For the work presented in this thesis, several schemes of WRF-Chem were tested, most importantly related to the choice in planetary boundary layer, but also concerning the sensitivity of the results to the choice in chemical mechanism. The considerations and results of these tests are briefly described in the following.
Other model options were changed in the course of this work, mainly because they rep-resent the latest developments and were recommended by the developers (Georg Grell, Ravan Ahmadov, personal communication, 2015). This concerns in particular the convec-tion scheme.
Planetary boundary layer
The choice in planetary boundary layer scheme is important for simulating aerosols and chemistry: it parameterizes processes in the boundary layer and calculates exchange coef-ficients, that are then used in the model to calculate the vertical mixing of air pollutants.
Its choice is “the most contested”, and the “results can differ significantly depending on its choice” (Peckham et al., 2013). This is why several different planetary boundary layer schemes were tested for Article 2, based on the results described in Article 1. The tested schemes include the Mellow-Yamada-Janjic scheme (MYJ, Janji´c, 1990, 1994), the Yonsai
Land surface model Noah LSM CORINE land use data tested: mosaic option Urban processes single layer UCM 3 categories: roofs, walls, trees parameters calculated for Berlin
Boundary layer MYJ, YSU, MYNN
Cumulus convection Grell, Grell-Freitas switched on for both domains Cloud microphysics Morrison double-moment
Radiation (sw+lw) RRTMG
Aerosols MADE/SORGAM chem opt=106
Chemistry RADM2 with KPP
Photolysis Madronich F-TUV
University Scheme (YSU, Hong et al., 2006) and the Mellor-Yamada Nakanishi & Niino scheme (MYNN, Nakanishi and Niino, 2004, 2006).
The results presented in Article 1 suggest that simulated NOx concentrations are too high during nighttime, and modelled mixing at nighttime might be underestimated. While only minor differences were found between the three different PBL schemes for grid cells outside of the urban area, the MYNN scheme was found to simulate a higher nighttime boundary layer in urban areas and slightly reduce the positive bias in simulated nighttime NOx concentrations, which is why this scheme was eventually chosen and is at the moment also recommended for future simulations with WRF-Chem over the Berlin-Brandenburg area. However, a lack of measurement data of temperature and wind speed profiles or planetary boundary layer height did not allow for a quantitative evaluation of simulated planetary boundary layer height over the urban area. Thus, additional measurement data would help to inform better the choice of the planetary boundary layer scheme in future studies.
Aerosols and chemistry
For simulating aerosols and chemistry, the Regional Acid Deposition Model chemical mech-anism (RADM2, Stockwell et al., 1990) with the Kinetic PreProcessor (KPP) and the MADE/SORGAM aerosol scheme - based on the Modal Aerosol Dynamics Model for Eu-rope (MADE, Ackermann et al., 1998; Binkowski and Shankar, 1995) and the Secondary Organic Aerosol Model (SORGAM, Schell et al., 2001) - were used. The priority was given to the KPP solver instead of the quasi-steady-state approximation (QSSA), because Forkel
is known that the MADE/SORGAM aerosol scheme underestimates the secondary organic aerosol contribution to PM (Ahmadov et al., 2012). Thus, this choice in chemistry and aerosol scheme is expected to result in an underestimation of simulated particulate matter.
The focus of this thesis is on nitrogen oxides and not on particulate matter, but the results for simulated particulate matter are in line with what is known from the literature (see Article 1). Thus, for simulating particulate matter and fully exploiting the benefits of an online-coupled model such as representing aerosol-cloud interactions, a different choice in aerosol and chemical mechanism would be recommended.
A sensitivity simulation has been done in the context of Article 2, testing the impact of a different choice in chemical mechanism on simulated NOx and O3 concentrations. The chemical mechanism tested was the Model for Ozone and Related Chemical Tracers chem-ical mechanism (MOZART, Emmons et al., 2010; Knote et al., 2014). The simulation was done for the month of July 2014, and used the settings described in Article 2. The results show that the differences in simulated O3 concentrations are quite substantial, while there is only a very small difference in simulated NOx concentrations: the difference between monthly mean simulated NOx is 0.4 µg m−3 for rural near-city and suburban stations, and essentially 0 for urban background stations. For O3, on the other hand, the difference in mean concentrations is of the order of ca. 30-35 µg m−3, with the simulation using the MOZART mechanism indicating the higher values. Looking at the simulated time series, a comparison with measurements shows that the simulation using the RADM2 mechanism generally underestimates observed peak values during episodes with higher O3 concentra-tions. However, compared to the simulation using the MOZART mechanism, it better captures the overall diurnal variation and periods with lower observed daily maxima. The simulation using the MOZART mechanism, on the other hand, better captures observed peak values during periods with high O3 concentrations, but overestimates concentrations overall. Both simulations do not capture the full amplitude of observed diurnal variations of O3 concentrations and do not seem to capture the (longer-term) variability of observed O3 concentrations.
These results are consistent with results reported in the literature, e.g. from the AQMEII phase 2 intercomparison study of online-coupled models (Im et al., 2015b). The authors find that, all over Europe, daily maximum 8-h average surface ozone concentrations below 50-60 µg m−3 are overestimated by all models in the intercomparison study, and concen-trations over 120-140 µg m−3 are underestimated. The study also included several model setups of WRF-Chem using the RADM2 mechanism. The model bias of daily maximum 8-h average surface ozone in Europe is even larger for WRF-Chem setups using other
the same pattern in model bias, but overall smaller biases at higher observed ozone levels.
Furthermore, a comparison of the RADM2 and MOZART mechanisms in WRF-Chem for a European domain (Mar et al., 2016) finds similar results as described above: the simulation using the MOZART chemical mechanism predicts O3 concentrations of up to 20 µg m−3 higher than the simulation using the RADM2 mechanism, and measurements are overestimated in the simulation using the MOZART mechanism and underestimated in the simulation using the RADM2 mechanism. At the same time, NOx concentrations simulated with both mechanisms were found to be relatively similar. In addition, O3 con-centrations simulated with both mechanisms were found to show a similar sensitivity to NOx.
Based on these results and to maintain an overall consistency among the different arti-cles contributing to this thesis, the studies presented in this thesis are all based on the RADM2 chemical mechanism. However, the results from the literature and the sensitivity test discussed above clearly show that further research is needed to better represent O3 concentrations at levels that are relevant for air quality management, as this seems to be a problem common to many WRF-Chem setups and, to a lesser extent, also to other online-coupled air quality models.