Modelling productivity of high-yielding cashew trees using bioclimatic indices and artificial neural networks

• Cashew productivity forecasted with ANN using climatic indices. • Productivity varied strongly across Benin’s agro-ecological zones. • Precipitation, temperature, humidity drove productivity trends. • ANN alone outperformed hybrid PCA or LASSO models. • Zone-specific models key to climate-resilient cashew forecasts. Cashew production plays a crucial role in the agricultural economy of West Africa, yet its productivity is highly sensitive to climatic variability. In the context of increasing climate uncertainty, reliable yield forecasting tools are essential for decision-making. This study aimed to predict annual cashew production in three agro-ecological zones (ZAE) of Benin using artificial neural networks (ANN), as well as hybrid models combining ANN with variable selection and dimensionality reduction techniques such as least absolute shrinkage and selection operator (LASSO) and principal component analysis (PCA). Climatic indices were computed and used as explanatory variables rather than raw meteorological variables to better capture the joint effects of climate on yield. Model performance was evaluated using a ranking approach based on multiple performance evaluation criteria, including coefficient of determination (R²), the root mean square error (RMSE) and its coefficient of variation (CV-RMSE). The results showed that cashew production varied significantly across ZAE, with ZAE III exhibiting the most stable yields (33.9 kg per tree, CV = 30.7%), ZAE IV showing greater variability (34 kg per tree, CV = 46.3%), and ZAE V having the highest but most fluctuating production (40.9 kg per tree, CV = 40.6%). Precipitation, temperature, and relative humidity were key drivers of cashew productivity, with notable spatial variability. Although the ANN model alone performed competitively, the addition of PCA or LASSO did not significantly enhance prediction accuracy. Overall, results highlight the potential of using synthesized climatic indices in machine learning models and the importance of zone-specific modeling strategies for improving cashew yield forecasting under changing climate conditions.