Assessing the Drought Effect on Crop Productivity for Precision Agriculture using Multimodal Remote Sensing Data and Deep Learning Models

Keywords: Drought, Sen4AgriNet, Precision Agriculture, Deep Learning
Abstract Type: Paper Abstract

Abstract

Vegetation drought, marked by insufficient moisture for sustaining healthy plant growth, is a critical factor in assessing ecosystem resilience and agricultural sustainability, especially among escalating climate change pressures. Traditional remote sensing models and techniques have revolutionized drought monitoring with large-scale, continuous datasets; however, these methods, often dependent on single-source data, are limited by insufficient spatial resolution, infrequent data collection, and an inability to capture multiple dimensions of drought impact. These limitations reduce their effectiveness for precise drought assessments, particularly in crop-sensitive areas where localized impacts, crop-specific responses, and timely updates are critical to understanding and managing drought conditions. Therefore, this study proposes a novel multi-sensor data fusion framework, integrating datasets from Landsat, Sentinel-1, Sentinel-2, SMAP, and GRACE, together with a benchmark dataset, Sen4AgriNet, to advance long-term monitoring of vegetation drought with enhanced accuracy using Deep Learning models. The framework is applied to the Southern France, a vital agricultural area known for its vineyards, wheat, and fruits, which faces increasingly severe droughts that strain water resources and reduce crop productivity. It begins with a crop classification phase to categorize crop types and estimate productivity, providing a foundation for analyzing how drought impacts specific crops over time. Next, drought indices, including the VCI and SMI, are applied to assess the temporal effects of drought on crop productivity, enabling a detailed time series analysis of drought stress. By leveraging Sen4AgriNet's high-resolution data, the framework delivers localized insights into drought impacts across different crop types, capturing variations that broader datasets may miss.