Impact of climatic factors on cotton production: A forecasting study using time series, machine learning, and regression models

Abstract

Abstract Cotton ( Gossypium hirsutum ) production is highly sensitive to climatic variability, including rainfall fluctuations, temperature extremes, drought, and flooding, which collectively introduce substantial uncertainty into agricultural output. This study investigates and forecasts cotton production using an integrated framework that combines classical time series models, machine learning techniques, and regression model‐based approaches. Specifically, Box–Jenkins ARIMA (autoregressive intergrated moving average), autoregressive moving average with exogenous variables (ARIMAX), exponential smoothing, Holt's exponential method, artificial neural networks, and multilayer perceptron (MLP) models are implemented and compared using standard metrics including root mean square error, mean absolute error, mean absolute percentage error, Akaike information criterion (AIC), and Bayesian information criterion (BIC). Empirical results indicate that while classical models outperform in all terms of information criteria (AIC/BIC), machine learning models, particularly MLP, achieve superior predictive accuracy based on error measures. To leverage the strengths of both approaches, an ensemble model combining ARIMA (1,2,2) and an MLP is developed, yielding improved forecasting performance compared with most individual models. Furthermore, the ARIMAX model incorporating exogenous variables (area, rainfall, and flood) highlights the dominant role of cultivated area in driving production, while rainfall exhibits a positive but less consistent effect, and flooding remains statistically insignificant. Complementary multiple regression and distributed lag analyses further confirmed the importance of area and rainfall, while capturing temporal lag effects of climatic influences. Forecast results reveal a divergence between modeling paradigms: classical time series models suggest a declining trend in cotton production, whereas machine learning models indicate potential recovery driven by nonlinear dynamics. The study provides actionable insights for policymakers and stakeholders to anticipate production trends under climate variability better and to design informed agricultural interventions.