Air pollution modeling over very complex terrain: An evaluation of WRF-Chem over Switzerland for two 1-year periods
The fully coupled chemistry module (WRF-Chem) within the Weather Research and Forecasting (WRF) model has been implemented over a Swiss domain for the years 2002 and 1991. The very complex terrain requires a high horizontal resolution (2 × 2 km2), which is achieved by nesting the Swiss domain into a coarser European one. The temporal and spatial distribution of O3, NO2 and PM10 as well as temperature and solar radiation are evaluated against ground-based measurements. The model performs well for the meteorological parameters with Pearson correlation coefficients of 0.92 for temperature and 0.88-0.89 for solar radiation. Temperature has root mean square errors (RMSE) of 3.30 K and 3.51 K for 2002 and 1991 and solar radiation has RMSEs of 122.92 and 116.35 for 2002 and 1991, respectively. For the modeled air pollutants, a multi-linear regression post-processing was used to eliminate systematic bias. Seasonal variations of post-processed air pollutants are represented correctly. However, short-term peaks of several days are not captured by the model. Averaged daily maximum and daily values of O3 achieved Pearson correlation coefficients of 0.69-0.77 whereas averaged NO2 and PM10 had the highest correlations for yearly average values (0.68-0.78). The spatial distribution reveals the importance of PM10 advection from the Po valley to southern Switzerland (Ticino). The absolute errors are ranging from - 10 to 15 μg/m3 for ozone, - 9 to 3 μg/m3 for NO2 and - 4 to 3 μg/m3 for PM10. However, larger errors occur along heavily trafficked roads, in street canyons or on mountains. We also compare yearly modeled results against a dedicated Swiss dispersion model for NO2 and PM10. The dedicated dispersion model has a slightly better statistical performance, but WRF-Chem is capable of computing the temporal evolution of three-dimensional data for a variety of air pollutants and meteorological parameters. Overall, WRF-Chem with the application of post-processing algorithms can produce encouraging statistical values over very complex terrain which are competitive with similar studies.