Time series and ensemble models to forecast banana crop yield in Tanzania, considering the effects of climate change

Abstract

Banana cultivation plays a pivotal role in Tanzania’s agricultural landscape and food security. Precisely forecasting banana crop yield is essential for resource optimization, market stability, and informed policymaking, particularly in the face of climate change. This study employed time series and ensemble models to forecast banana crop yield in Tanzania, offering crucial insights into future production trends. We utilized Seasonal ARIMA with Exogenous Variables (SARIMAX), State Space (SS), and Long Short-Term Memory (LSTM) models, chosen based on regression analysis and data exploration. Leveraging historical banana yield data (1961–2020) and relevant climate variables, we formulated an ensemble model using a weighted average approach. Our findings underscore the potential of time series and ensemble models for accurate banana crop yield forecasting. Statistical evaluation metrics validate their effectiveness in capturing temporal variations and delivering reliable predictions. This research advances agricultural forecasting by demonstrating the successful application of these models in Tanzania. It emphasizes the importance of considering temporal dynamics and relevant factors for precise predictions. Policymakers, farmers, and stakeholders can leverage this study’s outcomes to make informed decisions on resource allocation, market planning, and agricultural policies. Ultimately, our research bolsters sustainable banana production and enhances food security in Tanzania.