Mapping drought severity impact on arboriculture systems over Tadla and lower Tassaout plains in Morocco using Sentinel-2 data and machine learning approaches
Sabir Oussaoui, Abdelghani Boudhar, Abdessamad Hadri, Youssef Lebrini, Ismaguil Hanadé Houmma, Ismail Karaoui, El Mahdi El Khalki, Jamal-Eddine Ouzemou, Christophe Kinnard
Abstract
Severe droughts have affected the irrigated regions of Tadla and Lower Tassaout, Morocco, since 2019, peaking in September 2021. This study integrates Sentinel-2 satellite imagery with machine learning algorithms (MLAs) to quantify drought impacts on fruit tree systems. Three predictor scenarios were tested: M1 (Sentinel-2 bands and indices), M2 (added historical vegetation indices), and M3 (incorporated phenological metrics). Tree-based MLAs performed best, with Random Forest (RF) and Gradient Tree Boost achieving 95.94% and 94.09% accuracy under M3. RF-based analysis identified significant crop losses: 2,121 ha of citrus orchards and 12,127 ha of olive groves, with 16,276 ha moderately affected. However, groundwater and spring irrigation preserved 5,298 ha of olive trees and 7,216 ha of citrus orchards but led to declining aquifer levels. These findings highlight remote sensing and MLAs’ role in assessing drought impacts and balancing agricultural resilience with water sustainability.