24
7/18/2019 Jones Et Al 2013 Model Attribution http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 1/24 JOURNAL OF GEOPHYSICAL RESEARCH: ATMOSPHERES, VOL. 118, 4001–4024, doi:10.1002/jgrd.50239, 2013 Attribution of observed historical near-surface temperature variations to anthropogenic and natural causes using CMIP5 simulations Gareth S. Jones, 1 Peter A. Stott, 1 and Nikolaos Christidis 1 Received 19 July 2012; revised 7 January 2013; accepted 3 February 2013; published 21 May 2013. [1]  We have carried out an investigation into the causes of changes in near-surface temperatures from 1860 to 2010. We analyze the HadCRUT4 observational data set which has the most comprehensive set of adjustments available to date for systematic  biases in sea surface temperatures and the CMIP5 ensemble of coupled models which represents the most sophisticated multi-model climate modeling exercise yet carried out. Simulations that incorporate both anthropogenic and natural factors span changes in observed temperatures between 1860 and 2010, while simulations of natural factors do not warm as much as observed. As a result of sampling a much wider range of structural modeling uncertainty, we find a wider spread of historic temperature changes in CMIP5 than was simulated by the previous multi-model ensemble, CMIP3. However, calculations of attributable temperature trends based on optimal detection support  previous conclusions that human-induced greenhouse gases dominate observed global warming since the mid-20th century. With a much wider exploration of model uncertainty than previously carried out, we find that individually the models give a wide range of  possible counteracting cooling from the direct and indirect effects of aerosols and other non-greenhouse gas anthropogenic forcings. Analyzing the multi-model mean over 1951–2010 (focusing on the most robust result), we estimate a range of possible contributions to the observed warming of approximately 0.6 K from greenhouse gases of  between 0.6 and 1.2 K, balanced by a counteracting cooling from other anthropogenic forcings of between 0 and –0.5 K. Citation: Jones, G. S., P. A. Stott, and N. Christidis (2013), Attribution of observed historical near-surface temperature variations to anthropogenic and natural causes using CMIP5 simulations, J. Geophys. Res. Atmos. , 118 , 4001–4024, doi:10.1002/jgrd.50239. 1. Introduction [2] Successive reports of the Intergovernmental Panel on Climate Change (IPCC) have come to increasingly confi- dent assessments of the dominant role of human-induced greenhouse gas emissions in causing increasing global near-surface temperatures [e.g.,  IPCC , 2007a]. Furthermore, analyses of changes across the climate system, including of sub-surface ocean temperatures, of the water cycle, and of the cryosphere, show that as the observational evidence accumulates, there is an increasingly remote possibility that climate change is dominated by natural rather than anthropogenic factors [Stott et al., 2010]. Thus, the IPCC concluded in its fourth assessment report that “most of the observed increase in global average temperatures since Additional supporting information may be found in the online version of this article. 1 Met Office, Exeter, UK. Corresponding author: G. S. Jones, Met Office, FitzRoy Rd., Exeter, UK. (gareth.s.jones@metoffice.gov.uk) ©2013. Crown copyright. This article is published with the permission of the Controller of HMSO and the Queen’s Printer for Scotland. 2169-897X/13/10.1002/jgrd.50239 the mid-20th century is very likely due to the observed increase in anthropogenic greenhouse gas concentrations” [  IPCC , 2007a]. [3] However, despite the high level of confidence in the conclusion that greenhouse gases caused a substantial part of the observed warming, there remain many uncertainties that have so far limited the potential to be more precise about such attribution statements. These uncertainties are associ- ated with observational uncertainties caused by remaining  biases in data sets and gaps in global coverage [  Morice et al. , 2012] and modeling uncertainties, which limit the ability to define the expected fingerprints of change due to anthro-  pogenic and natural factors, and which result from errors in model formulation, deficiencies in model resolution, and inadequacies in the way external climate forcings are speci- fied. All attribution results are contingent on such remaining uncertainties and until now they have been explored in a relatively limited way. Many previous attribution studies have been limited to a single observational data set and a single climate model [e.g.,  Tett et al., 2002] or a rather limited ensemble of different climate models [e.g.,  Hegerl et al., 2000;  Gillett et al., 2002;  Huntingford et al., 2006] while another study used simple climate models to emu- late global mean temperatures from over a dozen models [Stone et al., 2007]. 4001

Jones Et Al 2013 Model Attribution

Embed Size (px)

DESCRIPTION

Jones et al 2013, Model Attriubtion to Climate Prediction Variability

Citation preview

Page 1: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 1/24

JOURNAL OF GEOPHYSICAL RESEARCH: ATMOSPHERES, VOL. 118, 4001– 4024, doi:10.1002/jgrd.50239, 2013

Attribution of observed historical near-surface temperature variations

to anthropogenic and natural causes using CMIP5 simulations

Gareth S. Jones,

1

Peter A. Stott,

1

and Nikolaos Christidis

1

Received 19 July 2012; revised 7 January 2013; accepted 3 February 2013; published 21 May 2013.

[1]   We have carried out an investigation into the causes of changes in near-surfacetemperatures from 1860 to 2010. We analyze the HadCRUT4 observational data setwhich has the most comprehensive set of adjustments available to date for systematic

 biases in sea surface temperatures and the CMIP5 ensemble of coupled models whichrepresents the most sophisticated multi-model climate modeling exercise yet carried out.Simulations that incorporate both anthropogenic and natural factors span changes inobserved temperatures between 1860 and 2010, while simulations of natural factors donot warm as much as observed. As a result of sampling a much wider range of structuralmodeling uncertainty, we find a wider spread of historic temperature changes in CMIP5than was simulated by the previous multi-model ensemble, CMIP3. However,

calculations of attributable temperature trends based on optimal detection support previous conclusions that human-induced greenhouse gases dominate observed globalwarming since the mid-20th century. With a much wider exploration of model uncertaintythan previously carried out, we find that individually the models give a wide range of 

 possible counteracting cooling from the direct and indirect effects of aerosols and other non-greenhouse gas anthropogenic forcings. Analyzing the multi-model mean over 1951–2010 (focusing on the most robust result), we estimate a range of possiblecontributions to the observed warming of approximately 0.6 K from greenhouse gases of 

 between 0.6 and 1.2 K, balanced by a counteracting cooling from other anthropogenicforcings of between 0 and –0.5 K.

Citation: Jones, G. S., P. A. Stott, and N. Christidis (2013), Attribution of observed historical near-surface temperature variations

to anthropogenic and natural causes using CMIP5 simulations, J. Geophys. Res. Atmos., 118, 4001–4024, doi:10.1002/jgrd.50239.

1. Introduction

[2] Successive reports of the Intergovernmental Panel onClimate Change (IPCC) have come to increasingly confi-dent assessments of the dominant role of human-inducedgreenhouse gas emissions in causing increasing globalnear-surface temperatures [e.g., IPCC , 2007a]. Furthermore,analyses of changes across the climate system, includingof sub-surface ocean temperatures, of the water cycle, andof the cryosphere, show that as the observational evidenceaccumulates, there is an increasingly remote possibilitythat climate change is dominated by natural rather thananthropogenic factors [Stott et al.,  2010]. Thus, the IPCC

concluded in its fourth assessment report that “most of the observed increase in global average temperatures since

Additional supporting information may be found in the online versionof this article.

1Met Office, Exeter, UK.

Corresponding author: G. S. Jones, Met Office, FitzRoy Rd., Exeter, UK.([email protected])

©2013. Crown copyright.This article is published with the permission of the Controller of HMSOand the Queen’s Printer for Scotland.2169-897X/13/10.1002/jgrd.50239

the mid-20th century is very likely due to the observedincrease in anthropogenic greenhouse gas concentrations”[ IPCC , 2007a].

[3] However, despite the high level of confidence in theconclusion that greenhouse gases caused a substantial partof the observed warming, there remain many uncertaintiesthat have so far limited the potential to be more precise aboutsuch attribution statements. These uncertainties are associ-ated with observational uncertainties caused by remaining

 biases in data sets and gaps in global coverage [ Morice et al.,2012] and modeling uncertainties, which limit the ability todefine the expected fingerprints of change due to anthro-

 pogenic and natural factors, and which result from errorsin model formulation, deficiencies in model resolution, andinadequacies in the way external climate forcings are speci-fied. All attribution results are contingent on such remaininguncertainties and until now they have been explored in arelatively limited way. Many previous attribution studieshave been limited to a single observational data set anda single climate model [e.g.,  Tett et al.,   2002] or a rather limited ensemble of different climate models [e.g.,  Hegerl et al.,  2000;  Gillett et al.,  2002;   Huntingford et al.,  2006]while another study used simple climate models to emu-late global mean temperatures from over a dozen models[Stone et al., 2007].

4001

Page 2: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 2/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Table 1.   Model Experiment Definitions as Used in the CMIP5

Archive   [Taylor et al., 2012], With Equivalent CMIP3 Terms

[ Meehl et al., 2007a]

Experiment Name Description CMIP3 Name

 piControl Pre-industrial control simulati on picntrlhistorical 20th century (1850–2005) 20c3m

forced by anthropogenic and natu-

ral factorshistoricalNat 20th century (1850–2005) forced by only natural factors

 N.A.

historicalGHG 20th century (1850–2005) forced by only greenhouse gas factors

 N.A.

historicalExt Extension to historical experimentfrom 2005 onward

A1B

rcp45 21st century (2005–2100) forced by RCP4.5 factors

"

[4] There have also been studies that have used simpletime series methods to determine contributions of forcingsto observed global temperatures [ Lockwood , 2008; Lean and 

 Rind , 2008;  Kaufmann et al., 2011;  Foster and Rahmstorf   ,2011]. While many such studies are broadly consistent with

studies using coupled climate models in finding a domi-nant role for greenhouse gases in explaining recent warming,the advantage of using climate models rather than simplestatistical relationships with forcings is that coupledatmosphere-ocean general circulation models (AOGCMs)attempt to simulate all the most important physical processesin the climate system that lead to a model’s response to a

 particular forcing. Responses are therefore emergent fromthe model and not imposed upon it [ Hegerl and Zwiers,2011]. Also climate models produce temperature variationsover space as well as time and this study analyzes thesespace-time variations.

[5] Increasingly in the climate science community hascome a realization of the importance of testing the robust-

ness of results to observational uncertainty. One study[ Hegerl et al., 2001] deduced that one type of observationaluncertainty, grid box sampling error, had little impact onthe attribution of anthropogenic influence on temperaturetrends. A more recent study [ Jones and Stott , 2011] ana-lyzed four global temperature data sets (GISS, NCDC, JMA,and HadCRUT3—see section 3) together with one climatemodel and concluded that the choice of observational dataset had little impact on the attribution of greenhouse gaswarming. Therefore, the conclusions that greenhouse gaseswere the dominant contributor to global warming over thelast 50 years of the 20th century are robust to that observa-tional uncertainty.

[6] Since that study, there have been two importantdevelopments. First, a new analysis of global tempera-tures, HadCRUT4, has been released that includes a muchmore thorough investigation of systematic biases in sea sur-face temperature measurements than carried out previously[ Morice et al., 2012]. While an overall global warmingtrend is still seen, the detailed nature of the time series haschanged, particularly in the middle part of the 20th cen-tury [Thompson et al., 2008]. It is therefore worth exploringhow this change affects attribution results. Second, resultsfrom the CMIP5 experiment have become available, themost complete exploration of climate model uncertaintyin simulating the last 150 years ever undertaken [Taylor 

et al.,   2012]. The new multi-model ensemble of opportu-nity includes a new generation of climate models with moresophisticated treatments of forcings, including aerosols andland use changes. As part of the experimental design, it alsoincludes ensembles of simulations with both anthropogenicand natural forcings, as well as alternative ensembles which

 just include natural forcings, and ensembles which include just changes in well-mixed greenhouse gases. Therefore, the

CMIP5 ensemble provides a much more thorough and up todate exploration of modeling uncertainty than available fromthe CMIP3 ensemble [ Meehl et al., 2007a].

[7] In this paper therefore we have the opportunity toundertake the widest exploration yet of the effects of model-ing uncertainty when applied to the most up-to-date observa-tional estimates of data until 2010. In section 2 we describethe CMIP model simulations, in section   3   we discuss theobservational data sets we use, in section  4   we compareresults from the CMIP3 and CMIP5 model ensembles, andin section  5   we compare the CMIP models with observa-tions. In section 6   we carry out optimal detection analysesand describe the results using standard techniques that have

 been widely applied to near-surface temperature data and

other climate data over the last 10 years [Tett et al.,  2002;Gillett et al.,  2002;  Nagashima et al.,  2006;  Zhang et al.,2007]. In section 7 we provide conclusions.

2. Climate Model Intercomparison Project

[8] The World Climate Research Programme’s CoupledModel Intercomparison Project phase 3 (CMIP3) and

 phase 5 (CMIP5) are arguably the largest collaborative effortfor bringing together climate model data for access by cli-mate scientists across the world. The CMIP3 repository[ Meehl et al.,   2007a] was a major contributor for modeldata used in studies assessed by the IPCC’s fourth assess-ment report   [ IPCC , 2007b] while the CMIP5 repository

[Taylor et al., 2012] will be used in many studies to beassessed by the IPCC’s fifth assessment report due to be

 published in 2013. Over 20 different institutions and groupshave used over 60 different climate models to produce sim-ulations for dozens of experiments to contribute data to

 both CMIP archives. All the climate models examined inthis study are atmosphere-ocean general circulation models(AOGCMs), where the ocean and atmosphere componentsare coupled together, covering a wide range of resolutionsand sophistication of physical modeling.

