Incorporating landscape patterns can improve global-scale models of land use and land cover change

Environmental and socioeconomic variables that are used to predict land use and land cover (LULC) change in most global-scale land use models miss some LULC change processes, including cropland expansion in frontier areas. We tested whether incorporating landscape metrics, which measure the spatial arrangement of landscapes and vary according to complex anthropogenic and environmental factors, in global- and continental-scale machine learning models of LULC change from 1992 to 2020 improved model performance when compared to socioeconomic and biophysical variables. Two types of LULC change were modelled: cropland expansion and forest loss. Including landscape metrics always improved the accuracy of cropland expansion models but enhanced the performance of forest loss models for only coarser resolution pixels of 80 × 80 km. Therefore, landscape metrics may provide additional insights for modelling LULC change beyond socioeconomic and biophysical variables, but further research is needed to assess the impact of spatial scale on their effectiveness.