Pixel-level flood loss assessment driven by a consistent high-resolution land-cover base map: Integrating semantic segmentation, hydrodynamic simulation, and ecosystem service valuation

Study region This study examines a typical piedmont depression in Xushui District, Baoding City, on the North China Plain, where fragmented land cover and mixed exposed elements require refined regional flood loss assessment. Study focus Regional flood loss assessment is often limited by inconsistent land cover datasets used for hydrodynamic parameterization, exposure mapping, and loss estimation. We propose a pixel-level framework driven by a consistent 0.8 m, 16-class land cover base map. Four EfficientNetV2-S-enhanced semantic segmentation models generated the map and supported unified assignment of roughness, economic value, ecosystem service value, and vulnerability, ensuring semantic, parametric, and scale consistency. Coupled with two-dimensional hydrodynamic simulations, the framework estimates flood intensity, exposure, direct economic losses, and environmental damage. E-UNet performed best, with mIoU of 76.83% and mPA of 89.64%. New hydrological insights for the region Land cover resolution, class definition, and temporal consistency affect inundation patterns, exposed assets, and direct economic loss and environmental damage estimates, while amplifying the influence of built-up areas and small water bodies on loss composition. Under two extreme flood scenarios, inundated areas from the proposed map differed from contemporaneous Esri-LC24 by 3.0%–5.0%. In Case 1, direct economic losses reached CNY 1.147–1.684 billion, and monetized environmental damage reached CNY 123–177 million, accounting for 8%–10% of total losses under the adopted ESV coefficients and Fusion map assumptions. The framework provides an updatable, transferable, and spatially consistent method for regional flood risk assessment.