Estimating Snow Mass in North America Through Assimilation of Advanced Microwave Scanning Radiometer Brightness Temperature Observations Using the Catchment Land Surface Model and Support Vector Machines

Abstract

To estimate snow mass across North America, brightness temperature observations collected by the Advanced Microwave Scanning Radiometer (AMSR‐E) from 2002 to 2011 were assimilated into the Catchment model using a support vector machine as the observation operator and a one‐dimensional ensemble Kalman filter. The performance of the assimilation system is evaluated through comparisons against ground‐based measurements and reference snow products. In general, there are no statistically significant skill differences between the domain‐averaged, model‐only (open loop, or OL) snow estimates and assimilation estimates. The assessment of improvements (or degradations) in snow estimates is difficult because of limitations in the measurements (or products) used for evaluation. It is found that assimilation estimates agree slightly better in terms of root‐mean‐square error and Nash‐Sutcliffe model efficiency with ground‐based snow depth measurements than OL estimates in 82% (56 out of 62) of pixels that are colocated with at least two ground‐based stations. Assimilation estimates tend to agree slightly better in terms of mean difference with reference snow products over tundra snow, alpine snow, maritime snow, and sparsely vegetated, snow‐covered pixels. Changes in snow mass via assimilation translate into improvements (e.g., by 22% on average in terms of root‐mean‐square error, relative to OL) in cumulative runoff estimates when compared against discharge measurements in 11 out of 13 snow‐dominated basins in Alaska. These results suggest that a support vector machine can potentially serve as an effective observation operator for snow mass estimation within a radiance assimilation system, but a better observational baseline is required to document a statistically significant improvement.

Application of machine learning techniques for regional bias correction of snow water equivalent estimates in Ontario, Canada

“…Linear regression and machine learning techniques have previously been used effectively across the geosciences for bias correction of global and regional climate model output (Teutschbein and Seibert, 2012;Li et al, 2010;Lary et al, 2009;Reichstein et al, 2019;Shen, 2018). Previous studies on the estimation of North American SWE using artificial neural networks and support vector machines also exhibit similar results, with machine learning techniques outperforming general linear models (Snauffer et al, 2018;Xue et al, 2018). However, recent work by Dixon et al (2016) and Ehret et al (2012) suggests that bias-correction methods have their own associated uncertainties which must be considered when applied to datasets like SNODAS.…”

“…Neural networks and support vector machines have also been effectively implemented for the purpose of bias correction in the geosciences (Lary et al, 2009). A paper by Xue et al (2018) also found that machine learning methods can act as effective operators at estimating North American snow mass. It is possible that these other machine learning techniques may offer further improvements to the methods examined here, and should be considered in additional follow-up work.…”

Exploring the Utility of Machine Learning-Based Passive Microwave Brightness Temperature Data Assimilation over Terrestrial Snow in High Mountain Asia

“…The synthetic observations of ∆T B between the 10.65 and 36.5 GHz (V and H) channels and between the 18.7 and 36.5 GHz (V and H) channels were simultaneously assimilated in the SVM-DA framework. The standard deviation of the ∆T B error was assumed to be 3 K based on Kwon et al [16], Xue et al [23], and Durand and Margulis [77]. Note that only SWE was directly updated during the assimilation and that other snow states such as snow ice content, snow liquid water content, and snow depth were adjusted based on the amount of SWE update to maintain physical consistency between snow properties in the prior and posterior ensembles from Noah-MP.…”

“…Through sensitivity experiments, Xue and Forman [22] supported the findings of Forman and Reichle [21] with respect to the use of SVMs within a DA framework. More recently, Xue et al [23] conducted T B spectral difference assimilation over snow-covered areas in North America using a well-trained SVM and presented promising initial results. Building on these prior efforts, we adopted the use of SVM regression algorithm in this study, but applying it to a different land surface model using a different set of boundary conditions in a different part of the globe where the accurate estimation of snow is very challenging.…”

Analyzing Machine Learning Predictions of Passive Microwave Brightness Temperature Spectral Difference Over Snow-Covered Terrain in High Mountain Asia

“…However, SWE sensitivity varies spatially and temporally. Recent studies have utilized SVM as an observation operator within data assimilation frameworks (Forman and Xue, 2017;Xue and Forman, 2017;Xue et al, 2018). If such a methodology is performed over HMA, then it is expected that the utilization of SVM within a data assimilation framework would benefit those areas the most that have high sensitivity to SWE or high sensitivity to other geophysical variables that have high crosscorrelation with SWE (such as snow depth and snow liquid water content).…”