A decomposition based imputation algorithm for long consecutive missing atmospheric pollution data and its application

With the intensification of environmental air pollution, the impact of air pollutants on both the ecological environment and human health has attracted widespread attention. However, due to the relatively late introduction of environmental monitoring systems, there were long consecutive missing values in early pollutant data. In this paper, we propose a decomposition-based imputation method for long consecutive missing pollution data. Firstly, wavelet coherence analysis is employed to investigate the periodic relationship between the pollution data and the relevant air data, decomposing them into periodic and non-periodic components. Then, machine learning and transfer learning are used to impute the periodic and non-periodic components, respectively. Furthermore, the effectiveness of the method is validated on artificially missing ${\mathrm{NO}}_2$ and ${\mathrm{SO}}_2$ concentration data from five regions of China. Comparison results show that the proposed method significantly outperforms some other imputation methods in the literature in terms of both mean absolute error and mean absolute percentage error. Finally, the proposed imputation method is applied in the study of accelerated aging of polycarbonate materials. Experimental results show that the predictive accuracy of the aging model is improved when using the imputed pollutant data.