Answers are grounded in this paper and cite their sources.
Climate variability is a critical challenge to smallholder agriculture in semi-arid zones, where rainfed cropping systems are exposed to recurrent rainfall deficits, delayed onsets, and rising temperatures. This study quantifies how climate variability shapes crop selection and rotation decisions of smallholder farmers in Bugesera District, Rwanda, and develops data-driven tools to support climate-smart policy. Bugesera is purposively selected because it is the warmest and one of the driest agro-ecological zones of Rwanda, with a coefficient of variation of seasonal rainfall above 25% reported by the Rwanda Meteorology Agency. We integrate primary household data from 240 farmers (sample size justified through Yamane’s formula with a 6.5% margin of error and corroborated by Cochran’s 1977 criterion) with secondary climate records (2018–2023) and combine panel-data econometrics (Fixed-Effects logit, selected via the Hausman test) with interpretable machine learning (Random Forest and Gradient Boosting, with SHAP explanations). Results show that rainfall variability and drought significantly reduce the probability of selecting water-demanding crops and increase the adoption of drought-tolerant crops such as sorghum, millet, and cassava. The Gradient Boosting classifier reaches 87.4% accuracy on the held-out test set and uncovers non-linear behavioral thresholds: a 28% coefficient of variation in seasonal rainfall triggers a sharp shift away from maize, and the effect of rainfall variability on crop choice is 1.8 times stronger for low-income than for high-income farmers. SHAP analysis confirms rainfall variability, drought occurrence, and farm size as the top predictors. We further show that machine learning and econometrics are complementary rather than competing: econometrics provides causal identification while interpretable machine learning reveals non-linear thresholds. These findings indicate that risk-minimization rather than yield-maximization is the dominant adaptive logic of smallholder farmers in Bugesera and support a data-driven climate-smart agricultural policy in which the predictive models can be embedded in extension and early-warning services to anticipate farmer behavior.
No takes yet. Share an insight, caveat, or question.
Dushime et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: