Development and Preliminary Validation of an EnKF-Like Image Assimilation System for the Common Land Model

Abstract. Accurate representation of both the location and magnitude of soil moisture anomalies in land surface model initial conditions is crucial for simulating land–atmosphere interactions. However, traditional point-based land data assimilation methods primarily adjust anomaly magnitude, with limited capability to improve spatial structure due to the single-column design of most land surface models. This study develops an assimilation approach that optimizes the spatial structure of soil moisture. For the Common Land Model (CoLM), soil moisture fields are treated as images, and a curvelet transform is introduced as the image observation operator. Ensemble methods are used to dynamically estimate errors in the image structure, and the background field is updated in image space using a Kalman filter framework, forming an EnKF-like land surface image assimilation system. Assimilation experiments show that this system effectively exploits the multi-scale spatial information contained in observations, improving soil moisture spatial patterns while reducing magnitude errors. After assimilation, the spatial correlation of surface soil moisture with GLDAS increases from 0.4 to 0.8, and the unbiased RMSE decreases from 0.12 to 0.06 m³/m³. Through vertical propagation, correlations rise from 0.35 to 0.55 at 10–40 cm and from 0.25 to 0.4 at 40–100 cm. Independent validation using in-situ stations shows correlation increases from 0.153 to 0.425 in China and from 0.142 to 0.504 in the United States. These results highlight the potential of the proposed system to improve land surface initial fields and strengthen weather and climate predictions.