Disaggrregation of remotely sensed soil moisture using neural networks

Currently hydrological models are being developed that can be used to predict soil moisture conditions. However, these models have the same tendencies as meteorological simulations: they suffer from drift due to nonlinearities in the dynamic system being modeled and due to roundoff errors in the computer hardware. We want use remotely sensed information to update the hydrological model. In order to sufficiently penetrate the soil to yield any useful information about the soil moisture of all put the very surface layer (< I cm) we need to choose from long wavelength microwave bands. Given the finite aperture of the antennas, this gives us a very low resolution (each pixel is 10s of km on each side). The problem we need to solve is how to match the low spatial resolution of the microwave sensor with the high resolution of the hydrological model. We have developed an Artificial Neural Network that is input the low-resolution remote sensor data along with information about the soil type, vegetation, and precipitation history at high resolution. The output is soil moisture information at high resolution. We can then use a Kalman filter to update the hydrological model.