SUPPLEMENTARY MATERIAL

Transcription

1 SUPPLEMENTARY MATERIAL Impacts of Air-Sea Interactions on Regional Air Quality Predictions Using a Coupled Atmosphere-Ocean Model in Southeastern U.S. Jian He, Ruoying He, and Yang Zhang * Department of Marine, Earth, and Atmospheric Sciences, North Carolina State University, Raleigh, NC, U.S.A. Impacts of Cumulus Parameterizations Table 1 in the main text shows the two simulations conducted to compare different cumulus parameterizations in this work, Wang et al. (2015a) (refer to as W15) and SEN1. The cumulus parameterization scheme used in this work is based on Grell 3D ensemble scheme (referred to as G3D, Grell and Devenyi, 2002; Grell and Freitas, 2014), which allows for a series of different assumptions that are commonly used in convective parameterizations and includes options to spread subsidence to neighboring grid points. In addition to the options listed in Table 1 of Wang et al. (2015a), W15 also includes prescribed SST forcing from NCEP by updating every 6-hour. SEN1 is conducted with the same model configurations as W15 but with a different cumulus parameterization scheme based on Grell and Freitas (2014) (referred to as GF scheme), which allows for subgrid scale convection representation. The differences in the model results between W15 and SEN1 can provide insights about the sensitivity of cumulus parameterization on model meteorological, cloud/radiative, and chemical predictions. Comprehensive model comparisons between GF scheme and G3D scheme and model evaluations are discussed as below. * Corresponding author. Tel: ; Fax: address: yang_zhang@ncsu.edu 1

2 Meteorological Predictions Figures S1a and b show the absolute differences in monthly-averaged meteorology, cloud/radiative variables, and chemical predictions between SEN1 and W15. Compared to W15, SEN1 predicts higher T2 over most land area, and lower T2 over part of oceanic area. The increases of T2 in SEN1 can be up to 0.76 o C and the decrease of T2 can be up to 0.4 o C, with a domain averaged increase of 0.07 o C. The increase of T2 over land and decrease of T2 over ocean are mainly due to an increase of SWD by 5.2 W m -2 over land and a decrease of SWD by 2.5 W m -2 over ocean. Q2 decreases in SEN1 over most of domain, with a domain averaged decrease of 0.3 g kg -1, indicating much drier conditions predicted in SEN1. GF scheme is designed to be less active as the grid size reduces to cloud resolving scales. Precipitation decreases in SEN1 over most of domain, with a domain averaged reduction of 3.9 mm day -1. The reduction of the total precipitation is mainly from the dominance of the decrease in nonconvective precipitation (by domain averaged of 6.9 mm day -1 ) over the increase in convective precipitation (by domain averaged of 3.0 mm day -1 ). This is because, comparing to the G3D scheme, the GF scheme predicts stronger convection which leads to stronger detrainment of cloud water and ice near the cloud top, drying the troposphere and reducing the grid-scale precipitation. As explained in Grell and Freitas (2014), less precipitation simulated by GF than G3D is attributed to the differences in autoconversion mechanism used in both schemes. Due to the stronger convection in SEN1, PBLH predicted by SEN1 also increases up to 185 m, with a domain averaged increase of 55.0 m. Due to higher T2 and lower Q2 predicted by SEN1, less water vapor can condense onto the CCN surface. As a result, SEN1 predicts smaller cloud droplets than W15. CDNC predicted by SEN1 varies throughout the domain, with increases up to 993 cm -3 and decreases up to 749 cm -3, 2

