Human-model hybrid Korean air quality forecasting system

Figures & data

Table 1. Comparison of world-wide national air quality forecasting systems by executing organization, start year, and target pollutants.

Country	Executing organization	Start year	Target pollutants
United States	EPA (http://airnow.gov/), National Oceanic Atmospheric Administration	‘99~	Air Quality Index (AQI) for O3 and PM2.5 divided into six categories (good, moderate, unhealthy for sensitive groups, unhealthy, very unhealthy, hazardous).
Germany	Rhenish Institute for Environmental Research (http://www.uni-koeln.de/math-nat-fak/geomet/eurad)	‘04~	AQI for five pollutants (O3, NO2, PM10, SO2, CO) divided into six categories (very good, good, satisfactory, sufficient, poor, very poor).
France	INERIS, METEO-France (http://www.prevair.org)	‘03~	Three pollutants (O3, NO2, PM,) with maps representing average concentration and maximum daily.
United Kingdom	DEFRA (http://uk-air.defra.gov.uk/forecasting/)	‘90~	Daily AQI for five pollutants (O3, NO2, SO2, PM10, PM2.5) using a combination of numbers (1–10) and words (very good, good, satisfactory, sufficient, poor, very poor) and colors (green/yellow/orange/red/purple).

Figure 1. Forecasting area at (a) the initial stage, (b) present.

Figure 2. Schematic diagram of the national air quality forecast system.

Table 2. Statistics measures used in the forecast and model evaluation (complied from Boylan and Russell, 2006; Zhang et al., 2012a).

Target	Measure	Mathematical Expression	Range
Forecast*	Accuracy		0% to 100%
	Critical Success Index (CSI)		0% to 100%
	False Alarm Rate (FAR)		0% to 100%
Model value**	Mean Fractional Bias (MFB)		−200% to +200%
	Mean Fractional Error (MFE)		0% to +200%
	Normalized Mean Bias (NMB)		−100% to +∞
	Root Mean Square Error (RMSE)		O to +∞

Notes: * N is the number of samples (by time and/or location). Where a, b, c, d represent simulated AQL while 1,2,3,4,5 represent observed AQL (a & 1: ‘good’, b & 2: ‘normal’, c & 3: ‘slightly bad’, d & 4: ‘bad’, e & 5: ‘very bad’). I, II, III, IV represent forecast nonexceedances that did occur, forecast exceedances that did not occur, forecast nonexceedances that did not occur, and forecast exceedance that did occur, respectively

** M and O are 24h averaged simulated and observed values in SMA .

Figure 3. AQM modeling domains for NAQFMS. (a) Three model domains that have resolutions of 27km, 9km, 3km, respectively. (b) Spatial distribution of observations in South Korea; the Box is study area.

Table 3. Summarization of modeling experiments’ information.

	Control run	Experiment1	Experiment2
	GFS	UM	GFS	Experiment3
Meteorological model	WRFv3.3 Short wave radiation: Goddard (Chou and Suarez, 1999) Long wave radiation: RRTM (Mlawer et al., 1997) Land surface model: NOAH (Chen and Duhia, 2001) Planetary boundary layer: YSU (Hong et al., 2006) Cumulus convection: Kain-Fritsch (Kain, 2004) Cloud microphysics: WSM3 (Hong et al., 1998; Hong et al., 2004)			MACC Reanalysis (Inness et al., 2013)
Foreign Inventory	MICS-Asia 2010	MICS-Asia 2010	INTEX-B
National Inventory	CAPSS (base year:2010)	CAPSS (base year:2010)	CAPSS (base year:2007)
Emission model	SMOKEv3.1 (anthropogenic) MEGANv2.04 (biogenic)
Chemical transport model	CMAQv4.7.1 Aerosol module: Aero5 Advection: Priecewise Parabolic Method Vertical diffusion: Eddy viscosity (K-theory)
Time step	1hr	1hr	1hr	3hr
Domain resolution	27km, 9km, 3km			80km

Figure 4. Comparison of pollutants concentration between observation and model prediction for daily (a) PM₁₀, (b) PM_2.5, (c) O₃, (d) NO₂, (e) SO_2.

