A Local Spatial Kriging Applied to the PM2.5 Concentration Estimation

Abstract

Air pollution in China has aroused special concerns from the general public in recent years. Accurately estimating air pollutant concentrations, especially fine particulate matter (i.e., PM2.5), is of importance to understand the distribution patterns of pollutants and then facilitate the control of their emissions and the reduction of public exposure. While estimating ambient PM2.5 concentrations based on monitoring station measurements is inherently complex, geostatistical spatial interpolation could provide a potential solution. However, standard Kriging may fail in yielding an accurate estimate as it assumes a universally homogeneous semi-variogram and lacks in considering the non-stationary spatial process that arises from anisotropy and non-randomized point pattern. This study proposes an algorithm named Point Pattern Local Anisotropy Kriging (PPLAKriging) that can derive a locally adaptive semi-variogram. Specifically, this algorithm determines an optimum neighboring search range in terms of the local distribution pattern of surrounding points followed by a local anisotropy detection. A geographical coordinates transformation is then performed for ease of calculation. To validate the proposed model in interpolating PM2.5 concentration, a regional experiment is conducted. The results show that there are substantial benefits in considering local features under different neighborhood environmental conditions. In the experiment, PPLAKriging proves to be superior to the standard ordinary Kriging, root-mean-squared-error decrease by 33.9%.