Weather forecasting using deep learning and seasonal autoregressive integrated moving average model

Abstract

A number of industries, including mining, agriculture, transportation, and disaster relief, rely heavily on weather forecasting. Time-series trends have been well-captured by conventional forecasting models like SARIMA (Seasonal Auto Regressive Integrated Moving Average). Deep learning methods have become effective instruments for identifying complex patterns and raising predicting precision in recent years. This paper suggests a method to improve the accuracy of weather forecasting by utilizing SARIMA models and deep learning. To capture spatial and temporal correlations in meteorological data, the integration of the SARIMA model and Long Short-Term Memory (LSTM) networks of deep learning architecture is investigated. The Nigeria Metrological Agency (NIMET) provided 20 years' worth of temperature, humidity, wind, and rainfall data for this study. According to the evaluation results, the LSTM had an RMSE of 41.00 for the features of the training dataset, whereas the proposed SARIMA had RMSEs of 0.59 for rainfall, 23.99 for temperature, 1.23 for wind, and 24.47 for relative humidity. This demonstrates unequivocally that SARIMA outperformed the LSTM model.