The Cloud Feedback Model Intercomparison Project (CFMIP) Progress and future plans

1. The Cloud Feedback Model Intercomparison Project (CFMIP) Progress and future plans Mark Webb (Hadley Centre) and CFMIP contributors (Hadley Centre,IPSL, NIES/CCSR,BMRC,GFDL, MPI, NCAR,UIUC) September 2006

2. Outline CFMIP background Low cloud feedbacks and spread in climate sensitivity Reducing uncertainty in low cloud feedbacks Future plans for CFMIP

3. CFMIP: Cloud Feedback Model Inter-comparison Project Modelling and Prediction of Climate variability and change • Set up by Bryant McAvaney (BMRC), Herve Le Treut (LMD) • WCRP Working Group on Coupled Modelling (WGCM) • Systematic intercomparison of cloud feedbacks in GCMs • Aim to identify key uncertainties • Link climate feedbacks to cloud observations • +/-2K atmosphere only and 2xCO2 ‘slab’ experiments • ISCCP simulator (Klein & Jakob, Webb et al) mandatory • Now data for 13 GCM versions from 8 modelling groups • Website shows data available, publications, plans, etc. • http://www.cfmip.net

4. Low cloud feedbacks and spread in climate sensitivity Modelling and Prediction of Climate variability and change Recent studies have attributed much of the spread in climate sensitivity in climate models to differences in low cloud feedbacks: Bony & Dufresne 2005 (15 coupled AR4 models) Wyant et al 2006 (3 CPT* models) Webb et al 2006 (9 CFMIP mixed layer models) * Low Latitude Cloud Feedback Climate Process Team (CPT)

5. Bony and Dufresne GRL 2005 Cloud radiative forcing (CRF) climate response in vertical velocity bins over tropical oceans (30N-30S) from 15 coupled climate models 8 higher sensitivity and 7 lower sensitivity Net CRF spread largest in subsidence regions, suggesting low clouds are ‘at the heart’ of cloud feedback uncertainties

6. LW cloud feedback (W/m2/K) SW cloud feedback(W/m2//K) Net cloud feedback (W/m2/K) LW cloud feedback (W/m2/K) SW cloud feedback(W/m2//K) Net cloud feedback (W/m2/K) Webb et al Climate Dynamics 2006 – CFMIP models Areas with small LW cloud feedbacks explain 59% of the NET cloud feedback ensemble variance Cloud feedbacks in these areas are indeed dominated by reductions in low level cloud amount (shown with ISCCP simulator)

7. Reducing uncertainty in low cloud feedbacks Two basic strategies: 1/ find relationships between climate feedbacks and observables and use these as a basis for constraining climate feedbacks 2/ understand the physics of feedback mechanisms in models and fix aspects which are not credible, hoping that model differences will reduce in the long term These strategies can be pursued in parallel

8. Williams and Tselioudis (Climate Dyn submitted) ISCCP observational cloud cluster regimes (20N-20S)

9. Williams and Tselioudis (CFMIP daily data) In the cloud regime framework, the mean change in cloud radiative forcing can be thought of as having contributions from: • A change in the RFO (Relative Frequency of Occurrence) of the regime • A change in the CRF (Cloud Radiative Forcing) within the regime (i.e. a change in the tau-CTP space occupied by the cluster/development of different clusters).

10. HadSM3 Cloud Response along GCSS Pacific Cross Section transect Mid-level cloud categories excluded. SW CRF response along Pacific transect in CFMIP and AR4 slab models. Cloud feedbacks are typically smaller than model biases. No clear relationship between bias and feedback

11. HadSM3 Cloud Response along GCSS Pacific Cross Section transect(Negative low cloud feedback case) Mid-level cloud categories excluded.

12. Low cloud response in HadSM3 transition regions • Deep and shallow convection weaken in the warmer climate (consistent with Held and Soden 2006 ) • Shallow convection typically detrains into two model layers in present day, but one level in the warmer climate • If a certain amount of water vapour is detrained into a single layer it will moisten that layer more than would be the case if it was spread over two layers • May explain why HadSM3 stratiform cloud fraction increases with weakening shallow convection • Hence the negative low cloud feedback in HadSM3 may be due to poor vertical resolution and so not credible

13. HadGSM1 Cloud Response along GCSS Pacific Cross Section transect

14. CFMIP Phase I continues… Modelling and Prediction of Climate variability and change Daily cloud diagnostics to be hosted by PCMDI To be made available as a community resource UKMO, MIROC, MPI, NCAR data currently in transit IPSL, Env Canada promised later this year Will allow many ISCCP cloud studies to be applied to a representative selection of climate models Plans to build cloud clustering into ISCCP simulator to support its routine use in model evaluation

