Revealing causes of a surprising correlation: snow water equivalent and spatial statistics from Calibrated Enhanced-Resolution Brightness Temperatures (CETB) using interpretable machine learning and SHAP analysis

Seasonal snowpack is a crucial water resource, making accurate Snow Water Equivalent (SWE) estimation essential for water management and environmental assessment. This study introduces a novel approach to Passive Microwave (PMW) SWE estimation, leveraging the strong, unexpected correlation between SWE and the Spatial Standard Deviation (SSD) of PMW Calibrated Enhanced-Resolution Brightness Temperatures (CETB). By integrating spatial statistics, linear correlation, machine learning (Linear Regression, Random Forest, GBoost, and XGBoost), and SHapley Additive exPlanations (SHAP) analysis, this research evaluates CETB SSD as a key feature to improve SWE estimations or other environmental retrievals by investigating environmental drivers of CETB SSD. Analysis at three sites—Monument Creek, AK; Mud Flat, ID; and Jones Pass, CO—reveals site-specific SSD variability, showing correlations of 0.64, 0.82, and 0.72 with SNOTEL SWE, and 0.67, 0.89, and 0.67 with PMW-derived SWE, respectively. Among the sites, Monument Creek exhibits the highest ML model accuracy, with Random Forest and XGBoost achieving test R 2 values of 0.89 and RMSEs ranging from 0.37 to 0.39 [K] when predicting CETB SSD. SHAP analysis highlights SWE as the driver of CETB SSD at Monument Creek and Mud Flat, while soil moisture plays a larger role at Jones Pass. In snow-dominated regions with less surface heterogeneity, such as Monument Creek, SSDs can improve SWE estimation by capturing snow spatial variability. In complex environments like Jones Pass, SSDs aid SWE retrievals by accounting for factors such as soil moisture that impact snowpack dynamics. PMW SSDs can enhance remote sensing capabilities for snow and environmental research across diverse environments, benefiting hydrological modeling and water resource management.