The effect of land cover changes on annual peak discharge is evaluated using a novel method combining principal component analysis (PCA) and random forest regression (RF). This approach calculates original feature contributions by multiplying absolute PCA loadings with RF-derived component importances, enabling direct attribution of discharge variations to specific land cover types in each subbasin, which represents an integration unexplored in prior PCA–RF applications. The PCA–RF method is applied in the basin upstream of Golestan Dam, Iran. It uses multitemporal Landsat imagery for NDVI-based land cover classification, annual instantaneous peak discharge data from hydrometric stations, and elevation data for subbasin delineation over a limited period (post-2003, due to data constraints). Despite limitations like scarce hydrometric data and a short collection period, the method analyzes relationships between land cover classes and floods in subbasins. Remote sensing provides...