A Method Using LSTM Networks to Impute Missing Temperatures in Temperature Datasets and to Predict Future Temperatures

Abstract

Listen

Translate article

With the growth of population and the development of industry, global warming has become a serious problem. It is necessary to focus on temperature change now. This paper designed and tested methods to impute the missing temperature in global temperature datasets and predict the future temperature. Imputation was divided into two steps, the first step is filling small gaps with equally spaced points for efficiency, and the second step is filling larger gaps with Long Short-Term Memory (LSTM) neural networks. This combination of imputation methods gave valid points with both accuracy and efficiency and successfully made some datasets continuous. Another LSTM network was used to predict future temperatures. The χ2 test for the goodness of prediction with real data in Australia showed that this prediction method using LSTM network is highly accurate.