Combined Time Series Prediction Model for Multi-Source Data Based on ARIMA-LSTM

In this paper, temperature trend assessment and prediction in different climatic zones will be carried out with the help of Pettitt's test, LSTM, and ARIMA. Pettitt's test was conducted for the southern and northern temperate zones as well as for the tropics in establishing the temperature mutation points between 1885 and March 2022. The points of mutation detected had 2008 for the southern temperate zone, 2009 for the northern temperate zone, and 2007 and 2008 for the tropics; hence, it did not tally with the expected mutation in March 2022 that led to the rejection of that claim. Based on the performed LSTM and ARIMA temperature prediction, it was shown that the former provided a result of 14.388°C by 2050 and 14.2048°C by 2100, while the latter returned 14.95°C by 2050 and 15.25°C by 2100. None of these models showed that the temperature would rise to 20°C over the next century. This would put it in 2898 according to ARIMA. It also showed, because of the lack of enough data to train it, the LSTM was overfitted, thus making the ARIMA the more reliable model under this circumstance.