An artificial neural network model for downscaling of passive microwave soil moisture

Abstract

In this paper, an Artitifial Neural Network (ANN) model was developed to downscale the soil moisture content from low resolution L-band passive microwave observation. Using the relationship between soil evaporative efficiency derived from MODerate resolution Imaging Spectroradiometer (MODIS) and soil moisture, the ANN model was used to downscale from 20 km×20 km observation to 1 km×1 km resolution over the whole area of 40 km×40 km. The method is tested using data collected during the National Airborne Field Experiment in 2005 (NAFE’05). The soil moisture variability in term of mean and standard deviation for the pixel to be disaggregated were proposed to be used in the ANN model for downscaling purpose. In this demonstration study, soil moisture data derived from 1 km resolution from the Polarimetric L-band Multibeam Radiometer (PLMR) were aggregated to 20 km resolution pixels, and subsequently downscaled using soil moisture statistics estimated from 1 km resolution data. The overall Root Mean Square Error (RMSE) difference between the measured and predicted soil moisture values varied between 1.8% v/v and 3.5% v/v across the complete range of typically experienced soil moisture conditions. The challenge of this model for real life practicality is presented in this paper and the suggestions are made at the end of this paper.