The Impact of Assumed Error Variances on Surface Soil Moisture and Snow Depth Hydrologic Data Assimilation

Abstract

Accurate knowledge of antecedent soil moisture (SM) and snow depth (SD) conditions is often important for obtaining reliable hydrological simulations of stream flow. Data assimilation (DA) methods can be used to integrate remotely sensed (RS) SM and SD retrievals into a hydrology model and improve such simulations. In this paper, we examine the impact of assumed model and observation error variance on stream flow estimates obtained by assimilating RS SM and SD data into a lumped hydrological model. The analysis is based on both synthetic and real DA experiments conducted within the Tuotuo River watershed at the headwaters of the Yangtze River. Synthetic experiments demonstrate that, when the true model error variance is small, DA is more sensitive to the overestimation of the error variance than to its underestimation. Conversely, if the true variance is large, DA is sensitive to the assumed model error variance but not the underestimation of the observation error variance. Given this sensitivity, the maximum a posteriori (MAP) estimation method is applied to accurately estimate model and observation error variances. In general, MAP is able to identify model and observation error parameters associated with an accurate stream flow analysis. However, its utility is somewhat limited by equifinality with regard to observation error statistics.