Abstract
Reliable long-term monitoring of ecological quality is essential for assessing progress toward the Sustainable Development Goals (SDGs) in fragile alpine regions, yet remote-sensing assessments often assume that factors explaining spatial patterns also drive temporal change. We tested this assumption in Sanjiangyuan, the headwater region of the Yangtze, Yellow and Lancang–Mekong rivers, using annual Landsat composites (1986–2024) and a modified remote sensing ecological index (MRSEI) with fixed normalization across years. We combined trend and class-centroid analyses with GeoDetector, variation partitioning, spatial-block cross-validation and pixel-level detrended correlations. Regional mean MRSEI increased significantly (0.0245 per decade); improving trends covered 78.1% of the valid area, including 36.6% with statistically significant improvement. A conventional RSEI showed the same direction but a steeper slope, and MRSEI agreed spatially with MODIS LAI and NPP (median r = 0.83 and 0.73). Good- and excellent-class centroids shifted northwestward by 56.08 and 20.51 km. Natural factors uniquely explained 41.81–54.41% of spatial variance and the human footprint only 0.02–0.59%, with negligible out-of-block predictive gain. Temporally, MRSEI anomalies were consistently associated with precipitation and soil moisture over 1986–2024, whereas the air-temperature association was weak and sampling-sensitive, and human-footprint associations were near zero. Spatial dominance therefore does not imply temporal control, and the fixed negative thermal direction of RSEI-type indices warrants testing in cold regions. These findings support the sustainable management of China’s Water Tower through targeted verification of degrading areas, drought-sensitive early warning and the monitoring of land degradation neutrality (SDG 15.3).