Data assimilation for validation of climate modeling systems

1
Data assimilation for validation of climate
modeling systems

• Pierre Gauthier

2
Validation of atmospheric models

• Transpose Atmospheric Model Intercomparison
  Experiments (AMIP)
• Comparison of atmospheric models against each
  other under same conditions (e.g., initial
  conditions provided by the same analysis)
• Short term forecasts as in NWP
• Intercomparison of data assimilation systems
• More difficult to carry out due to the added
  complexity coming from observations, data
  assimilation components, and atmospheric model
• Impact on both the analysis (information content)
  or on the forecasts

3
Schematic of the data assimilation process
(from Rodwell and Palmer, 2007)
4
Validation of atmospheric models against
observations

• Short-term forecast used by the assimilation
• Common ground provided by a short term forecast,
  the background state
• Sums up the information gained from observations
• Monitoring of averages of observations minus
  background is used to detect biases
• in new or existing observations
• the model itself (detection of biases)
  particularly if the observation dataset has been
  carefully quality controlled

5
Monitoring and quality control

• Statistics based on innovations (y -HXb) example
  from TOVS radiances

6
Using NWP to assess climate models(Rodwell and
Palmer, 2007)

• Impact of changes to climate models usually done
  by comparing several long climate runs with
  perturbed models
• Uncertainty associated with the physical
  processes used in the model (Stainforth et al.,
  2005)
• Assimilation produces analysis increments to
  correct the model forecast to bring it closer to
  the observations
• Reduced analysis increments is an indication that
  the model has improved its fit to observations
• Presence of spin-up can be associated with model
  differences with respect to what has been
  observed
• Examination of the physical tendencies in the
  early stages of the forecast can provide useful
  information about imbalances in the model

7
Schematic of the data assimilation process
(from Rodwell and Palmer, 2007)
8
Time series of precipitation rates averaged over
the Northern Hemisphere (Gauthier and Thépaut,
2001)
9
RMS error w.r.t. unperturbed model vs. Simulated
climate sensitivity
(from Rodwell and Palmer, 2007)
10
Comparing physical tendencies for different
processes in experiments with perturbed physics
Total tendency
Convection
Dynamicalcooling
Rodwell and Palmer (2007)
11
Conclusions

• Data assimilation and reanalyses
• often based on an adapted NWP suite for which the
  model short term forecasts have been thoroughly
  validated
• Using a climate model to do data assimilation
• provide detailed information about systematic
  departures from observations
• Examination of the physical tendencies associated
  with the first instants of a forecast can
• Indicate how imbalances in the physical processes
  may cause excessive model sensitivity which
  increase the uncertainty of climate predictions
• Observation datasets used for reanalyses could be
  valuable for studies on climate model validation
• Added value for the data prepared for reanalyses

12
Conclusion (contd)

• For coupled systems, the complexity is increasing
  and this approach is certainly to be encouraged
• Parameter estimation with coupled models (Sugiura
  et al., 2008) to adjust parameter related to
  surface fluxes
• Found it was necessary to adjust also other
  parameterizations
• (1) the wind sensitivity parameter in the ocean
• (2) the isopycnal diffusion coefficient in the
  OGCM,
• (3) the mixing length in an atmospheric boundary
  layer in the AGCM,
• (4) the relaxation time for large-scale
  condensation in the AGCM,
• (5) the range of relative humidity change in the
  AGCM,
• (6) the standard height for precipitation
  efficiency in the AGCM, and
• (7) the adjustment time for cumulus convection in
  the AGCM