Prediction of global temperature anomaly by machine learning based techniques

Abstract

In this work, anthropogenic and natural factors were used to evaluate and forecast climate change on a global scale by using a variety of machine-learning techniques. First, significance analysis using the Shapley method was conducted to compare the importance of each variable. Accordingly, it was determined that the equivalent CO2 concentration in the atmosphere was the most important variable, which was proposed as further evidence of climate change due to fossil fuel-based energy generation. Following that, a variety of machine learning approaches were utilized to simulate and forecast the temperature anomaly until 2100 based on six distinct scenarios. Compared to the preindustrial period, the temperature anomaly for the best-case scenario was found to increase a mean value of 1.23 degrees C and 1.11 degrees C for the mid and end of the century respectively. On the other hand, the anomaly was estimated for the worst-case scenario to reach to a mean value of 2.52 degrees C and 4.97 degrees C for the same periods. It was then concluded that machine learning approaches can assist researchers in predicting climate change and developing policies for national governments, such as committing firmly to renewable energy regulations.