Experiments with Assimilation of Fine Aerosols using GSI and EnKF with WRF-Chem

(on the need of assimilating satellite observations)

Mariusz Pagowski

Georg A. Grell

NOAA/ESRL

Importance of air quality/chemical forecasting and data assimilation

• Fine aerosols and ozone detrimental to health. PM2.5 is the single most critical factor affecting human mortality due to air pollution.

• Interdependence of meteorology and atmospheric chemistry: e.g. aerosols affect radiation and microphysics; improved model chemistry has a potential to positively influence meteorology.

• Because of these interdependence assimilation of chemistry has

a potential to improve assimilation of meteorology via regression relationships either in VAR or EnKF; and vice versa.• Potential largely unrealized yet because of the weaknesses of

parameterizations, deficiencies of retrieval and data assimilation methods.

Challenges in air quality/chemical forecasting and data assimilation

• Scarcity of observations, especially in the vertical both for model verification and data assimilation, also for aerosol species.

• Multiple state variables (in tens) and parameterizations have large uncertainties – to account for such errors ensembles would benefit from stochastic parameterization.

• Results very much dependent on source emissions, error estimates in surface emissions 50-200% – need for ensembles using correlated perturbations.

• Air quality tied to meteorology in the boundary-layer, ensembles tend to collapse and surface observations poorly assimilated.

Observations

AIRNow network IMPROVE and STN networks

Aerosol species: SO4, OC, EC;

24-hour averages every three days;

available after several weeks/months;

limited value for assimilation,

possibly for FGAT in diagnostic studies.

Total aerosol mass;

1-hr average available round the clock;

urban, suburban, rural sites, unknown;

suitable for r-t assimilation.

Regional Model

ARW WRF-Chem updated version 3.2.1

grid length ~60 km, 40 vertical levels

GOCART aerosol for computational reasons;

assimilate standard meteorological observations

(prepbufr) and AIRNow PM2.5 in 6-hr cycle with 1-hr window;

NMM lateral boundary conditions;

GSI: Background Error Statistics derived from

continuous forecasts in summer 2006 using NMC method;

EnKF: 50 ensembles initialized from NMM using background error statistics and perturbing

emissions;

June 1 - July 15, 2010Emission perturbations – an example

PM2.5 increment – examples

00 UTC 12 UTC

Experiments/EvaluationsAIRNOW PM2.5

Conclusions

• Large positive impact of assimilation, but fast error growth after the assimilation; from previous work: positive effect still present after 24 hours; potential of ensembles.

• Some advantage of EnKF over GSI in terms of total PM2.5 mass. Little impact of assimilations on individual aerosols species.

• Detrimental when ensemble filter determines speciation compared to a priori ratios -> GOCART physics/chemistry not reliable for this modeling domain. Still, GOCART attractive because of simplicity, speed.

(JGR, 2012:

http://www.agu.org/journals/jd/papersinpress.shtml#id2012JD018333)

• Negative effect of meteorology on aerosols due to regressions derived by the filter – analysis follows.

EnKF_Met vs. EnKF_no_Met

No_Met

Met

Time series of domain averaged PM2.5 concentration (after five-day spin-up)

EnKF_Met vs. EnKF_no_Met

No_Met

Met

PM2.5 concentration model level 23

Geopotential perturbation

EnKF_Met vs. EnKF_no_Met

No_Met

Met

PM2.5 concentration model level 23 + 12h

Geopotential perturbation

Conclusions

• Negative effect of meteorology/aerosol regressions derived by the filter:

model builds up high PM2.5 concentrations above PBL since no observations available to constrain the effect – possibly counteract with satellite observations, near future: experiments with AOD.