Quantifying the Sources of Inter-Model Spread in Climate Sensitivity. Peter Caldwell, Mark Zelinka , Karl Taylor, and Kate Marvel LLNL. Prepared by LLNL under Contract DE-AC52-07NA27344. LLNL-PRES-661065. Motivation.
Quantifying the Sources of Inter-Model Spread in Climate Sensitivity
Peter Caldwell, Mark Zelinka, Karl Taylor, and Kate Marvel
Prepared by LLNL under Contract DE-AC52-07NA27344.
Here we point out issues with their approach and re-evaluate their conclusions
Goal: Write ECS as a sum of terms representing contributions from each feedback process. This can be done as follows:
net feedback, λ=Σλi
Inter-model mean forcing
In essence, Planck ΔT is calculated first and the remainder of ECS is partitioned amongst remaining feedbacks with size proportional to relative magnitude
DB08 consider inter-model standard deviation in each Ti (see figure), but formally
so covariances between feedbacks should also be considered!
Fig: Normalized inter-model standard deviation of ECS partitioning. Reprinted from DB08.
Covariance between water vapor/lapse rate is well known (e.g. Cess, 1975) and covariance between other λi has been noted recently (Huybers, 2010; Mauritsen et al, 2013)
Fig: Covariance matrices for both ECS partitionings. Details are listed above.
Fig: Covariance matrices for both ECS partitionings.
Fig: Covariance matrices for ECS partitionings(right) and for λ=Σλi (left)
Since Vial doesn’t separate forcing variability from λ variability:
Vial eq 24:
Divides by λPlinstead of λ
Uses F instead of F
Fig: After partitioning ECS into components listed on x-axis, Vial computes Inter-model standard deviations and normalizes by the standard deviation of ECS. This is Fig. 6 from Vial et al. (2013)
Fig: Covariance matrices as on previous slides, but with WV+LR and SW+LW cld feedbacks combined
Held and Shell (JClim 2012, hereafter HS12) note that relative humidity (RH) tends to stay constant as climate changes, but water vapor feedback is computed as the change in specific humidity
We think HS12 is better because it:
Fig: Covariance matrices shown previously (left) and using HS12 definitions (right)
corr(λSW, λnet)2=0.69 (i.e. λSWexplains ~70% of λnet spread)so λSWshould play a dominant role in any reasonable partitioning of ECS
Fig:ECS calculated as –F/λnet (in blue) and as calculated using multi-model averages for all feedbacks except λSW