A Neural Network based Concept to Improve Downscaling Accuracy of Coarse Resolution Satellite Imagery for Parameter Extraction

Abstract

In countries where most of the economy depends on agriculture, the agriculture has to be observed closely because there is a need to timely monitoring the agriculture for better productivity that leads in good economy. Various parameters like land surface temperature (LST), soil moisture, precipitation, vegetation indices, humidity, etc should be monitored timely, as they directly or indirectly affect the agriculture that affect the economy. Here we are trying to downscale the LST extracted from the MODIS data. The challenge here is to handle, process, and downscale the data from MODIS low resolution (1000m) to high resolution (30m) equivalent to the Landsat. For this purpose, a back propagation neural network is used. The neural network is trained using downscaled MODIS LST data as input to provide a LANDSAT equivalent output at 30m resolution. RMSE is computed as an indicator of the performance of the approach.