Upload
tim-osborn
View
93
Download
0
Embed Size (px)
Citation preview
Pa#ern scaling using ClimGen: User needs
Changing precipita0on variability Interac0on between global & regional responses
Tim Osborn & Craig Wallace Clima&c Research Unit, University of East Anglia
April 2014
Pa:ern scaling, climate model emulators & their applica&on to the new scenario process
NCAR, Boulder, Colorado
Work supported by TOPDAD & HELIX EU projects
Pattern scaling: meeting user needs
Key requirements: • Explore spread (uncertainty?) of climate projec0ons
• Pre-‐CMIP3, CMIP3, CMIP5 mul0-‐model, QUMP perturbed parameters
• Generate projec0ons for un-‐simulated scenarios User needs: • Iden0cal formats for all scenarios (& observa0ons) • Flexible temporal, seasonal and geographic windowing/averaging
Pattern scaling: meeting user needs
Example na0onal average summer T & P changes Pink = CMIP3 distribu:on Open symbols = CMIP3 models
Key requirements: • Explore spread (uncertainty?) of climate projec0ons
• Pre-‐CMIP3, CMIP3, CMIP5 mul0-‐model, QUMP perturbed parameters
• Generate projec0ons for un-‐simulated scenarios
Natural variability
ΔT = 0.5, 1.5, 3
For global warming ΔT = 3 K (left panel) or 0.5, 1.5 and 3 K (right panel)
Based on Osborn et al. (under review) Climatic Change
Pattern scaling: meeting user needs
Example na0onal average summer T & P changes Pink = CMIP3 distribu:on Open symbols = CMIP3 models Brown = CMIP5 distribu:on Solid symbols = CMIP5 models
Key requirements: • Explore spread (uncertainty?) of climate projec0ons
• Pre-‐CMIP3, CMIP3, CMIP5 mul0-‐model, QUMP perturbed parameters
• Generate projec0ons for un-‐simulated scenarios
Natural variability
ΔT = 0.5, 1.5, 3
For global warming ΔT = 0.5, 1.5 and 3 K (right panel)
Based on Osborn et al. (under review) Climatic Change
Pattern scaling: meeting user needs
Key requirements: • Explore spread (uncertainty?) of climate projec0ons
• Pre-‐CMIP3, CMIP3, CMIP5 mul0-‐model, QUMP perturbed parameters
• Generate projec0ons for un-‐simulated scenarios
Natural variability
ΔT = 0.5, 1.5, 3
Example na0onal average summer T & P changes Pink = CMIP3 distribu:on Open symbols = CMIP3 models Brown = CMIP5 distribu:on Solid symbols = CMIP5 models Blue = QUMP distribu:on Black le#ers = QUMP models
For global warming ΔT = 0.5, 1.5 and 3 K (right panel)
Based on Osborn et al. (under review) Climatic Change
Pattern scaling: meeting user needs
Mul0ple climate variables (all monthly means, mostly land-‐only): • Near-‐surface temperature (mean, min, max, DTR) • Precipita0on & wet-‐day frequency • Cloud-‐cover (can es0mate sunshine hours or radia0on variables) • Vapour pressure (can es0mate other humidity variables) • SST is currently the only variable provided over the oceans
User needs: more derived variables, extreme events & variability • Hea0ng & cooling degree days (HDD & CDD) • Poten0al evapotranspira0on (PET, e.g. from Penman-‐Mon0eth) • Drought indicators (e.g. Standardised Precipita0on-‐Evapotranspira0on
Index, SPEI)
How to deal with climate (and weather) variability?
Climate variability in pattern scaling: (1) use observations
Sample from observed variability: • Realis0c for present-‐day • But doesn’t change when the mean climate changes
Design sampling to allow the separa0on of climate change and natural variability effects
• Use mul0ple 0me-‐shided sequences instead of single observed sequence
Climate variability in pattern scaling: (1) use observations
Sample from observed variability: • Realis0c for present-‐day • But doesn’t change when the mean climate changes
Design sampling to allow the separa0on of climate change and natural variability effects
• Use mul0ple 0me-‐shided sequences instead of single observed sequence
Climate variability in pattern scaling: (1) use observations
• Or generate slices represen0ng climate+variability for specific amounts of ΔT
Fig. S3 of Osborn et al. (under review) Climatic Change
Climate variability in pattern scaling: (2) perturb observations
Pahern-‐scale higher moments (e.g. standard devia0on, skew) • We divide GCM monthly precipita0on 0meseries by low-‐pass filter • Represent the high-‐frequency devia0ons with a gamma distribu0on • Scale changes in gamma shape parameter with ΔT
Fig. 1 of Osborn et al. (under review) Climatic Change
Rel
ativ
e ch
ange
in
Climate variability in pattern scaling: (2) perturb observations
Example applica0on • SE England grid cell, HadCM3 GCM, July precipita0on • For ΔT = 3°C, pahern-‐scaling gives 45% reduc0on in mean precipita0on • But also 62% reduc0on in gamma shape param. of monthly precipita0on
Fig. 1 of Osborn et al. (under review) Climatic Change
Observed sequence
Sequence x 0.55 Sequence x 0.55
Sequence x 0.55 & perturbed to have 62% lower shape
Is there agreement in GCM-simulated changes of variability?
