objects, mean object distance and st.dev from cluster centroid, and cluster uniformity [2] Cluster Count mean stdev Uniformity 1 2 0.75 0.00 1.00 2 24 1.37 0.60 0.56 3 46 1.20 0.42 0.65 4 53 1.43 0.55 0.62 5 33 1.42 0.43 0.70 Uniformity expresses cluster compactness and it is computed on the basis of the mean object distance and st.dev. from cluster centroid. The centroid per cluster is given in Table 3. Table 3. Cluster standardized centroids. lna H sd LR G HI 1-4.4-1.4-1.6-1.7-2.0 3.8 2-1.3-1.1-1.1-1.2-1.1-0.3 3 0.2-0.6-0.3-0.2-0.5-0.8 4 0.1 0.6 0.3 0.3 0.6 0.6 5 0.6 1.0 1.5 1.6 1.0-0.2 The graphical representation of standardized cluster centroids (cluster 1 that includes 2 objects was not considered) is given in Fig. 5 while the distance between pairs of cluster that indicates the degree of separability among clusters is presented in Table 4. The centroids represent the terrain signatures of the local authorities in the study area (Fig. 5, Table 3). Fig. 5. Graphic representation of cluster centroids. Table 4. Distances between centroids. Cluster 1 2 3 4 5 1 0 2 5.3 0.0 3 7.0 2.2 0 4 7.0 3.6 2.3 0 5 8.7 5.2 3.5 2.1 0 In order to evaluate the existence of statistically significant differences in the distances among the cluster centers the one-way analysis of variance (ANOVA) was used (Table 5). ANOVA calculates the between-cluster mean square and the withincluster mean square. The ratio of these two mean squares is the usual ANOVA F statistic [2]. The logic of the test argues that for the class (cluster) means to be significantly different, interclass (between cluster) variance must be of much greater magnitude than intraclass (within clusters) variance.
Table 5. Analysis of variance (MS: cluster mean square, DF: degrees of freedom) Attribute MS Error MS DF DF F P lna 23.5 4 0.41 153 57.2 0 H 25.5 4 0.36 153 70.3 0 sd 29.8 4 0.25 153 117.3 0 LR 31.6 4 0.2 153 157.6 0 G 24.8 4 0.38 153 65.3 0 HI 20.1 4 0.51 153 39.3 0 Although the significance levels for the F statistic are below.001, the values of F in the case of cluster analysis should be mostly interpreted in terms of comparing differentiation of the cluster means for individual variables. The spatial distribution of clusters (Fig. 6) indicates that the major part of Peloponnesus is occupied by the clusters 4 and 5 that include the steepest and the most elevated objects (Fig. 5, Table 3). Fig. 6. Spatial distribution of clusters (the cluster 1 was excluded due to its very small occurrence). 2.5 Landcover signature The landcover signature is defined per geometric signatures. So, for the DEM grid points that belong to clusters 2, 3, 4 and 5 (Fig. 6), the frequency per CLC is computed (Table 6). Note that the CLCs 4, 8, 34, 39, 43, and 44 present 0 occurrences in the study area and were excluded. The CLC frequency per cluster is given in Fig. 7. Table 6. The landcover signature per terrain class. Terrain classes (clusters) CLC 1 3 4 5 1 0.07 0.07 0.01 0.08 2 0.87 0.57 0.33 0.47 3 0.04 0.04 0.01 0.06 4 0. 0. 0. 0. 5 0. 0. 0. 0. 6 0.65 0.07 0.02 0. 7 0. 0.04 0.14 0. 8 0. 0. 0. 0. 9 0. 0.01 0.07 0. 10 0. 0.01 0. 0. 11 0.08 0.01 0.01 0. 12 14.29 5.53 0.98 1.33 13 6.96 1.31 0.11 0.28 14 0.07 0. 0.01 0.07 15 6.04 4.15 0.85 2.25 16 4.47 3.77 2.45 2.63 17 13.21 12.2 6.43 5.12 18 0. 0. 0.01 0. 19 0.25 1.09 0.50 0.35 20 15.99 15.3 5.87 4.66 21 13.84 15.1 11.3 9.62 22 1.14 2.39 7.28 6.70 23 1.39 1.49 3.18 2.13 24 0.80 1.20 3.07 2.80 25 1.47 2.87 6.56 8.89 26 1.09 4.15 9.06 6.34 27 1.06 3.27 6.37 7.98 28 9.96 19.0 25.4 25.4 29 3.84 4.26 7.60 8.68 30 0.97 0.18 0.09 0.19 31 0. 0.06 0.45 0.84 32 0.21 0.36 1.36 2.78 33 0.06 0.23 0.21 0.09 34 0. 0. 0. 0. 35 0. 0.17 0. 0.01 36 0. 0. 0.02 0.03 37 0.11 0.09 0. 0.04 38 0. 0.01 0. 0. 39 0. 0. 0. 0. 40 0.20 0.07 0.07 0.02 41 0.13 0.40 0.06 0.06 42 0.24 0.22 0.01 0. 43 0. 0. 0. 0. 44 0. 0. 0. 0. In Fig. 7 cluster CLCs are observed within the interval 11 to 32 in which the occurrence of the CLCs in the study area is maximized (as it can be seen in Fig. 3).
