Statistical analysis of systematic differences in the calculated pollutant concentrations of the models ECMWF/CAMS (regional reanalysis) and Polyphemus/ DLR

Kurzfassung

In the last two decades, air pollution was viewed as a very serious issue due to the development of infrastructure all over the world. Environmental stressors such as air temperature, radiation, humidity, wind, noise, pollens, and air pollutants (e.g., O3, NO2, PM10, PM2.5) can affect human health in a variety of ways. With the Copernicus Atmospheric Monitoring Service (CAMS) and the air quality in-situ measurements from the European Environmental Agency, a wealth of data of unprecedented quality and spatiotemporal resolution are available. These data are supplemented by available spatiotemporal high-resolution numerical models like chemical-transport models for the comprehensive description of the environmental conditions. Their advantages are constant coverage and high spatial and temporal resolution. However, it is very important to assess the model performances and comparability with in-situ or satellite observations. The main focus of this paper is to perform a comparison of the outputs of the Copernicus Atmosphere Monitoring Service (CAMS) – Europe Air Quality Reanalysis data and the chemical transport model POLYPHEMUS/DLR, with in-situ measurements (station data). The scope is to assess the discrepancies concerning the different chemical species and to provide statistical indicators like Mean Bias, FGE, RMSE, and Trend Analysis and correction weights describing the different characteristics of the models. Also, a Machine Learning approach was applied as an exploratory task, with the goal to predict concentrations at in-situ stations and to identify the influence of each parameters considered by the Polyphemus model. From the results, it was found that Polyphemus/ DLR model overestimates NO2, PM2.5, and PM10 and underestimates the O3, concentrations in urban and rural areas over the time window considered [June 2016 to Dec 2018]. CAMS outputs especially for PM10 and PM2.5 deviates from station observations though the outputs are corrected using EEA air quality station datasets. Overall, the parameters like surface temperature, boundary layer height and season were found to play a major role in both urban and rural regions. There are also significant changes in the influence of some parameters depending on location. This comparison study will help to understand the model performances (overestimation and underestimation) for each of the pollutants and help to select modelled data for health and air pollution-related research in the future.