Optimal dataset selection and driving mechanisms of land-use change in the Yellow River basin: integrating PLUS model, geodetector and random forest approaches

Land-use change embodies the complex interactions between human activities and terrestrial ecosystems, with significant implications for ecological security and sustainable development. This study applied the PLUS model to evaluate four commonly used land-use datasets (CLCD, ESA CCI LC, GlobeLand30 and MCD12Q1) in the Yellow River Basin (YRB) from 2000 to 2020. Results show that ESA CCI LC achieved the highest simulation accuracy, with both Kappa and Overall Accuracy exceeding 0.91, and was therefore selected for subsequent analyses. Based on this optimal dataset, we examined land-use dynamics and conducted multi-scenario projections. Expansion outputs from the PLUS model combined with Geodetector analysis revealed distinct drivers: grassland expansion is dominated by interactions between the vegetation index and population, farmland by precipitation and population, and forest expansion by the vegetation index and dry‒wet conditions. Furthermore, a novel application of the random forest model classified land-use transitions into transfer-in and transfer-out levels, enabling the precise identification of dominant drivers. The results indicate that farmland transitions are driven primarily by precipitation and population, forest transitions by the vegetation index and grassland transitions by population pressure. Between 2000 and 2020, constructed land increased by more than 22%, mainly through conversions from farmland and grassland, while grassland remained relatively stable, with a decadal growth rate exceeding 0.3%. By introducing vegetation index and dry–wet conditions into the modelling framework, simulation accuracy and explanatory power were significantly enhanced.