Spatial interpolation of regional PM 2.5 concentrations in China during COVID-19 incorporating multivariate data

During specific periods when the PM 2.5 variation pattern is unusual, such as during the coronavirus disease 2019 (COVID-19) outbreak, epidemic PM 2.5 regional interpolation models have been relatively little investigated, and little consideration has been given to the residuals of optimized models and changes in model interpolation accuracy for the PM 2.5 concentration under the influence of epidemic phenomena. Therefore, this paper mainly introduces four interpolation methods (kriging, empirical Bayesian kriging, tensor spline function and complete regular spline function), constructs geographically weighted regression (GWR) models of the PM 2.5 concentration in Chinese regions for the periods from January–June 2019 and January–June 2020 by considering multiple factors, and optimizes the GWR regression residuals using these four interpolation methods, thus achieving the purpose of enhancing the model accuracy. The PM 2.5 concentrations in many regions of China showed a downward trend during the same period before and after the COVID-19 outbreak. Atmospheric pollutants, meteorological factors, elevation, zenith wet delay (ZWD), normalized difference vegetation index (NDVI) and population maintained a certain relationship with the PM 2.5 concentration in terms of linear spatial relationships, which could explain why the PM 2.5 concentration changed to a certain extent. By evaluating the model accuracy from two perspectives, i.e., the overall interpolation effect and the validation set interpolation effect, the results showed that all four interpolation methods could improve the numerical accuracy of GWR to different degrees, among which the tensor spline function and the fully regular spline function achieved the most stable effect on the correction of GWR residuals, followed by kriging and empirical Bayesian kriging.