Evaluating the Performance of Land Use Products in Mountainous Regions: A Case Study in the Wumeng Mountain Area, China

Mountainous regions with complex terrain often generate mixed pixels, reducing the accuracy of land use classification. This study evaluates the applicability and performance of eight commonly used Land Use Products (LUPs) with resolutions from 10 m to 1000 m in a typical mountainous area of the Wumeng Mountains, using manually vectorized high-resolution imagery as reference data. The spatial patterns of classification inconsistency and topographic influences were systematically analyzed. Results show the following: (1) All LUPs underestimate Cropland, Forest, Built-up areas, and Waters, while overestimating Grassland. Most products perform poorly, with an Overall Accuracy (OA) below 0.7. (2) Classification performance varies by land cover type and location. In complex terrains, high fragmentation causes mixed pixels, leading to inconsistency. (3) XGBoost and SHAP analyses indicate elevation and Terrain Surface Complexity as dominant factors. Accuracy increases with elevation and is higher on shady than on sunny slopes. Greater surface complexity reduces performance. (4) Spatial resolution alone does not determine performance—algorithm design, sample distribution, and data fusion are critical. All eight LUPs show poor applicability in mountainous regions. Future research should emphasize terrain-stratified sample balancing, terrain-aware algorithms, and strategies to mitigate mixed pixel effects.