Accurate crop yield forecasting under fluctuating agro-climatic conditions is critical for sustainable agriculture and food security. This study presents a climate-resilient, data-driven framework for soybean (Glycine max L.) yield prediction using Artificial Neural Networks (ANNs), to capture delayed and nonlinear agro-climatic effects relevant to smallholder farming systems. The model synthesizes a diverse set of agro-meteorological and environmental variables, including temperature, rainfall, humidity, wind speed, fertilizer inputs, soil characteristics, and atmospheric pressure, captured between June and October 2022 to support robust predictive analysis. The ANN architecture was trained and evaluated using standardized agro-climatic datasets, achieving R² of 0.9439, MSE of 0.3253 t 2 /ha², and MAE of 0.1432 t/ha, RSME of 0.1804 t/ha, demonstrating strong predictive capability under variable environmental conditions. Comparative analysis showed that the ANN outperformed Multiple Linear Regression and Random Forest models, particularly in representing nonlinear rainfall–temperature–fertility interactions. Lag-time sensitivity analysis further revealed that rainfall and humidity shifts exert the most influential delayed effects on soybean yield. Principal Component Analysis and feature ranking identified fertilizer input and rainfall as primary yield drivers, with relative humidity and temperature playing secondary roles. These findings are consistent with established agronomic patterns in legumes and highlight the ANN’s ability to capture complex environmental interactions. The findings highlight the ANN model’s applicability for resource-limited, smallholder contexts, providing a practical decision-support tool for climate-smart agronomy and adaptive planning in semi-arid regions .
Bali et al. (Wed,) studied this question.