Retrieval of semi-arid region parameters from passive microwave measurements using neural networks within a Bayesian framework

The problem of recovering geophysical parameters from microwave measurements is examined. A multilayer perceptron neural network was built, and then trained with data produced from a passive radiative transfer model. The model is applicable to semi-arid regions, and produces dual-polarised temperatures for frequencies of 6.6, 10.7, and 37 GHz from specification of soil moisture, soil temperature, and vegetation moisture. Synthetic semi-arid terrains, characterized by physically realizable moisture and temperature values, were created and used to test the inversion process. The model generated brightness temperatures for these terrains were inverted by the trained neural network within the Bayesian framework. Microwave brightness temperatures obtained from the Seasat Scanning Multichannel Microwave Radiometer (SMMR) over the Sahara Desert were also inverted. The values of soil moisture, temperature and vegetation moisture recovered from the inversion produced contours that follow expected trends for that region