Evaluating Predictions of ELAPSE and Google Air View-based Mixed-effects LUR Models for Air Pollution in Copenhagen, Denmark

BACKGROUND AND AIM: Fine scale exposure-assessment is needed for epidemiological studies on long-term health effects of air pollution (AP). We aimed to evaluate a) predictions of hyperlocal Google Air View-based mixed-effects land use regression (G-LUR) models for long-term AP for 2018-2020, and of ELAPSE (Effects of Low-Level AP: A Study in Europe) project LUR models for 2010 in Copenhagen, Denmark, and b) agreement of predictions between these two models. METHODS: We analyzed concentrations and Spearman correlations of ultra-fine particles (UFP), nitrogen dioxide (NO2), and black carbon (BC) by G-LUR models that predicted AP across ~30,000 streets for 2018-2020, and of NO2, BC, and fine particulate matter (PM2.5) by ELAPSE models that predicted 2010 concentrations at 100m spatial resolution (ELAPSE10). Correlation of pollutants between the two models were also assessed. Using annual mean data for 2010 and 2019 monitored at regulatory network stations, 2019 predictions were estimated for ELAPSE (ELAPSE19) NO2 and PM2.5. RESULTS: According to the G-LUR predictions, the long-term mean (SD) was 14,120 (8,849) particles/cm³ for UFP, 16.8 (8.3) μg/m³ for NO2, and 1.1 (0.4) μg/m³ for BC. According to the ELAPSE19 predictions, these were 21 (3.4) μg/m³ for NO2, and 11 (1.3) μg/m³ for PM2.5. The mean (SD) for BC was 1.6 (0.3) μg/m³ based on ELAPSE10. The correlation amongst predictions was highest between BC and NO2 (0.79 for G-LUR; 0.64 for ELAPSE). Between G-LUR and ELAPSE predictions, the highest correlation was for G-LUR NO2 and ELAPSE BC (0.64). The estimates of NO2 between the two models were moderately correlated (0.63), while for BC this was 0.51. CONCLUSIONS: Air pollution is a public health concern in Copenhagen, Denmark. There was moderate correlation between BC and NO2 predicted by both models. The moderate correlation between G-LUR and ELAPSE predictions suggests that spatial patterns have been fairly stable over 10-years.