3.1. Meteorological Predictions
Table 3 and
Table S1 summarize the annual mean model performance statistics of the meteorological, radiative, and cloud predictions from the 2006 and 2011 baseline WRF-CAM5 simulations, respectively.
Figure 1 shows the spatial distributions of annual mean MBs of T2, Q2, WS10, and precipitation against NCDC dataset for both years. The model performs well in predicting P, T2, Q2, and WS10 for both years with annual mean MBs of −13.9 and −13.4 hPa for P, −0.9 and −1.1 °C for T2, 0.03 and 0.1 g∙kg
−1 for Q2, 0.3 and 0.4 m∙s
−1 for WS10, respectively. The corresponding annual mean MAGEs are 22.2 and 21.8 hPa, 1.9 and 2.0 °C, 0.8 and 0.8 g∙kg
−1, 1.0 and 1.0 m∙s
−1, respectively, and the corresponding annual mean RMSEs are 35.9 and 35.3 hPa, 2.7 and 2.7 °C, 1.2 and 1.1 g∙kg
−1, 1.3 and 1.3 m∙s
−1, respectively. The values of annual mean IOA are 0.7 for WS10 and 1 for P, T2, and Q2 in both years. The model reproduces the observed seasonal variations for P, T2, and Q2. For WS10, the observed domain-wide mean WS10 is the highest in spring, followed by winter, fall, and summer, the simulated domain-wide mean WS10 is the highest in winter, followed by spring, fall, and summer.
The largest discrepancy between the observed and simulated T2 occurs in summer with seasonal mean MBs of −1.3 °C in 2006 and −1.4 °C in 2011. Consistent with Ma
et al. ([
43]) and Zhang
et al. ([
44]), large cold bias occurs in the Tibet Plateau, where the Mt. Himalayas covered with thick snow is located (see
Figure 1). The large cold bias indicates the limited capability of WRF-CAM5 in reproducing observed snow cover and its rate of melting. Large biases (both warm and cold biases) in T2 also occur over Japan, South and North Korea. WS10 is slightly underpredicted in summer in 2006 and overpredicted in other seasons in 2006 and in all seasons in 2011. The overpredictions in WS10 may be responsible for the underpredictions of the chemical concentrations, which will be discussed in
Section 3.2. Despite overpredictions, WS10 predictions in this work have much lower biases comparing to most other WRF/Chem applications which reported MBs of 1.3–2.6 m∙s
−1 over East Asia for 2001 ([
45]), 2006 ([
46]), and 2010 ([
47]), and 0.03–1.2 m∙s
−1 for 2010 over North America ([
48]), and 1.0–1.4 m∙s
−1 for 2010 over Europe ([
48]). This is because the Monin-Obukhov surface layer scheme ([
18,
19]) can represent surface roughness and topographical features well, which leads to better representation of surface drag in the Bretherton and Park ([
20]) PBL scheme used in this work than that used in the YSU PBL scheme used in most previous WRF/Chem applications.
Emery
et al. ([
49]) proposed threshold values for satisfactory performance for several meteorological variables: MB within ±0.5 °C, MAGE of ≤ 2.0 °C, and IOA of ≥ 0.8 for T2, MB within ±1.0 g∙kg
−1, MAGE of ≤ 2.0 g∙kg
−1, and IOA of ≥ 0.6 for Q2, and MB within ± 0.5 m∙s
−1, RMSE of 2.0 m∙s
−1 and IOA of ≥ 0.6 for WS10. Note that such criteria were developed based on the meteorological simulations with the four dimensional data assimilation (FDDA). In this work, FDDA is not used to allow the meteorology-chemistry feedbacks to be investigated. Instead, T2, Q2, and WS10 are re-initialized every five days as a compromise to allow the simulation of feedbacks while periodically constraining the simulation with re-initialized meteorological conditions based on observations. As a result, the model agreement with observations is not expected to be comparable with simulations that use FDDA. The annual mean MBs of T2 are larger than the suggested threshold values by Emery
et al. ([
49]) but they are comparable or even better than the performance using MM5 (e.g., [
44,
50]) and WRF (e.g., [
47,
48,
51,
52]), and hence deemed acceptable. The annual mean values of MAGEs, IOAs, and RMSEs of T2, Q2, and WS10 are also within the suggested threshold values.
Figure 1.
Spatial distribution of annual mean MBs of T2, Q2, WS10, and precipitation against NCDC dataset for the 2006 and 2011 WRF-CAM5 simulations with M92.
Figure 1.
Spatial distribution of annual mean MBs of T2, Q2, WS10, and precipitation against NCDC dataset for the 2006 and 2011 WRF-CAM5 simulations with M92.
Precipitation is overpredicted against both surface observations from NCDC and the merged satellite and rain gauge data from GPCP, with annual-mean NMBs of 13.6% and 9.0%, for 2006 and annual-mean NMBs of 11.1% and 2.5% for 2011. In 2006, the largest overpredictions of precipitation occur in fall against NCDC, with an NMB of 23.8%, and in winter against GPCP, with an NMB of 35.1%. In 2011, the largest overpredictions of precipitation occur in winter against NCDC, with an NMB of 28.1%, and in fall against GPCP, with an NMB of 20.9%. Those biases are either consistent or better than other applications of WRF, WRF/Chem, or WRF-CAM5 over East Asia reported in the literature (e.g., [
45,
53,
54,
55]). The moderate overpredictions in some seasons against NCDC and GPCP in both years may be due to several possible reasons including neglecting the impact of effect of clouds on radiation through the changes of droplet and ice effective radii ([
56]) and overpredictions of convective precipitation intensity by the cumulus parameterization. The model captures well the observed seasonal variations in both NCDC and GPCP data.
