Dynamics of vegetation responses to desert locust (Schistocerca gregaria) plagues of 2019-2021 in Wabe Shebelle River Basin of Ethiopia

Abstract

Desert locust ( Schistocerca gregaria ) outbreaks are among the most destructive trans-boundary pests, causing extensive damage to vegetation, ecosystems, and livelihoods in arid and semi-arid regions. The spatiotemporal impacts of locust infestations on vegetation dynamics remain insufficiently quantified in Ethiopia, particularly using advanced time-series modeling approaches. This study aims to assess vegetation loss associated with the 2019–2021 desert locust upsurges in the Wabe Shebelle River Basin by integrating remote sensing data with a counterfactual modeling framework. Monthly vegetation indices, including Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), and Vegetation Condition Index (VCI), derived from MODIS data (2015–2022), were analyzed across 1,187 geo-referenced locations. These were combined with infestation timing and environmental variables such as rainfall, soil moisture, land surface temperature, and topography. Long Short-Term Memory (LSTM) neural networks were trained on pre-infestation time series to reconstruct baseline vegetation conditions, enabling attribution of observed deviations during infestation periods to locust activity. The results demonstrate robust model performance (R² ≈ 0.62–0.64) and reveal widespread, statistically significant vegetation loss, with over 70% of locations exhibiting notable NDVI decline. NDVI proved most sensitive to defoliation effects, while EVI captured more gradual stress and recovery patterns, and VCI reflected broader vegetation condition variability. Temporal analysis indicates immediate vegetation stress following infestation, followed by delayed and spatially heterogeneous recovery. These findings confirm that desert locust outbreaks constitute major ecological disturbances with persistent impacts on vegetation productivity. The study introduces a novel LSTM-based counterfactual framework that is scalable and transferable for operational monitoring and impact assessment, offering valuable insights for early warning systems and sustainable land management in locust-prone regions.