Forecasts of air quality depend on meteorological conditions, especially within the planetary boundary layer (PBL). Therefore, the air quality models are commonly coupled with meteorological models. Of special importance for the performance of the air quality model is the parametrization of turbulence in PBL both in the meteorological model, which provides input winds, temperature and moisture data and in the air quality model itself.
This thesis investigates the effect of the parametrization of vertical turbulent diffusion on the quality of the PM$_{10}$ particle concentration forecast in the CAMx (Comprehensive Air Quality Model with extensions) model over Slovenia. CAMx is coupled with the operational weather forecast model ALADIN/SI of the Environmental Agency of Slovenia (ARSO). Numerical simulations with the CAMx model were carried out for January 2015 when 3 episodes with increased concentrations of PM$_{10}$ particles were recorded. Model results are compared with the PM$_{10}$ observations at the measurement stations across Slovenia.
The role of the coefficient of vertical turbulent diffusion $K_v$ is analyzed using several parametrization schemes. Five studied schemes (i.e. Kv schemes) included the OB70, the YSU, the ACM2, the CMAQ and the MYJ scheme. The schemes differ from each other according to the applied locality approach and the method for calculation of structure functions which describe vertical turbulence under different stability conditions. Fields of the $K_v$ coefficients obtained by different schemes are used in two methods for the vertical turbulent transport, the ACM2 method and the K theory.
Results are evaluated using standard statistical measures such as BIAS, mean absolute error (MAE), root mean square error (RMSE) and the correlation coefficient. Statistical significance is checked by the Fisher z-test. The aim was to determine the most appropriate combination of schemes for the operational application in the CAMx model at ARSO.
The results show that the vertical turbulent mixing is most intense in the local scheme MYJ and using the $K_v$ coefficient from ALADIN/SI. In all schemes, the model has underestimated measured values of PM$_{10}$ particle concentrations by at least factor 2. The underestimation is presumably mainly due to a poor horizontal resolution of the model (4 x 4 km) and consequently inadequately represented stability and turbulence in the boundary layer, as well as due to averaging of emission sources within the model cell.
If the schemes are judged by RMSE and MAE scores, the best performance is obtained using the YSU scheme in combination with K theory. If the correlation coefficient is used as a criterion, the combination of the MYJ scheme and the ACM2 method appears most successful. The Fisher z-test for the difference between the correlation coefficient for various schemes shows that differences between various schemes are on average statistically significant.
|