As shown in
Figure 2, the model generally captures the observed spatial distributions of precipitation from GPCP in all seasons in both years. In 2006, precipitation is slightly to moderately overpredicted in all seasons, particularly over oceanic areas, resulting in a large annual mean overprediction over western Pacific. In 2011, a large overprediction also occurs over western Pacific in all seasonal and annual means, although the domain-wide mean precipitation is slightly or moderately underpredicted in summer and spring. Precipitation is moderately overpredicted over Taiwan and Japan in all seasons in both years, leading to the underpredictions of concentrations of gaseous and PM species which will be discussed in
Section 3.2. The GPCP merged dataset contains precipitation estimates from satellite and rain gauge observations. It gives comparable precipitation to NCDC data in terms of domain-wide mean and spatial distributions (see
Figure 2 and
Table 3 and
Table S1), but its coarse horizontal resolution of 2.5° × 2.5° cannot capture the considerable spatial variability at a finer grid resolution, especially in southeastern China, Japan, and over oceanic areas. Comparing to 2006 NCDC, 2006 GPCP gives higher precipitation in summer and fall but lower precipitation over land in winter and spring, leading to a slightly higher observed annual mean precipitation. GPCP gives higher precipitation in all seasons in 2011 than does NCDC, especially over oceanic regions. The observed seasonal variation for precipitation is the same for NCDC and GPCP in 2011, but somewhat different in 2006. The inconsistencies in magnitudes and seasonality between NCDC and GPCP indicate uncertainties in observations that also contribute in part to the model performance evaluation.
Table 3.
Performance statistics for meteorological predictions for 2006 WRF-CAM5 baseline simulation with M92.
Table 3.
Performance statistics for meteorological predictions for 2006 WRF-CAM5 baseline simulation with M92.
Variable | Data Source | Number | Mean Obs. | Sim. | Mean Sim. | R | NMB (%) | NME (%) | MB | MAGE | RMSE | FB | FGE | IOA |
---|
P (mb) | NCDC | 6946 | 939.7 | M92 | 925.8 | 0.96 | −1.5 | 2.4 | −13.9 | 22.2 | 35.9 | −0.02 | 0.03 | 1.0 |
T2 (°C) | NCDC | 10524 | 13.8 | M92 | 12.9 | 0.98 | −7 | 14 | −0.9 | 1.9 | 2.7 | 0.92 | −0.93 | 1.0 |
Q2 (g∙kg−1) | NCDC | 6945 | 8.0 | M92 | 8.0 | 0.98 | 0.4 | 11 | 0.03 | 0.8 | 1.2 | 0.00 | 0.13 | 1.0 |
WS10 (m∙s−1) | NCDC | 8010 | 3.1 | M92 | 3.4 | 0.54 | 11 | 32 | 0.3 | 1.0 | 1.3 | 0.10 | 0.29 | 0.7 |
Precip (mm∙day−1) | NCDC | 10131 | 2.7 | M92 | 3.0 | 0.68 | 14 | 62 | 0.4 | 1.7 | 3.1 | - | - | 0.8 |
GPCP | 15908 | 2.9 | M92 | 3.1 | 0.76 | 9 | 37 | 0.3 | 1.0 | 1.5 | −0.02 | 0.40 | 0.9 |
CCN (cm−2) | MODIS | 4917 | 0.8 | M92 | 0.5 | 0.78 | −33.8 | 40.9 | −0.3 | 0.3 | 0.7 | −0.2 | 0.3 | |
CDNC (cm−3) | 9111 | 143.0 | M92 | 101.0 | 0.63 | −29.3 | 36.2 | −41.9 | 51.7 | 65.2 | −0.4 | 0.5 | |
CF | 13398 | 0.6 | M92 | 0.6 | 0.81 | −12.0 | 17.2 | −0.1 | 0.1 | 0.1 | −0.2 | 0.2 | |
PWV (cm) | 13398 | 2.2 | M92 | 2.2 | 0.99 | −0.9 | 6.3 | 0.0 | 0.1 | 0.2 | 0.05 | 0.1 | |
LWP (g∙m−2) | 13398 | 110.3 | M92 | 48.0 | 0.87 | −56.5 | 56.6 | −62.3 | 62.4 | 65.6 | −1.0 | 1.0 | |
IWP (g∙m−2) | 13398 | 245.1 | M92 | 9.5 | 0.01 | −96.1 | 96.1 | −235.6 | 235.6 | 243.9 | −1.8 | 1.8 | |
AOD | 13070 | 0.3 | M92 | 0.2 | 0.70 | −35.7 | 43.6 | −0.1 | 0.1 | 0.2 | −0.5 | 0.6 | |
COT | 13398 | 16.3 | M92 | 8.2 | 0.84 | −50.0 | 50.3 | −8.2 | 8.2 | 8.9 | −0.8 | 0.8 | |
GLW (W∙m−2) | CERES | 13398 | 324.6 | M92 | 317.4 | 0.99 | −2.2 | 2.6 | −7.2 | 8.4 | 12.0 | −0.03 | 0.03 | |
SWD (W∙m−2) | 13398 | 183.4 | M92 | 204.9 | 0.91 | 11.7 | 11.8 | 21.5 | 21.6 | 25.0 | 0.1 | 0.1 | |
SWCF (W∙m−2) | 13398 | −51.7 | M92 | −42.0 | 0.90 | −18.7 | 21.2 | −9.7 | 10.9 | 13.3 | −0.3 | 0.3 | |
LWCF (W∙m−2) | 13398 | 29.1 | M92 | 18.5 | 0.68 | −36.4 | 36.5 | −10.6 | 10.6 | 11.6 | −0.5 | 0.5 | |
Figure 2.
Spatial distributions of simulated seasonal-average precipitation for the 2006 and 2011 simulations with M92 against the GPCP data.
Figure 2.
Spatial distributions of simulated seasonal-average precipitation for the 2006 and 2011 simulations with M92 against the GPCP data.
Figure 3.
Spatial distributions of annual mean observed and simulated CCN, CDNC, CF, PWV, CWP, and IWP for the 2006 and 2011 simulations with M92.
Figure 3.
Spatial distributions of annual mean observed and simulated CCN, CDNC, CF, PWV, CWP, and IWP for the 2006 and 2011 simulations with M92.