[9] The different experiments represent sets of simula-tions that had different scenarios of forcing factors applied[Taylor et al., 2012]. The basic experiments examined in thisstudy are piControl, historical, historicalNat, and historical-GHG (Table   1). The piControl experiment is a long timescale simulation with no variations in external forcings, suchas greenhouse gas concentrations, set to pre-industrial con-centrations/settings. The other experiments are parallel to the

 piControl but with different forcing factor variations applied,initialized from different times from the piControl. It should

 be noted that there are differences between the names of the experiments in CMIP5 and CMIP3. For instance thehistorical experiment was called 20C3M in CMIP3 [ Meehl et al.,   2007a]. The 20C3M CMIP3 experiments com-

 prised not only simulations driven by anthropogenicand natural forcing factors but also simulations driven by

4002

Page 3: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 3/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Table 2.   Institutions That Supplied Model Data to CMIP3 and

CMIP5 Repositories Used in This Studya

Institutions key

Atmosphere and Ocean Research Institute (The University of Tokyo), Japan

a

Beijing Climate Center, China Meteorological Administration,China

 b

Bureau of Meteorology, Australia cCanadian Centre for Climate Modelling and Analysis, Canada dCentre National Recherches Météorologiques, Météo-France,France

e

Centre Europeen de Recherches et de Formation f  Avancee en Calcul Scientifique, FranceCommonwealth Scientific and Industrial Research Organisation, gMarine and Atmospheric Research, AustraliaInstitute for Numerical Mathematics, Russia hInstitute of Atmospheric Physics, Chinese Academy of Sciences iInstitut Pierre Simon Laplace, France jJapan Agency for Marine-Earth Science and Technology, Japan k Max Planck Institute for Meteorology, Germany lMet Office Hadley Centre, UK mMeteorological Institute of the University of Bonn, Germany nMeteorological Research Institute, Japan oMe teorological R esearch Insti tut e of KMA, Korea p

 NASA Goddard Institute for Space Studies, USA q National Center for Atmospheric Research, USA r  National Institute for Environmental Studies, Japan s NOAA Geophysical Fluid Dynamics Laboratory, USA t Norwegian Climate Centre, Norway uQueensland Climate Change Centre of Excellence, Australia vTechnology and National Institute for Environmental Studies,Japan

w

Tsinghua University, China x

aThe key is used in Tables 3 and 4.

anthropogenic forcing factors only  [Stone et al., 2007]. To be consistent with the CMIP5 historical experiments, weuse only those 20C3M CMIP3 simulations that were driven

 by both anthropogenic and natural forcing factors. Thereare no historicalNat and historicalGHG type simulations inthe CMIP3 archive. In Figure 9.5 of  Hegerl et al.  [2007],historicalNat simulations are presented. The authors of 

 Hegerl et al.   [2007] retrieved the historicalNat simula-tions from the institutions concerned (Dáithí Stone, personalcommunication). We have tried to collect the same data[Supplementary Materials in  Hegerl et al.,  2007] from theinstitutions and will call them CMIP3 simulations for sim-

 plicity sake. Additional CMIP5 experiments used are histor-icalExt and rcp45 (Table 1), both extensions to the historicalexperiments [Taylor et al., 2012]. The 20C3M CMIP3 exper-iments are extended if an appropriate A1B simulation isavailable [ Meehl et al.,  2007a]. These experiments extendthe historical simulations by taking their initial conditions

from the end of the equivalent historical experiment, in thecase of CMIP5 in 2005. To avoid the use of confusingly dif-ferent terminology we will generally use the CMIP5 terms(Table 1).

2.1. Models

[10] A list of the different institutions involved in pro-ducing data for CMIP3 and CMIP5 is given in Table   2.Tables   3   and   4   list the models from CMIP3 and CMIP5used in this study indicating the institutions involved anddescribing which experiments were made, how many initialcondition ensembles were produced, and the length of the

 pre-industrial control simulations that were available.

[11] For a given model experiment, up to 10 initial condi-tion ensembles were produced. These took their initial con-ditions from the model’s piControl with the period betweensamples varying greatly from model to model. We do notconsider the impact of different periods between initial con-ditions, although some studies have tried to minimize theimpact of any sampling bias by selecting initial conditionsaccording to the ocean’s state [e.g.,  Jones et al., 2011a].

[12] The CMIP3 and CMIP5 simulations, a multi-modelensemble (MME), are often called an ensemble of oppor-tunity [ Allen and Ingram, 2002]. Ideally models shouldsample uncertainties (physical modeling, forcing, and inter-nal variability) as widely as possible [Collins et al.,  2010].In practice, an ensemble of opportunity, like CMIP3 andCMIP5, would not methodically sample the full range of 

 possible uncertainties. For instance, the models are not inde- pendent   [ Jun et al., 2008;  Abramowitz and Gupta, 2008; Masson and Knutti, 2011], with many sharing common com- ponents and algorithms [ Knutti et al., 2010], which can beseen in climate responses sharing common patterns [ Mas-

 son and Knutti, 2011]. As a result, the effective number of independent models in such an ensemble of opportu-

nity is less than the actual number [ Pennell and Reich-ler , 2011]. Ensembles of models in which parameters in

 physics schemes have been perturbed (so-called perturbed physics ensembles) have been used to sample model uncer-tainty in a more methodical manner [ Murphy et al.,  2004;Stainforth et al.,  2005] although they do not sample struc-tural model uncertainty, i.e., uncertainty due to processesnot incorporated in a particular model. Similarly the CMIP5models do not systematically sample forcing uncertainties[Taylor et al., 2012].

[13] Bearing in mind the caveat that there are limitationsto the statistical interpretation of ensembles of opportunity[ Knutti, 2010;  Pennell and Reichler , 2011;  von Storch and 

 Zwiers, 2013], we use the CMIP3 and CMIP5 MME spread

to examine the confidence of any agreement between theensembles and the observations   [Taylor et al., 2012]. Inthis study, we generally treat each model as being an equalmember of the MME. While such “one model one vote”[ Knutti, 2010] methods may underestimate the uncertainty inthe model spread, it is arguably the simplest approach to use[Weigel et al., 2010]. In any event, it is difficult to justify any

 particular weighting scheme based on a model’s base climatesince a measure of skill of the CMIP5 models to representmean climate bears little relation to the skill of the models insimulating the observed trend [ Jun et al., 2008].

[14] While the CMIP3 archive was about 36 TB in size,the CMIP5 archive is estimated to be 2.5–3 PB in or evenlarger in size   [Overpeck et al., 2011;   Taylor et al., 2012].In this study, because we are interested in near-surface tem-

 peratures only, we examine only a tiny fraction of the totalCMIP archive; monthly means of one climate variable on asingle level for up to 10 initial condition ensemble membersfor each of seven experiments from 46 models. In total, morethan 66 thousand model years equating to about 65 GB of storage are analyzed in this study.

[15] We have endeavored to retrieve the latest versions of the data available in the CMIP5 archives up to 1 March 2012in order to allow time to make the analysis and write up theresults. Due to CMIP5’s limited version control, it has beena non-trivial task to keep track of data set changes. Thus,

4003

Page 4: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 4/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Table 3.   CMIP3 Models Used in This Study, Their Institutions (“Inst.”, See Table 2), the Number of 

Initial Condition Ensemble Members for Each Experiment, and the Number of Years Available From

Each piControla

Model Inst. Number of Ensemble piControl LengthMembers

histori cal hist orical Nat

cccma_cgcm3.1 d - - 1000cccma_cgcm3.1(t63) d - - 350cnrm_cm3 e - - 500csiro_mk3.0 g - - 380csiro_mk3.5 g - - 500gfdl_cm2.0 t 3 [1] - 500gfdl_cm2.1 t 5 [3] - 500giss_aom q - - 250giss_model_eh q 5 [3] - 400giss_model_er q 9 [5] - 500inmcm3_0 h 1 [1] - 330ipsl_cm4 j - - 275miroc3.2_hires a,k,s 1 [1] - -miroc3.2_medres a,k,s 10 [10] 10 [0] 500miub_echo_g n,p 4 [3] 3 [0] 340mpi_echam5 l - - 506mri_cgcm2.3.2a o 5 [0] 4 [0] 500

ncar_ccsm3 r 9 [7] 5 [0] 230ncar_pcm1 r 4 [0] 4 [0] 350ukmo_hadcm3 m 3 [0] 4 [0] 340ukmo_hadgem1 m 4 [0] - 240

a Numbers in square brackets (e.g., “[1]”) represent numbers of ensembles members that cover period ending in2010. Names of models are same as used in CMIP3. There are some minor differences between what historical andhistoricalNat CMIP3 simulations we useand those used in Hegerl et al. [2007] (Table S9.1) and in Stone et al. [2007]

Table 4.   CMIP5 Models Used in This Study, Their Institutions (“Inst.”, See Table 2), the Number of 

Initial Condition Ensemble Members for Each Experiment, and the Number of Years Available From

the piControl of the Modela

Model Inst. Number of Ensemble Members piControl Length

Historical historicalNat historicalGHG

ACCESS1-0 g,c 1 [1] - - 250CCSM4 r 6 [6] 4 [0] 6 [0] 500CNRM-CM5 e,f 10 [10] 6 [6] 6 [6] 850CSIRO-Mk3-6-0 g,v 10 [10] 5 [5] 5 [5] 500CanESM2 d 5 [5] 5 [5] 5 [5] 995FGOALS-g2 i,x 4 [2] - 1 [1] 900FGOALS-s2 i 3 [3] - - 500GFDL-CM3 t 5 [1] 3 [0] 3 [0] 500GFDL-ESM2G t 3 [1] - - 500GFDL-ESM2M t 1 [1] 1 [0] 1 [0] 500GISS-E2-H q 5 [5] 5 [5] 5 [5] 240,240GISS-E2-R q 6 [5] 5 [5] 5 [5] 300,500HadCM3 m 7 [7] - - -HadGEM2-CC m 1 [1] - - 250HadGEM2-ES m 4 [4] 4 [4] 4 [4] 1030

IPSL-CM5A-LR j 5 [4] 3 [0] 1 [0] 950IPSL-CM5A-MR j 1 [1] - - 300MIROC-ESM a,k,s 3 [1] 1 [0] 1 [0] 530MIROC-ESM-CHEM a,k,s 1 [1] 1 [0] 1 [0] 250MIROC5 a,k,s 4 [4] - - 670MPI-ESM-LR l 3 [3] - - 1000MRI-CGCM3 o 3 [3] 1 [0] 1 [0] 500

 NorESM1-M u 3 [3] 1 [1] 1 [1] 500 NorESM1-ME u 1 [0] - - - bcc-csm1-1 b 3 [3] 1 [1] 1 [1] 500inmcm4 h 1 [1] - - 500

a Numbers in square brackets (e.g., ‘[1]’) represent numbers of ensembles members that cover period ending in2010. Names of models as used in CMIP5. GISS-E2-H and GISS-E2-R models provided two separate piControlsimulations, thus the two numbers in the piControl length column for the models.

4004

Page 5: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 5/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Table 5.   Forcing Factors Included in CMIP3 Historical Exper-

iments Additional to Greenhouse Gases, Sulfate Direct Effects,

Ozone, Solar Irradiance, and Stratospheric Volcanic Aerosol

Factorsa

Model SI CA LU

gfdl_cm2.0 N Y Ygfdl_cm2.1 N Y Y

giss_model_eh Y Y Ygiss_model_er Y Y Yinmcm3.0 N N Nmiroc3.2_hires Y Y Ymiroc3.2_medres Y Y Ymiub_echo_g Y N Nmri_cgcm2.3.2a N N Nncar_ccsm3 N Y Nncar_pcm1 N N Nukmo_hadcm3 Y N Nukmo_hadgem1 Y Y Y

aSI, sulfate indirect effects (first and/or second effects); CA, carbona-ceous aerosols (blackcarbon andorganic carbon); andLU, land usechanges(Y-factor included, N-factor not included).

we cannot guarantee that all the data used in this study were

up to date as of March 2012. Nonetheless the CMIP5 datarepository is a substantial undertaking and a great success,and without such a project such studies as this would not be

 possible. Since March 2012 more models have been addedto the CMIP5 archive, we hope to be able to include thesemodels in future analyses.

2.2. External Forcing Factors

[16] Exactly what forcing factors are applied, and howthey are modeled, for a given experiment (Table  1)   differssomewhat across the models (see model’s documentationfor more details). Details of which forcings were includedin the CMIP3 historical experiments were deduced fromTable 10.1 in   Meehl et al.   [2007b]. Information about

which forcings were included in the CMIP5 simulationswere extracted from the metadata contained within the data[Taylor et al., 2012], with additional information beingobtained from the institutions. The minimum criteria for models being included in the following analyses are that themodel’s historical experiments must include at least vari-ations in long-lived well-mixed greenhouse gases, directsulfate aerosol, ozone (tropospheric and stratospheric), solar,and explosive volcanic influences. Therefore, as stated ear-lier, we do not examine those CMIP3 historical experi-ments (20C3M) that did not also include natural forcings.Only historicalNat simulations that include changes in bothtotal solar irradiance and stratospheric volcanic aerosolsare examined. The long-lived well-mixed greenhouse gasessimulated in the historical and historicalGHG experimentsinclude concentration changes in carbon dioxide, methane,and nitrous oxides (or carbon dioxide equivalent), withsome variation across the models in which CFC speciesare included. All the historical simulations include directsulfate aerosol effects but how other non-greenhouse gasanthropogenic forcing factors are applied differs across themodels. Tables   5   and   6   give a summary of which modelhistorical simulations include the indirect effects of sulfateaerosols, the effects of carbonaceous aerosols (black carbonand/or organic carbon), and land use influences. Further details of the intricacies of the forcings and how they are

implemented in particular models are available from theindividual model’s documentation.