3 resulting in a domain averaged increase of 3.4 cm -3. The decreases of LWP and CF are associated with the decrease in large-scale precipitation, which is affected by the GF scheme. As a result, SEN1 predicts lower CF and LWP, with a domain averaged decrease of 7.0% and 19.4 g m -2, respectively. However, COT increases in SEN1, with a domain averaged increase of The increase of COT is likely due to the decrease of cloud effective radius from smaller cloud droplets in SEN1. Although both CDNC and COT increase over land, the significant decrease of CF and LWP over land can result in a decrease in cloud albedo, and therefore a decrease in SWCF over land and near coastal areas in SEN1. The increase of SWCF over remote ocean is mainly due to the increase of CDNC and COT over these regions. As a result, compared to W15, SEN1 predicts higher SWCF up to 49 W m -2 and lower SWCF up to 45.3 W m -2, with a domain averaged decrease of 1.7 W m -2. The decrease or increase in SWCF can result in an increase or a decrease, respectively, in SWD. Compared to W15, SEN1 predicts higher SWD by up to W m -2 and lower SWD by up to 51.8 W m -2, with a domain averaged increase of 1.1 W m -2. Tables S1a and b show the model performance of W15 and SEN1. Meteorological variables such as T2, RH2, and SST are well predicted in both W15 and SEN1, with NMBs within 6%, and with slightly better performance in SEN1. RMSEs of T2 and SST are very similar in W15 and SEN1, with a slightly lower RMSE of T2 in SEN1. The RMSE of RH2 is also lower in SEN1 than that in W15. Vertical evaluation of monthly-averaged temperature and specific humidity are shown in Figures 2a and b. As shown in Figure 2a, temperature profiles are well simulated compared to the NCEP reanalysis data. However, relatively large biases exist for specific humidity profiles. For example, both W15 and SEN1 overpredict specific humidity profiles (see Figure 2b) over Gulfport Youth Court (with a smaller bias by SEN1). But W15 predicts better specific humidity profile over Indian River Lagoon, Everglades, and Cape 3

4 Romain than SEN1. WS10 is moderately underpredicted against observations at the NCEI sites in both W15 and SEN1, with NMBs of -58.3% and -61.3%, respectively, whereas it is well predicted against the observations at the CASTNET sites, with NMBs of 5.4% and 4.9%, respectively, for W15 and SEN1. A slight shift in wind direction (i.e., WD10) to more westerly is predicted in both W15 and SEN1, with relatively good correlation coefficients of 0.6 for both cases. With GF in SEN1, PBLH is impacted significantly over ocean, with increasing NMBs from 0.2% in W15 to 16.2% in SEN1 against the NCEP reanalysis data. The biases in PBLH can be due in part to different methods for calculating PBLH in the NCEP models (e.g., the Global Forecast System and North American Model) and WRF. Also, Seidel et al. (2012) found that the NCEP reanalysis data showed deeper PBLH due to difficulty in simulating stable conditions compared with radiosonde observations. Therefore, the performance of PBLH here can only represent the deviation from the NCEP models. Both LHFLX and SHFLX are overpredicted in W15 and SEN1, which is mainly due to a lack of representations of the air-sea interactions. The uncertainties associated with the WRF/Chem s representations of convection-cloud-radiation may also contribute to the overpredictions. For example, Alapaty et al. (2012) reported that lack of the model treatments of subgrid cloud feedbacks in radiation calculation in WRF can explain overpredictions in shortwave radiation and precipitation. Even though the convection-cloud-radiation feedback is included in G3D/GF in WRF/Chem v3.6.1, uncertainties exist in the model representation of such feedbacks. Compared to W15, SEN1 improves the predictions of Precip over land (ocean) significantly, by reducing NMBs of 95.9% (335.2%) to 31.7% (211.5%) against GPCP. The RMSE of Precip against GPCP is substantially reduced from W15 (7.6 mm day -1 ) to SEN1 (2.8 mm day -1 ). CF is also improved by reducing NMBs from 23.7% in W15 to -1.0% in SEN1 over 4

