1 Aerosol Assimilation Based on NCEPs GSI Compared to Optimal Interpolation MethodYouhua Tang1,2, Mariusz Pagowski4,5, Tianfeng Chai1,2, Li Pan1,2, Pius Lee1, Barry Baker1,2, Rajesh Kuma6, Luca Delle Moncache6, Daniel Tong1,2,3, and Hyun-Cheol Kim12 1. NOAA Air Resources Laboratory, College Park, MD. 2. Cooperative Institute for Climate and Satellites, University of Maryland at College Park, MD. 3. Center for Spatial Information Science and Systems, George Mason University, Fairfax, VA. 4. NOAA Earth System Research Laboratory, Boulder, CO. 5. Cooperative Institute for Research in the atmosphere, Colorado State University at Fort Collins, CO. 6. National Center for Atmospheric Research, Boulder, CO.

2 Background of this workExisting National Air Quality Forecasting Capability (NAQFC) expanded to support PM2.5 prediction. However PM2.5 is not a single aerosol species but the combination of aerosols with diameter < 2.5 um and its source/sink are very uncertain. NCEP has the existing data assimilation tool Gridpoint Statistical Interpolation (GSI) to assimilate observation to adjust models’ initial condition. Pagowski et al (Geosci. Model Dev. 2014) expanded the 3D-Var method to assimilate surface PM2.5 and MODIS AOD (only for WRF-CHEM with GOCART aerosols). In this study we apply this GSI assimilation tool to CMAQ aerosols using surface PM and MODIS AOD and make the test runs in 4-cycle per day. This result is compared to a simpler optimal interpolation (OI) method (Tang et al 2015)

3 GSI 3D-var Cost Function𝐽= 1 2 𝑥𝑎−𝑥𝑏 𝑇B-1 𝑥𝑎−𝑥𝑏 𝐻𝑥𝑎−𝑂𝑜 𝑇O−1 𝐻𝑥𝑎−𝑂𝑜 +𝐽𝑐 Where xa and xb are the analyzed and background (a prior modeled) concentrations or AOD data, B and O are the background and observation error covariance matrices, H is the observational operator, Oo is the observations and Jc is the constraint terms. Optimal Interpolation method (Tang et al 2015) Where Xa and Xb are the analyzed and background ( a prior modeled) concentrations or AOD data, B and O are the background and observation error covariance matrices, H is the observational operator and HT is its matrix transpose operator, and Y is the observation vector

4 Aerosol Assimilation DiagramCMAQ v5.1 a prior aerosols from previous run AIRNow/AQS Surface PM2.5 GSI/OI Assimilation Satellite AOD (Terra/Aqua MODIS currently) OI uses CMAQ’s reconstruction method to estimate AOD while GSI has its own method for AOD via CRTM Adjusted Aerosol Concentrations MODIS AOD Observation (Terra and Aqua) Prediction Cycle 00Z 06Z 12Z 18Z AQS/AIRNow PM2.5

5 CMAQ mean aerosol sizes for each mode: DGATKN, DGACC, DGCORASO4I ANO3I ANH4I ACLI ASO4J ANO3J ANH4J ASO4K ANO3K ANH4K Sulfate Aerosol AECI AECJ Black Carbon Aerosol APOCI APNCOMI APOCJ AOTHRJ AXYL1J AXYL2J AXYL3J ATOL1J ATOL2J ATOL3J ABNZ1J ABNZ2J ABNZ3J AISO1J AISO2J AISO3J ATRP1J ATRP2J ASQTJ AALK1J AALK2J AORGCJ AOLGBJ AOLGAJ APAH1J APAH2J APAH3J APNCOMJ CRTM calculation of aerosol optical properties for each CMAQ aerosol species Organic Carbon Aerosol AFEJ AALJ ASIJ ACAJ AMGJ AKJ AMNJ ACORS ASOIL Dust Aerosol Ambient relative humidity ANAJACLJ ACLK ASEACAT Sea Salt Aerosol GOCART Aerosol Optical Properties used in GSI/CRTM CMAQv5.1 Aerosol Species

6 The model relative uncertainties used in the OI assimilation are varied in location and time

7 Model Standard Deviation and Length Scales used in GSI assimilation

8 Base CMAQ Model ConfigurationMeteorology: WRF-ARW 3.4 driven by the NCEP FNL (Final Global Analysis) 1×1 analysis field and was re-initialized every 24 hours Air Quality Model: CMAQ 5.1 in 12km horizontal resolution and 42 vertical layers up to 50hPa National Emission Inventory (base year 2011); HMS-BlueSky wildfire emission; RAQMS chemical lateral boundary condition Study period: July 2011

9 The base case generally under-predicted PM2The base case generally under-predicted PM2.5, especially over Western CONUS

10

11 Overall difference of GSI versus OI methods

12 OI adjustment is stronger, especially for its AOD adjustmentOI adjustment is stronger, especially for its AOD adjustment. GSI yields more moderate adjustment based on our current figuration.

13 Cross-section ComparisonOI adjustment is also stronger vertically, and yield sharper gradient near the PBL top.

14 Continuous CMAQ runs with GSI and OI adjustments for CONUS domainCMAQ Base: MB=-4.39 µg/m3 R=0.34 CMAQ OI: MB=-1.92 µg/m3 R=0.38 CMAQ GSI: MB=-2.04 µg/m3 R=0.48

15 Summary We tested the two assimilation method: optimal interpolation and GSI to adjust CMAQ’s initial conditions for July 2011 over the continental USA using AIRNow/AQS surface data ( every 6 hours) and MODIS AOD (once per day). Both of them yield better results than the base case without adjustments. In our current setting, the OI adjustment is stronger and changes the aerosol field more aggressively. GSI yields smoother adjustment with much higher correlation coefficient. CRTM’s AOD calculation through GOCART aerosol table is quite different from CMAQ’s own reconstructive method. We need investigate further in order to get the best AOD assimilation. The effect-lasting time of adjusted initial condition depends on many factors, such as the strength of local emission and of local wind.