基于RepFNet印度河上游流域积雪覆盖度的反演研究OA

Mapping the fractional snow cover(FSC)is significant for water resource management,especially in the upper reaches of the Indus River,where water resources are highly dependent on alpine snowmelt.This study proposed a RepVGG network-based FSC inversion model,RepFNet,which used FY-4A/AGRI remote sensing data and Landsat8 OLI images for FSC feature extraction.Moreover,this model incorporated upsampling and downsampling modules,an innovative attention mechanism,and a specific loss function,thereby enabling the FSC mapping of the upper reaches of the Indus River at 2 000-m resolution.The mapping results were validated using the MODIS data.The experimental results show that the RepFNet achieved a coefficient of determination of 0.667,a root mean square error(RMSE)of 0.090,a correlation coefficient(r)of 0.890,an explained variance score(EVS)of 0.683,and a Kappa coefficient of 0.468,remarkably outperforming the classical algorithms such as random forest and U-Net.Overall,the RepFNet model demonstrates excellent performance in FSC inversion,offering a novel technical solution for high-accuracy FSC monitoring.