Joint State and Parameter Estimation of Two Land Surface Models Using the Ensemble Kalman Filter and Particle Filter
Other literature type
Hendricks Franssen, Harrie-Jan
Land surface models (LSMs) contain a suite of different parameters and state variables to resolve the water and energy balance at the soil-atmosphere interface. Many of the parameters of these models cannot be measured directly in the field, and require calibration against flux and soil moisture data. In this paper, we use the Variable Infiltration Capacity Hydrologic Model (VIC) and the Community Land Model (CLM) to simulate temporal variations in soil moisture content at 5, 20 and 50 cm depth in the Rollesbroich experimental watershed in Germany. Four different data assimilation (DA) methods are used to jointly estimate the spatially distributed water content values, and hydraulic and/or thermal properties of the resolved soil domain. This includes the Ensemble Kalman Filter (EnKF) using state augmentation or dual estimation, the Residual Resampling Particle Filter (RRPF) and Markov chain Monte Carlo Particle Filter (MCMCPF). These four DA methods are tuned and calibrated for a five month data period, and subsequently evaluated for another five month period. Our results show that all the different DA methods improve the fit of the VIC and CLM model to the observed water content data, particularly if the maximum baseflow velocity (VIC), soil hydraulic (VIC) properties and/or soil texture (CLM) are jointly estimated along with the model states. In the evaluation period, the augmentation and dual estimation method performed slightly better than RRPF and MCMCPF. The differences in simulated soil moisture values between the CLM and VIC model were larger than variations among the data assimilation algorithms. The best performance for the Rollesbroich site was observed for the CLM model. The strong underestimation of the soil moisture values of the third VIC-layer are likely explained by an inadequate parameterization of groundwater drainage.