Monitoring aerosol concentrations and optical thickness over Europe - PARMA final report

Executive summary

The understanding of particulate matter levels over Europe as a whole is at present limited by the diversity of ground level measurement methods. This hampers a comparison of air quality levels between EU member states, and checking compliance with (proposed) EU limit values. In this study, satellite observations (MODIS) of aerosol optical thickness (AOT) have been used to improve the mapping of yearly average PM_2.5 concentrations in Europe in 2003. Two different approaches were followed to use AOT data for PM_2.5 mapping: a statistical approach and a data-assimilation approach. The AOT measurements from MODIS were first validated against AERONET data.

AOT validation results

The AOT data measured by the MODIS instruments on board of the EOS/Terra and EOS/Aqua platforms, have been compared to AOT measurements of the AERONET surface network. The spatial correlation between MODIS and AERONET observed yearly average AOT over Europe is 0.64, and 0.72 using the fraction of the MODIS derived AOT pertaining to small particles only (AOT_F). The temporal correlation between MODIS and AERONET observed AOT is generally high, with a mean correlation of 0.72 (0.77 median of all stations), with slightly lower correlation for the AOT_F. However, the results show that MODIS systematically overestimates the AOT. On average, the annual mean AOT (AOTF) observed by MODIS averaged over all validation stations is 0.30 (0.25) compared to 0.20 as obtained by the sun-photometers. A more or less constant bias was found between MODIS AOT and AOTF and AERONET, of 0.7 and 0.9, respectively. After correction of the AOT and AOTF values, through multiplication with 0.7 and 0.9, respectively, the MODIS data agree with AERONET within the uncertainty range of ±0.05±0.2AOT_F.

Statistical mapping results

A map of yearly average concentrations of PM2.5 has been constructed, through fitting modelled PM_2.5 (with the Lotos-Euros model) and measured AOTF fields to observed PM_2.5 concentrations. For this fitting, in the final stage of this study, also five EMEP stations were added to the eight rural background stations in the AirBase database, to obtain a more uniform spatial coverage within the fitting domain (Europe). Using both modelled PM_2.5 and measured AOT_F fields as explanatory variables for the yearly average PM_2.5 distribution, the RMS-errors decrease by about 25% compared to fitting with only one explanatory variable. The spatial correlation between fitted and observed yearly average PM_2.5 levels is 0.82, with a RMS-error of 2.8 μg/m³. Since the modelled PM_2.5 and measured AOT_F fields contribute about equally to the fitted map, the fitted map resembles the features of the modelled PM2.5 map and the AOT_F map in equal proportions. The number of stations considered in this fitting is limited however, and adding more stations may significantly alter the resulting map, depending on where these stations are located. For example, the gradient in AOT_F between Scandinavia and Spain appears not very realistic, with higher AOTF-values in Scandinavia than in Spain. This gradient is opposite (and more realistic) in the modelled field. Therefore, adding more stations in Scandinavia and Spain will reduce the weight attached to the AOT_F field and increase that attached to the modelled field.

Assimilation results

The assimilation was performed with a LOTOS-EUROS model version that was slightly updated to that used in the statistical approach. Also, the assimilation has been done for a more limited area to speed up calculation time. The assimilation of MODIS AOT_F in LOTOS-EUROS leads to AOT fields that are in better agreement with the AERONET measurements. Through assimilation, the timing and representation of spatial patterns in AOT improves substantially. Through the assimilation of MODIS AOT_F, PM_2.5 levels are increased by 2-3 microgram in central Europe, but, as a result of the assimilation method applied, less in regions closer to the model boundaries. The assimilation slightly increases the spatial correlation between measured and modelled yearly average PM_2.5 (from 0.88 to 0.91). In the assimilation, the emissions (primary and precursors of secondary aerosols) have been taken as free parameter. Therefore, as a result of the assimilation, changes in the emission strengths are found. However, not much value was attributed to this, as the differences between modelled and measured AOT are large, and these differences between modelled and measured AOT are not only attributable to uncertainties in emissions of aerosols and their precursors from known sources. The differences also stem from emissions that are not taken into account in the emission inventories (like windblown dust), and errors in the description in the chemical transformation, dispersion and deposition of aerosols. Moreover, there is a large uncertainty in the optical properties of the aerosols that determine the relationship between PM and AOT, which explains part of the differences between measured and modelled AOT values.

Comparison between statistical mapping and assimilation

Generally, the spatial features in the map based on the statistical mapping approach resemble that of the assimilation approach in the central part of the model domain (North-West Europe). It is apparent that the assimilation approach leads to lower PM_2.5 concentrations than the statistical mapping approach. The absolute difference is 2-3 μg/m³ in the centre of the domain and it increases to 5-7 μg/m³ near the boundaries of the domain. The reason is that in the statistical mapping approach, the absolute measured levels of the PM_2.5 are actively used in the mapping procedure, in contrast to the assimilation approach where PM_2.5 measurements are used only for validation. Because of this, the statistical mapping approach leads by definition to almost no bias compared to the ground-based measurements, while data-assimilation will provide a bias depending on the model underestimation of PM, the uncertainty in the conversion between AOT and PM, and the uncertainty assumed for the AOT data and model results. Both methods lead to a better description of the spatial gradients in the yearly average PM_2.5 field in Europe as compared to modelled results only. In the assimilation approach, the spatial correlation between yearly average PM_2.5 measurements and the assimilated PM_2.5 field was already high (0.88) in the area considered, and increases to 0.91 after assimilation, while in the statistical approach, which is applied to a larger domain and with a slightly different model version, it increases from 0.70 to 0.82. Highest concentrations of particulate matter are found in densely populated and industrialized areas, such as the Po-valley, the Benelux countries, the Ruhr area, areas in Central Europe and specific large cities in Europe. Largest uncertainties in both methods are related to missing aerosol sources and the optical properties of aerosols that determine the relation between AOT and PM.