A Neural Network-based disaggregation model, called DisaggNet, has been developed to address the feasibility of disaggregating low-resolution microwave remote sensing data to estimate soil moisture and to estimate associated errors as a function of the data resolution. DisaggNet was tested using microwave brightness temperature (TB) data from the Electronically Steered Thinned Array Radiometer (ESTAR), an aircraft-based instrument operating a 1.4 GHz that was flown on a near-daily basis over the experimental domain of the Southern Great Plains 1997 Hydrology Experiment. The method has been tested by first degrading the resolution of the ESTAR TB data from the original 800 m to lower resolutions by simple arithmetic averaging. DisaggNet combines the degraded TB data with higher-resolution precipitation, soil, vegetation, and topographic properties to estimate the spatial distribution of soil moisture at 800 m resolution. Through separate training procedures, the DisaggNet has been trained to estimate either ESTAR- or model-derived near-surface soil moisture. Performance of DisaggNet was evaluated by comparing these estimates with soil moisture estimates derived from the hydrology model or from the full-resolution ESTAR TB data. Testing was performed at five degraded resolutions (1.6, 3.2, 6.4, 12.8, 25.6 km) under a variety of soil moisture conditions. Experiments have provided insight into the contributions of various input data to the performance and design of the neural network.
Supplementary URL: http://www.caos.aamu.edu/HSCaRS/LinDANet/