In order to assess the similarity of the 4 vectors (CLCs frequency vectors) presented in Table 6, the correlation coefficient was computed (Table 7). Fig 7. CLCs frequency per cluster. Note that in Table 7, the frequency of the CLCs in the interval 1 to 44 was considered (only the 6 CLCs with zero occurrences in the study area were excluded). Table 7. Correlation coefficient per CLC vector. clust er 2 3 4 5 2 1 3 0.87 1 4 0.52 0.83 1 5 0.48 0.79 0.98 1 The description of the CLCs of level 3 is given in Table 8, in an attempt to assist the interpretation of Table 6 and Fig. 7. Table 8. Level 3 CLCs description Code Description 1 Continuous urban fabric 2 Discontinuous urban fabric 3 Industrial or commercial units 4 Road and rail networks & associated land 5 Port areas 6 Airports 7 Mineral extraction sites 8 Dump sites 9 Construction sites 10 Green urban areas 11 Sport and leisure facilities 12 Non-irrigated arable land 13 Permanently irrigated land 14 Rice fields 15 Vineyards 16 Fruit trees and berry plantations 17 Olive groves 18 Pastures Code Description 19 Annual crops associated with permanent crops 20 Complex cultivation patterns Land principally occupied by agriculture, 21 with significant areas of natural vegetation 22 Agro-forestry areas 23 Broad-leaved forest 24 Coniferous forest 25 Mixed forest 26 Natural grasslands 27 Moors and heathland 28 Sclerophyllous vegetation 29 Transitional woodland-shrub 30 Beaches, dunes, sands 31 Bare rocks 32 Sparsely vegetated areas 33 Burnt areas 34 Glaciers and perpetual snow 35 Inland marshes 36 Peat bogs 37 Salt marshes 38 Salines 39 Intertidal flats 40 Water courses 41 Water bodies 42 Coastal lagoons 43 Estuaries 44 Sea and ocean 49 NODATA 50 Sea and ocean 3 Discussion The parametric representation of local authorities on the basis of terrain attributes and the subsequent cluster analysis identified four terrain classes with significant occurrence (Table 2). The geometric signatures as identified by the cluster centroids (Fig. 5) indicate that the local authorities of cluster 5, 4, 3 and 2 are sorted in order of decreasing mean elevation, roughness, local relief, and slope (Table 3). An analogous conclusion can be derived from the attribute correlation matrix (Table 1). From the vulnerability point of view, local authorities of cluster 5 and 4 are more subject to high intensity of the erosion/deposition processes (high slopes) and receive greater mean annual rainfall height. Additionally the height variability and terrain diversity is greater and so landslide related hazards should be highest for the local authorities of clusters 5 and 4 [8]. The cluster 4 of intermediate mean elevation
presents the highest massiveness while the cluster 5 with highest elevation presents an intermediate HI. The interpretation given is that local authorities developed on mountain sides (cluster 4) are in more youthful erosion cycle in contrast to cluster 5, due to the presence of vertical tectonic movements due to faulting. An analogous conclusion is derived with respect to clusters 3 (greater mean elevation and less HI) and 2 (less mean elevation and higher HI). The size of local authorities decreases with increasing mean elevation, roughness, local relief and slope (Tables 1 and 3), indicating that in flat terrain land economic value and land productivity is higher (thus a smaller in size local authority can be survive from economic, population and administrative point of view). The interpretation of landcover signatures (Table 