Snow Depth Inversion Based on Simulated Pixel-Scale Ground Measurements

The absence of pixel-scale ground measurements presents a notable challenge in the creation and validation of passive microwave snow depth (SD) inversion models, due to the huge scale mismatch between ground measurements and remotely sensed pixels. The coarse spatial resolution of passive microwave remote sensing products further complicates the accurate representation of detailed SD information in space, particularly in mountainous regions. In this study, pixel-scale SD values were generated using regression kriging (RK) and simple averaging (SA) point-to-surface upscaling models, and their impact on the two SD downscaling inversion models, the Chang’s and best subset regression models were evaluated, respectively. The results indicate that the RK model exhibits superior upscaling accuracy and alignment with observed SD, yielding root-mean-square error (RMSE) and mean absolute error (MAE) values of 1.75 and 1.41 cm, respectively. The performance of the SA model is affected by the observed SD thickness within the snow quadrat, with RMSE and MAE values of 2.35 and 1.92 cm, respectively. However, the accuracy of the pixel-scale ground measurements does not significantly influence the accuracy of the Chang’s algorithm. Nevertheless, the best subset regression model, utilizing ascending brightness temperature data and the simulated pixel-scale ground SD measurements, achieves the ideal accuracy and suitability for complex terrain areas, with RMSE and MAE values of 2.04 and 1.53 cm, respectively. This study is expected to provide valuable insights for developing point-to-surface upscaling strategies and addressing scale effect challenges in SD inversion.