Comparative Model Study of Global Temperature Forecasting Based on the ARIMA-OLS Model

Abstract

Global warming is the most important climate issue for mankind today. Predicting the direction of future global temperature change is particularly important to develop relevant response strategies. To reasonably predict the future pace of global warming, this paper firstly collects global monthly average temperature data from Berkeley Earth, covering major land areas around the world and monthly sea surface temperatures from 2012 to 2022, to ensure wide spatial coverage and high data integrity. Based on the data's cyclical and stable growth characteristics, a time series model and a simple regression model are established. Then, temperature forecasts for 2050 and 2100 are generated using the ARIMA model and simple regression model separately. Results reveal that both models predict months when the average temperature exceeds 20 degrees Celsius in 2050 and 2100. However, only the simple regression model predicts that the average annual temperature will surpass 20 degrees Celsius in 2100. Considering the trend of global warming, this paper supports the superior predictive capability of the simple regression model. The paper contributes to climate analysis, providing valuable insights for understanding future temperature trends and guiding further research in this field.