Toward An EnsembleBased Data Assimilation System for Reactive
Toward An Ensemble-Based Data Assimilation System for Reactive Trace Gases: Application to NO 2 TEMPO Ronald C. Cohen UC Berkeley $ NASA ACMAP and GEOCAPE
(Some of) The new science from TEMPO will depend on the ratio of NO 2 (or O 3, …) at two locations or two times of day being more accurate than the absolute NO 2
Riyadh urban plume (OMI) Winds 15 -20 km hr-1 Riyadh L Valin et al. , GRL 2013
Riyadh L Valin et al. , GRL 2013
Model NOx lifetime vs. wind speed. L Valin et al. , GRL 2013
WRF-CHEM 1 km – 4 -Corners Plume NO 2 column OH Column
Four Corners Power Plants: WRF-Chem L Valin et al. , Atmos. Chem. Phys. 20
In addition to geophysical and chemical variation in the NO 2 column, there is noise and systematic errors. Most (all? ) data assimilation systems assume errors are Gaussian.
NO 2 profile shape PDF of systematic errors Russell et al. , Atmos Chem & Phys, 2011
Terrain Reflectivity (Albedo) PDF of systematic errors Russell et al. , Atmos Chem & Phys, 2011
Ensemble Kalman Filter-Data Assimilation Xueling Liu Also: A. P. Mizzi, J. L. Anderson, NCAR I. Fung, UC Berkeley
Data Assimilation System • Forecast model: WRF-Chem • Location: Denver • Time period: 2014/07/02 • Nested domain: 12 km 3 km • Anthr: National Emission Inventory 2011 • Bio: MEGAN • Meteorology input: NARR • Chemistry Mechanism: RADM 2 one-way nested domain Denver • Data Assimilation Algorithm: Ensemble Adjustment Kalman Filter (EAKF) (Anderson, 2001) Ø Updated chemical variables: NOx Emission, NO 2 Ø EAKF Algorithm: a Denver Ø Ø Adaptive Inflation Localization length: 10 km Frequency: 1 hour Observation uncertainty: 5~10% 1) Full Run: Emission uncertainty + Meteorological uncertainty (U, V, T, H 2 O(g)) NCEP Global Upper Air and Surface Weather Observations 2) Chem Run: Emission uncertainty
WRF-CHEM 1 km – 4 -Corners Plume NO 2 column OH Column
NOx Emissions • Can we detect 30%-underestimation of NOx emissions in Denver using TEMPO NO 2 column? • What is the optimal strategy for ensemble-based chemical data assimilation of short-lived species? Nature Run “True” Emission Assim. Run Prior Emission 70% ×truth (20 members) M “True” atmosphere Forecast model : xt 2= M (xt 1) M H TEMPO NO 2 column Observation simulator : y= H(x) Prior estimate • Step 1: Add -30% bias to true emissions • Step 2: simulate the 3 -D NO 2 concentration field Emissions:Prior – truth (mol/km 2/hr) Surface Conc:Prior – truth (ppmv) Data assimilation • Step 3: project to observation space for “observed” discrepancy TEMPO NO 2 column (2230): Prior – truth (molecules/cm 2)
NOx Emission Perturbed Emissions:Prior – truth (mol/km 2/hr)
NOx Emissions recovered reasonably when using one realization of the weather Emissions:Prior – truth (mol/km 2/hr) Emissions:Posterior – truth (mol/km 2/hr)
Weather substantially degrades performance Emissions:Prior – truth (mol/km 2/hr) Emissions and Meteorology Emissions:Posterior – truth (mol/km 2/hr) Emissions only Emissions:Posterior – truth (mol/km 2/hr)
RMSE time series of TEMPO NO 2 column in Denver Full Run RMSE (molecules/cm 2) Chem Run RMSE (molecules/cm 2)
Conclusions • A simple experiment with identical meteorology in nature run and assimilation rapidly converges a perturbation in emissions to true value. • Other runs (not shown) indicate (no surprise) that differences in PBL dynamics between two independent models are the primary source of failure for OMI assimilation to converge quickly. • Even in runs with similar meteorology, variance in the weather degrades the ability to recover accurate emissions in the assimilation. Next Steps • Include cloudy-scenes in the generation of TEMPO data to better describe the instrument sensitivity. • Attempt to focus meteorology assimilation on PBL only • Assess whether longer runs or larger spatial domain provide increased capability.
- Slides: 21