Figure 3 and
Figure 4 compare observed and simulated annual mean spatial distributions of radiation and cloud related variables for both years. The simulated CCN at a supersaturation (S) of 0.5% (CCN
0.5) are evaluated against MODIS observations over oceanic area (note that no CCN data are available over land areas). Moderate underpredictions of CCN occur for all months and seasons in 2006 and 2011, especially in winter, with annual mean NMBs of −33.8% in 2006, and annual mean NMBs of −32.2% in 2011 (
Table 3 and
Table S1 and
Figure 3). The model does not reproduce the seasonal variation of the CCN over oceanic area. The underpredictions in CCN are due to possible underpredictions of sea salt over coastal areas and uncertainties in the CCN retrievals. CDNC in warm clouds is moderately underpredicted in both years against the MODIS-derived CDNC by Bennartz ([
40]). The annual mean MB and NMB are −41.9 cm
−3 and −29.3% in 2006, respectively. The annual mean MB and NMB are −42.6 cm
−3 and −30.4% in 2011, respectively. The model generally reproduces the seasonal variation of CDNC for both years. As shown in
Figure 3, the largest underpredictions in CDNC occur over mainland China, Japan, and North and South Korea. CDNC depends strongly on several parameters such as CCN, updraft velocity, mass accommodation, and supersaturation in the AR-G02 parameterization. The underpredictions of CDNC are caused by possible underpredictions over land and moderate underpredictions over oceanic areas for CCN as well as possible underestimate in the fraction of activated particles by the AR-G02 parameterization due to the simplified calculation of maximum supersaturation and other assumptions and approximations used in the AR-G02 parameterization ([
51]) and the omission of CCN from insoluble particles such as mineral dust and black carbon through an absorption mechanism ([
55,
57]). The derived CDNC is based on MODIS retrievals of cloud properties such as cloud effective radius (CER), LWP, and COT, all of which are subject to uncertainties. As indicated by Bennartz ([
40]), the errors in CDNC can be up to 260%, especially for regions with low CF (<0.1) (e.g., northwestern China, see
Figure 3). The large uncertainties in derived CDNC may affect the evaluation of the simulated CDNC.
Figure 4.
Spatial distributions of annual mean observed and simulated COT, AOD, LWD, SWD, LWCF, and SWCF for the 2006 and 2011 simulations with M92.
Figure 4.
Spatial distributions of annual mean observed and simulated COT, AOD, LWD, SWD, LWCF, and SWCF for the 2006 and 2011 simulations with M92.
The simulated annual-mean CF agrees well with the MODIS data in terms of domain-wide mean with NMBs of −12.0% in 2006 and −5.7% in 2011. The spatial distributions of annual mean CF are overall consistent with observations except for northwestern China where underpredictions occur (
Figure 3). Relatively large discrepancies of the spatial distribution of simulated CF against MODIS occur in winter and spring in 2006 and 2011, particularly over northern China (Figures not shown). Such discrepancies may be due to the underpredictions of water and possible underpredictions of CCN over northern China in dry seasons (
i.e., winter and spring). The simulated and observed PWV is overall consistent in terms of both magnitudes and spatial distributions for both 2006 and 2011. Large uncertainties exist in the predictions in LWP, IWP, and COT. LWP is significantly underpredicted over the most of the domain, particularly over Tibet Plateau and Mongolia Plateau, in all seasons and annually (see
Table 3 and
Table S1, and
Figure 3), with annual mean NMBs of −56.5% and −49.8%in 2006 and 2011, respectively. The largest underprediction occurs in fall in 2006 and in winter in 2011; the best model performance occurs in summer for both years. Such underpredictions may be due to the model’s limitations in simulating cloud properties, aerosol-cloud interactions, as well as inaccuracies in satellite-derived LWP, which is highly uncertain ([
41]). For example, uncertainties exist in the parameterizations for cloud microphysics and aerosol-cloud interactions used in the model simulations. Zhang
et al. ([
51,
53]) compared two aerosol activation parameterizations and reported that the AR-G02 parameterization used in this work has a tendency to underpredict aerosol activation fraction and thus CDNC, CWP, and COT due to several limitations of the AR-G02 parameterization. The large uncertainties in the satellite retrieval of LWP may also affect the evaluation of the simulated LWP. Both observed and simulated LWP show the highest values in summer in 2006 and the highest values in summer and the second highest values in fall in 2011.
Comparisons with observations indicate that IWP is significantly underpredicted over most of the domain, with annual mean NMBs of −96.1% and −95.7% in 2006 and 2011, respectively. This is due mainly to the uncertainties in predicting cloud ice nuclei (IN) formation and growth and related variables such as cloud ice number concentrations and ice mixing ratio by the parameterizations for cloud microphysics and aerosol-cloud interactions, as well as uncertainties associated with satellite-derived IWP. Another possible reason lies in the underpredictions in the concentrations of aerosols aloft that can serve as IN (note that
Table 4,
Table 5,
Tables S2 and S3 show underpredictions of surface concentrations of PM
2.5 and PM
10), While both the derived and simulated IWP show the highest domain-wide mean in summer, the second highest mean occurs in fall for the satellite-derived IWP but in spring in the simulated IWP. The significant underpredictions of LWP and IWP also indicate a need of improvement for the model treatments for cloud droplet and ice nucleation. Comparing against the MODIS observations, COT is moderately to significantly underpredicted over the entire domain, with annual mean NMBs of −50.0% and −44.0%in 2006and 2011, respectively. The best performance for COT in summer in 2006 and 2011 coincides with the best performance for LWP because they are closely related. While the model reproduces the seasonal variation of COT in 2011, it gives somewhat different seasonal variation from the observations in 2006. The significant underprediction of COT is due not only to underpredictions of LWP, IWP, and CDNC, but also other reasons, e.g., the calculation of COT does not include contributions from graupel.
The MODIS-derived AOD peaks in spring due to the highest PM concentrations, followed by summer, fall, and winter in 2006 and 2011. The satellite-derived high AOD values in spring and summer in both years are attributed to several factors such as stagnant synoptic meteorological conditions, secondary aerosol formation, growth of hydrophilic aerosols due to enhanced relative humidity, and smoke aerosols from regional biomass burning ([
58]). Comparing with MODIS data, moderate underpredictions of AOD occur in all seasons with annual mean NMBs of −35.7% and f −45.2% in 2006 and 2011, respectively. The model reproduces the observed seasonal variations of AOD, with the highest in spring, followed by summer, fall, and winter. As shown in
Figure 4, while the model generally captures the spatial distributions of AOD in northern domain. AOD is significantly underpredicted over the southern domain in both years, where dust concentrations are low and PM
2.5 concentrations are significantly underpredicted (except for Hong Kong) (which will be discussed in
Section 3.2).
