IMPROVING AEROSOL DISTRIBUTIONS

Transcription

1 IMPROVING AEROSOL DISTRIBUTIONS BY ASSIMILATING SATELLITE- RETRIEVED CLOUD DROPLET NUMBER AND AEROSOL OPTICAL DEPTH Pablo Saide 1, G. Carmichael 1, S. Spak 1, P. Minnis 2, K. Ayers 2, Z. Liu 3, H.C. Lin 3, C. Schwartz 3 (1) University of Iowa, Center for Global and Regional Environmental Research (2) NASA LARC (3) NCAR pablo-saide@uiowa.edu 4 th IWAQFR Meeting, 12/14/2012

2 Motivation 2 Growing Evidence Better constrained aerosol distributions (mass, number, composition ) lead to improved PM predictions improved weather predictions Goals Maximize the information content in satellite observations to constrain aerosol distributions Make it operational Approach Develop techniques in the on-line WRF-Chem model with feedbacks (direct & indirect) between aerosols and meteorology First step embed in operational 3d-Var system --- GSI (Gridpoint Statistical Interpolation)

3 Challenge Cloud Ship tracks 3 AOD retrievals provide information to constrain aerosol mass. But in many regions there is a lack of AOD information. However cloud retrievals contain information on aerosols. MODIS fine AOD SE Pacific MODIS fine AOD CA (22 Jun 2008) CLOUDS

4 20º S VOCALS-REx stats Opportunity N d 4 N SEP marine Sc N d satellite retrievals show evidence of aerosol load and agree with observations Aerosol indirect effects simulated with some skill in regional modeling to date N d τ: Cloud optical depth r e : Cloud effective radius N d : Cloud droplet # N: Aerosol # conc. Hypothesis: variational assimilation with N d retrievals can improve below-cloud aerosols in models N Refs: Saide et al., ACP 2012, Bretherton et al., ACP 2010, Painemal and Zuidema, ACP 2011

5 We ve developed a technique to assimilate cloud satellite retievals to constrain aerosol distributions 5 Below cloud aerosols Model N & N d Assimilating N d N d changes during the forecast due to activation Forecast N & N d after assim Observed N d N d : Cloud droplet # N: Aerosol # conc. Log 3DVAR Adjoint of WRF Aerosol Activation Changes N and aerosol mass N: Aerosol Number N d : Cloud droplet number Ref: Saide et al., PNAS 2012

6 6 Assimilation results: + & - biases reduced 6 Assimilate MODIS Terra N d Aerosol mass and number are changed N d N WRF-Chem Guess WRF-Chem Assimilated MODIS Obs N d SO4 Ref: Saide et al., PNAS 2012

7 Ref: Saide et al., PNAS 2012 Daytime N d after assimilation 7 vs GOES and in-situ aerosol Half Soccer plot for 3 flights Large improvements during the first 2 days for all domain GOES Assimilation improves agreement with VOCALS-REx C130 aerosol number and mass observations A single MODIS N d Assimilation Forecast GOES10 OBS WRF-Chem Guess WRF-Chem Assimilated +22 hrs +5 hrs +22 hrs +5 hrs +22 hrs +5 hrs

8 8 Porting Nd assimilation to an operational DA tool: GSI MODIS Terra Background Assimilated PNAS method Assimilated GSI 1 st day 2 nd day 1 st day 2 nd day A single MODIS N d Assimilation using both methods followed by 2 day forecast vs GOES

9 9 Moving forward: Simultaneous N d and AOD assimilation N d and AOD provide complementary information Gridpoint Statistical Interpolation (GSI) can perform simultaneous DA of different datasets (e.g. ground PM2.5 and AOD) First, need to add MOSAIC (sectional) AOD assimilation (just GOCART before) in GSI: AOD and sensitivities computed with WRF-Chem optical averaging routine (Mie code + Internal Mixture) and its adjoint (TAPENADE) Aerosol water computed as in MOSAIC (use electrolytes) Add correlation between bins sizes by using smoothing filters Simultaneous assimilation of total and fine AOD OR AOD at all wavelengths Ref: Schwartz et al., JGR 2012, Liu et al., JGR 2011, Fast et al., JGR 2006

