Two-step combined algorithm for improving the accuracy of predicting methane concentration in atmospheric air based on the narx neural network and subsequent prediction of residuals

Abstract

Climate change in the Arctic is great and can have a significant inverse effect on the global climate, which determines the global significance of climate change in the Arctic. To date, many issues regarding the mechanisms responsible for the rapid melting of Arctic ice and permafrost degradation have not been resolved. It is not known when and what consequences these changes will lead to. Assessing the relationship between global warming and greenhouse gas emissions is an important environmental challenge. Among the main greenhouse gases, the evolution and climate-forming role of the carbon dioxide have been studied. The data on the methane subcycle of the carbon cycle is much less. In the paper, the authors propose a two-step combined algorithm (NARXR) to improve the accuracy of predicting methane concentration in atmospheric air based on the NARX neural network and subsequent prediction of the residuals. Two commonly used models based on artificial neural networks (ANN) for predicting time series are compared to determine the most appropriate base model. Nonlinear autoregressive neural network with external input (NARX) and Elman’s neural network are used. For the forecast, the authors use data on the methane concentration (CH4) in the atmospheric surface layer on the Arctic Island of Bely (Russia). Data is selected for a time interval of 192 hours, because it is characterized by significant daily fluctuations in the concentration of CH4. Values corresponding to the first 168 hours of the interval are used to train the ANN, and then concentrations are predicted for the next 24 hours. The proposed approach shows more accurate forecast results. © Subbotina I. E., Buevich A. G., Sergeev A. P., Shichkin A. V., Baglaeva E. M., Remezova M. S., 2020.