[17] There are a few notable oddities in the waysome model experiments have been set up which makethem different from the rest of the models in thearchive. The historicalGHG experiments for the CNRM-CM5, GFDL-CM3, MIROC-ESM, MIROC-ESM-CHEM,MRI-CGCM3, and NorESM1-M models include varia-

tions in ozone concentrations in addition to the well-mixed greenhouse gas variations. The CMIP3 modelsmiub_echo_g and mri_cgcm2_3_2a and the CMIP5 modelIPSL-CM5A-LR simulate volcanic influences by perturb-ing the shortwave radiation at the top of the atmo-sphere. Whereas the ukmo-hadcm3 (run1 and run2) andukmo-hadgem1 (run1) 20C3M simulations listed in CMIP3contained anthropogenic only forced simulations, theukmo-hadcm3 and ukmo-hadgem1 20C3M simulations weanalyze here include both anthropogenic and natural forc-ings [Stott et al., 2000, 2006b].

2.3. Simulation Details

[18

] Monthly mean near-surface air temperatures (TAS)were retrieved from the CMIP3 and CMIP5 archives.The historical CMIP3 simulations had start dates varying

 between 1850 and 1900 and end dates varying between 1999and 2002. To enable a continuation of the CMIP3 histor-ical simulations to near-present day, we use any availableA1B SRES scenario simulations [ Meehl et al.,  2007a] thatare continuations of the 20C3M experiments. For CMIP5,the “historic” period is defined as starting in the mid-19thcentury and ending in 2005, so to extend the historical sim-ulations up to 2010, we append to it the historicalExt exper-iment, or if not available the rcp45 experiment (Table   1)[Taylor et al., 2012]. There are some minor differences

 between the different representative concentration path-

ways (RCPs) anthropogenic emissions and concentrationsTable 6.  Same as Table 5 but for CMIP5 Models

Model SI CA LU

ACCESS1-0 Y Y NCCSM4 N Y YCNRM-CM5 Y Y NCSIRO-Mk3-6-0 Y Y NCanESM2 Y Y YFGOALS-g2 Y Y NFGOALS-s2 N Y NGFDL-CM3 Y Y YGFDL-ESM2G N Y YGFDL-ESM2M N Y YGISS-E2-H Y Y YGISS-E2-R Y Y YHadCM3 Y N NHadGEM2-CC Y Y YHadGEM2-ES Y Y YIPSL-CM5A-LR Y Y YIPSL-CM5A-MR Y Y YMIROC-ESM Y Y YMIROC-ESM-CHEM Y Y YMIROC5 Y Y YMPI-ESM-LR N N YMRI-CGCM3 Y Y Y

 NorESM1-M Y Y N NorESM1-ME Y Y N bcc-csm1-1 N Y Ninmcm4 N N N

4005

Page 6: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 6/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

during the first 10 years of the 21st century [van Vuurenet al.,   2011], but these are very small compared to thedifferences over the whole century. Which RCP experi-ment is chosen to extend the historical experiment to 2010is unlikely to be important. There are bigger differences

 between the forcing factors in the CMIP3 SRES and theCMIP5 RCP experiments over the first few decades of the21st century, but differences in the climatic responses are

also relatively small [ Knutti and Sedlᡠcek , 2013]. How thesolar and volcanic forcing factors are applied during this

 period will also differ across the models, so there will beadditional forcing uncertainties due to these choices. Whilethe official CMIP5 guidance was for the historicalNat andhistoricalGHG simulations to cover the mid-19th centuryto 2005 period  [Taylor et al., 2012], a number of insti-tutions supplied historicalNat and historicalGHG data toCMIP5 to cover the period up to 2010 (Table  4). None of these CMIP3 historicalNat simulations have data beyond theyear 2000.

3. Observed Near-Surface Temperatures[19] Measurements of near-surface temperatures have

 been used to create the longest global scale diagnosticsof observed climate going back to the mid-19th century.In this study we analyze HadCRUT4 produced by theMet Office and Climate Research Unit, University of EastAnglia [ Morice et al., 2012], GISS produced by NASAGoddard Institute for Space Studies [ Hansen et al.,  2006],

 NCDC produced by NOAA’s National Climatic Data Center [Smith et al., 2008], and JMA produced by the Japan Mete-orological Agency [http://ds.data.jma.go.jp/tcc/tcc/products/gwp/gwp.html accessed 8/8/2011;   Ishii et al.,   2005]. Thedata sets differ in how raw data have been quality controlled,and in the homogenization adjustments and bias corrections

made. They also differ in their coverage. While HadCRUT4and JMA have areas of missing data where no observationsare available (i.e., no infilling outside of grid boxes withdata), GISS extrapolate over data-sparse regions using datawithin a radius of 1200 km, and NCDC use large area aver-ages from low-frequency components of the data and spatialcovariance patterns for the high frequency components toextrapolate data. The data sets incorporate land air temper-atures and near-surface temperatures over the ocean (e.g.,sea surface temperatures) into a global temperature record.The main data set we analyze is HadCRUT4, an update toHadCRUT3 [ Brohan et al., 2006] with additional data, biascorrections, and a sophisticated error model. The data set is

 provided as an ensemble that samples a number of uncertain-ties and bias corrections that are correlated in time and space,as well as statistical descriptions of the other uncertainties[ Kennedy et al., 2011; Morice et al., 2012]. For the purposesof this study, we use the median field of the HadCRUT4 biasrealizations (see Morice et al. [2012] for more details) as the

 best estimate of the data set. We plan to do a thorough inves-tigation using HadCRUT4’s error model in a future study.The HadCRUT4, NCDC, and JMA data sets have a grid-ded spatial resolution of  5°   5° and GISS a resolution of 1° 1°. The periods covered by the data sets are 1850–2010for HadCRUT4, 1880–2010 for GISS and NCDC, and1891–2010 for JMA.

4. Comparison of CMIP3 With CMIP5 Models

[20] We first compare the CMIP3 and CMIP5 multi-model ensemble (MME). Annual mean spatial fields arecreated by averaging up the monthly gridded data (January– December) and when comparing spatial patterns, model dataare projected onto a 5° 5° spatial grid.

4.1. Differences in CMIP3 and CMIP5 Variability

[21] For consistency when comparing the CMIP3 andCMIP5 piControl experiments, we limit the length of sim-ulations being examined to that of the shortest piControlsimulation, 230 years (Tables 3  and 4). Often a model sim-ulation with no changes in external forcing (piControl) willhave a drift in the climate diagnostics due to various fluximbalances in the model [Gupta et al.,  2012]. Some stud-ies attempt to account for possible model climate drifts,for instance Figure 9.5 in   Hegerl et al.   [2007] did notinclude transient simulations of the 20th century if thelong-term trend of the piControl was greater in magnitudethan 0.2 K/century (Appendix 9.C in  Hegerl et al.  [2007]).Another technique is to remove the trend, from the transient

simulations, deduced from a parallel section of piControl[e.g.,  Knutson et al.,  2006]. However whether one shouldalways remove the piControl trend, and how to do it in

 practice, is not a trivial issue [Taylor et al.,   2012;  Guptaet al., 2012]. Only two of the CMIP model simulations ana-lyzed in this paper, giss-model-eh and csiro-mk3-0 (bothfrom CMIP3), have trends with magnitude greater than0.2 K/century (Figure S1 in the supporting information). Wechoose not to remove the trend from the piControl from par-allel simulations of the same model in this study due to theimpact it would have on long-term variability, i.e., the possi-

 bility that part of the trend in the piControl may be long-terminternal variability that may or may not happen in a par-allel experiment when additional forcing has been applied.

The overall range of CMIP5 piControl trends has a smaller spread than the CMIP3 piControl trends, and all are lower in magnitude than about 0.1 K/century. While the varianceof the annual mean TAS of the first 230 years of piControl(Figure S1) has a lower spread in CMIP5 than in CMIP3, thedifferences are not statistically significant. The models withthe very highest variability, which are all in CMIP3, havelower variability when the drift in the piControl is removed(Figure S1).

[22] The interannual variability across latitudes in theCMIP3 and CMIP5 models is shown in Figure 1. The spreadof the models is greatest around the tropics and high latitudeswith the range across the models being quite large in places.For CMIP3, the tropical variability is very weak in few mod-

els and very strong in others; for CMIP5, there is closer agreement in the tropics. The reduction in the range acrossthe models of tropical variability in CMIP5 is probably dueto a reduction in the range of El Ni Qno-Southern Oscillationvariability across the models relative to CMIP3 [Guilyardiet al., 2012; Kim and Yu, 2012]. Variability across the mod-els over the high latitudes is similar between CMIP3 andCMIP5 apart from one model forming an outlier in CMIP5.

4.2. Historic Transient Experiments

[23] Global annual mean TAS for simulations for thehistoric period from 1850 to 2010 for CMIP3 (Table 3)  and

4006

Page 7: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 7/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

CMIP3

90S 50S 0 50N 90N

Latitude

0.0

0.5

1.0

1.5

   Z  o  n  a   l  s   t .   d  e  v . ,

   K

cccma_cgcm3_1cccma_cgcm3_1_t63cnrm_cm3csiro_mk3_0csiro_mk3_5gfdl_cm2_0gfdl_cm2_1giss_aom

giss_model_e_hgiss_model_e_rinmcm3_0ipsl_cm4miroc3_2_medresmiub_echo_gmpi_echam5mri_cgcm2_3_2a

ncar_ccsm3_0ncar_pcm1ukmo_hadcm3ukmo_hadgem1

CMIP5

90S 50S 0 50N 90NLatitude

0.0

0.5

1.0

1.5

   Z  o  n  a   l  s   t .   d  e  v . ,   K

ACCESS1-0CCSM4CNRM-CM5CSIRO-Mk3-6-0CanESM2FGOALS-g2FGOALS-s2GFDL-CM3

GFDL-ESM2GGFDL-ESM2MGISS-E2-HGISS-E2-RHadGEM2-CCHadGEM2-ESIPSL-CM5A-LRIPSL-CM5A-MR

MIROC-ESM-CHEMMIROC-ESMMIROC5MPI-ESM-LRMRI-CGCM3NorESM1-Mbcc-csm1-1inmcm4

Figure 1.   Standard deviation of zonal annual mean TASfrom piControls for CMIP3 (top) and CMIP5 (bottom). The

first 230 years of each model’s piControl used to calculateannual latitudinal zonal means and then the standard devia-tion for each latitude was calculated.

CMIP5 models (Table  4) are shown in Figure  2.  Includedin the figure are the individual historical, historicalNat, andhistoricalGHG simulations for CMIP3/5 together with theCMIP3 and CMIP5 ensemble averages. While for figure 9.5in Hegerl et al.  [2007] the simple average of all the histor-ical and historicalNat simulations was shown together witheach individual simulation, here we take a slightly differ-ent approach. As for some models, there are as many as 10initial condition ensemble members in the CMIP5 archive,and for other models, there is as few as one single simula-tion for a given experiment, a simple average would givemost weight to the model with the most ensemble mem-

 bers. Therefore to avoid this bias toward models with mostensemble members, here we calculate the weighted aver-age that gives equal weight to each model [Santer et al.,2007] regardless of how many ensemble members a modelhas (see the supporting information for details). A simula-tion, one of an ensemble for a given model, will have aweight which is the inverse of the number of ensemble mem-

 bers for that model multiplied by the inverse of the number of different models. The resultant weighted average is theequivalent of taking the average of all the models’ ensemble

averages. To estimate the statistical spread of the MME, wecreate a cumulative probability distribution, after rankingthe simulations, with the probability assigned to each simu-lation equal to its weight (as described above). The cumu-lative probability distribution can then be calculated andsampled at whatever percentiles are of interest to obtainMME ranges. A more basic analysis could just look at theaverage and the percentile range of the ranked simulations,

which would be equivalent to simply setting the weightsto be the inverse of the total number of simulations. How-ever by given equal weight to each model, rather than toeach simulation, the statistical properties of the MME willnot be dominated by those models with many ensemblemembers. The weighted ensemble means for CMIP3 andCMIP5, separately, are shown as thick lines in Figure  2.

[24] All the CMIP3 and CMIP5 historical experiments(see Figures S2 and S3 for the responses for each modelshown separately) show warming over the historic period.The spread in the increase in TAS across the models is larger than the spread across the initial condition ensemble for any of the individual models. The historicalNat simulationsshow little overall warming due to the combined influences

of solar and volcanic activity. In most models, the historical-GHG experiment warms more than the historical experimentconsistent with the other anthropogenic factors—sulfate andcarbonaceous aerosols, ozone, land use—and natural fac-tors having an overall net cooling influence. However, theCCSM4, GFDL-ESM2M, and bcc-csm1 models have histor-icalGHG simulations that warm by 2000 by similar amountsto the equivalent historical simulations (Figure S3) whichimplies that the non greenhouse gas forcing factors havelittle or no net warming/cooling influence over the whole

 period. These models were the only CMIP5 models that pro-vided historical and historicalGHG experiments that did notinclude the indirect effects of sulfate aerosols effects in thehistorical simulations, i.e., the effects of aerosols to make

clouds brighter (first indirect effect) or longer lasting (secondindirect effect).

[25] The gradual warming seen in the historical simula-tions is punctuated with short periods of cooling—of varyingdegrees—from the major volcanic eruptions [ Driscoll et al.,2012]. Far less obvious is any response to solar irradiancechanges with little evidence of an “11 year” cycle in theweighted mean of the historical or historicalNat simula-tions, supporting previous examinations of the response of climate models to solar forcing that suggest they have aweak global mean response to the “11 year” solar cycle[ Jones et al., 2012].

