Agricultural drought assessment using deep learning and multi-sensor remote sensing data integration

Abstract

This study proposes a Deep Learning Multi-Layer Perceptron Neural Network (DLMLPNN) model-based Drought Index (DI), capable of handling large amounts of remotely sensed data from different sensors, for the Gaya district of South Bihar. Apart from Synthetic Aperture RADAR (SAR) data from Sentinel-1, the multispectral data from Sentinel-2 was to generate vegetation (NDVI) and moisture indices (NDMI) for the Gaya district in South Bihar. Further, rainfall data from the Tropical Rainfall Measuring Mission (TRMM) along with surface soil moisture data from the Soil Moisture Active Passive (SMAP) satellite, thermal data from Landsat-8 Operational Land Imager (OLI) and CH₄ and O₃ concentration data from Sentinel-5P are used. These remote sensing datasets were used as input for training the DLMLPNN to predict the Normalized Differential Moisture Index (NDMI) as an indicator of soil moisture. It was observed that the model estimated the NDMI with R² statistics of 0.87 and 0.852 in the training and testing phases respectively. The NDMI gave a high correlation of more than 60 % with the ground collected Volumetric Soil Moisture (VSM). Feature Importance (FI) score was also computed to find out the contribution of each parameter used in the estimation of soil moisture. Based upon the weightage of each parameter used in the estimation of NDMI, a novel DI of the Gaya region was prepared for 2023. This index is the first of its kind for multi-sensor and multi-parameter drought analysis for the region and can be used to indicate drought conditions in other drought-prone areas.