Diagnostic Analysis of a Data Assimilation Framework for Improving Snow Mass Estimation in Complex Terrain

Contemporary data assimilation (DA) systems tend to violate the underlying Gaussianity, unbiasedness, and error independence assumptions related to model and observation errors. An innovation is the difference between the observed measurement and the model predicted measurement. In this study, we use an innovation analysis to diagnose assumption violations within the DA system. An Ensemble Kalman Filter is used to assimilate brightness temperature (spectral difference) observations, using well trained support vector machines as the observation operator, within a land surface model to improve terrestrial snow water equivalent estimates over complex terrain. Divergence of the normalized innovations from the standard normal distribution highlighted the effect of bias in the model forcings on assimilated estimates. Results indicated that, in general, the Advanced Microwave Scanning Radiometer-2 brightness temperature (spectral difference) error standard deviation as related to SWE over complex terrain is better represented by 6 Kelvin. This value is larger than previously published values and is justified by the complexity of the topography in the study domain.