[26] With the increased number of models available inCMIP5 compared to CMIP3 there is a wider variety of 

responses in CMIP5 than CMIP3. The spread in TAS in thefirst decade of the 21st century for the historical CMIP5simulations is somewhat wider than the CMIP3 simulations(Figure  2). The spread is almost identical up to 1960 butthen widens for CMIP5 relative to CMIP3. A non-parametricKolmogorov-Smirnov test [ Press et al., 1992, p. 617] doesnot rule out that the CMIP3 and CMIP5 distributions of 2001–2010 mean temperatures are drawn from the same

 population distribution, i.e., the distributions are not signif-icantly different, but the test is not robust to differencesin the tails of distributions which is what seems to be themain difference between CMIP3 and CMIP5. One of the

4007

Page 8: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 8/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

1860 1880 1900 1920 1940 1960 1980 2000

Year

-0.5

0.0

0.5

1.0

1.5

-0.5

0.0

0.5

1.0

1.5

-0.5

0.0

0.5

1.0

1.5

   T  e  m  p  e

  r  a   t  u  r  e  a  n  o  m  a   l  y   (   K   )

historical

1860 1880 1900 1920 1940 1960 1980 2000

Year

   T  e  m  p  e  r  a   t  u

  r  e  a  n  o  m  a   l  y   (   K   )

historicalNat   CMIP3CMIP5

1860 1880 1900 1920 1940 1960 1980 2000

Year

   T  e  m  p  e  r  a   t  u  r  e  a

  n  o  m  a   l  y   (   K   )

historicalGHG

Figure 2.   Global annual mean TAS for both CMIP3 andCMIP5 for historical (top), historicalNat (middle), andhistoricalGHG (bottom) individual ensemble members (thinlight lines). The weighted ensemble average for both CMIP3(blue thick line) and CMIP5 (red thick line) are esti-mated by finding the average of the model ensemble means(supporting information). TAS annual means shown withrespect to 1880–1919.

striking differences between the CMIP3 and CMIP5 MMEis the stronger cooling some of the models have for thehistorical experiments around the 1950s to 1980s (FiguresS2 and S3). These differences indicate that some changes inmodeling are potentially increasing the variety of tempera-ture climate responses in the CMIP5 historical simulations.

This is despite the similarity of the range of the transient cli-mate response (TCR) seen in CMIP3 and CMIP5 [ Andrewset al., 2012]. Differences in the way forcing factors areapplied in the CMIP models and the resulting uncertaintyin the radiative forcings [ Forster and Taylor , 2006;  Forster et al.,  2013] may also be contributing to the wider spreadin TAS responses in CMIP5. For instance a higher propor-tion of the CMIP5 models include land use changes and a

wider range of aerosol influences than in CMIP3 (Tables  5and 6). Also the 2007 IPCC assessment [ Forster et al., 2007]estimated that historic total solar irradiance (TSI) increaseswere half that estimated by the previous report leading toa recommendation for CMIP5 to use a TSI reconstructionwhich had a smaller increase over the first half of the 20thcentury than those used by CMIP3. On the other hand, someof the CMIP5 models have very similar variations in histori-cal TAS to that of their earlier generation model counterpartsin CMIP3, for example giss_model_e_r and GISS-E2-R or ncar_ccsm3_0 and CCSM4 models (Figures S2 and S3).

[27] The multi-model weighted average of the CMIP3 his-torical ensemble is very similar to the CMIP5 historicalweighted average (Figure   2), which given the wide range

of responses in the individual models in CMIP5 comparedto CMIP3, and the differences in the models and forcingfactors applied, is perhaps surprising. The weighted meansfor CMIP3 and CMIP5 historicalNAT simulations are alsovery similar with the mean cooling following the major volcanic eruptions being very similar even though there arelarge differences between individual models.

5. Comparison of CMIP Models WithObserved Temperatures

[28] In this section we compare the CMIP modelswith observed near surface temperatures. When compar-ing models with observations it is important to treat the

model data in as similar a way as possible as the observeddata [Santer et al.,   1995]. All data are projected onto a5° 5° spatial grid and then monthly anomalies, relative to1961–1990, are masked by HadCRUT4’s spatial coverage.Annual means (January to December) are calculated for agridpoint if at least 2 months of data are available (seethe supporting information for more details of the pre-

 processing). Imposing HadCRUT4 spatial coverage on alldata means that the other observational data sets, in partic-ular GISS and NCDC, have reduced coverage [ Jones and Stott , 2011] but this will allow a consistent comparison

 between the observations and the models.

5.1. Global Annual Mean Temperatures

[29] Global annual mean near-surface temperatures for each model and the four observational data sets are shown inFigure 3 for CMIP5 MME (See Figure S7 in the supportinginformation for the equivalent figure for the CMIP3 mod-els). Global mean anomalies were calculated by removingthe average of the global annual means over the 1880–1919

 period (see the supporting information). As has been notedelsewhere, [ Jones and Stott , 2011;  Morice et al., 2012], allobservational data sets track each other relatively closelyover the 1850–2010 period. Each model’s historical sim-ulations warm up more than, and tracks more closely theobservations, than the equivalent historicalNat simulations.

4008

Page 9: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 9/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

   T  e  m  p  e  r  a   t  u  r  e  a  n  o  m  a   l  y ,   K

-0.5

0.0

0.5

1.0

1.5

1880 1900 1920 1940 1960 1980 2000

ACCESS1-0

1880 1900 1920 1940 1960 1980 2000

CCSM4

1880 1900 1920 1940 1960 1980 2000

CNRM-CM5

-0.5

0.0

0.5

1.0

1.5

1880 1900 1920 1940 1960 1980 2000

CSIRO-Mk3-6-0

-0.5

0.0

0.5

1.0

1.5   CanESM2 FGOALS-g2 FGOALS-s2

-0.5

0.0

0.5

1.0

1.5GFDL-CM3

-0.5

0.0

0.5

1.0

1.5   GFDL-ESM2G GFDL-ESM2M GISS-E2-H

-0.5

0.0

0.5

1.0

1.5GISS-E2-R

-0.5

0.0

0.5

1.0

1.5   HadCM3 HadGEM2-CC HadGEM2-ES

-0.5

0.0

0.5

1.0

1.5IPSL-CM5A-LR

-0.5

0.0

0.5

1.0

1.5   IPSL-CM5A-MR MIROC-ESM MIROC-ESM-CHEM

-0.5

0.0

0.5

1.0

1.5MIROC5

-0.5

0.0

0.5

1.0

1.5   MPI-ESM-LR MRI-CGCM3

1880 1900 1920 1940 1960 1980 2000

NorESM1-M

-0.5

0.0

0.5

1.0

1.5

1880 1900 1920 1940 1960 1980 2000

NorESM1-ME

-0.5

0.0

0.5

1.0

1.5

1880 1900 1920 1940 1960 1980 2000

bcc-csm1-1

1880 1900 1920 1940 1960 1980 2000

inmcm4 observationshistoricalhistoricalNathistoricalGHG

Figure 3.   Global annual mean TAS variations, 1850–2010, for the CMIP5 historical, historicalNat andhistoricalGHG experiments and the four observational data sets. TAS annual means shown with respectto 1880–1919. Model and observed data have same spatial coverage as HadCRUT4.

A comparison with simulations that have complete spa-tial coverage (supporting information) shows that the vastmajority,  >95% of the historical simulations, warm up lessover the historic period when masked by the observational

coverage. This is predominantly caused by the high north-ern latitude warming from greenhouse gas warming in themodels being masked by the lack of coverage in HadCRUT4in that region. The largest reduction in the linear trend

4009

Page 10: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 10/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

1860 1880 1900 1920 1940 1960 1980 2000

Year

-0.5

0.0

0.5

1.0

1.5

   T  e  m  p  e  r  a   t  u  r  e  a  n  o

  m  a   l  y   (   K   )

(a)CMIP3CMIP5

observations

1860 1880 1900 1920 1940 1960 1980 2000

Year

-0.5

0.0

0.5

1.0

1.5

   T  e  m  p  e  r  a   t  u  r  e  a  n  o  m  a   l  y   (   K   )

(b)

1860 1880 1900 1920 1940 1960 1980 2000-0.2

0.0

0.2

0.4

0.6

0.8

1.0HadCRUT4GISSNCDCJMA

1860 1880 1900 1920 1940 1960 1980 2000

Year

-0.5

0.0

0.5

1.0

1.5

   T  e  m  p  e  r  a   t  u  r  e  a  n  o  m  a   l  y   (   K   )

(c)

Figure 4.   Global annual mean TAS for CMIP3 (thin blue

lines) and CMIP5 (thin red lines) for (a) historical, (b)historicalNat, and (c) historicalGHG ensemble members,compared to the four observational data sets (black lines)— also shown individually in the insert of Figure   4a. Theweighted ensemble average for CMIP3 (blue thick line)and CMIP5 (red thick line) are estimated by given equalweight to each model’s ensemble mean (supporting infor-mation). All model and observed data have same spatialcoverage as HadCRUT4. TAS anomalies with respect to1880–1919 period.

 between 1901 and 2010, in an individual simulation due toapplying the observational coverage mask, is  – 0.18 K per 110 years demonstrating the importance of comparing likewith like. The historicalGHG simulations consistently warmmore than the observations across all the CMIP5 models(Figure 3).

[30] Many of the model’s historical simulations (Figur e 3)capture the general temporal shape in the observed TAS,

an increase from the 1900s to the 1940s, then flattening or even cooling to the 1970s, then increase to the present day.The spread in a given model’s warming across the ensem-

 ble is relatively small over the whole period compared tothe spread in warming across the models. This indicates thatover the 100 year timescale differences in the forcing factorsapplied to the models and their responses are more impor-tant than internal variability, although on shorter timescalesthe opposite may be the case [Smith et al.,  2007;  Hawkinsand Sutton, 2009].

[31] Global annual mean TAS for the MME for bothCMIP3 and CMIP5 for the historical, historicalNat andhistoricalGHG simulations are given in Figure   4   together with the CMIP3 and CMIP5 model weighted averages and

the four observational data sets. As suggested in previousanalyses [e.g.,   Stott et al.,   2000] and as documented inFigure 9.5 in Hegerl et al. [2007], the historical simulationsdescribe the variations of the observed annual mean near surface temperatures fairly well (Figure S8 in the support-ing information is an alternative version of Figure 4 showingthe overall spread of the CMIP3 and CMIP5 simulationscombined). Linear trends for the observed data sets, and themodel experiments are given in Table 7 for different periods.While all of the observational data sets show similar timehistories (Figure 4), there is a small spread in linear trendswith the GISS data set having a slightly smaller trend thanthe other observations [ Jones and Stott , 2011; Morice et al.,2012] over 1901–2010. The observational data sets show a

linear warming of between 0.64 and 0.75 K per 100 yearsover 1901–2010 while the spread of the central  95%  of thehistorical simulations trends is 0.33–1.11 K per 100 years.It should be noted that linear trends in these circumstancesare just summary statistics and do not imply linear climatechanges are expected or observed.

[32] While the observed trend over the first half of the20th century is higher than the historical MME mean, it iswithin its ensemble spread, but not the historicalNat ensem-

 ble spread (Table   7). The spread of the historical MMEtrends over the 1951–2010 and 1979–2010 periods encom-

 pass the observed trends while the historicalNat MME donot. One must be careful not to draw too many conclusionsabout the significance of differences just between the model

mean and the observations. The process of averaging manymodel’s simulations reduces substantially any internal vari-ability in the ensemble mean of the historical experiments,leaving an average model response that is predominantlya forced signal with almost no internal variability eventhough the observations still contain internal variability.So, leaving aside issues of observational uncertainty, thedifference between multi model mean and observations ismainly due to mean model error and observational internalvariability [Weigel et al., 2010]. This means it is important toaccount for the MME spread, and measures of internal vari-ability, when comparing models with observations [Santer 

4010

Page 11: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 11/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Table 7.   Global Mean Linear Trends for the Observed Data Sets and Both CMIP3 and CMIP5 MME (K per 100 Years)a

1901–2010 1901–1950 1951–2010 1979–2010 2001–2010

HadCRUT4 0.72 1.02 1.09 1.78 0.35GISS 0.64 0.81 1.05 1.55 0.10

 NCDC 0.75 0.95 1.14 1.60 0.17JMA 0.74 0.90 1.06 1.27 0.16historical 0.65 (0.33,1.11) 0.65 (0.24,1.11) 1.23 (0.63,1.93) 2.11 (0.91,3.23) 1.87 (–0.47,4.92)hi storic alNat 0.00 (–0.13,0.13) 0.43 (0.08,0.78) –0.14 (–0.58,0.15) 0.16 (–0.79,1.07) 0.07 (–2.49,2.43)

historicalGHG 1.09 (0.81,1.59) 0.37 (0.05,0.72) 1.93 (1.47,2.74) 2.07 (1.26,3.13) 1.93 (0.41,4.14)

aThe average of the MME trends together with the 2.5–97.5 % range (in brackets) are given for the CMIP experiments (given equal weight to eachmodel). All observations and model simulations have same temporal-spatial coverage as HadCRUT4. Trends calculated for a period when less than 10years have missing data, apart from the 2001–2010 when trend is calculated only when all 10 years are available.

et al.,  2008], and not just contrast the MME mean with anobservational data set [e.g., Wild , 2012].

[33] A comparison of the variability of the global meanof the models with the observations on different timescalesis shown in Figure 5 as a power spectral density (PSD) plot(see also Figures S10 and S11 for the individual modelsPSDs). The method used is described elsewhere [ Mitchell et al.,  2001;  Allen et al.,  2006;  Stone et al.,  2007;  Hegerl 

et al., 2007]. The spectra contain variance from internal vari-ability and the response to external forcings, as the datahas not been de-trended. The CMIP3 and CMIP5 histori-cal MME encompass the variability of all four observationaldata sets on all the timescales examined. The historicalNatMME starts to diverge from the observations after peri-odicities of 20 or so years and for periodicities of about35 years no historicalNat simulations have variability aslarge as observed. Together with Figure 4 this is strong evi-dence that observed temperature variations are detectableover internal and externally forced natural variabilityon the longer timescales, whereas on timescales shorter than 30 years changes are indistinguishable [ Hegerl and 

 Zwiers, 2011].

