Levee Soil Moisture Assessment Based On a Backpropagation Neural Network Using Synthetic Aperture Radar Data

a methodology of soil moisture assessment around a levee based on a backpropagation neural network and using Synthetic Aperture Radar (SAR) data is developed. Soil moisture changes along a levee over a period of time can help to monitor and find potential failure indicators such as slides and sand boils around the levee. Several analytical and empirical models have shown relationships between SAR backscatter and soil moisture. The algorithm includes three steps (1) segmentation of levee area into a 300 meter buffer zone from levee centerline, and removal of the trees using a threshold; (2) extracting the backscatter and texture features from each pixel within the buffer; (3) soil conductivity calculation using a backpropagation neural network. We use SAR data from UAVSAR and conductivity measurements as ground-truth data. With 10% of the data set allocated to testing samples, the result shows an RMSE around 8 mS/m and the correlation 0.64.