Soil moisture retrieval using SMOS brightness temperatures and a neural network trained on in situ measurements

An algorithm using in situ measurements for training a neural network (NN) to retrieve soil moisture (SM) from SMOS observations is discussed. The in situ data are measurements of the SM content in the 0-5 cm depth layer from the SCAN, SNOTEL and USCRN networks. It is shown that this approach can be used to retrieve SM at continental scale in North America. The NN retrieval (NNinSitu) is evaluated against in situ data not used during the training phase and against maps of the SMOS level 3 SM product and ECMWF SM models. NNinSitu SM values are closer to ECMWF values for wet areas. A method to use NNs as a tool to classify in situ sites representative of the remote sensing observations scale is briefly discussed.