LWD and SWD agree well with observations in terms of spatial distributions and magnitudes over the entire domain for all seasons in both years (
Figure 4 and
Table 3 and
Table S1), with annual mean NMBs of −2.2% and 11.7% and R of 1.0 and 0.9 in 2006, and annual mean NMBs of −2.8% and 12.7% and R of 1.0 and 0.9 in 2011. The model generally reproduces the seasonal variation of LWD and SWD in both years. The small underpredictions of LWD and moderate overpredictions of SWD may be due likely to the underpredictions of aerosol direct radiative forcing (indicated by underpredictions of PM and AOD) and underpredictions of cloud radiative forcing (indicated by underpredictions of LWP, IWP, and COT).
The model biases in prediction of clouds parameters such as LWP, IWP, COT, and CF, directly affect the radiative forcing at top-of-atmosphere, aqueous-phase chemistry, and wet scavenging. As shown in
Figure 4 and
Table 3 and
Table S1, LWCF is moderately-to-significantly underpredicted over the entire domain, with annual mean NMBs of −36.4% and −30.9% in 2006 and 2011, respectively. The largest underprediction occurs in winter in both years. Despite the underpredictions, the model reproduces the observed seasonal variations of LWCF, with the highest in summer, followed by spring, fall, and winter. The model is capable of capturing the spatial distributions and seasonal variations of SWCF, but underpredicts its magnitudes. The domain-wide annual mean MB and NMB are −9.7 W∙m
−2 and −18.7%, respectively, in 2006. The domain-wide annual mean MB and NMB are −8.4 W∙m
−2 and −15.5%, respectively, in 2011. Moderate underpredictions of SWCF occur in both years over the entire winter and spring, especially over the southern domain and oceanic areas, leading to moderate undepredictions in annual mean SWCF (
Figure 4). The moderate underpredictions of LWCF and SWCF may be due to the uncertainties associated with the predictions of cloud properties such as CDNC, CF, cloud albedo, and incoming radiation at the top-of-the atmosphere. The biases in meteorological, radiative, and cloud variable predictions will in turn affect gas-phase chemistry and secondary aerosol formation, as discussed below.
3.2. Chemical Predictions
The annual mean performance statistics for surface chemical concentrations and column mass for 2006 and 2011 are given in
Table 4,
Table 5 and
Table S2.
Figure 5 compares the simulated and observed surface mixing ratios (or concentrations) of CO, NO, NO
2, SO
2, and O
3, and concentrations of PM
2.5 and PM
10 from various datasets in Hong Kong, Taiwan, Japan, South Korea, and mainland China for 2006. The results in 2011 are similar to those in 2006, thus not shown. The mixing ratios of CO and NO
2 at the Hong Kong sites are slightly-to-moderately underpredicted, with annual mean NMBs of −17.8% and −2.4% in 2006, and annual mean NMBs of −26.9% and −13.3% in 2011, respectively. The mixing ratios of NO at the Hong Kong sites are significantly underpredicted, with annual mean NMBs of −86.8% and −83.1% in 2006 and 2011, respectively. The possible reasons for underpredictions in NO and CO include underpredictions in their emissions, overpredictions in WS10, and overestimations of planetary boundary layer (PBL) height. The mixing ratios of O
3 at the Hong Kong sites are moderately overpredicted due likely to insufficient of titration by NO, with annual mean NMBs of 22.4%, and 33.9% in 2006 and 2011, respectively. The mixing ratios of SO
2, and mass concentrations of PM
2.5, and PM
10 in Hong Kong are moderately to significantly overpredicted, with annual mean NMBs of 265%, 154%, and 79% in 2006, and 478%, 85.5%, and 31.3%, in 2011, respectively. The overpredictions of PM
2.5 and PM
10 are clearly caused by the significant overprediction of SO
2, which is due likely to the overestimation of SO
2 emissions in the MEIC used and overpredictions in precipitation in the Hong Kong area (see
Figure 2), which may lead to the overpredictions of SO
42− and NH
4+. Another possible reason for overpredictions of PM
2.5 and PM
10 may be the overestimation of emissions of dust particles that are transported from nearby mega cities, such as Guangzhou and Shenzhen. As shown in
Figure 5, most simulated values of CO, NO
2, and O
3 at the Hong Kong sites are within a factor of two of the observations, with relatively higher
R values of 0.41, 0.46, and 0.52, respectively. However, most simulated values of NO, SO
2, PM
2.5, and PM
10 are beyond a factor of two of the observations, with low
R values of 0.3, 0.19, 0.31, and 0.3, respectively. At the Hong Kong sites, the model reproduces the seasonal variations of CO in both years and those of PM
2.5 and PM
10 in 2006, but fails to reproduce the seasonal variations of NO, NO
2, SO
2, and O
3 in both years.
At the Taiwan sites, the mixing ratios of O
3 are moderately overpredicted with annual mean NMBs of 18.9% and 16.6% in 2006 and 2011, respectively. The concentrations of all other species are moderately to significantly underpredicted, with annual mean NMBs of −48.0% and −41.1% for CO, −86.7% and −79.6% for NO, −47% and −36.4% for NO
2, −66.0% and −62.2% for SO
2, −49.7% and −49.5% for PM
2.5, and −65.8% and −62.7% for PM
10 in 2006 and 2011, respectively. The significant underpredictions for CO, NO
x, SO
2, PM
2.5, and PM
10 are associated with the underestimations for anthropogenic emissions and the overpredictions for WS10 (Figure not shown) and precipitation in this region (see
Figure 2). Moderate underpredictions of cloud amounts and significant underpredictions of LWP (
Table 3 and
Table S1) might affect the aqueous-phase chemistry in cloud, which is a major source for SO
42−. The significant underpredictions for NO and NO
2, and moderate overpredictions for O
3 indicate possible underestimations in NO
x emissions, leading to insufficient NO
x for titration of O
3. The concentrations of SO
2, PM
2.5, and PM
10 are largely overpredicted in Hong Kong but underpredicted in Taiwan, indicating some problems in the spatial distributions of the emissions of SO
2 and primary PM in the MEIC emission inventory. As shown in
Figure 5, at the Taiwan sites, most simulated values are within a factor of two of the observations for CO and O
3 with R values of 0.39 and 0.33, but fall beyond a factor of two of the observations for NO, NO
2, SO
2, PM
2.5, and PM
10 with R values of 0.66, 0.2, 0.05, 0.22, and 0.25, respectively. At the Taiwan sites, the model reproduces the seasonal variations of O
3 and PM
10 in 2006, and those of CO, O
3, PM
2.5, and PM
10 in 2010.