6, Fig. 7) of the 4 terrain classes on the basis of correlation coefficient (Table 7) indicates that the terrain classes 5 and 4, as well as 3 and 2 are mostly correlated with respect the CLCs. In other words CLCs are terrain dependent since 5 and 4 and 3 and 2 classes are more terrain alike (Fig. 5). A more detailed examination is performed on the basis of Tables 6, 8 and Table 9 (terrain classes on the basis of Corine Level 1). It is seen that the occurrence of artificial surfaces and agricultural areas in local authorities is decreased with increasing mean, elevation slope and terrain variability. The opposite is valid for forest and semi natural areas class. Table 9. Level 1 CLCs for the 4 terrain classes. Level 1 2 3 4 5 Artificial surfaces 1.71 0.81 0.59 0.61 Agricultural areas 76.3 60.9 35.7 33.0 Forest and semi natural areas 20.8 37.1 63.4 66.1 Wetlands 0.11 0.27 0.02 0.07 Water bodies 0.57 0.69 0.14 0.09 4 Conclusion The geometric signatures indicate that the local authorities are sorted in four terrain classes in decreasing mean elevation, roughness, local relief, and slope. The landcover signatures defined per terrain class on the basis of CLCs were proved to be terrain dependent. These signatures might be used in order to determine local authority vulnerability and assist the planning at a local authority level. Acknowledgements: This work was financed by the European Social Fund (ESF), Operational Program for Educational and Vocational Training II (EPEAEK II), and particularly the Program New graduate programs of University of Patras. References: [1] Corine land cover 2000 (CLC2000) 100 m ver. 5, European Environment Agency { EEA, Copenhagen}, 2005, http://dataservice.eea.eu.int/ [2] Mather P., Computer Processing of Remotely- Sensed Images, John Wiley and Sons, 2004 [3] Miliaresis, G., Extraction of bajadas from digital elevation models & satellite imagery, Computers & Geosciences, Vol.27, 2001, No.10, pp. 1157-1167. [4] Miliaresis, G., Argialas, D., Extraction & delineation of alluvial fans from DΕΜs & Landsat TM images, Photogrammetric Engineering & Remote Sensing, Vol.66, No.9, 2000, pp.1093-1101. [5] Miliaresis G., Illiopoulou P., Clustering of Zagros Ranges from the Globe DEM representation, Int. Journal of Applied Earth Observation & GeoInformation, Vol.5, No.1, 2004, pp. 17-28. [6] Miliaresis G., Kokkas N., Segmentation and terrain modeling of extra-terrestrial chasmata, Journal of Spatial Sciences, Vol.49, No.2, 2004, pp. 89-99. [7] Miliaresis G., Paraschou Ch., Vertical accuracy of the SRTM DTED Level 1 of Crete, Int. Journal of Applied Earth Observation & GeoInformation, Vol.7, No.1, 2005, pp. 49-59. [8] Miliaresis G., Sabatakakis N., Koukis G., Terrain pattern recognition and spatial decision making for regional slope stability studies, Natural Resources Research, Vol.14, No.2, 2005, pp. 91-100. [9] Mugnier, C., Grids & datum: the Hellenic Republic, Photogrammetric Engineering & Remote Sensing, Vol.68, No.12, 2002, pp. 1237-38. [10] Petrova, E., Vulnerability of Russian regions to natural risk: experience of a quantitative assessment, Natural Hazards and Earth System Sciences, Vol.6, No.1, 2006, pp.49-54. [11] Pike R., Wilson S., 1971, Elevation-relief ratio, hypsometric integral and geomorphic areaaltitude analysis, Geological Society of America Bulletin, Vol.82, 1971, pp.1079-1084. [12] Strahler A., Hypsometric (area-altitude) analysis of erosional topography, Geological Society of America Bulletin, Vol.63, 1952, pp.1117-1142.