Table 4. List of false forecast dates classified by four factors including emission, meteorology, model, and forecaster with false alarm dates during 1 September 2013 ~ 31 June 2015.

Factors (number of days)	Date
Emission (16)	’13.12.5, ’13.12.30, ’14.1.22, ’14.1.30, ’14.2.17, ’14.2.25, ’14.3.4, ’14.3.15, ’14.4.16, ’14.4.23, ’14.5.2, ’14.5.13, ’14.5.14, ’14.7.15, ’14.7.27, ’15.2.5
Meteorology (14)	’13.11.2, ’13.11.24, ’13.12.20, ’14.1.17, ’14.1.21, ’14.2.28, ’14.3.1, ’14.3.16, ’14.6.2, ’14.6.17, ’15.1.16, ’15.3.3, ’15.3.8, ’15.3.18
Model limit (5)	’13.10.29, ’13.11.16, ’14.2.23, ’14.4.12, ’15.1.24
Forecaster (20)	’13.9.9, ’13.12.3, ’13.12.22, ’14.1.3, ’14.1.20, ’14.1.25, ’14.2.16, ’14.2.21, ’14.2.24, ’14.2.27, ’14.4.10, ’14.4.15, ’14.4.25, ’14.5.30, ’14.6.19, ’15.1.5, ’15.1.14, ’15.4.11, ’15.4.24, ’15.6.13

Figure 5. Two types of synoptic patterns for the false-forecasting high PM events in winter. The Siberian high pressure is expanded and turned into the migratory high pressure from Lake Baikal (a) to East China Sea and (b) to northeastern China (Reference: The COMET Program: www.meted.ucar.edu/oceans/northwall).

Figure 6. Comparison of the simulated both PM₁₀ concentrations and surface wind speed in SMA for control run(GFS) and Exp. 1(UM).

Figure 7. False forecasting case of February 27, 2014. Spatial distribution of the simulated PM₁₀ concentration for (a) control run and (b) Experiment 1; (c) daily variation of the observed PM₁₀ concentration in SMA with the simulated PM₁₀ concentration for control run, Experiment 1 and 2.

Figure 8. False forecasting case of June 02, 2014. (a) 850hPa weather chart (source: KMA); (b) time series of the simulated and observed hourly precipitation; (c) hourly variation of the simulated and observed PM₁₀ concentration in SMA.

Figure 9. Comparison of observed and predicted 24-h average PM₁₀ concentration from control run(GFS, MICS-Asia/CAPSS2010) and Exp. 2(GFS, INTEX-B/CAPSS2007) during (a) the evaluation period and (b) all false forecasting cases.

Table 5. The comparison of forecasting accuracy in terms of PM₁₀ and PM_2.5 mass, and major PM_2.5 components depending on emission inventory when the forecasts were failed for 2014-year.

		PM10		PM2.5		Sulfate		OC		EC
		MC*	IC**	MC	IC	MC	IC	MC	IC	MC	IC
NMB	All	−42.8	−47.0	−20.3	−25.3	−35.3	−41.1	−33.9	−53.0	−10.2	−55.1
	False	−49.6	−47.9	−32.6	−31.7	−49.2	−45.0	−45.8	−60.2	−26.6	−66.5
MFB	All	−54.1	−64.1	−27.4	−38.7	−34.9	−45.4	−35.6	−67.3	−9.4	−67.4
	False	−67.9	−69.9	−43.2	−45.4	−65.7	−60.8	−55.6	−84.6	−36.0	−99.2
MFE	All	65.4	76.1	37.1	46.1	66.3	72.3	58.3	77.1	51.2	79.7
	False	74.9	80.6	50.7	56.4	86.1	85.5	68.1	86.5	58.2	99.4
RMSE	All	42.6	43.8	25.4	26.8	7.0	7.6	2.8	3.1	1.0	1.2
	False	68.2	68.3	46.1	49.7	13.6	14.4	5.3	5.6	1.7	2.1

