A Framework for Detecting and Managing Non-Point-Source Pollution in Agricultural Areas Using GeoAI and UAVs

Abstract

This study proposes a novel framework for detecting and managing non-point-source (NPS) pollution in agricultural areas using unmanned aerial vehicles (UAVs) and geospatial artificial intelligence (GeoAI). High-resolution UAV imagery, combined with the YOLOv8 instance segmentation model, was employed to accurately detect and classify various NPS sources, such as livestock barns, compost heaps, greenhouses, and mulching films. The spatial information, including the area and volume of detected objects, was analyzed to track temporal changes and evaluate management strategies. The framework integrates remote sensing, deep learning, and geographic information system (GIS) analysis to enhance decision-making processes, providing detailed insight into NPS pollution dynamics over time. This approach not only improves the efficiency of NPS monitoring but also facilitates proactive management by offering precise location and environmental impact data. The results indicate that this framework can significantly improve resource allocation and environmental management practices, particularly in agriculture-dominated regions susceptible to NPS pollution, thereby contributing to the sustainable development of these areas.