Mitigating baseline contamination in satellite-derived water frequencies for robust river shrinking assessment under recurrent droughts

Climate change is increasingly threatening global inland freshwater availability, driving more frequent and severe hydrological droughts. Monitoring spatial variations in surface water dynamics is essential for sustainable water resource management. Time series of satellite optical multi-spectral imagery at high spatial resolution, such as Sentinel-2 MSI, enable quantitative tracking of river shrinking by deriving percentage water content (PWC) through a fuzzy transformation of the normalized difference water index (NDWI) to compute monthly water frequency anomalies (MWFAs). However, establishing accurate reference baselines (climatology) remains challenging under climate change conditions: including previous drought years in standard multi-year mean calculations skews the baseline, artificially making subsequent drought years appear less dry. One option is to use a climatological reference baseline that covers only a limited number of the years analysed. To resolve the reference baseline limitation, this study evaluates and compares various robust statistical methods, such as statistical trimmers, Winsorized means, and median-based estimators, to compute river climatology while filtering out past extreme low-flow events. Applied to dense Sentinel-2 time series across dynamic river reaches, statistical adjustments successfully isolated multi-year drought anomalies from baseline estimations. Comparative performance results demonstrate that standard averaging underestimates surface water loss during recurring drought conditions, whereas trimmed climatologies preserve an uncorrupted reference of normal hydrological conditions, accurately capturing the true magnitude of channel narrowing and river shrinking. By mitigating baseline contamination caused by prior extreme events, this approach substantially improves the sensitivity of satellite-based methodology for quantitative drought anomaly detection and the tracking of river shrinking. Ultimately, integrating robust statistical estimators into remote sensing workflows enhances long-term hydromorphological monitoring and supports more effective climate adaptation and water management strategies, offering a general framework adaptable to other optical satellite imagery at high spatial resolution.