Predicting 21st century global agricultural land use with a spatially and temporally explicit regression-based model

The extensive alteration of the earth's land cover during the anthropocene had widespread, and in some cases unknown, effects on terrestrial and atmospheric conditions and processes. Predicting future changes to the earth system therefore mandate a future-predicting framework of land use dynamics. However while future-predicting earth surface and atmospheric models tend to explicitly incorporate projected climatic conditions they all but ignore or overly simplify land use dynamics. As most surface and atmosphere dynamics models use gridded input datasets, and land use is a highly spatially-dynamic phenomena, a need clearly arise for spatially explicit representation of future land use dynamics. While a number of such datasets exists at regional and country scales, no fully gridded future-predicting global land use model and database has been reported to date. Here we present the Global Land Use Dynamics Model (GLUDM), a gridded and temporally explicit agricultural land use predictor. GLUDM calculates the relative area of a land use category (e.g. cropland) in each grid-cell by generating unique regression coefficients in each grid-cell based on local historic trends and global population dynamics. Spatial expansions or abandonment of agricultural land is simulated by propagating excesses or deficiencies in agricultural areas between neighboring grid-cells. This spatial connectivity is restricted by topographic, latitudinal and urban characteristics. A validation analysis shows that GLUDM corresponds well to observed land use distribution. GLUDM-predicted global cropland area dynamics between 2005 and 2100 are described herein. Globally, 18% increase in cropland area is predicted between 2005 and 2050 which corresponds very well to previous estimations. Following 2050, a general decrease in cropland area is predicted. The results reveal new insights about global cropland dynamics, demonstrating, for example, that changes in its spatial distribution will be highly heterogeneous, at both micro and macro scales, in some locations worldwide.