• Mul0-‐model mean of 22 CMIP3 GCMs • Normalised change in gamma shape of July precipita0on
Units: % change / K
Fig. 1 of Osborn et al. (under review) Climatic Change
Is there agreement in GCM-simulated changes of variability?
• Mul0-‐model mean of 20 CMIP5 GCMs • Normalised change in gamma shape of July precipita0on
Units: % change / K
Based on Osborn et al. (under review) Climatic Change
Is there agreement in GCM-simulated changes of variability?
• Mul0-‐model agreement of 22 CMIP3 GCMs • Frac0on of models showing increased gamma shape of July precipita0on
Units: fraction
Based on Osborn et al. (under review) Climatic Change
Is there agreement in GCM-simulated changes of variability?
• Mul0-‐model agreement of 20 CMIP5 GCMs • Frac0on of models showing increased gamma shape of July precipita0on
Units: fraction
Based on Osborn et al. (under review) Climatic Change
Transform observed rainfall series by factors given by range of ΔT from 0 to 6K Count frequency of short droughts in each transformed series Estimate uncertainty
UK drought frequency vs.
global ΔT
Does pattern-scaling emulate GCM/RCM behaviour?
HadCM3 GCM HadRM3 RCM
Can we treat global and regional changes independently? • Separa0on into global ΔT & regional paherns is convenient • Especially for the treatment of uncertain0es
Can we treat global and regional changes independently? • Separa0on into global ΔT & regional paherns is convenient • Especially for the treatment of uncertain0es
Simple example: Estimating conditional PDFs of UK drought frequency, using HadRM3 RCM pattern-scaling results and the Wigley & Raper (2001) PDFs of ΔT
Simple example: Estimating conditional PDFs of UK drought frequency, using HadRM3 RCM pattern-scaling results and the Wigley & Raper (2001) PDFs of ΔT
Can we treat global and regional changes independently? • Separa0on into global ΔT & regional paherns is convenient • Especially for the treatment of uncertain0es
Estimating conditional PDFs of UK drought frequency
Can we treat global and regional changes independently? • Separa0on into global ΔT & regional paherns is convenient • Especially for the treatment of uncertain0es
Can we treat global and regional changes independently? • Separa0on into global ΔT & regional paherns is convenient • Especially for the treatment of uncertain0es
• But can I combine ΔT derived from a par0cular climate sensi0vity with any of the GCM paherns?
• Or are the normalised change paherns of high sensi0vity GCMs systema0cally different from those of low sensi0vity GCMs?
Rank correla0on between temperature and ECS for CMIP3
Are the normalised change patterns of high sensitivity GCMs systematically different from those of low sensitivity GCMs?
Osborn et al. (in preparation) Rank correlation for 22 GCMs
>80% significant correlations shown
Rank correla0on between temperature and ECS for QUMP
Are the normalised change patterns of high sensitivity GCMs systematically different from those of low sensitivity GCMs?
Osborn et al. (in preparation) Rank correlation for 17 GCMs
>80% significant correlations shown
Rank correla0on between temperature and ECS for CMIP3, CMIP5 & QUMP
Are the normalised change patterns of high sensitivity GCMs systematically different from those of low sensitivity GCMs?
Osborn et al. (in preparation) Rank correlation for 52 GCMs
>80% significant correlations shown
Conclusions: meeting user needs with pattern scaling
Exploring the uncertainty of climate projec0ons: • Given wide mul0-‐model ensemble ranges, sufficient to approximately
emulate plume of future regional changes Increasing demand for emula0on to include variability & represent extremes: • Need to treat variability with care, sufficient sampling etc. • Can pahern-‐scale higher order parameters (e.g. standard devia0on,
skew) and perturb observed variability accordingly • More complicated changes (e.g. shid in ENSO behaviour) cannot,
however, be captured Systema0c differences between normalised paherns from low and high sensi0vity models complicates the separate treatment of uncertainty in global ΔT and regional climate change