Notes: *MC: MICS-Asia/CAPSS2010, ** IC: INTEX-B/CAPSS2007

Figure 10. Mean fractional error (MFE) and mean fractional bias (MFB) for major components (Sulfate, OC, and EC) from control run (MICS-Asia) and Exp. 2(INTEX-B).

Figure 11. The comparison of the predicted PM from control run (MICS-Asia/CAPSS2010), Exp. 2(INTEX-B/CAPSS2007) and Exp. 3(MACC) concentrations with observations in the case of 2 May, 2014.

Figure 12. The daily variation of the predicted and observed (a) PM₁₀ mass, (b) sulfate, (c) OC, and (d) EC on 15 March, 2 May, 17 June, 15 and 27 July in 2014 (Histogram: Observation, red dot: MACC, blue dot: CMAQ of control run).

Figure 13. Comparison of the number of cases with 24-h average PM₁₀ for the false alarms between observation and model.

Boylan, J. W., and A. G. Russell. 2006. PM and light extinction model performance metrics, goals, and criteria for three-dimensional air quality models. Atmos. Environ. 40(26):4946–4959. doi:10.1016/j.atmosenv.2005.09.087

 Web of Science ®Google Scholar

Zhang, Y., M. Bocquet, V. Mallet, C. Seigneur, and A. Baklanov. 2012a. Real-time air quality forecasting.Part I: History, techniques, and current status. Atmos. Environ. 50:632–655.

 Google Scholar

Chou, M.-D. and M.J. Suarez. 1999. A solar radiation parameterization for atmospheric studies. Technical report series on global modeling and data assimilation. NASA, TM-1999-104606(15).

 Google Scholar

Mlawer, E. J., S. J. Taubna, P. D. Brown, M. J. Lacono, and S. A. Clough. 1997. Radiative transfer for inhomogeneous atmospheres: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res. 102. doi:10.1029/97JD00237

 Web of Science ®Google Scholar

Chan, F., and J. Duhia. 2001. Coupling and advanced land surface-hydrology model with the Penn State-NCAR MM5 modeling system Part I: model implementation and sensitivity. Mon. Weather Rev. 129:569–585. doi:10.1175/1520-0493(2001)129%3C0569:CAALSH%3E2.0.CO;2

 Web of Science ®Google Scholar

Hong, S.-Y. and Y. Noh. 2006. A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Weather Rev. 134:2318–2341. doi:10.1175/MWR3199.1

 Web of Science ®Google Scholar

Kain, J. S. 2004. The Kain-Fritsch convective parameterization: An update. J. Appl. Meteorol. 43: 170–181. doi:10.1175/1520-0450(2004)043%3C0170:TKCPAU%3E2.0.CO;2

 Google Scholar

Hong, S.-Y., H.–M. H. Juang, and Q. Zhao. 1998. Implementation of prognostic cloud scheme for a regional spectral model. Mon. Weather Rev. 126:2621–2639. doi:10.1175/1520-0493(1998)126%3C2621:IOPCSF%3E2.0.CO;2

 Web of Science ®Google Scholar

Hong, S.–Y., J. Dudhia, and S.–H. Chen. 2004. A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation. Mon. Weather Rev. 132:103. doi:10.1175/1520-0493(2004)132<0103:ARATIM>2.0.CO;2

 Web of Science ®Google Scholar

Inness, A., F. Baier, A. Benedetti, et al. 2013. The MACC reanalysis: an 8 yr data set of atmospheric omposition. Atmos. Chem. Phys. 13(8):4073–4109. doi:10.5194/acp-13-4073-2013

 Web of Science ®Google Scholar