With the settings described above, a control run is conducted by applying the BEP scheme with the chemical and dynamical options displayed in Tab. 5 and Tab. 6. The configuration files ‘namelist.wps’ and ‘namelist.input’ used for executing the chemical simulations are added to the appendix as Figure A.5 and A.6. The WRF-Chem runs are performed on a 382 Tflop SGI based high performance computing system which is located at the NOAA Environmental Security Computing Center (NESCC), Fairmont, West Virginia, which is the location of NOAA's newest High Performance Computing Data Center. An urbanized WRF-Chem run with full online chemistry for one domain with 3 km resolution simulating a 9 day period is executed in about 4 hours computing time when using 240 nodes. The ratio between computational time (wall-time) and real-time accounts for 541 . A comparable run with the previously explained setup using the KEA environment would imply a ratio of
1
15 . Considering a spin up time of 2 days, the modelling time period is extended to begin
on August 9 0000 h and to end on August 18 2003 0000 h. The evaluation of the WRF- Chem run is executed by comparing the model output for selected variables from the full period at one grid cell (Fig. 27) with the mean of observations within that grid cell.
Fig. 27: Urban area of Stuttgart (Source: Google Earth) with meteorological and air quality measurement stations (red dots). The square in the middle indicates the WRF-Chem grid cell of 9 km² used for evaluation of the chemical model runs. For comparison, a mean of 3 stations (blue dots) is calculated.
Tab. 8: Measurement stations used for evaluation of the urbanized WRF-Chem run
Station Height NN [m] Measurment Height [m] Provider
Stuttgart Schwabenzentrum 250 20 Envir. Prot. Agency Stuttgart Stuttgart Bad Cannstadt - Gnesener Str. 235 3.5 LUBW
6. The WRF model
73 Each station is located at about the same geographical height, whereas the measurement heights differ by several meters. The average measurement height is calculated to 8.6 m. The first model level however refers to an atmospheric column of ~11 m depth. By averaging over three different heights, the mean concentration of pollutants in the urban canopy is approximated by including observations close to the ground and close to local emission sources, as well as concentrations in the upper canopy layers. The results for the evaluation of 5 compounds (O3, NO, NO2, CO and PM10) are illustrated in Fig. 28. Due to
the high fluctuation rates of hourly concentration observed within an urban environment, diurnal variations are calculated for both model and measurement as a basis for model evaluation. Modelling results are displayed for the multi-layer approach (Urban) and for a simplified bulk approach (No_Urban).
Fig. 28: Average daily trend of modelled concentration at an urban grid cell using the multi-layer approach (red) and the simple bulk approach (grey) in comparison to the mean of equivalent observations from 3 measurement stations located within that pixel (Fig. 27). From left to right: Ozone (a), NO (b), NO2 (c), NOx (d), CO (e) and PM10 (f).
With regard to Fig. 28, the multi-layer model is able to reproduce the mean concentrations of ozone (40 ppb) as observed throughout the time period and shows a quite good agreement with the diurnal cycle (a). The coefficient of determination for hourly output amounts to 77 %. For NOand NO2, the morning concentrations can be better reproduced
than during the day (b,c). The low daytime values can be caused by an overestimation of photolysis rates or overestimated turbulent mixing near the ground. Low NO values might also be the cause of the slight overestimation of Ozone during the day. The modelled NOx
(d) on average underestimates the observations by 13 ppb (~30 %). With respect to carbon 0 20 40 60 80 100 0 2 4 6 8 10 12 14 16 18 20 22 0 O3 [p p b ] Urban OBS No_Urban 0 10 20 30 40 50 60 0 2 4 6 8 10 12 14 16 18 20 22 0 N O [p p b ] b) 0 10 20 30 40 50 60 0 2 4 6 8 10 12 14 16 18 20 22 0 N O2 [p p b ] c) 0 20 40 60 80 100 0 2 4 6 8 10 12 14 16 18 20 22 0 N Ox [p p b ] d) 0 100 200 300 400 500 600 700 800 0 2 4 6 8 10 12 14 16 18 20 22 0 C O [p p b ] e) 0 20 40 60 80 100 0 2 4 6 8 10 12 14 16 18 20 22 0 PM 10 [u g m -3] f)
6. The WRF model
74
monoxide (e), WRF-Chem on average underestimates the observations by about 7 ppb (~3 %) and the diurnal variation is smoothed. The peak concentration seen in the observations however are overestimated which can be due to generally higher modelled background concentrations of CO being transported into the domain from boundary conditions retrieved from the global chemistry model MOZART (Chapter 6.3.4). Particulate matter is underestimated by almost 40 %.
When switching off the urban canopy model for the chemistry run and using a simple bulk urban parameterization, the simulated concentrations change in comparison with the observations. The diurnal trends are captured as well when using the bulk approach but mean concentrations tend to underestimate the multi-layer modelled concentrations of NO, NO2 and PM10 and overestimate ozone concentrations.
The effect of an urban canopy in trapping pollutants is simulated better when using a multi-layer canopy model but still the observations cannot be reproduced in their full intensity, even for averaged diurnal variations. The reasons for this can be manifold. On the one hand, schemes in the model representing turbulent exchange or photolysis rates are not configured in the proper manner, so that certain fractions of pollutants are removed. Another problem lies in the selection of the emission inventory which is causing insufficiencies in the modelling result due to coarse resolution. The short modelling period and short spin up time could also be a source of probable errors. However this could not be proven when extending the modelled timeframe by two weeks. The major part of the discrepancy between the model and observations however is ascribed to the fact of comparing the model output for a 3 by 3 km grid cell with one, or at least the average of 3 point measurements. Whereas the model accounts for an average value for an urban grid cell, the measurement stations are predominately located near places with high traffic loads and close to major emitters. This makes it difficult to characterize one single measurement as representative for a larger area and is especially true when studying processes in heterogeneous environments such as urban areas. Although the difference between the multi-layer approach and a simple urban parameterization is smaller than expected, the use of an urban canopy model is mandatory for discussing changes in surface properties based on the effect of urban planning strategies on local air quality.