Fusion of remotely sensed data for soil moisture estimation using relevance vector and support vector machines

摘要: A data assimilation DA methodology that uses two state-of-the-art techniques, relevance vector machines RVMs and support SVMs, is applied to retrieve surface 0–6 cm soil moisture content SMC at a depth of 30 cm. SVMs are known for their robustness, efficiency sparseness provide statistically sound approach solve inverse problems thus build statistical models. Here, we model produces acceptable estimations by using inexpensive readily available data. The study area this research the Walnut Creek watershed in Ames, south-central Iowa, USA. were obtained from Soil Moisture Experiments 2002 SMEX02 conducted Iowa. combines remotely sensed inputs with field measurements, crop physiological characteristics, temperature, water-holding capacity meteorological two-step estimate both i.e. SVMs. First, RVM used retrieves SMC. This information serves as boundary condition second step model, which estimates An exactly similar routine followed an SVM estimation results models compared statistics show perform better root mean square error RMSE =  0.014 m3 m−3 when 0.017 reduced computational complexity more suitable real-time implementation. Cross-validation techniques optimize model. Bootstrapping check over/under-fitting uncertainty estimates. Computations good agreement actual measurements coefficients determination R 2 equal 0.92 0.88. Statistics indicate generalization capability indexes IoAs 0.97 0.96.