State and bias estimation for soil moisture profiles by an ensemble Kalman filter: Effect of assimilation depth and frequency

An ensemble Kalman filter for state estimation and a bias estimation algorithm were applied to estimate individual soil moisture profiles in a small corn field with the CLM2.0 model through the assimilation of measurements from capacitance probes. Both without and with inclusion of bias correction, the effect of the assimilation frequency, the assimilation depth, and the number of observations assimilated per profile were studied. Assimilation of complete profiles had the highest impact on deeper soil layers, and the optimal assimilation frequency was about 1–2 weeks, if bias correction was applied. The optimal assimilation depth depended on the calibration results. Assimilation in the surface layer had typically less impact than assimilation in other layers. Through bias correction the soil moisture estimate greatly improved. In general, the correct propagation of the innovations for both the bias‐blind state and bias filtering from any layer to other layers was insufficient. The approximate estimation of the a priori (bias) error covariance and the choice of a zero‐initialized persistent bias model made it impossible to estimate the bias in layers for which no observations were available.