Predictive modelling of historical and recent land-use patterns

Land-use is constrained both by physical and socio-economic factors. The aim of this study is to evaluate how and to what extent physical factors control the land-use pattern in a Central European rural landscape. Moreover, it intends to compare different temporal situations in order to find out, whether the relations between physical features (geomorphology, geology) and the land-use pattern have changed during the last 150 years. To achieve this, logistic regression was applied to model two temporal states (1850 and 2000) of the land-use pattern. The statistical models were implemented in a raster GIS to obtain spatially explicit prediction maps. The model performance was evaluated using the AUC value and the Kappa index. In a first step, models for the single land-use types were calculated including several physical parameters as independent variables. This was done for the recent situation as well as for the situation in 1850. In a second step historical land-use and usability parameters were added as predictors for the present landuse pattern. The study area, located in northern Hesse, Germany, experienced several land-use transitions over the past 150 years. The most significant was the conversion of cropland into grassland or meadow orchards, respectively. Others were afforestation of common pastures and the expansion of residential, industrial and traffic route areas. The modelling results show that the land use pattern of the study area is fundamentally determined by physical factors. The degree of determination is at present lower than it was 150 years ago. Models comprising the additional parameters, e. g. historical land-use, show improved prediction accuracy. By comparing the single land-use type models valuable information is gained for a better understanding of land-use changes and the underlying processes.