Machine learning and time-series approaches for forecasting bacterial blight of pomegranate

Bacterial blight, caused by Xanthomonas axonopodis pv. punicae, poses a significant threat to pomegranate (Punica granatum L.) cultivation, resulting in considerable yield losses. The disease exhibits pronounced seasonal variability, which is largely influenced by environmental and climatic factors. Understanding the temporal behaviour of disease severity and its relationship with meteorological factors is essential for effective disease surveillance, forecasting, and management. This study analysed long-term surveillance data on bacterial blight severity from pomegranate-growing regions in Maharashtra, India (2013-2024), along with meteorological parameters. This study employed statistical methods, time-series models (ARIMA, SARIMA, and VAR), and Machine learning (ML) based regression models using seasonal assessments of disease severity based on Standard Meteorological Weeks (SMWs). The results indicated weak associations between disease severity and weather variables, with temperature showing a weak positive correlation (r = 0.18), while relative humidity exhibited a moderate inverse relationship (r = -0.33). Time-series analysis revealed a clear temporal dependence in disease progression, with the non-seasonal ARIMA (2,1,1) model providing the best fit (R² = 0.691; RMSE = 0.085), whereas seasonal components were not retained in the final model. The multivariate VAR model further enhanced biological interpretability by integrating weather variables, achieving a comparable accuracy (R² = 0.732; RMSE = 0.239). Among ML-based regression models, LightGBM achieved the best prediction (R² = 0.776; RMSE = 0.566). The explainable ML analysis consistently identified temperature as the dominant driver of bacterial blight severity. Overall, the integrated analytical framework provides a robust understanding of bacterial blight dynamics by combining surveillance data, temporal modelling, and predictive analytics. These findings provide a basis for data-driven forecasting systems, enabling timely interventions and improved disease management under varying climatic conditions.