A Model for forecasting land productivity decline

Abstract

Sub-Saharan Africa (SSA) faces significant challenges in agricultural sustainability due to its vulnerability to climate change, which directly threatens crop productivity and economic stability. While existing crop simulation models have demonstrated potential in optimizing water and soil resources, their application within SSA’s unique agro-climatic conditions has remained limited. This study sought to bridge this gap by developing a robust farm productivity decline simulation model tailored specifically to SSA’s agricultural landscape. By leveraging predictive analytics, the model forecasted productivity trends and provided actionable insights to mitigate yield gaps, reduce food insecurity, and enhance land-use strategies. Through scenario modelling based on climate variability and resource availability, this study provides practical information on yield decline, fostering resilience against climate-induced shocks. The trained model demonstrated strong predictive performance, achieving a Train 𝑅2 of 0.86 and Test 𝑅2 of 0.84, indicating a high explanatory power in capturing the relationships between agro-climatic variables and farm productivity decline. The Cross-validation 𝑅2 of 0.83 and an Out-of-Bag (OOB) Score of 0.83 further validated the model’s robustness, ensuring it generalized well across different data partitions without overfitting. Additionally, Pearson correlation coefficients of 0.93 for training and 0.92 for testing confirmed a strong linear relationship between the observed and predicted productivity values, reinforcing the model’s reliability in capturing real-world agricultural trends. Collectively, these evaluation metrics highlight the model’s effectiveness in forecasting productivity decline while maintaining high predictive accuracy and generalization across unseen data. The forecasting system provided 20 years of future projections, revealing that Kenya’s food basket regions are expected to experience a significant productivity decline starting around 2026, with the most substantial reductions projected to occur by 2040. The model’s Mean Squared Error (MSE) ranged between 0.001 and 0.002, further confirming its ability to generalize well on unseen data. To facilitate accessibility and usability, the model was deployed through a Next.js-based web interface, with a Python-powered backend, leveraging containerized architecture via Docker for enhanced scalability and efficiency. This study contributes to the integration of advanced agricultural modeling into real-world farming practices, supporting scalable climate adaptation strategies aimed at safeguarding food security and economic resilience across SSA.