Data assimilation using support vector machines and ensemble Kalman filter for multi-layer soil moisture prediction

Abstract

Hybrid data assimilation (DA) is a method seeing more use in recent hydrology and water resources research. In this study, a DA method coupled with the support vector machines (SVMs) and the ensemble Kalman filter (EnKF) technology was used for the prediction of soil moisture in different soil layers: 0–5 cm, 30 cm, 50 cm, 100 cm, 200 cm, and 300 cm. The SVM methodology was first used to train the ground measurements of soil moisture and meteorological parameters from the Meilin study area, in East China, to construct soil moisture statistical prediction models. Subsequent observations and their statistics were used for predictions, with two approaches: the SVM predictor and the SVM-EnKF model made by coupling the SVM model with the EnKF technique using the DA method. Validation results showed that the proposed SVM-EnKF model can improve the prediction results of soil moisture in different layers, from the surface to the root zone.