Multi-Level Modelling of Global Warming Based on Machine Learning Algorithms

Abstract

Today, abnormally high temperatures around the world are a wake-up call for global warming. Global warming is a problem that needs to be addressed urgently. In order to assess past temperature changes, this paper constructs indicators of data growth rates over a certain period of time. Firstly, this paper analyses the changes in growth rates after 1800. In addition, this study uses the Mann-Kendall method to determine the trend of the growth rate, and the ridge regression model, ARIMA model, and LSTM model are constructed to predict the future temperature change, respectively. The results show that the LSTM model is the most reliable, whereas the ARIMA model is only suitable for short-term forecasts due to the lack of a follow-up dataset and is less effective in predicting global temperatures below 20°C. The results of the ARIMA model are summarised as follows, with polynomial fitting results being affected by the extremes of the data, and performing the worst. The results show that the temperature will reach 20°C around 2310 and within 2398~2439 according to the ridge regression model and LSTM model, respectively.