 |
<< Back to WCRP CMIP3 Subprojects
Estimated noise covariance from multi-model ensemble of unforced control simulations
| PI: |
Chris E. Forest |
| Institution: |
Massachusetts Institute of Technology |
| Additional Investigators: |
Peter Stone (MIT), Andrei Sokolov (MIT), Bruno Sanso (UC-Santa Cruz), Doug Nychka (NCAR) |
| Abstract: |
Unforced variability of climate change detection patterns on century time-scales must be estimated from long control simulations because observational records are too short. To date, we have relied on only a few models for such estimates owing to availability. This multi-model project will enable us to estimate the unforced variability from a significantly larger number of models. We would like to participate in the “multi-model analysis activity” by estimating the noise covariance that is required for the climate change detection problem for surface, upper-air and deep-ocean temperatures. One key issue is the short length of unforced control runs for estimating the noise covariance matrix. Typically, a climate model control run is a few hundred years long and only yields a small number of independent 50 or 100 year segments that can be used as noise realizations. We would create a multi-model ensemble from the available control simulations. A primary question is whether the estimates of C_N from the individual models are distinguishable from each other. If not, the multi-model ensemble should provide a significantly larger number of degrees of freedom to improve the estimated C_N. If they do differ, this would be an equally interesting result, and their differences would be one source of uncertainty in the detection analyses.
This would require the longest available time-series from the control run data as available for each coupled model. We would need the monthly mean (or seasonal or annual) of surface 2m air, upper-air (on standard pressure levels), and deep-ocean temperatures. |
| Publications: |
Add Publication
|
<< Back to WCRP CMIP3 Subprojects
|
|
