Adaptive Thresholding (AT) for snowmelt detection with Calibrated Enhanced-Resolution Brightness Temperatures (CETB): Timing and regional patterns for case study of Alaska

Accurate detection of snowmelt timing is critical for understanding hydrologic processes and monitoring environmental changes in heterogeneous snow-dominated regions. This study introduces a new data-driven framework, termed Adaptive Thresholding (AT), which optimizes enhanced-resolution passive microwave brightness temperature (Tb) thresholds from the 37 GHz vertically polarized SSM/I sensor (CETB dataset) and diurnal amplitude variation (DAV) thresholds for snowmelt detection. The approach integrates Tb histogram analysis and DAV time series to identify melt transitions across diverse snow-covered environments. Across a regional domain surrounding Fairbanks, Alaska, our results show that the optimized Tb and DAV thresholds are generally consistent with commonly used legacy values, while the AT method yields improved performance across varying snow conditions. Using AT with these optimized thresholds, we evaluated melt onset date (MOD) estimates under seven thresholding scenarios, ranging from fixed legacy values to fully adaptive configurations, across nine monitoring sites in Interior Alaska from 2003 to 2007. Validation against reference data showed that AT achieved the highest accuracy, with mean absolute errors (MAE) as low as ≤ 1.0 day and the lowest standard deviations across sites, in contrast to legacy methods, which produced MAE values exceeding 6–8 days at several locations. Results also revealed that MOD estimates varied with snow class and terrain, representing the influence of local conditions on melt timing. Comparative MOD maps demonstrated that optimized thresholds captured spatial melt gradients more realistically than legacy methods. These results reveal the advantages of adaptive, physically interpretable thresholding for remote sensing of snowmelt in heterogeneous terrain and support its application to large-scale monitoring systems.