Trends in measured NO2 and PM. Discounting the effect of meteorology.
Publikasjon: Trends in measured NO2 and PM. Discounting the effect of meteorology.
Serie: Eionet Report - ETC/ACM, 2018/9
Utgiver: ETC/ACM, Bilthoven
This report documents a study on long-term trends in observed atmospheric levels of NO2, PM10 and PM2.5 based on data from the European Environmental Agency (EEA) Airbase v8 (EEA, 2018). The main aim is to evaluate to what extent the observed time series could be simulated as a function of various local meteorological data plus a time-trend by a Generalized Additive Model (GAM). The GAM could be regarded an advanced multiple regression model. If successful, such a model could be used for several purposes; to estimate the long-term trends in NO2 and PM when the effect of the inter-annual variations in meteorology is removed, and secondly, to “explain” the concentration levels in one specific year in terms of meteorological anomalies and long-term trends. The GAM method was based on a methodology developed during a similar project in 2017 looking at the links between surface ozone and meteorology. The input to the study consisted of gridded model meteorological data provided through the EURODELTA Trends project (Colette et al., 2017) for the 1990-2010 period as well as measured data on NO2, PM10 and PM2.5 extracted from Airbase v8. The measurement data was given for urban, suburban and rural stations, respectively. The analysis was split into two time periods, 1990-2000 and 2000-2010 since the number of stations differ substantially for these periods and since there is reason to believe that the trends differ considerably between these two periods. The study was focused on the 4-months winter period (Nov-Feb) since it was important to assure a period of the year with consistent and homogeneous relationships between the input explanatory data (local meteorology) and the levels of NO2 and PM. For NO2, this period will likely cover the season with the highest concentration levels whereas for PM high levels could be expected outside this period due to processes such as secondary formation, transport of Saharan dust and sea spray. When measured by the R2 statistic, the GAM method performed best for NO2 in Belgium, the Netherlands, NW Germany and the UK. Significantly poorer performance was found for Austria and areas in the south. For PM10 there were less clear geographical patterns in the GAM performance. Based on a comparison between the meteorologically adjusted trends and plain linear regression, our results indicate that for the 1990-2000 period meteorology caused an increase in NO2 concentrations that counteracted the effect of reduced emissions. For the period 2000-2010 we find that meteorology lead to reduced NO2 levels in the northwest and a slight increase in the south. The amount of observational data is much less for PM than for NO2. For the 1990-2000 period the number of sites with sufficient length of time series is too small to apply the GAM method on a European scale. For the 2000-2010 period, we find that the general performance of the GAM method is poorer for PM10 than for NO2. With respect to the link between PM10 and temperature, the results indicate a marked geographical pattern with a negative relationship in central Europe and a positive relationship in Spain, southern France and northern Italy. For PM10 during 2000-2010, the vast majority of the estimated trends are found to be negative. The difference between the GAM trend and the plain linear regression, indicates that meteorology lead to increased PM10 levels in the southern and central parts and decreased levels in the north. For PM2.5 it turned out that the amount of data in the entire period 1990-2010 was too small to use the GAM method in a meaningful way on a European scale. Only a few sites had sufficient time series and thus more recent data are required.