VEGETATION INDEX DYNAMICS (NDVI AND SAVI) IN RESPONSE TO CLIMATIC VARIATIONS IN THE SENEGALESE PEANUT BASIN (1984-2025)
Authors: François Ngor SENE, Ibra FAYE, Djiby YADE and Babacar NDAO
François Ngor SENE: Université Assane SECK de Ziguinchor, Département de Géographie, Laboratoire de Géomatique et d’Environnement (LGE), BP 523, Ziguinchor, Sénégal.
Ibra FAYE: Université Assane SECK de Ziguinchor, Département de Géographie, Laboratoire de Géomatique et d’Environnement (LGE), BP 523, Ziguinchor, Sénégal.
Djiby YADE: Université Assane SECK de Ziguinchor, Département de Géographie, Laboratoire de Géomatique et d’Environnement (LGE), BP 523, Ziguinchor, Sénégal.
Babacar NDAO: Université Assane SECK de Ziguinchor, Département de Géographie, Laboratoire de Géomatique et d’Environnement (LGE), BP 523, Ziguinchor, Sénégal.
ABSTRACT
The Senegalese peanut basin, located in the Sudano-Sahelian zone, experiences significant climatic variability that strongly influences vegetation dynamics and agricultural performance. This study analyses the spatiotemporal evolution of the NDVI and SAVI indices over the period 1984-2025, using Landsat and Sentinel-2 imagery processed through Google Earth Engine, combined with CHIRPS and TerraClimate climatic datasets. An inter-sensor harmonization procedure was applied to correct the additive bias at the Landsat/Sentinel-2 transition (2017-2022), confirming minimal residual bias (< 0.02 NDVI units). Results reveal a general upward trend in vegetation indices, structured across three phases: stagnation during persistent drought (1984-1999), progressive recovery (2000-2019), and a marked acceleration after 2020, with SAVI peaks reaching 0.55-0.60. BFAST structural break analysis confirms two statistically significant breakpoints. Seasonality is tightly coupled to the rainfall regime, with a vegetation peak in September-October. Precipitation and RAI emerge as the dominant climatic drivers (RF permutation importance ΔMSE ~ 9.0×10⁻⁴ and 8.0×10⁻⁴, respectively), while mean temperature exerts a moderate negative effect (r = -0.14). Among seven models tested (GLM, Random Forest, XGBoost, GAM, SVR, Elastic Net, and a Mean Ensemble), Random Forest delivers the best predictive performance (R² = 0.323, RMSE = 0.0487), confirming the non-linear nature of climate-vegetation relationships. Bootstrap prediction intervals (B = 200) demonstrate appropriate RF calibration with ~95% coverage. Analysis of climate-unexplained NDVI residuals reveals no significant non-climatic trend (Mann-Kendall τ = 0.026, p = 0.378), supporting climate as the primary driver of observed dynamics. ARIMA-based projections to 2040 suggest relative NDVI stability under a baseline scenario (S1: ~0.15-0.20), with a marginal decrease under a pessimistic scenario (S2: -10% precipitation, +0.5°C per decade). SAVI proves more discriminating than NDVI in sparse vegetation areas due to its soil-background correction. These findings contribute to understanding Sahelian vegetation resilience and offer an operational framework for agroecological monitoring and sustainable land management under climate change.
Keywords: NDVI, SAVI, remote sensing, climatic variability, peanut basin, Sahel, Random Forest, BFAST, inter-sensor harmonization