A Non-Sigmoidal-Curve-Dependent Dynamic Threshold Method Improves Precipitation Phase Partitioning in the Northern Hemisphere

Abstract

Given the significant impact of precipitation phase transitions on water and energy balances, accurate phase partitioning is essential for hydrological modeling. Many commonly used precipitation phase partitioning methods (PPMs) rely on sigmoidal curve assumptions to determine thresholds, leading to biased partitioning results. Here we developed a non-sigmoidal-curve-dependent dynamic threshold method (NSDT) to establish time-varying and spatially varying thresholds for classifying precipitation into rain, snow, and sleet in the Northern Hemisphere. The NSDT avoids curve-fitting errors by directly calculating thresholds from snowfall and rainfall frequency curves. In this method, relative humidity and elevation are the two most influential variables to precipitation phase, and single-threshold and dual-threshold strategies are employed separately across different relative humidity ranges. The results show that station thresholds derived from NSDT have marked spatial variability. Furthermore, the NSDT performs well and robustly, with accuracy exceeding 80% over the wet-bulb temperature range [−10°C, 10°C] at each elevation range, relative humidity subinterval, and sub-time period. The NSDT outperforms six commonly used PPMs, especially at high elevations. Regarding the wet-bulb temperature range of [−4°C, 4°C], NSDT exhibits accuracy improvements ranging from 1.0% to 11.8% (0.4%–14.5%) across all elevation (relative humidity) subintervals compared to other PPMs. Overall, the NSDT method developed herein improves precipitation phase partitioning, which is expected to enhance the simulation accuracy of land surface models and hydrological models and provide a theoretical basis for a more accurate understanding of hydrological processes.