[34] Figur e 6 shows a summary of three statistical indica-tors for the CMIP simulations compared with HadCRUT4,on a Taylor diagram   [Taylor , 2001]. The Taylor diagramenables the simultaneous representation of the standard devi-ation of each simulation and HadCRUT4’s global annualmean TAS, the root mean square error (RMSE) and cor-relation of the simulations with HadCRUT4. The period1901–2005 is used, to increase the number of simulationsthat can be examined, with global annual means having their whole period mean removed. Perhaps unsurprisingly the his-toricalNat (green points in Figure   6) simulations have thelowest standard deviation and the lowest correlation withHadCRUT4. None of the historicalNat simulations have aRMSE lower than 0.2 K. All the historicalGHG simulations

have correlations with HadCRUT4 around 0.8 and RMSEsup to 0.4 K. The historical simulations have some of thesimulations with the lowest RMSE with correlations withHadCRUT4 varying from just above 0.4 up to just below0.9. While the historicalNat simulations are clustered awayfrom the other simulations, there is some overlap betweenthe clusters of historical and historicalGHG simulations.

5.2. Continental-Scale Mean Temperatures

[35] Climate changes from internal variability and exter-nal forcings would not be expected to be uniform acrossthe globe [Santer et al.,   1995]. We examine annual mean

temperatures over sea, land and six continental land areas.We group pre-defined regions used by the IPCC in a reporton climate extremes   [SREX , 2012] into six continentalregions (Figure 7 insert). These SREX areas (Figure 3.1 andTable 3.A-1 in  SREX  [2012]) do not always align perfectlywith common geographic or political definitions of the conti-nents, but for convenience we group and call the areas NorthAmerica, South America, Africa, Europe, Asia, Australasia

and Antarctica (insert in Figure   7). All data, models andHadCRUT4, are processed in the same way to constructthe global and regional land and global ocean temperatures.We use the proportion of land area in each of HadCRUT4’sgrid boxes to deduce which grid boxes, in HadCRUT4 andthe models, to use. Only those grid boxes where there is25% or more land in HadCRUT4 are used to calculate landtemperatures and only those grid boxes with   0%   land areused to calculate ocean temperatures (see the supportinginformation for further details).

[36] The observed (HadCRUT4) data coverage acrossthe regions changes substantially over the period beingexamined (Figure S6). Europe has the least amount of 

10 100

Period, years

0.0001

0.0010

0.0100

0.1000

1.0000

10.0000

   P  o  w  e  r   S  p  e  c   t  r  a   l   D  e  n  s   i   t  y ,   K   2  y  r  -   1

2

HadCRUT4GISSNCDCJMA

historical 5-95%iles

historicalNat 5-95%iles

Figure 5.   Power spectral density for 1901–2010 period for  both CMIP3 and CMIP5 simulations and the observations.Analysis on annual mean data as shown in Figure 4. Tukey-hanning window of 97 years in length applied to all data.The central  90%  ranges of the historical and historicalNatmulti-model ensemble are shown separately as shaded areas.The 5–95% ranges are calculated given equal weight to eachmodel (see section 4.2). The HadCRUT4, GISS, NCDC, andJMA global mean near surface temperature observations areas shown in the key.

4011

Page 12: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 12/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-1

-0.99

-0.95

-0.9

-0.8

-0.6

-0.4

-0.2   0 0.2

0.4

0.6

0.8

0.9

0.95

0.99

1

C  o  r  r  e  l   a  t  i   o  n  

0.1

0.2

0.30.40.50.60.70.8

0.6 0.5 0.4 0.3 0.2 0.1 0.1 0.2 0.3 0.4 0.5 0.60.0St. Dev., K

0.0

0.1

0.2

0.3

0.4

0.5

0.6

historical

historicalNathistoricalGHG

HadCRUT4

Figure 6.   Taylor diagram [Taylor , 2001] for global annual mean TAS for the CMIP5 models comparedwith HadCRUT4 for 1901–2005 period. Historical, historicalNat, and historicalGHG model data havesame spatial coverage as HadCRUT4. Shown are lines of equal standard deviation (dashed), correlation(dash-dot), and centered root mean square error (RMSE dotted). Each colored point represents summarystatistic for a single simulation, so some models have more simulations in the diagram than others. Asonly one variable is being examined the data in the diagram has not been normalized.

variation, increasing from   60%   in 1860 to near   100%   bythe 1950s. The other continental regions have much lower observational coverage increasing from less than  10% in the1860s to around 80% by the 1960s. Antarctica has very littlespatial coverage even during recent times and zero before

the 1950s, so the Antarctica temperatures are shown withanomalies with respect to the 1951–1980 period.

[37] Figure 7  shows TAS for HadCRUT4 and the modelweighted CMIP MME average and  5 – 95%  (equal weightgiven to each model; supporting information) range for the

   T  e  m  p  e  r  a   t  u  r  e  a  n  o  m  a   l  y ,

   K

-1.0-0.5

0.0

0.5

1.0

1.5

2.01880 1920 1960 2000

Global

1880 1920 1960 2000

Global Land

-1.0-0.5

0.0

0.5

1.0

1.5

2.01880 1920 1960 2000

Global ocean

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0North America South America

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

Europe

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0Africa

1880 1920 1960 2000

Asia

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

1880 1920 1960 2000

Australasia

-1.0

-0.5

0.0

0.5

1.0

1.52.0

1880 1920 1960 2000

Antarcticahistorical 5-95%ileshistoricalNat 5-95%ilesHadCRUT4

Figure 7.   Global, land, ocean, and continental annual mean temperatures for CMIP3 and CMIP5historical (red) and historicalNat (blue) MME and HadCRUT4 (black). Weighted model means shown asthick dark lines and 5–95% ranges shown as shaded areas. Continental regions as defined in insert. Tem-

 peratures shown with respect to 1880–1919 period apart for Antarctica which is shown with respect to1951–1980 period mean.

4012

Page 13: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 13/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-90 0 90 180

-90

-45

0

45

90-90 0 90 180-90 0 90 180-180 -90 0 90 180

-90

-45

0

45

90

20%-90

-45

045

90

15%32%12%-90

-45

045

90

47%-90

-45

0

45

90

70%44%90%-90

-45

0

45

90

24%

-90 0 90 180-90

-45

0

45

90

45%

-90 0 90 180

48%

-90 0 90 180

51%

-180 -90 0 90 180-90

-45

0

45

90

-2 -1 0 1 2

Trend in K per period

1901-2010 1901-1950 1951-2010 1979-2010

   H  a   d   C   R   U   T   4

   h   i  s   t  o  r   i  c  a

   l

   h   i  s   t  o  r   i  c  a   l   N  a   t

   h   i  s   t  o  r   i  c  a   l   G   H   G

Figure 8.   Spatial linear trends for four periods, 1901–2010, 1901–1950, 1951–2010, and 1979–2010.Trend shown as temperature change over period, with each period having different lengths. HadCRUT4trends (first row), weighted ensemble averages for CMIP3 and CMIP5 historical (second row),historicalNat (third row), and historicalGHG (fourth row) simulations. Black boxes placed around grid

 points where the central   90%   range of the simulations (with equal weight given to each model— supporting information) do not encompass the observed trend. Numbers in bottom left-hand corner of 

 panels give the percentage number of non-missing data grid points that are highlighted with a black box.

historical and historicalNat experiments for the globe, land

only, ocean only, and the seven continents. Historical simu-lations generally capture observed variations over land andocean, both showing more warming over land than sea.HistoricalNat simulations show little overall warming trendand diverge from the historical experiment by the 1950s.While the continental regions have more interannual vari-ability than globally, they all show a warming over the whole

 period in HadCRUT4. Antarctica, which only covers the period after 1950, shows some warming but also a great dealof interannual variability; nevertheless, the anthropogeniccomponent of warming has been detected in the region in aspatial-temporal analysis [Gillett et al., 2008].

5.3. Spatial Temperature Trends

[38] As indicated by looking at the continental scalemean TAS variations, observed and modeled tempera-tures are not warming uniformly across the globe. Thiscan further be investigated by examining spatial linear trends. Figure   8   shows spatial trends for four different

 periods for HadCRUT4 and the historical, historicalNat,and historicalGHG CMIP simulations. The periods exam-ined, 1901–2010, 1901–1950, 1951–2010, and 1979–2010,capture the most important periods in temperature changessince the start of the 20th century. For each simulationat each grid point, linear trends are calculated where nomore than five consecutive years have missing data (see the

supporting information). Linear trends are calculated for 

each simulation and then the weighted model average iscalculated for each model experiment. In Figure   8,   grid

 boxes are outlined where the central 90% range (where equalweight is given to each model rather than each individualsimulation—section 4.2  and the supporting information) of the CMIP MME does not encompass HadCRUT4’s trend[ Karoly and Wu,   2005;   Knutson et al.,   2006]. This indi-cates where 95%  or more of the simulated trends are larger than observed (or where   95%   or greater have trends lessthan observed) and thus where there is some inconsistency

 between the CMIP MME and the observations. This can belooked at as a simple detection and attribution analysis on thegrid point level [ Karoly and Wu, 2005], where an observedtrend is inconsistent with the historicalNat MME then it can

 be said to be detected, and where it is also consistent with thehistorical MME it is attributed. However, this is not a strongattribution statement [ Hegerl and Zwiers, 2011], especiallyconsidering the limitations of an ensemble of opportunity[ Knutti, 2010;  Pennell and Reichler , 2011;  von Storch and 

 Zwiers, 2013].[39] Over the 1901–2010 period (Figure 8  first column),

HadCRUT4 warms almost everywhere, except for a smallregion south of Greenland, with slightly higher warmingat the higher latitudes. The historical MME average has asimilar pattern of warming with not quite as much warm-ing at the higher latitudes, with only a small number of 

4013

Page 14: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 14/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-2

-1

0

1

2

3

4

5

   K  p  e  r   1   1   0  y  e  a  r  s

90S 60S 30S 0 30N 60N 90N

1901-2010

-2

-1

0

1

2

3

4

5

   K  p  e  r   5   0  y  e  a  r  s

1901-1950HadCRUT4GISSNCDCJMA

-2

-1

0

1

2

3

4

5

   K  p  e  r   6   0  y  e  a  r  s

1951-2010

-2

-1

0

1

2

34

5

   K  p  e  r   3   2  y  e  a  r  s

1979-2010

90S 60S 30S 0 30N 60N 90N

Latitude

historical 5-95% rangehistoricalNat 5-95% range

Figure 9.   Latitude zonal TAS trends for observationsand CMIP3 and CMIP5 simulations for four periods,1901–2010, 1901–1950, 1951–2010, and 1979–2010. For each simulation, the average of the trends for each latitudeis calculated, then the 5–95% (equal model weighted) rangeof the available ensemble members for each latitude is esti-mated. All observed data sets and simulations have samespatial coverage as HadCRUT4. Where the other observeddata sets have less than 50%  of the coverage of HadCRUT4for a given latitude, the zonal mean is set to a missingdata value.

grid boxes showing an inconsistency between HadCRUT4and the central  90%  range of the MME. The historicalNatmean shows little change, generally a slight cooling, withonly 12%  of grid points showing trends that are consistentwith HadCRUT4—the cooling off Greenland. The histori-calGHG MME generally warms more than observed almosteverywhere, except for some regions in the ocean south of Australia and southern South America.

[40] The 1901–1950 period (Figure   8   second column)sees HadCRUT4 warming significantly more than the histor-ical MME in the north Pacific and parts of the Atlantic andcooling more in the tropical Pacific. Many areas of the his-toricalNat and historicalGHG MME do not warm as muchas HadCRUT4.

[41] The HadCRUT4 warms over the 1951–2010 period(Figure 8  third column) with a strong signal over land andthe Northern Hemisphere. The historical simulations cap-ture this land warming trend, although the MME range isinconsistent in places over Eurasia and Africa. Some parts

of the Pacific, in HadCRUT4, do not warm as much as  95%of the historical simulations. The historicalNat simulationsare largely inconsistent with HadCRUT4 over 1951–2010,showing an overall cooling trend. A large proportion of theglobe for historicalGHG is warming more than HadCRUT4.

[42] While much of the globe is warming in HadCRUT4for the 1979–2010 period (Figure   8  fourth column), thereare some interesting features. There are areas of cooling

over the Southern Ocean and in a distinct “v” shape in thePacific and strong warming over the south Pacific and someareas in Africa and Eurasia. These areas are outside the cen-tral   90%   of both the historical and historicalGHG MMEtrends. The historicalNat shows inconsistencies over muchof the globe again generally showing less warming thanHadCRUT4. The HadCRUT4 patterns of warming/coolingover the Pacific may be related to changes in the PacificDecadal Oscillation [ Mantua and Hare, 2002] from a warm

 phase to a cool phase over the 1979–2010 period [Trenberthet al., 2007]. Several of the historical simulations have pat-terns similar to the Pacific cooling seen in HadCRUT4, butonly one simulation (third ensemble member of MIROC-ESM) cools with a similar magnitude.

[43] Latitudinal zonal trends were calculated for eachobservational data set and simulation for the four periodsshown in Figure 9.  We only calculate the zonal trend at agiven latitude for the observational data sets where they have50% or more of the number of grid points that HadCRUT4has. This is to reduce the impact of coverage differences,

 particularly from the NCDC and JMA data sets which haveless coverage at high latitudes than HadCRUT4, on compar-isons between the observed zonal trends. The central  90%range (where equal weight is given to each model) of the his-torical and historicalNat are shown as shaded regions and thefour observational data sets as colored lines—HadCRUT4as the black line. The spread in the observational data setsgives an indication of the uncertainties in the observations.

Reducing the spatial coverage of model simulations to thatof HadCRUT4 increases the spread across the MME inthe higher latitudes (compare with Figure S4 where themodel data have full coverage). The observations gener-ally warm more than historicalNat across the latitudes for three of the periods, but not over the 1901–1950 period.Alternatively the observations lie within the spread of thehistorical simulations for most of the latitudes and periods

 being examined.[44] The observed cooling over the Southern Ocean

during the 1979–2010 period seen in both Figures  8  and  9may be related to changes in the Southern Annual Mode[Trenberth et al., 2007], although there may also be forcingcontributions to these patterns of change [ Karpechko et al.,

2009]. This cooling is not captured by most of the historicalsimulations, but a few do show cooling trends of a similar magnitude as HadCRUT4.