10 GSI MOSAIC AOD: Total AOD Monterey, CA 10 preliminary results Study case: ARCTAS-CARB Assimilation every 3 hours on a 3 hour window for 4 days using Terra and Aqua total and fine AOD Overall good behavior MODIS Terra total AOD WRF-Chem Guess WRF-Chem Assimilated Monterey

11 11 Assimilate AOD at all wavelengths versus just at 550nm MODIS Observation Only assimilating 550 nm Assimilating all wavelengths AOD at 550 nm AOD at 1240 nm Drifting away from observation rather than getting closer! Assimilating all wavelengths changes the angstrom exponent (size distribution information!)

12 Number of droplets Fine AOD Total AOD Simultaneous Number of droplet and total + fine AOD assimilation for ARCTAS-CARB 12 MODIS Background Assimilated

13 3 hour forecast AOD 21Z (Aqua) AOD 18Z (Terra) AOD assimilation for fires coming from Central America MODIS Background Assimilated AQS PM2.5 Aeronet Wave Assimilation Guess

14 Conclusions and future work 14 Cloud (N d ) assimilation A data assimilation method to improve below-cloud accumulation aerosol using cloud retrievals is developed and tested, showing improvement vs. satellite and in-situ observations with up to 48 hours influence This technique can be now used within the GSI assimilation system Coupling cloud (N d ) and AOD assimilation Development of GSI MOSAIC AOD with positive results, incorporating fine AOD OR total AOD at different wavelengths Future work Compare N d +AOD assimilation to in situ data (ARCTAS-CARB, CALNEX) Include products from other platforms (other LEO and GEO AOD, AAOD) Analyze effect of assimilated aerosol on meteorology and chemistry (e.g. O3) through WRF-Chem aerosol feedbacks Moving forward with full adjoint of 3-d WRF-Chem

15 Acknowledgments 15 CGRER, The University of Iowa VOCALS-REx scientists MODIS aerosol and cloud and Aeronet teams WRF, WRF-Chem and GSI developers Questions?

16 16 WRF-Chem indirect effects validation for VOCALS-REx Ref: Saide et al., ACP 2012

17 17 Supporting slide Assimilation r e obs τ obs N model Forward Adjoint approach N d obs N d model Sensitivities N assim 3DVAR Data assimilation τ: Cloud optical depth r e : Cloud effective radius N d : Cloud Number of drops N: Aerosol number conc. Factor per bin & phase Assimilate Aerosol mass for each species WRF Forecast Sensitivities computed using the adjoint of the WRF-Chem mixactivation routine obtained with TAPENADE (WRF-Chem adjoint) 3DVAR assimilation using lognormal errors for different scales 5D correlation lengths: lat, lon, height, bins(1:8), phase (dry/wet) Optimize for aerosol number concentration, and update mass concentrations accordingly (same phase, same bin) Ref: Saide et al., PNAS 2012

18 Supporting slide assimilation equation 18 Ref: Saide et al., PNAS 2012

19 Study case: Southeast Pacific 19 during VOCALS REx Assimilate extensive stratocumulus cloud deck Coastal higher Nd, remote zone with lower Nd MODIS Terra N d Base WRF-Chem N d Case: Oct. 16 th, 15:00 UTC Ref: Saide et al., PNAS 2012

20 Comparison to in-situ aerosol from 20 VOCALS-REx observations Do assimilation using GOES N d before a flight starts Assimilation improves agreement with aerosol number and mass observations N and SO4 for 1 longitudinal flight Half Soccer plot for 3 flights Ref: Saide et al., PNAS 2012, Wood et al., ACP 2011

21 Angstrom exponent at nm 21 MODIS Observation Only assimilating 550 nm MODIS OBS Background Assim AOD at all Wavelengths Assim AOD only at 550 nm Assimilating all wavelengths changes the angstrom exponent (size distribution information!)

22 3 hour forecast AOD 21Z (Aqua) AOD 18Z (Terra) AOD assimilation for fires coming from Central America MODIS Background Assimilated AQS PM2.5 Aeronet Wave Assimilation Guess