5 land, and from 48.7% in W15 to 42.3% in SEN1 over ocean, compared to CERES SYN1deg observations. LWP is more underpredicted with NMBs from -35.0% in W15 to -80.7% in SEN1 over land but improved substantially over ocean with NMBs from 304.6% in W15 to 35.1% in SEN1. The significant decrease of LWP over ocean is likely due to smaller cloud effective radius associated with higher CDNC, resulting from less precipitation in SEN1. The cloud effective radius is not included in the model output. However, LWP is proportional to both the COT and the effective radius, since SEN1 gives higher COT than W15, the decrease of LWP in SEN1 can be due to the smaller cloud effective radius in SEN1. Also, the increase of CDNC is usually associated with a decrease in cloud effective radius. Total precipitation is reduced in SEN1, resulting in more aerosols that can be activated to increase CDNC. The performance of COT is improved over land with NMBs reducing from -70.3% in W15 to -39.5% in SEN1 whereas it is degraded over ocean, with NMBs increasing from -21.8% in W15 to 64.6% in SEN1. The large overpredictions of COT over ocean are likely due to the smaller cloud effective radius in SEN1, which indicates the uncertainties in the treatments of cloud dynamics and thermodynamics. Compared to MODIS data, PWV over land is more underpredicted, with NMBs from -0.5% in W15 to -5.5% in SEN1, and the performance of PWV over ocean is from overprediction by 3.2% in W15 to underprediction by 4.2% in SEN1. The performance of AOD over land is slightly degraded with NMBs from -10.8% in W15 to -11.5% in SEN1 and the performance of AOD over ocean is slightly improved with NMBs from -1.0 in W15 to -0.3% in SEN1. The predictions of CCN5 are improved in SEN1, with NMBs from 21.1% in W15 to -0.8% in SEN1. The decreases of CCN5 in SEN1 are mainly due to the lower aerosol number concentrations in SEN1. The overpredictions of CDNC are largely due to the uncertainties in the observations as there are only a few grid cells that contain observations. 5

6 Model predictions of radiative variables such as LWD, SWD, and OLR are comparable in W15 and SEN1, with slightly better performance of LWD over ocean (NMBs of 0.8% vs. 0.3% for W15 and SEN1, respectively), SWD over land (NMBs of -2.4% vs. 0.5%), and OLR over land (NMBs of -13.9% vs. -7.0%) and ocean (NMBs of -22.9% vs %) in SEN1. On the other hand, there are significant changes in SWCF and LWCF. Compared to W15, the domain averaged SWCF predicted by SEN1 decreases from W m -2 to W m -2 over land (with NMBs reduced from 57.2% to 45.1%), and it increases from W m -2 to W m -2 over ocean (with NMBs increased slightly from 108.9% to 112.5%). The domain averaged LWCF predicted by SEN1 decreases from 53.2 W m -2 to 37.7 W m -2 over land (with NMBs reduced significantly from 61.7% to 18.9%), and it decreases from 77.8 W m -2 to 74.5 W m -2 over ocean (with NMBs reduced from 152.1% to 141.4%). The improvements of SWCF and LWCF over land are attributed to the improvement of cloud variables (e.g., CF and COT) over land. The large overpredictions of SWCF and LWCF over ocean are attributed to the inaccurate predictions clouds over ocean, indicating the model uncertainties in the cloud dynamics and thermodynamics. Impacts on Chemical Predictions As shown in Figure S1b, compared to W15, SEN1 predicts higher surface CO and SO 2, with domain averaged of 6.2 ppb and 0.06 ppb, respectively. The increased mixing ratios of CO and SO 2 are likely due to the lower chemical loss through oxidation by lower OH levels and less wet deposition resulted from lower precipitation in SEN1. The increase of surface mixing ratios of NO 2 over land is likely due to less wet deposition, and the decrease of surface mixing ratio of NO 2 over ocean is likely due to the vertical mixing over ocean (e.g., higher PBLH). The increase of surface O 3 mixing ratios over eastern land areas and east coast in SEN1 is likely due to the increase of NO 2 surface mixing ratios and the decrease of surface mixing ratios of O 3 over 6

