In this study, we performed the two sensitivity tests of emission input data and meteorological fields in PM10 forecasting simulations based on air quality model. First, a set of CAPSS-2003 and CAPSS-2010 emission data has been employed to perform emission sensitivity tests of how reasonably PM10 forecasting simulations reflect to the different CAPSS (Clean Air Policy Support System) emissions developed for the base year of 2003 and 2010. In comparison of CAPSS-2003 with CAPSS-2010, CAPSS-2010 has shown more detailed point, line, and area sources of PM10, showing more reasonable input data. In simulating PM10 concentrations by WRF-CMAQ, CAPSS-2003 simulation showed extremely lower concentrations by factor of even less than 25 than observations, while CAPSS-2010 simulation showed more reasonable lower simulations of concentrations by factor of 6.4. This is clearly due to the emission amount estimated in CAPSS system. In particular, simulations over residential area greatly improved, indicating more detailed and recent emission input data showed more reasonable results. In this study, the estimated factors identified here were proposed to be used as ‘scaling factor’ for optimizing the emissions of PM10 for the realistic PM10 simulation that can be applicable in southeast area in Korea.
As a second sensitivity test, reanalysis meteorological field-FNL data, and forecasting field-GFS, have been respectively employed in this study, and used as an initial and boundary conditions of meteorological fields to investigate the PM10 results. We first selected four multiple cases for simulations where the differences between forecasting and measures PM10 concentrations exist higher than reference level. The simulation results indicated that the biases between simulations and forecasts mostly estimated even 50㎍/㎥ but both simulated mostly less than observations, mainly due to the strong surface wind speed induced by low pressure system located over north-east area of Korean peninsula. Sensitive test of temperature showed no significant bias between FNL and GFS, and therefore FNL with weaker but closer to observation to some degree, is more proper meteorological variables for PM10 simulations in Korea. Statistical significance level also showed higher in FNL than GFS fields. We also analyzed time series, spatial distributions, and vertical cross section, and concluded that FNL generated better results of PM10 forecasting simulations, although FNL and GFS showed both underestimated. In particular spatial distributions and vertical cross section of daily mean concentrations over Seoul metropolitan area showed better results by employing FNL data. Therefore we concluded that higher wind speed predicted than observation is a main controlling factor to miss the higher level PM10 forecasts, particularly houly prediction, indicating that surface wind speed is one of the most important factors for the improvement of PM10 prediction over Korean Peninsula.