Decision support system for crop damage estimation based on waterlogging detection using synergy of remote sensing and machine learning

Abstract

Agriculture, the backbone of the economy, faces numerous challenges, with waterlogging being a prominent threat to crop yield. Traditionally, detecting waterlogged areas in agricultural fields has relied on ground observation techniques, which are often time-consuming and prone to imprecision. Remote sensing technology has emerged as a pivotal tool in agricultural monitoring, offering extensive data on land surface conditions. In this paper, we propose a pixel-based decision support system that identifies waterlogged areas in agricultural fields, utilizing remote sensing data and machine learning. The model inputs include a range of crop types and parcel polygons that assist in delineating vegetation regions and identifying areas impacted by waterlogging. Employing Sentinel-2 satellite data, the machine learning model is trained with a dataset that spatially distinguishes these conditions. To further refine the model's ability to differentiate between vegetated and waterlogged areas, 20 vegetation indices are computed, thereby enhancing its accuracy. With this proposed method the model achieved a precision of 0.9. The fusion of this model with detailed crop classification and yield prediction maps facilitates a precise estimation of the damage caused by waterlogging.