Land cover spatial analysis using geostatistical tools on remote sensing data.

In order to characterize the spatial pattern of different land cover types (TCS) we applied geostatistical analysis on a land cover map obtained from a classification of a satellite image of a sector of the Buenos Aires province. The map included 7 land cover types (TCS): high cover grassland (PAC), low cover grassland (PBC), agriculture (CUL), wetland vegetation and/or water (VPA), forest (MON), humid praire (PRH) and urban areas (URB). We evaluated the spatial components of each TCS on the basis of omnidirectional experimental variograms, adjusted to specific models. Due to the high heterogeneity of the TCS we used a strategy based on sampling the thematic image with different degree of resolution and extension. The spatial distribution of the TCS in the image is irregular but no random; while the size of each TCS is quite stable but different among each other. The models that best adjust the variograms are different depending on the TCS: exponential for PBC, PAC, PRH and CUL; spherical for URB, random for VPA and exponential or random for MON. The geostatistical parameters for the model calculated for each TCS, show structural differences in the spatial distribution at a patch (non explained proportion of the model), landscape (model) and regional scale (differences among sectors), which may be related to land use. The use of geostatistical analysis on information derived from satellite images allows the identification and differentiation of the TCS’s spatial pattern, and may be used to develop strategies for the analysis and monitoring of land use.

Key words: variogram, spatial pattern, remote sensing, land use, environmental assessment