15. CFMIP Phase II – looking further ahead Modelling and Prediction of Climate variability and change The CFMIP Phase I experimental protocol will ideally become part of the IPCC requirements for AR5 CFMIP Phase II will aim to improve our understanding of cloud feedback mechanisms over the next 5 years by: - using idealised climate experiments (e.g. patterned SST forcing) - exploring sensitivities to model physics - developing more detailed model diagnostics

16. Proposed new diagnostics for CFMIP Phase II Modelling and Prediction of Climate variability and change CFMIP Cloudsat/CALIPSO Simulator (C3S) Sub-timestep information - Cloud condensate budget terms - Physics increments for temperature, humidity, etc Detailed diagnostics at key locations as used in GCSS/ARM studies (e.g GPCI)

17. CFMIP CloudSat/CALIPSO simulator (C3S) Modelling and Prediction of Climate variability and change Under development by Alejandro Bodas-Salcedo and Marjolaine Chiriaco (Hadley Centre, LMD/IPSL) A package to simulate radar and lidar reflectivities from within models to support model validation using CloudSat/CALIPSO data

18. B A C3S/CloudSat comparison with NWP model(A. Bodas-Salcedo) Transect trough a mature extra-tropical system Strong signal from ice clouds Strong signal from precipitation Cloud and precip not present in obs B A

19. C3S LIDAR comparison with GLAS / ICESat data (M. Chiriaco LMD/IPSL) Lidar signalsimulated from LMDZ outputs z, km Lidar signalobserved from GLAS spatial lidar z, km signal latitude • Evaluation of the vertical structure of the atmosphere in models, at global scale Indicates excessive reflectivities from cloud ice in this climate model

20. Cloud water budget / process diagnostics (Tomoo Ogura NIES Japan) Condensation + Precipitation + Ice-fall-in + Ice-fall-out (1) (2) Cumulus Conv. + Advection + + mixing adjustment (3) (4) (5) - Transient response of terms (1)…(5) to CO2 doubling is monitored every year. - Terms showing positive correlation with cloud water response contribute to the cloud water variation. Qc response Qc increase related to the source term Source term response Qc decrease related to the (1)or…(5) source term

21. Cloud water budget diagnosis in MIROC (Tomoo Ogura NIES Japan) Correlation coefficient (when positive) (5)Dry conv. adj. [Cloud water vs ] (1)Condens-Precip (3) Advection (2) Ice fall in+out (4)Cumulus ΔCloud water (2xco2 - 1xco2, DJF) sigma 0.2 0.2 0.6 0.6 1.0 1.0 90S 60S 30S EQ 30N 60N 90N 90S 60S 30S EQ 30N 60N 90N contour=3e-6 [kg/kg] -1 0 1 cloud decrease increase Decrease in mid-low level sub-tropical cloud water correlates with decrease in large-scale condensation

22. Detailed diagnostics at key locations Modelling and Prediction of Climate variability and change CFMIP Phase II is an opportunity to align diagnostics in a range of climate models with those in use in other areas such as GCSS, ARM and the low cloud CPT We could provide data from AMIP and idealised climate warming simulations – for example: 3 hourly or timestep data at selected locations - GPCI points - ARM sites - could include forcing data for SCMs/CRMs - other suggestions welcome…

23. Summary A number of studies now point to low cloud feedbacks being a key uncertainty in climate models Daily cloud/radiation/ISCCP simulator diagnostics from CFMIP Phase I will be available from PCMDI by the end of the year Reductions in uncertainty due to cloud feedbacks will require links to observations and also understanding and criticism of feedback mechanisms in models CFMIP Phase II is an opportunity to align diagnostics in a range of models with those in use in GCSS/ARM/CPT Doing so should allow a wider range of expertise to be applied to analysis of multi-model cloud climate feedbacks

24. Hawaii – California transect: IPCC AR4 workshop March 2005 mark.webb@metoffice.gov.uk

25. Alternative experimental setups Modelling and Prediction of Climate variability and change Cess +/-2K fixed season experiments are not a quantitative guide to coupled model feedbacks - no seasonal cycle - no high latitude amplification of warming Alternative options include: - continuing use of mixed layer experiments - re-running sections of AR4 coupled experiments with extra diagnostics - ‘patterned SST perturbation’ experiments as developed by Brian Soden (possibly AMIP+)

26. Proposed climate physics sensitivity tests Modelling and Prediction of Climate variability and change Consistent implementation of simplified physics in climate models to identify which contribute most to spread in cloud feedbacks – e.g. - suppression of interactive cloud radiative properties - simplified mixed-phase feedback processes - simplified clouds and precipitation - consistent treatment of shallow convection Also possible tests of the effects of differing deep convective entrainment/detrainment on climate sensitivity New process diagnostics will aid the interpretation of these experiments

