Space-time dynamics of rainfed wheat yield in Ethiopia: Insights from remote sensing

Publication Type

Journal Article

Journal Name

ISPRS Journal of Photogrammetry and Remote Sensing

Name of Author

• Habtamu S. Gelagay, Université de Montpellier
• Louise Leroux, Université de Montpellier
• Lulseged Tamene, International Center for Tropical Agriculture (CIAT)
• Zewdu Abro, International Centre of Insect Physiology and Ecology Nairobi
• Siyabusa Mkuhlani, International Institute of Tropical Agriculture (IITA)
• Kindie Tesfaye, International Maize and Wheat Improvement Center, CIMMYT
• Wuletawu Abera, International Center for Tropical Agriculture (CIAT)
• João Vasco Silva, Universität Hohenheim
• Marc Corbeels, Université de Montpellier

Publication Date

12-1-2026

Abstract

Ethiopia is the largest wheat producer in sub-Saharan Africa, yet national food security efforts are constrained by coarse administrative level yield data that lack spatial and temporal detail. To bridge this gap, we present a spatio-temporally explicit assessment of rainfed wheat productivity, stability, and trends across Ethiopia from 2016 to 2023 using a multi-scale, data-driven framework. We integrated Sentinel-1-and Sentinel-2-derived biophysical, biochemical, physiological, phenological, and structural metrics as proxies for the integrated effects of genotype, environmental, and management factors (G × E × M) on wheat yield. These metrics were modeled using an ensemble machine learning approach to estimate wheat yield, while XGBoost was used to interpret drivers of yield variability through SHAP and permutation-based variable importance analyses. To reduce extrapolation risk, the yield estimates were restricted to regions of the feature space well represented in the training data using an area of applicability approach, while temporal feature stability analysis was used to assess the consistency predictive logic across the study period. Pixel-level trends were assessed using the Mann-Kendall test. Model performance reached an R² of 0.56 (RMSE = 0.85 t ha⁻¹) under random cross-validation and declined to 0.29 under spatial block cross-validation, with approximately 80% of the rainfed wheat area falling within the model’s demonstrated domain of applicability. The national mean rainfed wheat yield was 2.8 t ha⁻¹ over the study period, with pronounced spatial and interannual variability. Chlorophyll concentration and seasonal amplitude emerged as dominant drivers of yield variability, with strong context-dependent effects across agroecological zones. Our productivity-stability analysis identified resilient intensification hotspots representing 42% of high-yielding pixels, alongside vulnerable “double-burden” regions characterized by low productivity and low stability. While agronomically meaningful yield improvements were detected, statistically significant yield gains remained limited, contrasting with yield trends reported in official statistics. Our approach provides a scalable framework to support targeted interventions, inform policymaking, and strengthen national efforts toward wheat self-sufficiency and food security.

Keywords

Area of applicability, Earth observation, Machine learning, Rainfed wheat, Smallholder agriculture, Yield mapping, Yield variability

Recommended Citation

Gelagay, H., Leroux, L., Tamene, L., Abro, Z., Mkuhlani, S., Tesfaye, K., Abera, W., Silva, J., & Corbeels, M. (2026). Space-time dynamics of rainfed wheat yield in Ethiopia: Insights from remote sensing. ISPRS Journal of Photogrammetry and Remote Sensing, 242, 298-315. https://doi.org/10.1016/j.isprsjprs.2026.08.037

Creative Commons License

This work is licensed under a Creative Commons Attribution 4.0 International License.