A site-optimised multi-scale GIS based land use regression model for simulating local scale patterns in air pollution

摘要

Standard Land Use Regression (LUR) models rely on one universal equation for the entire city or study area. Since this approach cannot represent the heterogeneous controls on pollutant dispersion in central, urban and suburban areas effectively the models are not transferable. Further, if different land use types are not adequately sampled in the measurement campaign, model estimates of local-scale pollutant concentrations may be poor. In this study, this deficiency is overcome with a site-optimised multi-scale GIS based LUR modelling approach developed. This approach is used to simulate nitrogen dioxide (NO2) concentrations in Auckland at three scales (central business district (CBD), urban, and suburban). The simulated NO2 distribution clearly shows a higher concentration of pollution along arterial roads and motorways as expected. Areas of limited dispersion (such as among high-rise buildings of the CBD) are also identified as high pollution areas. Predictor variables vary between scales; no single variable is common to all the scales. The leave-one-out cross validation (LOON) revealed that the multi-scale LUR model achieved an R-2 of 0.62, 0.86 and 0.73, respectively, at the CBD, urban, and suburban scales. The corresponding LOOCV root-mean-square-errors (RMSE) were 5.58.3.53 and 4.41 mu g.m(-3) respectively. Based on these statistical measures the multi-scale LUR model performs slightly better than the universal kriging (UK) model and the standard LUR model, and significantly better than the inverse distance weighting (EDW) and ordinary kriging (OK) models. When evaluated against external observations at eight fixed regulatory monitoring stations, the multi-scale CUR model out-performed all four of the other models considered and achieved an R-2 value of 0.85 with the lowest RMSE (8.48 mu g.m(-3)). This approach offers a robust alternative for modelling and mapping spatial concentrations of NO2 pollutants at multi-scales in large study areas with distinct urban design and configurations. (C) 2019 Elsevier B.V. All rights reserved.