In Japan, the mixing ratios of CO are significantly underpredicted, with annual mean NMBs of −63.7% and −55.4% in 2006 and 2011, respectively. The mixing ratios of NO and NO2 are also significantly underpredicted, with annual means of −93.7% and −89.0% for NO and annual means of −59.0% and −46.1%, for NO2 in 2006 and 2011, respectively. The large underpredictions may be caused by underestimations of anthropogenic CO and NOx emissions, overpredictions of WS10, and overestimations of PBL height (PBLH). The underpredictions in NOx mixing ratios may have resulted in insufficient titration of O3 by NO. The mixing ratios of SO2 are also significantly underpredicted with annual mean NMBs of −56.9% and −41.4% in 2006 and 2011, respectively. Similar to the results at the Taiwan sites, only the annual mean mixing ratio of O3 is overpredicted in Japan, with NMBs of 11% and 14.9%, respectively.
Table 4.
Performance statistics for surface chemical concentrations and column mass abundance of gaseous species for 2006.
Table 4.
Performance statistics for surface chemical concentrations and column mass abundance of gaseous species for 2006.
Variable | Data Source | Number | Mean Obs. | Mean Sim. | R | NMB (%) | NME (%) | MB | MAGE | RMSE | FB | FGE |
---|
CO (µg·m−3) | HK | 8760 | 855.7 | 703.6 | 0.41 | −17.8 | 33.8 | −152.1 | 289.0 | 380.7 | −0.23 | 0.37 |
CO (ppm) | TW | 324 | 0.5 | 0.2 | 0.39 | −48.0 | 50.1 | −0.2 | 0.2 | 0.3 | −0.57 | 0.62 |
JP | 1390 | 0.5 | 0.2 | 0.03 | −63.7 | 64.1 | −0.3 | 0.3 | 0.4 | −0.87 | 0.88 |
SK | 731 | 0.6 | 0.2 | 0.28 | −62.0 | 62.3 | −0.4 | 0.4 | 0.4 | −0.82 | 0.83 |
Col. CO (1018 molec.·cm−2) | MOPPIT | 13398 | 2.0 | 2.0 | 0.94 | 0.3 | 7.5 | 0.01 | 0.2 | 0.2 | 0.00 | 0.08 |
NO (µg·m−3) | HK | 8758 | 103.5 | 13.8 | 0.30 | −86.6 | 86.7 | −89.7 | 89.8 | 109.2 | −1.57 | 1.57 |
NO (ppb) | TW | 324 | 5.8 | 0.8 | 0.66 | −86.7 | 86.7 | −5.0 | 5.0 | 6.6 | −1.47 | 1.47 |
JP | 2670 | 7.8 | 0.5 | −0.02 | −93.7 | 94.3 | −7.3 | 7.4 | 10.3 | −1.55 | 1.63 |
NO2 (µg·m−3) | CH | 40 | 125.8 | 16.0 | 0.04 | −87.3 | 87.3 | −109.8 | 109.8 | 118.3 | −1.59 | 1.59 |
HK | 8760 | 62.1 | 60.6 | 0.46 | −2.4 | 41.3 | −1.5 | 25.7 | 33.2 | −0.09 | 0.43 |
NO2 (ppb) | TW | 324 | 15.2 | 8.1 | 0.20 | −47.0 | 52.7 | −7.2 | 8.0 | 10.2 | −0.57 | 0.69 |
JP | 2670 | 12.2 | 5.0 | −0.07 | −59.0 | 70.5 | −7.2 | 8.6 | 10.6 | −0.8 | 1.00 |
SK | 732 | 17.4 | 9.8 | 0.16 | −43.3 | 60.7 | −7.5 | 10.6 | 13.0 | −0.49 | 0.80 |
Col. NO2 (1015 molec.·cm−2) | SCIAMACHY | 13398 | 2.3 | 2.5 | 0.91 | 7.6 | 34.2 | 0.2 | 0.8 | 1.8 | 0.07 | 0.36 |
SO2 (µg·m−3) | CH | 2600 | 101.5 | 67.1 | −0.12 | −33.9 | 66.9 | −34.4 | 67.9 | 88.9 | −0.56 | 0.82 |
HK | 8760 | 21.8 | 79.9 | 0.19 | 265.8 | 273.0 | 58.0 | 59.6 | 78.6 | 1.03 | 1.07 |
SO2 (ppb) | TW | 324 | 4.3 | 1.5 | 0.05 | −66.4 | 74.1 | −2.9 | 3.2 | 3.8 | −1.05 | 1.10 |
JP | 2612 | 2.7 | 1.2 | −0.18 | −56.9 | 72.8 | −1.6 | 2.0 | 2.6 | −0.66 | 0.99 |
SK | 732 | 5.0 | 3.6 | 0.32 | −28.3 | 52.5 | −1.4 | 2.6 | 3.5 | −0.31 | 0.63 |
Col. SO2 (DU) | SCIAMACHY | 13398 | 0.2 | 0.3 | 0.87 | 62.9 | 103.5 | 0.1 | 0.2 | 0.4 | −0.15 | 0.72 |
Col. HCHO (1015 molec.·cm−2) | SCIAMACHY | 13398 | 5.3 | 6.1 | 0.83 | 15.0 | 26.1 | 0.0 | 0.8 | 1.9 | 0.06 | 0.25 |
O3 (µg·m−3) | HK | 8760 | 35.8 | 43.8 | 0.52 | 22.4 | 87.5 | 8.0 | 31.3 | 46.0 | −0.35 | 0.95 |
O3 (ppb) | TW | 324 | 31.3 | 37.2 | 0.33 | 18.9 | 28.6 | 5.9 | 8.9 | 10.9 | 0.18 | 0.26 |
JP | 2355 | 31.6 | 35.2 | 0.48 | 11.1 | 23.3 | 3.5 | 7.4 | 9.2 | 0.13 | 0.23 |
SK | 732 | 25.5 | 36.0 | 0.44 | 40.9 | 52.9 | 10.4 | 13.5 | 16.3 | 0.34 | 0.45 |
TOR (DU) | OMI | 13398 | 30.7 | 33.6 | 0.95 | 9.5 | 9.7 | 2.9 | 3.0 | 3.4 | 0.09 | 0.09 |
PM2.5 (µg·m−3) | HK | 8757 | 40.8 | 103.4 | 0.31 | 153.6 | 167.2 | 62.7 | 68.2 | 111.8 | 0.64 | 0.77 |
TW | 324 | 31.7 | 15.9 | 0.22 | −49.7 | 52.7 | −15.8 | 16.7 | 20.9 | −0.62 | 0.66 |
