Forecasting rice blast disease severity using weather-dependent regression and time series models

Abstract

Rice (Oryza sativa) is a staple food crucial for food security and economic stability, especially in developing countries. However, rice cultivation faces significant challenges, with rice blast disease, caused by the fungal pathogen Magnaporthe oryzae, being one of the most severe threats, potentially leading to yield losses of up to 30 %. This study aims to develop and apply regression models, including stepwise regression, multiple linear regression (MLR) and ARIMA (Autoregressive Integrated Moving Average), to predict the severity of rice blast disease based on weather parameters. Weekly data over 7 years (2017-2023) were collected from the Paddy Breeding Station at Tamil Nadu Agricultural University, Coimbatore, encompassing various weather factors such as temperature, relative humidity, rainfall and solar radiation. Data pre-processing included handling missing values, detecting outliers and creating time-lagged variables. The study revealed distinct seasonal patterns in rice blast incidence, with peak occurrences observed from mid-November to late January. Among the regression models, the ARIMA model incorporating weather variables as external regressors demonstrated superior performance with an R-squared value of 0.92, compared to 0.55 for stepwise regression and 0.57 for MLR. Accurate predictions of rice blast outbreaks could enable farmers and agricultural managers to implement timely and targeted disease management strategies, reducing dependence on broad-spectrum fungicides and minimizing crop losses. This study contributes to data-driven agriculture and disease management, potentially leading to more effective, economically viable and environmentally sustainable rice cultivation practices.