Climate change, vegetation dynamics, and maize production in Mozambique: evidence from method of moments quantile regression and machine learning

This study examines the impacts of climate variability and vegetation dynamics on maize production across 10 provinces in Mozambique using a balanced panel dataset covering the period 2002–2023. Following confirmation of cross-sectional dependence, second-generation panel techniques are employed. The Kao and Westerlund cointegration tests reveal a stable long-run relationship among the variables. Results from the Method of Moments Quantile Regression (MMQREG) indicate significant distributional heterogeneity in climate–maize relationships. Rainfall and the Normalized Difference Vegetation Index (NDVI) positively influence maize output, with stronger effects observed in lower-producing provinces, whereas temperature exerts a negative and statistically significant impact, particularly at the median and upper quantiles. Mediation analysis further confirms that NDVI partially transmits the effects of climatic variables to maize production, providing empirical support for the proposed climate–vegetation–production pathway. Interaction analysis shows that adequate rainfall and healthier vegetation conditions help mitigate temperature-induced stress. In addition, regional analysis across Northern, Central, and Southern Mozambique reveals substantial spatial heterogeneity, with semi-arid provinces exhibiting greater climate sensitivity. Complementary machine learning analysis identifies the K-Nearest Neighbors (KNN) model as the most accurate predictor of maize production, while temperature, NDVI, and rainfall emerge as the most influential predictors. The findings underscore the importance of accounting for both distributional and spatial heterogeneity in climate–agriculture linkages and highlight the need for climate-smart, region-specific interventions and data-driven forecasting tools to strengthen maize productivity and resilience in Mozambique.