7 remote ocean is likely due to the more convection in SEN1. The decrease of surface mixing ratio of O 3 over southwestern areas of the domain is likely due to the more chemical loss through oxidation with alkenes (e.g., isoprene and terpene) under more stable and warmer conditions over these regions. Compared to W15, SEN1 predicts higher SO 2-4 by up to 1.0 μg m -3 and lower SO 2-4 by up to 1.6 μg m -3. The increase of SO 2-4 over oceanic areas and some land areas is mainly due to the increase of surface mixing ratios of SO 2 and decrease of wet deposition. The decrease of SO 2-4 concentration over some land areas is mainly due to the decrease of chemical production from lower OH levels over these regions. The increase of SOA concentrations is mainly due to the increase of ASOA (by domain-average of 0.22 µg m -3 ) and BSOA (by domainaverage of 0.37 µg m -3 ). Higher concentrations of ASOA can be attributed to less wet deposition and higher gaseous precursors of SOA due to less wet deposition. Higher concentrations of BSOA can be attributed to less wet deposition and higher biogenic emissions resulted from higher SWD and T2 in SEN1. Compared to W15, SEN1 predicts higher PM 2.5 and PM 10 up to 4.9 µg m -3, with a domain averaged increase of 0.74 and 0.18 µg m -3, respectively. Unlike PM 2.5 concentrations that increase over most of the domain, the concentrations of PM 10 decrease over remote ocean, which is mainly due to the decrease of sea-salt concentrations resulted from lower WS10 in SEN1. Table S1b shows the statistical performance for chemical predictions. As total precipitation is reduced in SEN1, the concentrations of gases and aerosols are less underpredicted or more overpredicted, with better performance for surface SO 2 (with NMBs changed from -43.7% in W15 to -23.0% in SEN1 against SEARCH), EC (with NMBs changed from -71.2% in W15 to % in SEN1 against SEARCH), OC (with NMBs changed from -44.1% in W15 to -0.4% in SEN1 against SEARCH), TC (with NMBs changed from -16.2% in W15 to 1.0% in SEN1 7

8 against STN and from -40.3% in W15 to 1.5% in SEN1 against SEARCH), PM 2.5 (with NMBs changed from -22.2% in W15 to -12.0% in SNE1 against, and from -29.3% in W15 to -25.1% in SEN1 against STN), PM 10 (with NMBs changed from -65.4% in W15 to -62.1% in SEN1 against AIRS-AQS), and column SO 2 (with NMBs changed from -68.8% in W15 to -63.8% in SEN1 against SCIAMACHY). Compared to WRF-CMAQ simulations (Yu et al., 2014), SEN1 gives better agreement for SO 2-4, EC, and PM 2.5 against, and for SO 2-4 and NH + 4 against CASTNET. Chemical Predictions from SEN1 Figure S3 shows the monthly-averaged surface predictions of the concentrations of chemical species from SEN1, which serves as the baseline results for the absolute differences between SEN2 and SEN1 in Figure 6 and between SEN3 and SEN1 in Figure 7 in the main text of this manuscript in which those changes are discussed in detail. 8

9 Table S1a. Statistical Performance of Meteorological, Cloud, and Radiative Variables. W15 SEN1 SEN2 SEN3 Species/ Datasets obs NMB NME NMB NME NMB NME NMB NME Variables sim Corr sim Corr sim Corr sim Corr (%) (%) (%) (%) (%) (%) (%) (%) NCEI T2 ( o C) NDBC SEARCH RH2 (%) NCEI SEARCH NCEI WS10 (m s -1 ) 1 NDBC SEARCH GPCP (land) Precip GPCP (ocean) (mm day -1 ) TMPA (land) TMPA (ocean) SWD (W m -2 ) 3 CERES (land) CERES (ocean) CASTNET SEARCH OLR CERES (land) (W m -2 ) CERES (ocean) SWCF CERES (land) (W m -2 ) CERES (ocean) LWCF CERES (land) (W m -2 ) CERES (ocean) CF (%) CERES (land) CERES (ocean) COT CERES (land) CERES (ocean) LWP CERES (land) (g m -2 ) CERES (ocean) AOD MODIS (land) MODIS (ocean) CCN5 (10 8 cm -2 ) MODIS (ocean) CDNC Bennartz (2007) NARR (land) PBLH (m) NARR (ocean) SST ( o OAFlux (ocean) C) NDBC LHFLX (W m -2 ) OAFlux (ocean) SHFLX (W m -2 ) OAFlux (ocean)

