Chemical Data Assimilation using CAM and DART:

1
Chemical Data Assimilation using CAM and
DART Tests with CO Remote-Sensed Measurements
Avelino Arellano, Jr. and Peter Hess Atmospheric
Chemistry Division, NCAR Kevin Raeder and
Jeffrey Anderson Data Assimilation Research
Section, NCAR
2
Goal

• Provide a consistent and likely representation
  of CO distribution
• Develop a spatially and temporally robust
  estimates of CO emissions.

• (In the context of model improvement through
  parameter estimation)

Problem Current estimation procedure (or
inversion) rely on GCTMs to map the emission
(surface fluxes) to observable CO state
variables. As such, errors need to be reasonably
accounted for. A promising technique is a joint
data assimilation and parameter estimation using
Ensemble Kalman Filter (EnKF).
3
MODELS

• DART
• http//www.image.ucar.edu/DAReS

• CAM3.1 with CO as a tracer
• Used standard CAM3.1 with FV dycore (4ox5o)
  and a simplified CO chemistry,
• mainly based on CAMs carbon aerosol
  package. Emissions and sinks of CO
• prescribed.

4
OBSERVATIONS
For T,U,V NCEP BUFR (includes radiosonde
T,U,V ACARS data T,U,V SATwind U,V, etc) For CO
NASA Terra MOPITT CO Retrievals (used 500 hPa
subset, dynamic avg kernels)
e.g.
5
To check the validity and performance of current
DART-CAM-CO implementation
Conduct Observing System Simulation Experiments
(OSSEs)
Given that we know the true state and we have a
best guess of the probability distribution of the
initial state, can we reproduce the truth by
assimilating various synthetic observations?
6
A. Generate Initial Ensembles (80 members)

• For CAM variables ? taken from previous CAM
  climatological runs
• (Kevin Raeder)
• For CO ? generated by running CAM (CO) with FV
  dycore for 1 week using
• MOZARTv2 initial field and the ensemble
  of CAM/CLM2 initial conditions.
• (i.e. variability of CO generated from
  the ensemble alone).

7
B. Generate Synthetic Observations

• Generating synthetic obs is easily facilitated in
  DART (as one of the DART tools)
• 1) Took initial ens 40 and run CAM-CO with
  prescribed sources and sink.
• Assumed that generated model states are the
  true states.
• We sample the true states using the NCAR BUFR
  and MOPITT CO obs location and time (truth).
• Perturbed the sample by adding a Gaussian noise
  with variance represented
• by the obs instrument error variance
  (synthetic obs).

Synthetic Obs Truth (example 07-Jan-2003 500
hPa)
8
C. Carry out 3 OSSEs

• Using the same initial ensembles (1st 20 members)
  for T,U,V,CO and the same CO sources and sinks
  prescribed in generating synthetic observations,
  we carry out the following experiments for a
  7-day period with 20-member ensemble
• Forecast gtRun DART-CAM-CO w/ no
  assimilation.
• Analysis T,U,V gtRun DART-CAM-CO w/
  assimilation of T,U,V only.
• Analysis T,U,V,CO gtRun DART-CAM-CO w/
  assimilation of T,U,V and CO.

9
Initial Results
RMSEs for both assimilation approaches to the
prescribed observation RMSEs.
10

• Similar improvements in RMSE and RMS.
• RMSEs for both assimilation approaches to the
  prescribed observation RMSEs.

11
Assimilation of T,U,V also provides better match
of modeled atmospheric pressure with observations.
12
Bias for both assimilation approaches to the mean
bias of the observations relative to the truth.
13
RMSEs for both assimilation approaches to the
prescribed observation RMSEs.
Assimilating T,U,V alone provides large
constraints to CO.
14
(No Transcript)
15
Relative Comparison in Model Space
Assimilation able to reasonably reproduce the
true state even for surface CO (which is not
currently observed)
Note Variability of CO attributed to T,U,V.
Emissions are fixed.
16
Summary

• The setup for DART-CAM (CO) has been
  implemented using synthetic
• CO observations based on MOPITT retrievals.
• Initial results show the potential of current
  setup for model evaluation and
• longer assimilation studies.
• Simultaneously constraining T,U,V and CO in a
  GCTM offers an opportunity
• for model improvements (i.e. source
  parameter estimation).
• Challenges
• a) limitation in run time ( increasing
  overhead with additional observation)
• b) what is the optimal number of ensembles
  to use?
• c) explore sensitivity of assimilation
  parameters
• Future research
• a) Assimilation of real obs (MOPITT CO
  retrievals and/or radiances)
• b) Joint CO data assimilation and CO source
  estimation

17
Acknowledgements
NSF ITR Grant 115912 NCAR MOZART group NCAR
MOPITT Tim Hoar (IMAGe) Francis Vitt (ACD) Louisa
Emmons (ACD)
18
OSSEs
model advance
obs
Case 1 Forecast
ensemble state variables
06
12
18
24
Case 2 Assimilation (T,U,V)
assimilation
Posterior
Prior
06
12
18
24
Case 3 Assimilation (T,U,V, CO)
06
12
18
24
19

• Overarching Goal
• Provide a consistent and likely representation of
  CO distribution
• Develop spatially and temporally robust emissions
  (surface fluxes) of
• tropospheric CO. Has important implication to
  biogeochemistry.

20

• Overarching Goal
• Provide a consistent and likely representation of
  CO distribution
• Develop spatially and temporally robust emissions
  (surface fluxes) of
• tropospheric CO. Has important implication to
  biogeochemistry.
• (model improvement through parameter
  estimation)

Current estimation procedure (or inversion) rely
on GCTMs to map the source parameters to
observable CO state variables. As such, errors
need to be reasonably accounted for.