PM10 (µg·m−3) | CH | 1030 | 98.6 | 97.8 | 0.09 | −0.8 | 58.0 | −0.8 | 57.2 | 74.5 | −0.12 | 0.60 |
HK | 8760 | 58.7 | 105.1 | 0.30 | 79.0 | 104.7 | 46.4 | 61.5 | 103.6 | 0.33 | 0.60 |
TW | 324 | 57.6 | 19.7 | 0.25 | −65.8 | 66.1 | −37.9 | 38.1 | 44.1 | −0.95 | 0.96 |
JP | 2719 | 23.4 | 13.3 | −0.01 | −42.9 | 52.9 | −10.0 | 12.4 | 15.1 | −0.56 | 0.67 |
SK | 789 | 47.9 | 30.2 | 0.27 | −37.0 | 46.0 | −17.7 | 22.0 | 26.5 | −0.50 | 0.58 |
Table 5.
Performance statistics for surface chemical concentrations of gaseous species for 2006 against EANET.
Table 5.
Performance statistics for surface chemical concentrations of gaseous species for 2006 against EANET.
Variable | Region | Number | Mean Obs. | Mean Sim. | R | NMB (%) | NME (%) | MB | MAGE | RMSE | FB | FGE |
---|
NO (ppb) | CH | 24 | 3.0 | 0.4 | 0.53 | −86.8 | 86.9 | −2.6 | 2.6 | 3.5 | −1.23 | 1.23 |
JP | 12 | 3.7 | 1.3 | −0.43 | −66.2 | 66.2 | −2.4 | 2.4 | 2.5 | −0.99 | 0.99 |
NO2 (ppb) | CH | 24 | 17.6 | 4.3 | 0.61 | −75.4 | 75.4 | −13.3 | 13.3 | 14.7 | −1.37 | 1.37 |
JP | 12 | 3.7 | 1.3 | −0.43 | −66.2 | 66.2 | −2.4 | 2.4 | 2.5 | −0.99 | 0.99 |
SO2 (ppb) | CH | 48 | 11.0 | 6.3 | 0.41 | −42.2 | 54.7 | −4.6 | 6.0 | 8.3 | −0.66 | 0.82 |
JP | 106 | 0.6 | 0.4 | 0.62 | −29.8 | 59.3 | −0.2 | 0.4 | 0.5 | −0.76 | 0.98 |
SK | 36 | 2.5 | 0.1 | 0.75 | −96.9 | 96.9 | −2.4 | 2.4 | 2.6 | −1.88 | 1.88 |
O3 (ppb) | JP | 118 | 41.7 | 34.9 | 0.26 | −16.2 | 32.4 | −6.8 | 13.5 | 16.1 | −0.14 | 0.38 |
SK | 36 | 37.1 | 30.0 | 0.17 | −19.1 | 28.2 | −7.1 | 10.5 | 13.1 | −0.20 | 0.31 |
PM2.5 (µg m−3) | JP | 24 | 11.7 | 4.4 | 0.07 | −62.0 | 62.0 | −7.2 | 7.2 | 8.7 | −0.87 | 0.87 |
SO4 (µg m−3) | JP | 120 | 4.3 | 2.3 | −0.05 | −47.8 | 63.8 | −2.0 | 2.8 | 3.7 | −0.46 | 0.82 |
SK | 35 | 7.7 | 1.6 | −0.36 | −79.6 | 79.6 | −6.1 | 6.1 | 7.7 | −1.2 | 1.2 |
PM10 (µg m−3) | CH | 48 | 71.0 | 41.3 | 0.53 | −41.8 | 56.0 | −29.7 | 39.7 | 50.3 | −0.68 | 0.71 |
JP | 117 | 21.7 | 13.1 | 0.25 | −39.9 | 51.9 | −8.7 | 11.3 | 14.1 | −0.52 | 0.68 |
SK | 36 | 50.0 | 9.5 | 0.03 | −81.0 | 81.0 | −40.5 | 40.5 | 43.3 | −1.35 | 1.35 |
Figure 5.
Simulated and observed surface mixing ratios of CO, NO, NO2, SO2, and mass concentrations of PM2.5 and PM10 for the 2006 simulation with M92.
Figure 5.
Simulated and observed surface mixing ratios of CO, NO, NO2, SO2, and mass concentrations of PM2.5 and PM10 for the 2006 simulation with M92.
The concentrations of PM
10 are significantly underpredicted, with annual mean NMBs of −42.9% and −35.0% in 2006 and 2011, respectively. The largest underpredictions occur in summer, followed by fall, winter, and spring in both years. The smallest underprediction of PM
10 in spring among all seasons reflects the impact of the long-range transport of dust particles from China. The significant undeprediction of PM
10 in summer, on the other hand, may be due to several reasons including the overprediction of wet deposition resulted from overpredicted precipitation, and the underestimation of anthropogenic emissions of primary aerosol particles and precursor gases for secondary aerosols. As shown in
Figure 5, at the Japan sites, most simulated O
3 mixing ratios are within a factor of two of the observations with
R of 0.48. Most simulated values for other species are beyond a factor of two of the observations with
R values of 0.03, −0.02, −0.07, −0.18, and −0.01 for CO, NO, NO
2, SO
2, and PM
10, respectively. The negative
R values may indicate inconsistent or even opposite spatial distributions and/or temporal variations of the simulated concentrations of those species, due likely to spatial and temporal variations in emissions and the large biases in simulated meteorological fields over Japan as shown in
Figure 1. At the Japan sites, the model reproduces the seasonal variations of O
3 but fails to reproduce those of CO, NO, NO
2, SO
2, and PM
10 in both years.