27. Publications using CFMIP data Modelling and Prediction of Climate variability and change Williams and Tselioudis 2006 GCM intercomparison of global cloud regimes - Part I: Evaluation of present-day cloud regimes. Clim. Dyn. Submitted. Williams and Tselioudis 2006 GCM intercomparison of global cloud regimes - Part II: Climate change response and evaluation of the change in the radiative properties of cloud regimes. Clim. Dyn. Submitted. Ogura et al 2006 Climate sensitivity of a general circulation model with different cloud modelling assumptions. J. Meteor. Soc. Japan Submitted. Tsushima et al 2006 Importance of the mixed-phase cloud distribution in the control climate for assessing the response of clouds to carbon dioxide increase: a multi-model study. Clim. Dyn. 27 Webb et al 2006 On the contribution of local feedback mechanisms to the range of climate sensitivity in two GCM ensembles. Clim. Dyn. 27 Ringer et al 2006 Global mean cloud feedbacks in idealized climate change experiments. Geophys. Res. Lett. 33 Williams et al 2006 Evaluation of a component of the cloud response to climate change in an intercomparison of climate models. Clim. Dyn. 26

28. CFMIP: Website http://www.cfmip.net Modelling and Prediction of Climate variability and change

29. Further links between feedbacks and observables Modelling and Prediction of Climate variability and change New diagnostics will provide opportunities for new links e.g. CloudSat/CALIPSO data may constrain models with strong mixed-phase cloud feedbacks due to excessive amounts of cloud ice This will be an ongoing area of research, and the focus of a joint ENSEMBLES/CFMIP meeting in Paris 11th-13th April 2007

30. CFMIP Phase II timescales Modelling and Prediction of Climate variability and change Concrete project proposal by Jan 2007 Aim for endorsement by WGCM Mar 07 Joint CFMIP/ENSEMBLES meeting Paris April 2007 Development of diagnostics / pilot studies 2007-2008 Systematic model inter-comparison with new model versions 2008-2009

31. Bony and Dufresne (2005) Interannual sensitivity of CRF to SST in vertical velocity bins over tropical oceans (30N-30S) from 15 coupled climate models 8 higher sensitivity and 7 lower sensitivity and observed (ISCCP FD) Models underpredict low cloud sensitivity, but higher sensitivity models less so.

32. ERBE /ERA SW ERBE /ERA LW ERBE/ NCEP SW ERBE/ NCEP LW Avg RMS Avg Climate Sens.& Range 1.2 1.3 1.5 1.8 1.7 1.0 1.2 1.0 1.2 1.1 1.6 1.3 1.9 2.1 2.2 0.9 1.1 1.0 1.1 1.2 1.2 1.2 1.4 1.5 1.5 3.8K 2.9 -4.4K 2.1 2.4 1.7 2.2 3.4 1.1 1.0 1.6 1.4 1.3 2.3 3.0 2.2 2.8 3.7 1.0 1.0 1.9 1.5 1.1 1.6 1.9 1.9 2.0 2.4 3.0K 2.3 -3.5K Williams et al Climate Dynamics 2006 (CFMIP) RMS-differences of present-day variability composites against observations for 10 CFMIP/CMIP model versions. The five models with smallest RMS errors tend to have higher climate sensitivities. (Consistent with Bony & Dufresne 2005)

33. CFMIP CloudSat/CALIPSO simulator (Alejandro Bodas-Salcedo) Simulated CloudSat reflectivities from UKMO forecast model • The simulator consists of 5 steps: • Orbital simulation • Sampling • Preprocessing • Subgrid sampling of cloud overlaps • Radar reflectivity calculated using code provided by Matt Rogers (CSU) Outputs: - Reflectivity from clouds and precipitation (without attenuation) - Total reflectivity, accounting for attenuation by gases, clouds and precipitation - Products obtained from inputs at gridbox and subgrid scales

34. ex. lidar profile (log) ACTSIM CALIPSO simulator – Marjolaine Chiriaco (LMD/IPSL) ex. particles concentration profile (/m3) ice ACTSIM Lidar Equation ex. water content profile (g/m3) • Optical properties: P(p,z), Qsca(z), Qabs(z) (532 nm) • Lidar calibration • Multiple scattering ice liquid snow

35. CFMIP Phase II C3S inter-comparison Modelling and Prediction of Climate variability and change Initially we will provide an A-train orbital simulator to allow climate modellers to save model cloud variables co-located with CloudSat/CALIPSO overpasses Initially sampled data would be submitted to CFMIP and both simulators run centrally As the approach matures we plan to integrate this package with the ISCCP simulator so that it can be run in-line as part of the model development cycle

36. Reducing uncertainty by understanding feedbacks 1/ Try to understand how physical cloud climate feedback processes are operating in models 2/ Ask ‘Is this behaviour credible?’ 3/ Develop new schemes with more credible cloud-climate feedback behaviour in mind 4/ Differences between model feedbacks may reduce in the longer term This approach has much in common with that of the low latitude cloud feedback CPT