Identifying the most important spatially distributed variables for explaining land use patterns in a rural lowland catchment in Germany

Land use patterns arise from interactive processes between the physical environment and anthropogenic activities. While land use patterns and the associated explanatory variables have often been analyzed on the large scale, this study aims to determine the most important variables for explaining land use patterns in the 50 km(2) catchment of the Kielstau, Germany, which is dominated by agricultural land use. A set of spatially distributed variables including topography, soil properties, socioeconomic variables, and landscape indices are exploited to set up logistic regression models for the land use map of 2017 with detailed agricultural classes. Spatial validation indicates a reasonable performance as the relative operating characteristic (ROC) ranges between 0.73 and 0.97 for all land use classes except for corn (ROC = 0.68). The robustness of the models in time is confirmed by the temporal validation for which the ROC values are on the same level (maximum deviation 0.1). Non-agricultural land use is generally better explained than agricultural land use. The most important variables are the share of drained area, distance to protected areas, population density, and patch fractal dimension. These variables can either be linked to agriculture or the river course of the Kielstau.