Impacts of Assimilation of Air Quality Data from Geostationary Platforms on Air Quality Forecasts Gregory Carmichael(1), Pablo E. Saide (NCAR), Meng Gao (1), Maryam Abdioskouei (1), Arlindo da Silva (NASA GSFC), R. Brad Pierce (NOAA NESDIS-STAR), David G. Streets (ANL), Jhoon Kim, and Myungje Choi (Yonsei University), Chul H. Song (GIST), Greg Thompson, Trude Eidhammer (NCAR), and SEAC4RS, KORUS-AQ and ORACLES science teams (1) Center for Global & Regional Environmental Research, University of Iowa, Iowa City, Iowa, USA

Models Constrained with Observations Play increasing Important Roles in Research and Applications Trend toward closer linkages of weather, atmospheric composition, and climate related services Information needed at higher resolution to address societal needs Improvements require advances in observing systems and model/assimilation systems

Testing the Impact of GOCI AOD Assimilation UIOWA/NCAR WRF-Chem forecasting system WRF-Chem with MOSAIC aerosols and a Reduced Hydrocarbon chemistry (Pfister et al. JGR 2014), including simplified SOA formation (Hodzic and Jimenez, GMD 2011) GFS and MACC meteorological and chemical boundary conditions KORUS-AQ anthro (Jung-Hun Woo) and QFED fire emissions AOD data assimilation using GSI (Saide et al., ACP 2013). MODIS and GOCI data were assimilated simultaneously every three hours. (Saide et al., GRL, 2014) Four days of forecasts were available for the outer domain, 2 days for the inner domain Output available to other users: All species and meteorology along DC-8 flight track on KORUS-AQ repository, full model outputs available by request (saide@ucar.edu) Images:http://bio.cgrer.uiowa.edu/KORUS-AQ/wrf_fullchem_2016-MM-DD/pmenu.html

GOCI AOD Data Significantly Improves Predictions of AOD (over MODIS-only and no-assimilation)

With GOCI AOD assimilation prediction skill is high for AOD (evaluation against AERONET) but surface PM is biased high (all Korean sites)) AOD PM10 PM2.5

Example of AOD assimilation impact during KORUS-AQ (May 24th flight, Chinese haze over Yellow sea) Day-3 Day-2 Day-1 Observed AOD (GOCI) WRF-Chem Day-1

DIAL-HSRL vs WRF-Chem May 24th (Extinction [1/km]) (Day-1)

AOD comparison May 24th (GOCI V2, DIAL-HSRL, 4STAR, WRF-Chem)

AOD comparison May 24th (GOCI V2, DIAL-HSRL, 4STAR, WRF-Chem)

In-situ aerosol comparison (May 24th) Ambient Extinction AMS mass Fine aerosol produces less extinction per unit mass than observed, probably related to hygroscopic growth? Aerosol not observed by AMS (“other PM2.5”) too high in the model? Emissions too high? Assigned to fine aerosol size bins, thus removed more slowly than in reality?

Aerosol hygroscopic growth [ f(RH) ] Base model (OA and “other PM2.5” do not take up water) f(RH) @ 550nm f(RH) @ 550nm No fine “other PM2.5” and OA slightly hydrophilic (κ = 0.14)

TOTAL OMI NO2 COLUMNS with Pandora OVERLAID High Resolution Model Profiles Improve Retrievals TOTAL OMI NO2 COLUMNS with Pandora OVERLAID OMI_WRF-Chem: Using 4 km WRF-Chem to calculate the Air Mass Factor instead of GMI (resolution: ~110 km) OMI_GMI NO2 Standard Product OMI_WRF-Chem NO2 Product OMI NO2 vs. Pandora Pandora 2-hour means co-located to valid daily OMI overpasses New OMI_WRF-Chem product shows much better agreement with total column measurements from Pandora NO2 (slope near unity) Streets and Goldberg

Emission Inversion method for Rapid Update of Emissions Simultaneously improve the model fit to various measurements: SEAC4RS in-situ (OA, CO, extinction) and remote sensing (DIAL-HSRL Extinction) from 2 flights AERONET AOD (1, 2, 3, 4) Ground based CO (1, 5, 6) Satellite AOD (NASA NNR) Constrain hourly RIM fire emissions from NASA QFED BB emissions Performed at 4km resolution, for 2 met boundary conditions (FNL, NARR) 26 Aug flight Ext [1/Mm] U of Reno AERONET AOD Saide et al., GRL 2015a

Impacts of constrained emissions Maximum changes shown comparing simulation with initial and constrained emissions Impacts can be substantial

Emissions update tested in ORACLES WRF with aerosol-aware microphysics (Thompson and Eidhammer, JAS 2014, Saide et al. JGR 2016) at 12km resolution Smoke emissions constrained in near-real time using Saide et al. (GRL 2015) over 6 regions for a 4 day window 12 km Simulations turning on and off fires to assess aerosol-cloud-radiation interactions One cycle a day, 3 day forecasts Surface Smoke Extinction Number droplet change by smoke Temp. change due to smoke (~1.5 km)

Daily evolution of smoke emissions 1) 3) 2) 4) -Region 6 (Mozambique) it’s the only showing substantial increases in emisions due to the inversion - Different regions show different trends in the emissions, with decrease for 1 and 3 (Congo), increase in 6 Mozambique, and roughly constant for the rest Aug Sept 5) 1 2 Prior Constrained 3 4 5 Tendency Aug Sept

