The potential to narrow uncertainty in regional climate predictions Ed Hawkins, Rowan Sutton NCAS-Climate, University of Reading IMSC 11 – July 2010
Motivation Adaptation planners would like quantitative projections of future climate on regional scales, especially for the next few decades —these projections exist but have (large) uncertainties Questions: —what are the largest sources of climate uncertainty on regional scales? —does this vary with region, lead time and climate variable? —are the dominant uncertainties potentially reducible?
European temperature projections
European temperature predictions
European temperature projections
European temperature predictions
Uncertainty in temperature projections Model uncertainty Scenario uncertainty Internal variability Global mean temperature Hawkins & Sutton, BAMS, 2009 – also see Cox & Stephenson (2007) CMIP3 projections Internal variability – spread in residuals from smooth fits to projections Scenario uncertainty – spread between multi-model mean of smooth fits Model uncertainty – spread around multi-model means of smooth fits Relative to
Hawkins & Sutton, BAMS, 2009 – also see Cox & Stephenson (2007) Uncertainty in temperature projections Model uncertainty Scenario uncertainty Internal variability CMIP3 projections British Isles (UK) mean temperature Internal variability – spread in residuals from smooth fits to projections Scenario uncertainty – spread between multi-model mean of smooth fits Model uncertainty – spread around multi-model means of smooth fits Relative to
A different representation Global mean temperature Hawkins & Sutton, 2010, Clim. Dyn.
A different representation British Isles mean temperature
Maps of uncertainty – temperature Hawkins & Sutton, BAMS, 2009
Precipitation uncertainties Global mean precipitation Hawkins & Sutton, 2010, Clim. Dyn.
Precipitation uncertainties Model uncertainty Scenario uncertainty Internal variability Global, decadal meanEuropean DJF, decadal mean Sahel JJA, decadal meanSE Asia JJA, decadal mean
Maps of uncertainty – DJF precipitation
Signal-to-noise ratios Signal-to-noise ratio (S/N) for JJA projections Hawkins & Sutton, 2010, Clim. Dyn.
Signal-to-noise ratios Signal-to-noise ratio (S/N) for JJA projections without model uncertainty with model uncertainty Hawkins & Sutton, 2010, Clim. Dyn.
Longer time means
Caveats Uncertainty estimates – only 3 scenarios used – only 15 models used – Internal variability estimate relies on GCMs Wide range in GCM estimates
Internal variability in CMIP3 GCMs Discussion:
Caveats Uncertainty estimates – only 3 scenarios used – only 15 models used – Internal variability estimate relies on GCMs Wide range in GCM estimates Spread ≠ skill Progress in climate science may increase uncertainty – carbon cycle feedbacks, ice sheet and land-use change uncertainties… Simple trend model used
Using ANOVA instead Thanks to Stan Yip, Chris Ferro, David Stephenson
Uncertainty in global ozone recovery Charlton-Perez et al. (2010), ACPD Global mean ozone CCMVal-2 intercomparison
Uncertainty in tropical evergreen tree cover for the Amazon Poulter et al., (2010), Glob. Change Bio.
Reducing uncertainty – decadal climate prediction June 1991June 2001 June 1995 Thanks to Jon Robson Retrospectively predicting North Atlantic upper ocean heat content Decadal climate prediction allows us to test our climate models in making predictions, to identify processes causing errors and may help predict some internal variability for up to a decade Observations GCM predictions
Summary Model uncertainty and internal variability are the dominant sources of uncertainty in regional climate projections for next few decades. —Uncertainty is potentially reducible with progress in climate science —Internal variability more important for precipitation than temperature —Scenario uncertainty is almost negligible in the tropics for precipitation Potential for reduction in uncertainty for precipitation appears smaller —Adaptation decisions will need to be made with low S/N predictions for precipitation, even with a perfect model! Climate impact modellers need to use more than one GCM! Could estimate potential value of climate science investments to reduce uncertainty, compared to economic savings from less costly adaptation Interactive website:
Robustness of internal variability CONTROL TRANSIENT
Fractional uncertainty comparison Using CMIP3 data, model uncertainty is clearly the dominant contribution for decadal predictions Cox & Stephenson schematic Hawkins & Sutton, BAMS, 2009 Using CMIP3 projections uncertainty mean signal