Similar to the statistical results for Japan and Taiwan sites, the mixing ratios of CO, NO
2, and SO
2 in South Korea are moderately to significantly underpredicted, with annual mean NMBs of −62.0%, −43.3%, and −28.3% in 2006, and −51.9%, −37.2%, and −21.6% in 2011, respectively. The mixing ratios of surface O
3 in South Korea are moderately overpredicted in spring, fall, and winter with NMBs of 14.5%, 41.8%, and 27.1% in 2006 and 10.0%, 37.9%, and 12.2% in 2011, and significantly overpredicted in summer, with NMBs of 83.2% in 2006, and 68.7% in 2011, respectively. The mass concentrations of PM
10 are moderately underpredicted, with annual mean NMBs of −37.0% and −31.7% in 2006 and 2011, respectively. The moderate to significant underpredictions of the mixing ratios of CO, NO
2, and SO
2, and mass concentrations of PM
10 are possibly due to the underestimation of the emissions of these species, overpredictions in WS10 (see
Table 3 and
Table S1), and the overpredictions of precipitation over South Korea (see
Figure 2). As shown in
Figure 5, at the South Korea sites, most simulated O
3 values are within a factor of two of the observations with an
R value of 0.44. Most simulated values for CO, NO
2, SO
2, and PM
10 are beyond a factor of two of the observations with
R values of 0.28, 0.16, 0.32, and 0.27, respectively. The model reproduces the seasonal variations of O
3 and PM
10 in 2011, but does not capture well the seasonal variations of other species in 2011 and all species in 2006.
The API-derived NO2 concentrations are significantly underpredicted with annual mean NMBs of −87.3% in 2006 and −48.0% in 2011. The API-derived SO2 concentrations are moderately overpredicted in summer and moderately to significantly underpredicted in other seasons in 2006. They are moderately overpredicted in spring and summer and moderately to significantly underpredicted in fall and winter in 2011. The overpredictions and underpredictions in SO2 concentrations compensate, leading to annual mean NMBs of −33.9% in 2006 and −18.2% in 2011. The API-derived PM10 concentrations are overall well produced with annual mean NMBs of −0.8% and −3.6% in 2006 and 2011, respectively. In 2006, PM10 is well simulated in summer and fall, but moderately underpredicted in winter and ovepredicted in spring. In 2011, PM10 is moderately overpredicted in spring and underpredicted in other seasons.
The performance of PM
10 in this work is consistent or even better than those from WRF/Chem applications over East Asia reported in the literature ([
53,
59]). As shown in
Figure 5, over mainland China, many simulated values of NO
2, SO
2, and PM
10 are beyond a factor of observations with
R values of 0.04, −0.12, and 0.09 for NO
2, SO
2, and PM
10, respectively. The model fails to reproduce the seasonal variation of those species over mainland China. The simulated spatial distributions of PM
10 in
Figure 6 show that the major dust source regions in East Asia are located in northwestern and northern China, and southern Mongolia, which is consistent with previous studies ([
60,
61]). In the dust season (spring), large amounts of dust particles are generated in these regions and transported to eastern China, southeastern China, South Korea, and Japan. The model overpredicts the particle concentrations in northern China, especially over the dust source regions in northwestern China, but underpredicts those in southern China, leading to a very low R value (
i.e., 0.09 in 2006 and 0.2 in 2011) for PM
10. The overpredictions of the concentrations of coarse aerosol particles in northern China might be associated with the overestimations of dust emissions in dust source regions, while the underpredictions of the concentrations of PM
10 in southern China might be due to the underestimation of anthropogenic emissions for primary particles and the precursor gaseous species for secondary aerosols, as well as the overprediction of precipitation (see
Figure 2) (which may lead to overpredictions of wet deposition of chemical species in this region).
Figure 7 compares annual mean concentrations of PM
2.5 and its major components such as SO
42−, NH
4+, Cl
−, and Na
+ at the THU and MY sites in Beijing, China. Although WRF-CAM5 uses a relatively simple aerosol module that is based on the modal approach and does not simulate nitrate, the simulated mass concentrations of PM
2.5 and SO
42− agree well with the observations at both sites, indicating a good skill of WRF-CAM5 in simulating site-specific PM
2.5 and SO
42−. However, the concentrations of Na
+ and Cl
− at both sites are significantly underpredicted because anthropogenic sources predominate at both sites, and the anthropogenic emissions of Na
+ and Cl
− are not included in the emission file.
As shown in
Table 5 and
Table S3, the concentrations of NO, NO
2, SO
2, O
3, SO
42−, PM
2.5, and PM
10 are underpredicted at all EANET sites in mainland China, Japan, South Korea including urban, rural and remote sites, which is consistent with surface evaluation using data from other surface networks. The underpredictions in NO
x and SO
2 may be caused by underestimations in their total emissions and/or vertical allocations of the total emissions, as well as strong cloud lofting. The underpredictions in SO
42− concentrations may be caused by several reasons such as the underpredictions of SO
2 and the overprediction of precipitation. The underpredictions in PM
2.5 and PM
10 may result from the underestimations of emissions of their gaseous precursors and primary PM species such as black carbon, organic carbon, and mineral dust. Unlike Hong Kong where the underpredictions of NO
x level lead to the overpredictions in O
3 due to insufficient titration, the underestimations of NO
x emissions at the EANET sites may contribute to the moderate underpredictions of O
3 concentrations, because most of the EANET sites are located at rural and remote sites where O
3 chemistry is NO
x-limited as shown in Liu
et al. ([
62]). At some urban sites where O
3 chemistry is VOC-limited or both NO
x- and VOC-limited, underestimations of VOC emissions may also contribute to the underpredictions of O
3.