AERONET over fire regions 1) Gorongosa (Mozambique) Aug Sept 2) Lubango (Angola) 2 1 Aug Sept Acknowledgements to the AERONET Team, specially Brent Holben, Joel Schafer and David Giles

HSRL2 data from ER-2 aircraft (Sept 22nd) PRELIMINARY FIELD DATA (Backscattering) WRF-AAM Extinction day 1 WRF-AAM AOD Qualitative AOD comparison Acknowledgements to the HSRL2 and ER-2 teams, specially Sharon Burton and Rich Ferrare

Above Cloud AOD (ACAOD) retrievals ACAOD retrievals by Meyer et al. (JGR 2015) produced in near-real time during September 2016 MODIS Terra QC1 (Observation) MODIS Terra QC2 (Observation) WRF-AAM Day 1 (Forecast) Acknowledgements to Kerry Meyer and Steven Platnick for the ACAOD data and interpretation Data: ftp://windhoek.nascom.nasa.gov/pub/ridgway/MOD06ACAERO/2016/ ftp://windhoek.nascom.nasa.gov/pub/ridgway/MYD06ACAERO/2016/

Models Constrained with Observations Play increasing Important Roles in Research and Applications Trend toward closer linkages of weather, atmospheric composition, and climate related services Information needed at higher resolution to address societal needs Improvements require advances in observing systems and model/assimilation systems Information from TEMPO, GEMS and others, by themselves (and together with other platforms) will advance a spectrum of research  applications

Supporting slides

OMI NO2 tropospheric columns at 1 OMI NO2 tropospheric columns at 1.33 km resolution in the Baltimore-Washington metropolitan area Goldberg et al., 2017; ACPD OMI NO2 NASA product OMI_CMAQ NO2 product OMI_CMAQ scaled NO2 product OMI_CMAQ scaled & spatial weighted NO2 product (a)(b): Use CMAQ instead of GMI to calculate the Air Mass Factor (AMF) (b)(c): Scale CMAQ based on D-AQ aircraft observations, and then calculate AMF (c)(d): Spatial weight based on the variability within CMAQ [Kim et al., 2016; GMD] (d) Is much better than (a) when compared to in situ observations! Key takeaways: Model resolution plays a significant role in the calculation of air mass factors (high resolution = better) Accuracy of model simulation is critical in generating robust satellite observations If emissions or chemistry are way off, satellite will be unrealistic Spatial weighting helps satellite match urban-scale variability better

Constraining biomass burning emissions for improved prediction skill and assessing smoke-weather interactions WRF with aerosol-aware microphysics (AAM) (Thompson and Eidhammer, 2014) and WRF-Chem emissions. Inversion based on Saide et al. (GRL 2015b) using WRF tracers (no adjoint, no ensembles) Plans for using it operationally for the NASA ORACLES and NOAA FIREX field experiments Run 2 forecasts needed by the inversion method (with and without feedbacks) so that aerosol feedbacks impacts are also forecasted Notes: Figure upper right: regions where scaling factors are applied to biomass burning smoke emissions Figure at center-right: Time series of emissions on each region showing prior and constrained emissions. Note that scaling of emissions varies through time and region considerably, so scaling all emissions for all times by a single factor would not work Bottom figure: Shows the improve fitting to observations after re-running with the constrained emissions

Improved Forward Models & Assm. Sys. Data Management & Discovery The Growing Interest in Improving Air Quality Predictions/Services and the Role of Atmospheric Composition in Weather and Climate Applications Offer Great Opportunities for Our Community Evolving Obs Systems Improved Forward Models & Assm. Sys. Improved Predictions Requires Continuous Research Enhanced Data Management & Discovery Systems Improved Emissions

Technique already Extended to include Assimilation of Surface PM2 Technique already Extended to include Assimilation of Surface PM2.5 Observations enable improved Predictions Haze JAN 2013 R no DA 0.67 R DA 0.95 PM2.5 Observations are growing rapidly Impacts of assimilating surface PM2.5 Mean over all surface observation Health impacts from winter haze; e.g., Gao et al., Science Tot. Env., (2015)

Forecasting system Move from WRF-Chem to WRF with aerosol-aware microphysics (AAM) (Thompson and Eidhammer, 2014) including primary emissions (anthropogenic, biomass burning, sea-salt, dust) (Saide et al., JGR 2016) Perform an inversion based on Saide et al. (GRL 2015b) instead of data assimilation for initial conditions. - Go from over 100 aerosol species to only 2 but still have a treatment of aerosol-cloud-radiation interactions

Forecasting system WRF with aerosol-aware microphysics (AAM) (Thompson and Eidhammer, 2014, Saide et al. 2016) at 12km resolution Smoke emissions constrained in near-real time using Saide et al. (GRL 2015) over 6 regions for 8 hour intervals Simulations turning on and off fires to assess aerosol-cloud-radiation interactions Tracers for 15 days of smoke emissions used for assessing age 12 km VIIRS Fires Aug-Sept 2016 FIRMS Web Fire Mapper

Some of the forecasting products

Some of the forecasting products Smoke tracer Cross-sections Above Cloud AOD Temperature change due to smoke Mean Age

In-situ comparison DLH RH Black Carbon CO