Integrating weather variables and AI models for forecasting major pests in jute: applications in climate-smart crop management

Jute crop suffers a substantial amount of physical and economic loss every year due to the infestation of several insect pests, such as yellow mite ( Polyphagotarsonemus latus Banks) and jute semilooper ( Anomis sabulifera Guen), at different stages of crop growth. This study utilizes data on the mean incidence of yellow mite and jute semilooper at different days after sowing (DAS) from 2013 to 2023, along with weather variables, collected at the AINP-JAF, UBKV Centre, Cooch Behar, West Bengal. The results indicate that the incidence of jute semilooper follows a seasonal pattern, with most peaks occurring at approximately 45 DAS. Additionally, the mean incidence of yellow mite is found to be significantly positively correlated with maximum temperature and negatively correlated with minimum and maximum relative humidity at a 2-week lag. This suggests that dry weather with high temperatures 2 weeks prior contributes to higher yellow mite infestations at the current time. A similar correlation is observed for jute semilooper infestation. Various time series and machine learning models, including Autoregressive Integrated Moving Average (ARIMA), ARIMA-T, Seasonal ARIMA (SARIMA), SARIMA-T, ARIMA with exogenous variables (ARIMAX), SARIMA with exogenous variables (SARIMAX)-T, Random Forest, Support Vector Regression (SVR), and TDNNX, are applied to the training dataset from 2013 to 2022. The models are validated using the test data for the year 2023, based on root mean square error (RMSE) and root median square error (RMdSE) values. For yellow mite, TDNNX is found to be the best fitted model followed by SVR and SARIMAX-T in terms of RMSE and RMdSE values. Similarly, for jute semilooper, TDNNX is found to be the best fitted model followed by Random Forest and SARIMA. Finally, pest incidence forecasts for yellow mite and jute semilooper are obtained for 2024 using the forecasted and average weather data, applying the TDNNX model.