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

Abstract

This study explores the use of a support vector machine (SVM) as the observation operator within a passive microwave brightness temperature data assimilation framework (herein SVM-DA) to enhance the characterization of snow water equivalent (SWE) over High Mountain Asia (HMA). A series of synthetic twin experiments were conducted with the NASA Land Information System (LIS) at a number of locations across HMA. Overall, the SVM-DA framework is effective at improving SWE estimates (~70% reduction in RMSE relative to the Open Loop) for SWE depths less than 200 mm during dry snowpack conditions. The SVM-DA framework also improves SWE estimates in deep, wet snow (~45% reduction in RMSE) when snow liquid water is well estimated by the land surface model, but can lead to model degradation when snow liquid water estimates diverge from values used during SVM training. In particular, two key challenges of using the SVM-DA framework were observed over deep, wet snowpacks. First, variations in snow liquid water content dominate the brightness temperature spectral difference (ΔTB) signal associated with emission from a wet snowpack, which can lead to abrupt changes in SWE during the analysis update. Second, the ensemble of SVM-based predictions can collapse (i.e., yield a near-zero standard deviation across the ensemble) when prior estimates of snow are outside the range of snow inputs used during the SVM training procedure. Such a scenario can lead to the presence of spurious error correlations between SWE and ΔTB, and as a consequence, can result in degraded SWE estimates from the analysis update. These degraded analysis updates can be largely mitigated by applying rule-based approaches. For example, restricting the SWE update when the standard deviation of the predicted ΔTB is greater than 0.05 K helps prevent the occurrence of filter divergence. Similarly, adding a thin layer (i.e., 5 mm) of SWE when the synthetic ΔTB is larger than 5 K can improve SVM-DA performance in the presence of a precipitation dry bias. The study demonstrates that a carefully constructed SVM-DA framework cognizant of the inherent limitations of passive microwave-based SWE estimation holds promise for snow mass data assimilation.

Exploration of Synthetic Terrestrial Snow Mass Estimation via Assimilation of AMSR‐E Brightness Temperature Spectral Differences Using the Catchment Land Surface Model and Support Vector Machine Regression

“…Controllability is related to the set of training data and the inability of the SVM to accurately predict snow ΔT b when the given inputs that are outside of the prediction space implicit in the training data (Ahmad et al, 2019). As a result, the model error would no longer correlate back to the corresponding error in the SVM-based observation operator, which can lead to spurious error correlations that ultimately degrade the model estimate (Kwon et al, 2019). To avoid this, prior SWE is updated only when the standard deviation of the prior SVM-predicted ΔT b is greater than 0.05 K based on heuristics outlined in Kwon et al (2019).…”

“…As a result, the model error would no longer correlate back to the corresponding error in the SVM-based observation operator, which can lead to spurious error correlations that ultimately degrade the model estimate (Kwon et al, 2019). To avoid this, prior SWE is updated only when the standard deviation of the prior SVM-predicted ΔT b is greater than 0.05 K based on heuristics outlined in Kwon et al (2019).…”

“…Alternatively, a machine learning technique in the form of physically informed support vector machine (SVM) regression can be employed (Forman & Reichle, 2014; Kwon et al., 2019; Xue & Forman, 2017; Xue et al., 2018). It has been shown that a SVM was able to accurately capture the temporal and spatial variability in the modeled T b or Δ T b , and thus, holds potential to improve snow estimates across large spatial scales.…”

“…Unlike previous works that simultaneously assimilating a fixed number of Δ T b channels (Kwon et al., 2019; Xue et al., 2018), a priori modeled SWE is used as an indicator to determine which Δ T b channels should be assimilated into the model. In addition, a simple “data‐thinning” strategy is also explored to help mitigate high‐frequency error (e.g., changes in snow temperature not related to snow mass) embedded in the synthetic AMSR‐E PMW Δ T b observations.…”

Prediction of Active Microwave Backscatter Over Snow-Covered Terrain Across Western Colorado Using a Land Surface Model and Support Vector Machine Regression

“…Support vector machine (SVM) regression is a supervised machine learning algorithm that maps the input space into higher dimensional feature space using a kernel function [80], [81]. SVM regression has been utilized in hydrological research for spatial pattern recognition [82], [83], classification [44], [84], and temporal prediction [54], [56]- [58], [85], [86]. The study here focuses on predicting C-band backscatter over snow-covered terrain using SVM regression.…”