Retrieval of soil moisture using a dynamic learning neural network trained with a 1-dimensional hydrology/radiobrightness model

The authors present a retrieval approach that uses satellite radiobrightness to infer surface parameters of prairie grassland based on a Dynamic Learning Neural Network (DLNN) trained with a 1-dimensional Hydrology/Radiobrightness (1dH/R) model. The parameters of interest include the temperatures and moisture contents of the soil and canopy. To evaluate the feasibility of the retrieval approach, the authors conduct two case studies. The first study is based on products from the 1dH/R model that are used for the model validation. The second study is based on results from a 60-day summer dry-down run of the 1dH/R model with a vegetation coverage of 100%. In each case, they utilize about 95% of the 1dH/R model predictions to train the DLNN and the rest of the predictions are used to evaluate the DLNN retrievals. The training data include horizontally and vertically-polarized, brightnesses at 1.4, 19, and 37 GHz, and the corresponding temperatures and moisture contents of the soil and canopy. In the first study, they find that differences between DLNN retrievals and desired quantities (1dH/R model products) are less than 0.1 Kelvin for the canopy temperature, 1.2 Kelvins for the soil temperature (uppermost 5 mm), 0.0001 kg/m/sup 2/ for the canopy water content, and 1.2% for the soil moisture content (by volume). In the second study the corresponding differences are smaller as expected.