Time-series Global Temperature Prediction and Visualization

Abstract

Temperature prediction is a prevalent and practical research in recent years. However, due to the performance of mathematical models or the application of visualization technology, previous work either lacked accuracy and intuitively visualized results. This paper designs a pipeline for global temperature forecast, which has relatively ideal accuracy and visualized results, and makes prediction of the temperature of each country/ region from 2023 to 2100 based on the earth surface temperature dataset packaged by The Berkeley Earth Surface Temperature Study. The mathematic model used by this paper is Prophet, which is a high-performance time series forecasting model and extremely applicable for dataset with high periodicity. The forecast results are generated by using PyEcharts in the form of HTML pages, where the worldwide average temperature, the years, a drag-operable month setting bar and the colored temperature distribution map are shown intuitively. Experimental results based on the testing dataset indicate that the mean absolute error of this method is 0.735°C. Forecast results on wider dataset indicate that the worldwide average temperature in the future may continue growing at the speed of around 0.0248°C per year.