Evaluating the impact of parameter estimation on data assimilation performance: A case study for soil moisture simulation

Data assimilation provides a way to improve deterministic model accuracy by combining model predictions with observations. The increasing availability of remotely sensed land surface variables has led to the increased need of data assimilation methods for hydrologic applications. The successful use of data assimilation methods, however, is dependent on unbiased model state predictions. Therefore, the presence of bias errors must be separately addressed for the success of a data assimilation system. Here we explore the use of parameter estimation techniques to reduce the inherent biases in model predictions. In this study, we employ the newly developed optimization infrastructure in the NASA Land Information System (LIS), which includes a suite of optimization algorithms ranging from gradient-based methods such as Levenberg-Marquardt (MINPACK) to global search methods such as a Genetic Algorithm (GA) and the Shuffled Complex Evolution Algorithm (SCE-UA). The LIS framework also includes a comprehensive sequential data assimilation infrastructure that can employ multiple land surface models and multiple observations with advanced techniques such as an Ensemble Kalman Filter (EnKF), in an interoperable manner. We employ this unique system to conduct a suite of Observing System Simulation Experiments (OSSEs) for the simulation of surface soil moisture. Here we focus on soil moisture given the large interest in the topic as a result of the Soil Moisture Active Passive (SMAP) mission planned for launch in March 2013. The experiments will focus on quantifying the added value of estimating model parameters prior to data assimilation using the suite of optimization algorithms in LIS.