ARIMA approach for temperature and rainfall time series prediction in Punjab

The present study aims to explore the effectiveness of Seasonal Autoregressive Integrated Moving Average (SARIMA) models in forecasting meteorological time series data exhibiting seasonal patterns. We compared the performance of SARIMA models with different configurations and evaluate their forecasting accuracy using real-world meteorological datasetsfor three different agroclimatic zones of Punjab (sub mountainous region, central region and south west region) was analyzed to forecast mean monthly maximum air temperature, minimum air temperature and rainfall. The weather data was used from 1984-2022 for sub-mountainous zone (Ballowal Saunkhri), 1970-2022 for Central zone (Ludhiana) and 1977-2022 for south west zone (Bathinda). The results provide insights into the suitability and limitations of SARIMA models for meteorological forecasting and offer practical recommendations for practitioners and researchers in the field. The goodness of fit was tested against residuals using Ljung-Box test. The accuracy of the model was tested using Mean Absolute Error (MAE) and root square mean error (RMSE). The model achieved Mean Absolute Errors (MAE) ranging from 0.61 to 0.78 for maximum temperature, 0.74 to 0.49 for minimum temperature, and 32.12 to 45.44 for rainfall, with lower MAE values indicating higher predictive accuracy. The fitted model was able to capture dynamics of the temperature time series and produce a sensible forecast. However, the model was unable to forecast rainfall series efficiently.