10 Species/ Variables Table S1b. Statistical Performance of Chemical Species. W15 SEN1 SEN2 SEN3 Datasets obs NMB NME NMB NME NMB NME NMB NME sim Corr sim Corr sim Corr sim Corr (%) (%) (%) (%) (%) (%) (%) (%) CO SEARCH SO 2 CASTNET SEARCH NO 2 SEARCH HNO 3 CASTNET Max 1-h O 3 CASTNET AIRS-AQS Max 8-h O 3 CASTNET AIRS-AQS Hourly O 3 AIRS-AQS SEARCH STN NH 4 CASTNET SO 4 STN CASTNET NO 3 STN CASTNET Na Cl EC SEARCH OC SEARCH TC STN SEARCH PM 2.5 STN SEARCH PM 10 AIRS-AQS Col.CO MOPITT Col.NO 2 SCIMACHY Col. SO 2 SCIMACHY TOR OMI/MLS MODIS (land) AOD MODIS (ocean) * The units for all surface gaseous and aerosol species are μg m -3 except for CO (ppb), SO 2 (ppb), NO 2 (ppb) against SEARCH, and O 3 (ppb). The units for column CO and NO 2 are molecules cm -2 and for column SO 2 and TOR are DU. 10

11 SEN1 W15 Figure S1a. Absolute differences in the monthly-averaged predictions of major meteorological and cloud/radiative variables between SEN1 and W15. 11

12 SEN1 W15 Figure S1b. Absolute differences in the monthly-averaged predictions of surface concentrations of chemical species between SEN1 and W15. 12

13 Beaufort (BFT142), NC (34.88, ), CASTNET Holiday ( ), FL (28.20, ), AQS Blackwater NWR (BWR139), MD (38.44, ), CASTNET St. Marks Wildlife Refuge ( ), FL (30.09, ), AQS Indian River Lagoon (IRL141), FL (27.85, ), CASTNET Gulfport Youth Court ( ), MS (30.39, ), AQS 13

14 Chassahowitzka (CHAS1), FL (28.75, ), Everglades (EVER1), FL (25.39, ), Brigantine (BRIG1), NJ (39.47, ), St. Marks (SAMA1), FL (30.09, ), 14

15 Cape Romain (ROMA1), SC (32.94, ), IMRPOVE Swanquarter (SWAN1), NC (35.45, ), Figure S2a. Comparison of simulated monthly-averaged temperature profiles with NCEP over coastal sites. 15

16 Beaufort (BFT142), NC (34.88, ), CASTNET Holiday ( ), FL (28.20, ), AQS Blackwater NWR (BWR139), MD (38.44, ), CASTNET St. Marks Wildlife Refuge ( ), FL (30.09, ), AQS Indian River Lagoon (IRL141), FL (27.85, ), CASTNET Gulfport Youth Court ( ), MS (30.39, ), AQS 16

17 Chassahowitzka (CHAS1), FL (28.75, ), Everglades (EVER1), FL (25.39, ), Brigantine (BRIG1), NJ (39.47, ), St. Marks (SAMA1), FL (30.09, ), 17

18 Cape Romain (ROMA1), SC (32.94, ), IMRPOVE Swanquarter (SWAN1), NC (35.45, ), Figure S2b. Comparison of simulated monthly-averaged specific humidity profiles with NCEP over coastal sites. 18

19 SEN1 Figure S3. Spatial distribution of surface concentrations of chemical species by SEN1 (WRF/Chem standalone without ocean coupling). 19