5.4. Models Including Indirect Aerosol Effects

[45] We have not made any judgements about the qualityor skill of the model’s response, such as how well modelssimulate an observed climatology. But one criteria of “quality” that could be used is how complete the esti-mates of forcing factors are incorporated by each model.For instance do models with a broader use of available forc-ing factors respond substantially differently to those with

4014

Page 15: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 15/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

1860 1880 1900 1920 1940 1960 1980 2000

Year

-0.5

0.0

0.5

1.0

1.5

   T  e  m  p

  e  r  a   t  u  r  e  a  n  o  m  a   l  y   (   K   )

DirectDirect+IndirectObservations

Figure 10.   Global annual mean TAS for CMIP historicalsimulations with and without indirect aerosol effectscompared with the observations (black lines). Showingthe weighted average (thick dark lines) and 5–95% range(thin light lines) for the simulations including direct sulfateaerosol effects only (blue) and those including both directand indirect effects (red). Data processed as in Figure 4.

a smaller collection of forcing factors? Arguably the mostdominant grouping of forcing factors after greenhouse gasconcentrations are sulfate aerosols [ Haywood and Schulz ,2007]. How the different aerosol processes are modeledvary considerably across the models [ Forster et al.,  2007],which can produce differences in the climatic responses of models (e.g., temperatures over the Atlantic region) [ Boothet al., 2012]. It could be reasonable to weight down simula-tions that do not include important physical processes, suchas indirect aerosol mechanisms [Weigel et al., 2010].

[46] Figure   10   compares the TAS variations over 1850–2010 for the CMIP3 and CMIP5 historical simulations

that include only the direct effects of sulfate aerosols andthose that include both indirect and direct effects. By theend of the 20th century, the weighted mean of those modelsonly incorporating aerosol direct effects warms up consider-ably more than the models that also include aerosol indirecteffects. This is largely because simulations that include thedirect effect only do not cool as much over the 1940–1979

 period. Both sets of simulations have a similar rate of increase thereafter. Maps of spatial trends (Figure S5) showthat the weighted average of the simulations including directeffects only warms up more at the higher northern latitudesover the 1901–2010 period. Over the 1940–1979, period theweighted average of the simulations that include both directand indirect effects has a strong cooling in the Northern

Hemisphere centered around the midlatitudes, over whichregion the average of models including just direct effectscontinues to warm. This indicates that including the indi-rect effects of sulfate aerosols in models produces a distinctsignal in TAS over the latter half of the 20th century whilenot including indirect effects produces a net warming biasover the whole century [Wild , 2012].

[47] Figure  10 also shows the global annual mean TASfor the different observational data sets. By the start of the21st century the observations are cooler than most of thesimulations that only include direct aerosol effects but arein the center of the distribution of models that include both

direct and indirect aerosol effects. It is intriguing to consider whether selecting a priori models with a more “complete”set of forcing factors may provide a more accurate represen-tation of past climate changes than models with a limited setof forcings. It should be remembered that there are still manyuncertainties in how factors that can influence climate aremodeled and in their radiative forcings and whether impor-tant factors are still not being included which could mean

any apparent agreement between models and observationsmay be fortuitously due to a cancelation of errors.

6. Detection Analysis on CMIP5 Models

[48] Of the CMIP5 models available, only eight have therequired historical, historicalNat, and historicalGHG sim-ulations covering the period 1901–2010 (BCC-CSM1-1,CNRM-CM5, CSIRO-Mk3-6-0, CanESM2, GISS-E2-H,GISS-E2-R, HadGEM2-ES, and NorESM1-M; Table  4) tocarry out a detection and attribution analysis that seeksto partition the observed temperature changes into contri-

 butions from greenhouse gases only, from other anthro-

 pogenic forcings and from natural forcings. We only usethe HadCRUT4 data set in this detection analysis. We planto examine the impact of choice of observational data setand uncertainties in HadCRUT4 (see Section  3) in a futuredetection study.

6.1. The Method of Optimal Detection

[49] We apply a standard implementation of optimaldetection, the methodology of which has been extensivelycovered elsewhere [ Allen and Stott ,   2003;   Hasselmann,1997;  Hegerl et al.,   1996;   Stott et al.,  2003]. The follow-ing description is adapted from  Jones et al.  [2008, 2011b]and  Jones and Stott   [2011]. The method is a linear regres-sion of simulated climate pattern responses against observed

climate changes, optimizing by weighting down modes of variability with low ratio of signal to noise. Patterns arefiltered (which can be spatial, temporal, or both), by pro-

 jecting onto an estimate of the leading orthogonal modes of internal climate variability, which have been whitened. This

 produces “optimal fingerprints” where modes of variabilitywith high signal-to-noise ratios (SNR) have more promi-nence than those with low SNR. The basis of orthogonalmodes—empirical orthogonal functions (EOFs)—is usuallydeduced from model simulations.

[50] Scaling factors for different forcing factors arededuced by regressing the filtered observed changes(response variable) against the forced climate changes opti-mal fingerprints (explanatory variables), allowing for noise

in both (the total least squares technique):

 y –  0  =

 I Xi=1

ˇi( xi –   i) (1)

where y  is the observed pattern and x i  the  ith simulated pat-tern of  I  signals. The scaling factors to be estimated in theregression are  ˇ i.   0   and   i  are the estimate of the internalvariability in the observations and in the simulated temper-atures, respectively. If more than one ensemble member isavailable, then   xi   is the ensemble average, although then

4015

Page 16: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 16/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

the model noise is scaled to allow for the changes in thenoise characteristics [ Allen and Stott , 2003]. An indepen-dent estimate of noise is used to estimate the uncertaintiesin  ˇi   and for a residual consistency test. The consistencyof the linear regression to over/under fitting is examined bycomparing the residual of the regression with the expectedvariance, as estimated from an independent estimate of noisevariability, using an  F  test [ Allen and Tett , 1999] at a two-

sided significance level of  10%. Uncertainties in the scalingfactors (ˇi) of the simulated patterns give a measure of whether a particular forcing factor is detected, by testing thenull hypothesis that the scaling factors are consistent with avalue of zero, i.e., a detection occurs when the scaling factor uncertainty range does not cross zero, and attributed by alsotesting the null hypothesis that the scaling factor is consis-tent with a value of one, i.e., is the response inconsistent withthat expected from the model. If all known major forcingfactors are included in a multiple signal analysis, then con-fidence in an attribution may be strengthened [ Hegerl et al.,2007], by showing consistency with the understanding of the

 physics and helping to exclude alternative factors as beingthe sole drivers for the observed changes. The use of such

an attribution consistency test [ Hasselmann, 1997; Allen and Tett , 1999], while not a “complete attribution assessment”[ Hegerl et al., 2007], may be considered a practical approachwithin the analysis framework we use [ Mitchell et al., 2001].If the factors are not detected or attributed, the pattern andmagnitude of the forced responses may have errors, other important forcing factors may have not been included, or amode of internal variability is not being simulated appropri-ately. Or alternatively the assumption that the responses todifferent forcing factors can be linearly added together may

 be inappropriate. The key point of the detection method isthat the uncertainty in the magnitude of the forcing factor and/or climate response can be compensated for by the scal-ing factor (ˇi). However this method is unable to account for 

uncertainties in forcing and response space-time patterns. Avariant of this method, using “error-in-variables,” attemptsto account for inter-model differences [ Huntingford et al.,2006], however to avoid adding further complexity, we donot use that method here.

[51] The values of the scaling factors will have a depen-dency on the number of EOFs used (or truncation in EOFspace). The fewer the EOFs used, then the lower the over-all variability of the observations and simulated patterns thatare captured. But the higher the number of EOFs used, themore modes of variability are added that add little informa-tion about the signals. The absolute maximum EOFs that can

 be used are determined by the number of degrees of free-dom in the creation of the EOFs. But if ranges of EOFs give

unbounded scaling factors or fail the residual  F  tests, then alower EOF truncation can be chosen.

[52] We are interested in investigating the different con-tributions to observed changes from greenhouse gasesonly, non-greenhouse gas anthropogenic factors and naturalinfluences, but we only have the historial, historicalGHG,and historicalNat experiments to use. To deduce the scal-ing factors for greenhouse gases only (G), the other anthropogenic forcings (OA), and natural influences (N),we make a transformation on the historical, historical-

 Nat, and historicalGHG scaling factors—  historical, ˇhistoricalNat

and  ˇhistoricalGHG, respectively—as calculated in a multiple

signal regression (equation (1)) [as described in  Tett et al.,2002], i.e.,

ˇG  =  ˇhistorical + ˇhistoricalGHG

ˇOA =  ˇhistorical

 N  =  ˇhistorical + ˇhistoricalNat

9>=>;

(2)

[53] A basic assumption of the optimal detection analysisis that the estimate of internal variability used is comparablewith the real world’s internal variability. As the observa-tions are influenced by external forcing, and we do not havea non-externally forced alternative reality to use to test thisassumption, an alternative common method is to comparethe power spectral density (PSD) of the observations withthe model simulations that include external forcings. Wehave already seen that overall the CMIP5 and CMIP3 modelvariability compares favorably across different periodicitieswith HadCRUT4-observed variability (Figure 5). Figure S11(in the supporting information) includes the PSDs for eachof the eight models (BCC-CSM1-1, CNRM-CM5, CSIRO-Mk3-6-0, CanESM2, GISS-E2-H, GISS-E2-R, HadGEM2-

ES and NorESM1-M) that can be examined in the detectionanalysis. Variability for the historical experiment in mostof the models compares favorably with HadCRUT4 over the range of periodicities, except for HadGEM2-ES whosevery long period variability is lower due to the lower over-all trend than observed and for CanESM2 and bcc-cm1-1whose decadal and higher period variability are larger thanobserved. While not a strict test, Figure S11 suggests thatthe models have an adequate representation of internalvariability—at least on the global mean level. In addition,we use the residual test from the regression to test whether there are any gross failings in the models representation of internal variability.

[54] Two types of analysis are examined here. The first

analysis uses samples of the eight model’s piControl tocreate a common EOF basis with which to analyze allthe data with [Gillett et al.,   2002]. In conjunction withthis method, averages are constructed of the three sig-nals from the simulations, the first a simple average andthe second a weighted average giving equal weight toeach model (following the method described in  Huntingford et al. [2006], see also the supporting information). The sec-ond analysis looks at the six models that have sufficient sim-ulated data to estimate their own EOF basis (CNRM-CM5,CSIRO-Mk3-6-0, CanESM2, GISS-E2-H, GISS-E2-R, andHadGEM2-ES). These are the models with long-enough

 piControl and enough ensemble members in the experimentsto allow the generation of a sufficiently large number of EOFs to represent internal variability.

[55] Two periods are examined, 1901–2010 and1951–2010, using 10 year mean anomalies (relative to

 periods being examined) filtered onto 5000 km spatial scales[Stott and Tett , 1998]. The decadal means are calculatedfor the simulations by averaging annual means and thenmasking by HadCRUT4’s 10 year means. The techniquefor creating the filtered data is the same as described in

 previous detection analyses [Tett et al.,   2002;   Stott et al.,2006b;  Jones et al.,  2011b] apart from using annual meanscalculated over January to December (further details in thesupporting information).

4016

Page 17: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 17/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-4

-2

0

2

4  5 10 15 20 25 5 10 15 20 25

-4

-2

0

2

45 10 15 20 25

 S  c  a

l  i  n g

f   a c  t   or  s 

-4

-2

0

2

4

-4

-2

0

2

4

-4

-2

0

2

4

5 10 15 20 25

EOF truncation no.

-4

-2

0

2

4

5 10 15 20 25

G

OA

N

-4

-2

0

2

4

-4

-2

0

2

4

5 10 15 20 25

   S  c  a   l   i  n  g   f  a  c   t  o  r  s

Figure 11.   Optimal detection scaling factors (ˇG, ˇOA and  N) for 1901–2010 period with 10 year means projected onto spherical harmonics. Scaling factors (5–95%  ranges) for each signal shown over rangeof available EOF truncations. Analysis using model common EOF basis is based on eight models. Thetriangle symbols represent EOF truncations where the residual consistency test fails at the 10% two-sidedsignificance level. Number of degrees of freedom that can be examined is 24.

6.2. Analysis of 1901–2010 Period

[56] Results for the 1901–2010 period using a commonEOF basis are shown in Figure 11  and using each model’s

own EOF basis in Figure 12.[57] A common EOF basis is constructed, following

Gillett et al. [2002], with the eight models that have histori-cal, historicalNat, historicalGHG, and piControl simulationsto project the observational and model data onto. The tech-nique of using a common EOF basis created from a rangeof climate models has been used in a number of studies[Verbeek ,  1997;  Barnett ,1998;  Stouffer et al.,  2000;  Hegerl et al., 2000;  Gillett et al.,  2005;  Zhang et al., 2006;  Santer et al., 2007; Christidis et al., 2010]. Equal length (240 years)segments from each of the eight models’ piControls areused to draw 110 year lengths, overlapping by 10 yearsexcept between models, to create a covariance matrix. Thisis repeated for a different 240 year long segment from the

models’ piControls. The first noise estimate is used to createthe common EOF basis and the second for uncertainty esti-mates and residual testing. We do not examine the sensitivityof the results to using other methods of creating a commonEOF basis [Stott et al., 2006a; Gillett et al., 2012b].