Figure 6.
Simulated PM10 concentrations overlaid with API-derived observations for 2006 and 2011 simulations with M92. The observational data are denoted as dots.
Figure 6.
Simulated PM10 concentrations overlaid with API-derived observations for 2006 and 2011 simulations with M92. The observational data are denoted as dots.
Figure 7.
Simulated and observed surface concentrations of PM2.5 and PM2.5 composition at THU and MY sites in Beijing, China for the 2006 simulation with M92. WRF-CAM5 does not simulate NO3−.
Figure 7.
Simulated and observed surface concentrations of PM2.5 and PM2.5 composition at THU and MY sites in Beijing, China for the 2006 simulation with M92. WRF-CAM5 does not simulate NO3−.
Figure 8 compares the spatial distributions of annual mean simulated column mass abundance with satellite observations. The simulated column CO is comparable with observations from MOPITT in terms of spatial distributions and magnitudes for both years with annual mean NMBs of 0.3% in 2006 and −1.8% in 2011 (
Table 4 and
Table S2). Moderate underpredictions occur in spring, and slight or moderate overpredictions occur in other seasons. While the satellite-derived CO column abundances show the highest in spring, followed by winter, fall, and summer, the simulated CO column abundances show the highest in winter, followed by spring, fall, and summer for both years. CO is a slowly reacting gas in the atmosphere with a sink reaction of CO + OH and the secondary formation through the oxidation of volatile organic carbons (VOCs). Its fate is mainly affected by emissions, transport, and deposition processes. Great biomass burning activities, which are large contributors to CO emissions, over South and Southeast Asia in spring were reported ([
63]). The CO emissions from biomass burning in spring over Southeast Asia may lead to a relatively high background CO concentrations over East Asia that WRF-CAM5 cannot reproduce because it does not represent such biomass burning emissions, leading to moderate underpredictions in column CO abundances in spring. Column abundances of NO
2 agree well with observations against SCIAMACHY in terms of spatial distributions, magnitudes, and seasonal variations, with small underpredictions in spring, and small to moderate overpredictions in other seasons. The annual mean NMBs are 7.6% in 2006 and 0.4% in 2011. The good performance of the column abundances of NO
2 but the significant underpredictions of surface mixing ratios of NO and NO
2 suggest some uncertainties in the vertical distribution of NO
x emissions used in the model.
Column SO
2 is moderately to significantly overpredicted in all seasons in 2006 with NMBs of 18.7% to 137%. In 2011, column SO
2 is moderately overpredicted in fall and winter and underpredicted in spring and summer. The annual mean NMBs of column SO
2 are 62.9% in 2006 and −14.5% in 2011. As shown in
Figure 8, significant overpredictions of column SO
2 occur in North China Plain, indicating significant overpredictions of anthropogenic emissions of SO
2 and/or possible uncertainties in the vertical allocation of SO
2 emissions, namely, more SO
2 emissions should have been allocated to surface layer rather than upper layers. Another possible reason for overpredictions in SO
2 aloft is the inefficient conversion into sulfate by cloud chemistry. While the SO
2 column abundance is overpredicted for all seasons in 2006 and fall and winter 2011, the mixing ratios of SO
2 at surface are significantly underpredicted, indicating the uncertainties in the vertical distribution of SO
2 emissions. This also indicates vigorous cloud lofting. Overpredictions in precipitation shown in
Table 3 and
Table S1 also contribute to the underpredictions in surface SO
2 concentrations in both years. While the model fails to reproduce the observed seasonal variations of the column SO
2 abundances, there are a large fraction of missing values and the reported overall error in the SO
2 retrievals is 45%–80% for annual averages over polluted regions ([
64]), the relatively poor data quality and inaccuracies in the retrieval algorithms for SO
2 would affect the evaluation of column SO
2. The situation is somewhat different in 2011. As shown in
Table S2, SO
2 is moderately to significantly underpredicted at the surface (except for Hong Kong) and also aloft in 2011. In such a case, possible reasons for underpredicted surface and aloft SO
2 include overpredicted precipitation, underestimation in total SO
2 emissions, and uncertainties in the SO
2 retrieval.
In 2006, the column HCHO is slightly to moderately overpredicted with an annual mean NMB of 15.0%. In 2011, the column HCHO is slightly underpredicted in spring and winter and slightly overpredicted in summer and fall, leading to an annual mean NMB of −0.3%. The satellite-derived HCHO column abundances show the highest in summer, followed by spring, fall, and winter, the simulation shows the highest in summer, followed by fall, spring, and winter for both years. As shown in
Figure 8, the overpredictions occur mainly in central China, Napel, northeastern India, and northern Burma, Thailand, Laos, and Vietnam. Possible reasons for such overpredictions may include uncertainties in HCHO emissions, biogenic emissions that can produce secondary HCHO, and satellite retrievals. In particular, De Smedt
et al. ([
65]) reported the errors in HCHO retrievals of (0.5–2.0)× 10
15 molecules∙cm
−2 are on the same order of magnitudes or even larger than the MBs in the simulated HCHO column for both years.
Figure 8.
Spatial distributions of annual mean observed and simulated column CO, NO2, SO2, HCHO, and TOR for the 2006 and 2011 simulations with M92.
Figure 8.
Spatial distributions of annual mean observed and simulated column CO, NO2, SO2, HCHO, and TOR for the 2006 and 2011 simulations with M92.
The simulated TOR agrees well with observed TOR with annual mean NMBs of TOR are 9.5% in 2006 and 7.9% in 2011. TOR is slightly underpredicted for summer and overpredicted for other seasons. The largest overprediction occurs in winter with an NMB of 26.5%, and an MB of 6.7 DU in 2006 and an NMB of 24.0%, and an MB of 6.2 DU in 2011. The satellite derived O3 column abundances show the highest in summer, followed by spring, fall, and winter, where the simulated column abundances show the highest in spring, followed by summer, winter, and fall in both years. The model performs better in predicting column TOR than surface O3 concentration, indicating that TOR predictions largely depend on O3 from upper atmosphere.