TEMPERATURE FORECASTING WITH SARIMA MODELS: A TIME SERIES APPROACH

Abstract

In a country like Brazil, which is an agricultural exporter, climate analysis and monitoring are essential for the success of crops. This study analyzes the application of the Seasonal Autoregressive Integrated Moving Average (SARIMA) model in the modeling and forecasting of air temperature time series (average, minimum, and maximum) in the municipality of Francisco Beltrão, Paraná, for the period from 2014 to 2024. Initially, an exploratory analysis of the series was performed, followed by stationarity tests, temporal decomposition, and evaluation of the dependency structure using autocorrelation and partial autocorrelation functions. The SARIMA models were fitted according to the Box-Jenkins methodology, with selection based on the Akaike information criterion. The adequacy of the models was evaluated using error metrics and residual analysis. The results indicate strong annual seasonality in the series and a positive predictive result of the modeling, especially for short- and medium-term forecasts. However, a limitation was observed in the representation of extreme climatic events. It was concluded that the SARIMA model constitutes an appropriate and efficient tool for the analysis and forecasting of air temperature on a local scale, and can support agricultural planning and territorial management in climatically sensitive regions.