[58] The common EOF basis has 24 degrees of freedom(estimated from the number of independent 110 year lengthsegments, multiplied by 1.5) [ Allen and Smith, 1996] and soa maximum EOF truncation possible of 24 (Table 8). Usingthis maximum truncation can explain 93.6% of the observedvariance [Tett et al.,  2002] and more than  95%  of the his-torical variance for each of the models. Figure   11  shows

the scaling factors for the eight models across the range of maximum EOFs being used. CNRM-CM5, CanESM2, and

 NorESM1-M give fairly consistent scaling factors across the

range of EOFs. HadGEM2-ES is more varied and fails theresidual consistency test for most of the maximum trun-

cation choices. The other models give poorly constrainedresults. Figure   13a—left-hand side—shows the results for the eight models at the truncation of 24. The same truncationis used to enable a consistent comparison when using thecommon EOF basis. Five of the models fail to detect green-house gases (G), with CSIRO-Mk3-6-0, GISS-E2-H, GISS-E2-R, and bcc-csm1 having very large ranges of scaling

factors and HadGEM2-ES failing the residual consistencytest. Of the three models that do detect G only, CNRM-

CM5 has scaling values consistent with unity, CanESM2has values lower, and NorESM1-M has values higher thanone. Only NorESM1-M detects other anthropogenic forc-

ings (OA), with a value significantly greater than 1, withCanESM2 and HadGEM2-ES both not detecting OA but

with negative best estimates. CanESM2, HadGEM2-ES, and

 NorESM1-M detect natural forcings (N) with values consis-tent with 1. We use two alternative ways of creating averagetemporal spatial patterns across the models. The first is justsimply averaging up all the available ensemble membersfrom the eight models for the historical (45 members), his-toricalNat (32 members), and historicalGHG (32 members)and treating the ensemble members as if being sampled fromthe same model [Gillett et al.,   2002]. The second methodaverages the ensembles for each model then averages the

4017

Page 18: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 18/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-4

-2

0

2

45 10 15 20 25 5 10 15 20 25

-4

-2

0

2

45 10 15 20 25

 S  c  a

l  i  n g

f   a c  t   or  s 

-4

-2

0

2

4

5 10 15 20 25

   S  c  a   l   i  n  g   f  a  c   t  o  r  s

5 10 15 20 25

EOF truncation no.

-4

-2

0

2

4

5 10 15 20 25G

OA

N

Figure 12.   Same as Figure 11 but using individual model basis (i.e., analysis of each model using EOF basis produced by that model).

models means [Santer et al., 2007], thus given equal weightto each model rather than to each ensemble member, with

the noise estimates being scaled appropriately. For the sim- ple model average (Simple avg) and weighted model average(Weighted avg), G is robustly detected across the choice of truncation with values consistent with 1 (Figures 11 and 13).

 N is also detected with values consistent with 1, but OAis not detected in both cases. For both “Simple avg” and“Weighted avg,” the residual consistency test fails over wideranges of EOF truncations (Figure 11), although not for themaximum truncation.

[59] The sensitivity of the results to what model is exam-ined may be a consequence of the common EOF basis

 being composed of relatively small amount of data with lowdegrees of freedom, although the captured variance in the

model’s historical simulations is between 95.2% and 96.7%,suggesting most modes of variability should be captured.

Problems of reliably estimating scaling factors are moreapparent when using the EOF basis produced from eachmodel. For each model, covariance matrices are constructedfrom the model’s piControl and intra-ensemble variability[Tett et al., 2002]. These are used to create the EOF basis for the model and noise estimates for uncertainty and residualtesting. The scaling factors for the six models able to con-struct their own EOF spaces are shown in Figure  12. Eachof the six models has different length piControl’s and a dif-ferent number of historical, historicalNat and historicalGHGensembles to be used to create intra-ensemble variabilityestimates [Tett et al., 2002], so it is not possible to make thesame choices for which parts of the noise estimates are used

Table 8.   CMIP5 Models Used in the Detection Analysis Using the Models Own EOF Basis (Individual

EOF Basis) and a Common EOF Basis for Both Periods Being Examined, 1901–2010 and 1951–2010 a

Model 1901–2010 1951–2010

Trunc. Cap.Var.(%) Trunc. Cap.Var.(%)

Obs. Mod. Obs. Mod.

 Individual EOF BasisCNRM-CM5 26 93.7 94.8 45 97.8 99.2CSIRO-Mk3-6-0 21 91.1 94.2 41 97.7 98.9CanESM2 18 90.0 91.9 24 95.7 97.7GISS-E2-H 14 80.2 81.7 28 95.4 96.6GISS-E2-R 20 92.6 91.2 39 96.7 98.0HadGEM2-ES 16 89.7 89.7 28 97.2 99.1

Common EOF BasisCNRM-CM5 24 93.6 96.7 24 96.7 99.2CSIRO-Mk3-6-0   " "   95.2   " "   98.7CanESM2   " "   95.8   " "   98.9GISS-E2-H   " "   95.5   " "   98.6GISS-E2-R    " "   95.6   " "   98.6HadGEM2-ES   " "   96.4   " "   98.9

 NorESM1-M   " "   95.2   " "   98.1 bcc-csm1   " "   96.6   " "   99.0Simple Avg.   " "   96.5   " "   99.1Wgt. Avg.   " "   96.5   " "   99.0

aEach rowgives themaximum truncation of EOFbeingused (Trunc.), thepercentage captured variance (Cap.Var.)for HadCRUT4 (Obs.), and the historical experiment ensemble average (Mod.) when projected onto the truncatedEOF basis.

4018

Page 19: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 19/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-1

0

1

2

   S  c  a   l   i  n  g   f  a  c   t  o  r

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

N  o  r  E  S  M  1  - M  

b  c  c  - c  s  m  1  

a)

S  i  m   p  l  e   a  v   g  

W   e  i   g  h  t  e  d   a  v   g  

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

GOAN

CommonEOF basis

IndividualEOF basis

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

   T  e  m  p  e  r  a   t  u  r  e   T  r  e  n   d  s

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

N  o  r  E  S  M  1  - M  

b  c  c  - c  s  m  1  

b)

S  i  m   p  l  e   a  v   g  

W   e  i   g  h  t  e  d   a  v   g  

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

   T  e  m  p  e  r  a   t  u  r  e   T  r  e  n   d  s

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

N  o  r  E  S  M  1  - M  

b  c  c  - c  s  m  1  

c)

S  i  m   p  l  e   a  v   g  

W   e  i   g  h  t  e  d   a  v   g  

C  N  R  M  - C  M  

5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

CommonEOF basis

IndividualEOF basis

   (   K  p  e  r   1   1   0  y  e  a  r  s   )

   (   K  p  e  r   6   0  y  e  a  r  s   )

Figure 13.   Optimal detection results for 1901–2010 period, 10 year means projected onto spherical harmonicsfor chosen maximum EOF truncation (Table   8). Results

using common EOF basis on left and using the model’s ownEOF basis on the right. (a) Scaling factors (ˇG,  ˇOA, and

 N), (b) linear trends of scaled reconstructed temperaturesover whole period, and (c) linear temperature trend over sub-period 1951–2010 while using scaling factors shown inFigure 13a. Best estimates shown with central  90%   uncer-tainty range. Trends in decadal global mean HadCRUT4shown as thin horizontal black lines. Triangles at top of 

 panels indicate where analysis fails residual consistency test.

in the analysis for each model. The supporting informationgives details of the choices made. The number of degrees of freedom, NDOF, is therefore different for each model’s anal-ysis. The maximum EOF truncation, set to the model’s EOF

 basis NDOF, will also be different for each model, varying between 14 and 26 (see Table   8  for details together withthe captured observed variance for that truncation). BothCNRM-CM5 and CanESM2 detect G, OA, and N, althoughCanESM2 finds a negative value for OA and fails the resid-ual consistency test over much of the lower EOF truncations.While CSIRO-Mk3-6-0 detects G, this is not a robust resultas it is not detected for all the other choices of truncation(Figure 12). The scaling factors shown in Figure 13a (right-hand side) are for the maximum truncations allowed for eachmodel (Table 8). Unfortunately the CMIP5 simulations do

not yet have available the millennia length control simula-tions that are needed for a robust estimation of the modes of variability needed for such analyses and therefore the mostreliable method of estimating scaling factors and attributabletemperature trends is arguably to use the common EOF basistechnique. This also has the advantage that the observations

 projected onto this basis do not change when compared withthe different models.

[60] Reconstructed temperature trends [ Allen and Stott ,2003], best estimates and 5–95% ranges, are shown inFigures 13 b and 13c. Where G is detected, it is the dominantattributed forcing to the observed temperature trend. WhereOA is also detected, G is significantly warmer than theHadCRUT4 trend. For the cases where the attributed trendfor G is low, OA has either small cooling or even a warm-ing contribution—as when using CanESM2. The modelaverage analyses give similar attributed trends. For instancethe “Weighted avg” analysis finds ranges of attributed trendsfor G of 0.50–1.14 K, OA of   – 0.41–0.24 K, and N of 0–0.01K over the 1901–2010 period (given as 5–95%ranges) compared to the observed trend of 0.76 K per 110 years. Examining the trends over the sub-period of 

1951–2010 shows that G had a slightly larger relative contri- bution to observed trends, e.g., the “Weighted avg” analysisgives attributed trends (given as K per 60 years) for G of 0.49–1.13K, OA of  – 0.31–0.18 K, and N of  – 0.002–0.003 K with the HadCRUT4 trend of 0.65 K per 60 years.

6.3. Analysis of 1951–2010 Period

[61] We next examine the 1951–2010 period, the part of the instrumental record that is best observed. Figures 14 and15 show the results for the common EOF and the individualmodel EOF analyses, respectively.

[62] For the common EOF basis set, the shortness of the period together with the amount of data increases thenumber of degrees of freedom and EOFs that are available

for use. Using the eight climate models provides noise esti-mates that have 48 degrees of freedom which is thus themaximum EOF truncation that can be examined. Figure 14shows the scaling factors for the choices of EOF truncationup to the maximum allowable value. Many of the modelsgive unconstrained values for the higher choice of trunca-tion so to compare across the models, we choose an EOFtruncation of 24 (Table 8) to examine the scaling factors in aconsistent manner (Figure 16). The observed variance cap-tured by using the first 24 EOFs is 96.7% and is above  98%for each of the model’s historical simulation. Greenhousesgases (G) are detected in every model except GISS-E2-H.The CNRM-CM5, CSIRO-Mk3-6-0, HadGEM2-ES, and

 NorESM1-M models detect G with values consistent with 1, but CanESM2, GISS-E2-R, and bcc-csm1 detect G withvalues significantly lower than 1. Other anthropogenic forc-ings (OA) are detected by CSIRO-Mk3-6-0, HadGEM2-ES,and NorESM1-M, and natural (N) is detected by CanESM2,GISS-E2-H, HadGEM2-ES, NorESM1-M, and bcc-csm1.Only NorESM1-M detected all three signals with valuesconsistent with 1. Again in those models where OA isdetected, G has a value consistent with 1, and in thosemodels where G has a value less than 1, OA is not detected.

[63] Both analyses of model averages (“Simple avg” and“Weighted avg”) for the 1951–2010 period show detectionof G and OA with values consistent with 1 and only the

4019

Page 20: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 20/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-4

-2

0

2

45 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

-4

-2

0

2

45 10 15 20 25 30 35 40 45 50

 S 

 c  al  i  n g

f   a c  t   or  s 

-4

-2

0

2

4

-4

-2

0

2

4

-4

-2

0

2

4

5 10 15 20 25 30 35 40 45 50

EOF truncation no.

-4

-2

0

2

4

5 10 15 20 25 30 35 40 45 50

G

OA

N

-4

-2

0

2

4

-4

-2

0

2

4

5 10 15 20 25 30 35 40 45 50

   S  c  a   l   i  n  g   f  a  c   t  o  r  s

Figure 14.   Optimal detection scaling factors (ˇG,  ˇOA, and    N) for 1951–2010 period with 10 year means projected onto spherical harmonics. Scaling factors (5–95%  ranges) for each signal shown over range of available EOF truncations. Analysis using model common EOF basis is based on eight models.The triangle symbols represent EOF truncations where the residual consistency test fails at the   10%two-sided significance level. Number of degrees of freedom that can be examined is 48.

“Weight avg” case is N detected. While the conclusion of detection of greenhouse gases with scaling factor consis-tent with 1 is robust to truncation choice (Figure 14), exactvalues of scaling factors are sensitive to the number of EOFsincluded. For instance G scaling factors generally reduce inamplitude the larger the truncation when the weighted modelaverage is taken (Figure 14).

[64] The temperature trends of the reconstructed globalmean-scaled patterns are shown in Figure   16 b. For the

common EOF analysis (left-hand side of Figure   16 b), therange of the best estimate of the attributed linear trend for G for the different individual models that detect G varies

 between 0.63 and 1.48 K over the 60 year period, withonly CanESM2 showing G warming less than the observedtrend of 0.64 K per 60 years. This attributed warming isoffset by OA cooling in all the models, up to   – 0.80K per 60 years in magnitude, apart from CanESM2 and bcc-csm1models which show near-zero trends. The average models

-4

-2

0

2

45 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

-4

-2

0

2

45 10 15 20 25 30 35 40 45 50

 S  c  al  i  n g

f   a c  t   or  s 

-4

-2

0

2

4

5 10 15 20 25 30 35 40 45 50

   S

  c  a   l   i  n  g   f  a  c   t  o  r  s

5 10 15 20 25 30 35 40 45 50

EOF truncation no.

-4

-2

0

2

4

5 10 15 20 25 30 35 40 45 50G

OA

N

Figure 15.   Same as Figure 14 but for individual EOF analysis.

