Point-to-Surface Upscaling Algorithms for Snow Depth Ground Observations

To validate the accuracy of snow depth products retrieved from passive microwave remote sensing data with a high confidence level, verification method based on points ground observation is subject to great uncertainty, due scale effect. Thus, it necessary use point-to-surface transformation obtain relative truth at pixel scale. In this study, by using observations different scales, upscaling methods are conducted simple average (SA), geostatistical, Bayes maximum entropy (BME), and random forest (RF) algorithms. addition, cross-validation leave-one-out employed results. The results show that SA algorithm seriously inadequate for estimating variation in space, only suitable regions relatively flat terrain small depth. BME can introduce prior knowledge perform kernel smoothing observed data, result superior geostatistical RF algorithms, especially when insufficient, outliers appear. study expected provide reference developing observations, further solve uncertainties caused effects other land surface parameter inversion validation, data.