Fitting SARIMA models to temperature time series to make predictions until 2050.
The project involves forecasting temperature trends until 2050 using Seasonal Autoregressive Integrated Moving Average (SARIMA) models. It starts with data preprocessing, including handling missing values and detecting seasonality. Stationarity checks are conducted using the Augmented Dickey-Fuller test to ensure data suitability for SARIMA modeling.
The model selection process involves analyzing the Auto-Correlation Function and Partial Auto-Correlation Function plots to determine optimal parameters (p, d, q, P, D, Q, m). Various SARIMA configurations are evaluated using Akaike Information Criterion and Bayesian Information Criterion to balance model complexity and accuracy.
The selected SARIMA model is then trained and validated on historical temperature data. Residual analysis and Ljung-Box tests confirm the model’s goodness of fit and independence of residuals. The model’s predictive accuracy is measured using Mean Absolute Error and Root Mean Square Error.
Long-term forecasts until 2050 are generated, capturing both seasonal fluctuations and upward temperature trends, indicative of climate change. Prediction intervals are calculated to estimate forecast uncertainty, with the intervals widening over time. The results are visualized to illustrate the projected temperature increase, providing insights into future climate scenarios.