4020

Page 21: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 21/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

-1

0

1

2

   S  c  a   l   i  n  g   f  a  c   t  o  r

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

N  o  r  E  S  M  1  - M  

b  c  c  - c  s  m  1  

a)

S  i  m   p  l  e   a  v   g  

W   e  i   g  h  t  e  d   a  v   g  

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

GOAN

CommonEOF basis

IndividualEOF basis

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

   T  e  m  p  e  r  a   t  u  r  e   T  r  e  n   d  s

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

N  o  r  E  S  M  1  - M  

b  c  c  - c  s  m  1  

b)

S  i  m   p  l  e   a  v   g  

W   e  i   g  h  t  e  d   a  v   g  

C  N  R  M  - C  M  5  

C  S  I  R  O  - M  k  3  - 6  - 0  

C  a  n  E  S  M  2  

G  I  S  S  - E  2  - H  

G  I  S  S  - E  2  - R  

H  a  d  G  E  M  2  - E  S  

Common

EOF basis

Individual

EOF basis

   (   K  p  e  r   6   0  y  e  a  r  s   )

Figure 16.   As Figure   13   but showing results for 1951–2010 period. Table 8 gives the EOF truncations used.

analyses (“Simple avg” and “Weighted avg”) give warm-ings attributed to greenhouse gases greater than observedwith some offsetting due to other anthropogenic forcings.For instance “Weighted avg” analysis gives attributed trends(5–95%   trend range over 60 years) for G of 0.66–1.22 K,OA of  – 0.51–0 K, and N around 0 K.

[65] In contrast, the individual EOF basis analysis(Figure  15   and right-hand side of Figure   16) has a wider range of scaling factors across the models than for the com-

mon EOF basis. The scaling factors for CNRM-CM5 andCanESM2 are unbounded for the higher EOF truncations, sowe use lower EOF truncations (Table 8) to show their scalingfactors in Figure 16.  The best estimates of the attributed Gtrends, of the five models that detect G, range from 0.86to 2.06 K per 60 years, while the OA trends vary between – 1.36 and 0.24 K per 60 years. It is possible that whereas theoptimal detection regression procedure is able to scale grosserrors in a model’s transient climate response or net forcing,errors in model patterns for these models (which the regres-sion scaling cannot account for) may be biasing results, amodel bias that is less important when models are aver-aged. We therefore consider the more stable results using themulti-model mean, in particular “Weighted avg,” the mostrobust of the analysis.

[66] These analyses have a number of sources of sensi-tivity due to choices made in the methodology. For instancethe choice of what data to be used to create the model noiseestimates and EOF bases is arbitrary to a certain extent.Swapping the noise estimates so that the estimates used tocreate the EOF basis are now used for uncertainty estimatesand residual consistency testing (and vice versa) gives gen-erally similar results but with some interesting differences(supporting information). For instance the most strikingdifference is for the simple and weighted average scalingfactors with the common EOF basis where OA is now

consistently detected in both 1901–2010 and 1951–2010 periods (Figures S12 and S13).

[67] Three studies have also used optimal detection anal-yses on observed near-surface air temperatures for periodsending in 2010 [Stott and Jones, 2012; Gillett et al., 2012a,2012b]. While the studies produce a variety of results, dueto the different analysis choices, a broadly consistent con-clusion is the detection of G with attributed warming near 

or greater than observed, thereby adding confidence to theresults reported here.   Gillett et al.   [2012b]   examined aselection of seven CMIP5 models, some of which are thesame used in this analysis—CNRM-CM5, CSIRO-Mk3-6-0,CanESM2, HadGEM2-ES, NorESM1-M, and bcc-csm1— and HadCRUT4 for the period 1861–2010 using a commonEOF basis—created with different criteria than we use. InGillett et al.  [2012b]  and our study, G and OA are scaleddown for both CanESM2 and HadGEM2-ES while the other models have scaling values for G consistent or greater than1. The “Simple avg” result for both periods examined is sim-ilar to the model average result in Gillett et al. [2012b] withG detected and the scaling factors for both G and OA consis-tent with 1. The differences in the results, however, indicate

the sensitivity of the results to analysis choices, even whensimilar model and observational data is used in the studies.

[68] The results presented here are somewhat differentto previously reported detection studies on centennial timescales. The last IPCC assessment [ Hegerl et al.,   2007]reported on studies that compared different models withobserved changes over the 20th century [ Nozawa et al.,2005;   Stott et al.,   2006a], using similar methodologies asdescribed in this analysis that showed “. . . that there is arobust identification of a significant greenhouse warmingcontribution to observed warming that is likely greater thanthe observed warming over the last 50 years . . . ” For the

 period 1951–2010 examined in this analysis, we find a wider range of attributed greenhouse warming across a variety

of models than assessed in   Hegerl et al.   [2007] for the1950–1999 period.

7. Conclusions and Discussion

[69] Our analysis of the HadCRUT4 observational dataset and the CMIP5 multi-model ensemble supports previousanalyses in showing that the observed warming seen sincethe mid-20th century cannot be explained by natural forc-ings and natural internal variability alone. This conclusionstill holds even though we find a wider spread of modeledhistoric global mean temperature changes than in the CMIP3ensemble. This wider spread is possibly associated with agreater exploration of modeling uncertainty than hithertoincluding a much wider exploration of aerosol uncertaintythan previously carried out in models that now include amuch more sophisticated treatment of aerosol physics.

[70] Despite this wider spread of model results, calcu-lations of attributable temperature trends based on optimaldetection support previous conclusions that human-inducedgreenhouse gases dominate observed global warming sincethe middle part of the 20th century. It is the first timethat eight different climate models have been used in thistype of space-time TAS analysis with the consequencethat a wider range of aerosol and other forcing uncer-tainty is explored. We find a wider range of warming from

4021

Page 22: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 22/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

greenhouse gases and counteracting cooling from the directand indirect effects of tropospheric aerosols and other non greenhouse gas anthropogenic forcings than previouslyreported in detection studies of decadal variations in near-surface temperatures.

[71] Analyzing 1951–2010 (thereby concentrating on the part of the instrumental record that is best observed andwhen the external forcings are most dominant), and using the

multi model mean (Weighted avg), we estimate a range of  possible contributions to the observed warming of approx-imately 0.6 K (to nearest 0.1 K) from greenhouse gases of 

 between 0.6 and 1.2 K, balanced by a counteracting cool-ing from other anthropogenic forcings of between 0 and  – 0.5K. Our comprehensive set of sensitivity studies has shownthat there remains some dependence on model and method-ological choices, such as associated with choice of EOFtruncation and which set of data is used to estimate the EOF

 basis onto which model signals are projected. When modelsignals are averaged and when a common EOF basis is used,we get the most stable results.

[72] There remain continuing uncertainties associatedwith the methodology, in particular the extent to which

the methodology can compensate for model errors. Whilethe regression-based methodology can compensate for grossmodel errors in transient climate response and net radia-tive forcing by scaling the overall pattern of response toa particular forcing, it cannot compensate for errors in themodel patterns themselves since the whole pattern is either scaled up or down. Further work is required to understandthe extent to which model error might be influencing resultsand to develop the methodology to cope with this in amulti-model setting.

[73]   Acknowledgments.   We acknowledge the Program for ClimateModel Diagnosis and Intercomparison and the World Climate ResearchProgramme’s Working Group on Coupled Modelling, which is responsi-

 ble for CMIP, and we thank the climate modeling groups for producingand making available their model output. For CMIP, the U.S. Departmentof Energy’s Program for Climate Model Diagnosis and Intercomparison

 provides coordinating support and led development of software infrastruc-ture in partnership with the Global Organization for Earth System SciencePortals. We give thanks to the hard work of the institution modelers, data

 processors, and those who developed and maintain the gateways. Themodel data used in this study were obtained from  http://www-pcmdi.llnl.gov/ipcc/about_ipcc.php  (CMIP3) and  http://cmip-pcmdi.llnl.gov/cmip5/and associated gateways (CMIP5). We wish to thank Dáithí Stone for discussions about analysis of CMIP3 data in the IPCC 2007 report as wellas providing access to additional data. We acknowledge our colleagues whoretrieved some of the data from the CMIP archives including Jamie Kettle-

 borough and Peter Good. Thanks also goes to Piers Forster, Tim Andrews,Mat Collins, Glen Harris, Ben Booth, David Sexton, John Kennedy,Colin Morice, Nathan Gillett, and Nathan Bindoff for information, discus-sions, and comments. The work of the authors was supported by the JointDECC/Defra Met Office Hadley Centre Climate Programme (GA01101).We are also grateful to the reviewers, David Karoly, Gabi Hegerl, andFrancis Zwiers, for their helpful comments and insights.

References

Abramowitz, G., and H. Gupta (2008), Toward a model spaceand model independence metric,   Geophys. Res. Lett.,   35, L05705,doi:10.1029/2007GL032834.

Allen, M. R., and W. J. Ingram (2002), Constraints on future changes inclimate and the hydrological cycle, Nature, 419, 224–232.

Allen, M. R., and L. A. Smith (1996), Monte Carlo SSA: Detectingirregular oscillations in the presence of coloured noise,  J. Climate,   9,3373–3404.

Allen, M. R., and P. A. Stott (2003), Estimating signal amplitudes inoptimal fingerprinting, part I: Theory,  Clim. Dyn.,   21 (5–6), 477–491,doi:10.1007/s00382-003-0313-9.

Allen, M. R., and S. F. B. Tett (1999), Checking for model consistency inoptimal fingerprinting, Clim. Dyn., 15, 419–434.

Allen, M. R., et al. (2006), Quantifying anthropogenic influence on recentnear-surface temperature change, Surv. Geophys., 27 (5), 491–544.

Andrews, T., J. M. Gregory, M. J. Webb, and K. E. Taylor (2012), Forcing, feedbacks and climate sensitivity in CMIP5 coupledatmosphere-ocean climate models,   Geophys. Res. Lett.,   39, L09712,doi:10.1029/2012GL051607.

Barnett, T. P. (1998), Comparison of near-surface air temperature variabilityin 11 coupled global climate models,  J. Climate, 12, 511–518.

Booth, B. B. B., N. J. Dunstone, P. R. Halloran, T. Andrews, and N. Bellouin (2012), Aerosols implicated as a prime driver of twentieth-century North Atlantic climate variability,   Nature,   484, 228–232,doi:10.1038/nature10946.

Brohan, P., J. J. Kennedy, I. Haris, S. F. B. Tett, and P. D. Jones(2006), Uncertainty estimates in regional and global observed tempera-ture changes: A new dataset from 1850,  J. Geophys. Res.,  111, D12106,doi:10.1029/2005JD006548.

Christidis, N., P. A. Stott, F. W. Zwiers, H. Shiogama, and T. Nozawa(2010), Probabilistic estimates of recent changes in temperature: Amulti-scale attribution analysis, Clim. Dyn., 34(7–8), 1139–1156.

Collins, M., et al. (2010), Climate model errors, feedbacks and forcings:A comparison of perturbed physics and multi-model ensembles,   Clim.

 Dyn., 36 (9–10), 1737–1766.Driscoll, S., A. Bozzo, L. J. Gray, A. Robock, and G. Stenchikov (2012),

Coupled Model Intercomparison Project 5 (CMIP5) simulations of climate following volcanic eruptions,  J. Geophys. Res.,   117 , D17105,doi:10.1029/2012JD017607.

Forster, P., et al. (2007), Changes in atmospheric constituents and in radia-tive forcing, in   Climate Change 2007: The Physical Science Basis.Contribution of Working Group I to the Fourth Assessment Report of   the Intergovernmental Panel on Climate Change, edited by S. Solomon,et al., pp. 129–234, Cambridge University Press. Cambridge, UnitedKingdom and New York, NY, USA.

Forster, P. M., et al. (2013), Evaluating adjusted forcing and model spreadfor historical and future scenarios in the CMIP5 generation of climatemodels, J. Geophys. Res., 118, 1139–1150, doi:10.1002/jgrd.50174.

Forster, P. M. D., and K. E. Taylor (2006), Climate forcings andclimate sensitivities diagnosed from coupled climate model integrations,

 J. Climate, 19(23), 6181–6194.Foster, G., and S. Rahmstorf (2011), Global temperature evolution

1979–2010,  Environ. Res. Lett., 6 (4), 044022, doi:10.1088/1748-9326/6/4/044022.

Gillett, N., R. Allan, and T. Ansell (2005), Detection of external influenceon sea level pressure with a multi-model ensemble,  Geophys. Res. Lett.,32, L19714, doi:10.1029/2005GL023640.

Gillett, N. P., et al. (2002), Detecting anthropogenic influencewith a multi-model ensemble,   Geophys. Res. Lett.,   29(20), 1970,doi:10.1029/2002GL015836.

Gillett, N. P., et al. (2008), Attribution of polar warming to human influence, Nat. Geosci., 1, 750–754.

Gillett, N. P., G. M. Flato, J. F. Scinocca, and K. von Salzen (2012a),Improved constraints on 21st-century warming derived using 160years of temperature observations,   Geophys. Res. Lett.,   39, L01704,doi:10.1029/2011GL050226.

Gillett, N. P., V. K. Arora, D. Matthews, and M. R. Allen (2012b),Constraining the ratio of global warming to cumulative CO2   emissionsusing CMIP5 simulations,  J. Climate,   doi:10.1175/JCLI-D-12-00476.1(in press)

Guilyardi, E., et al. (2012), A first look at ENSO in CMIP5,  CLIVAR Exchanges, 17 (1), 29–32.

Gupta, A. S., et al. (2012), Climate drift in the CMIP3 models,  J. Climate,25, 4621–4640.

Hansen, J., et al. (2006), Global temperature change,  Proc. Natl. Acad. Sci.

U. S. A., 103(39), 14,288–14,293.Hasselmann, K. (1997), Multi-pattern fingerprint method for detection andattribution of climate change, Clim. Dyn., 13, 601–612.

Hawkins, E., and R. Sutton (2009), The potential to narrow uncer-tainty in regional climate predictions,  Bull. Am. Meteorol. Soc.,  90 (8),1095–1107.

Haywood, J., and M. Schulz (2007), Causes of the reduction in uncer-tainty in the anthropogenic radiative forcing of climate betweenIPCC (2001) and IPCC (2007),   Geophys. Res. Lett.,   34, L20701,doi:10.1029/2007GL030749.

Hegerl, G. C., P. D. Jones, and T. P. Barnett (2001), Effect of observa-tional sampling error on the detection of anthropogenic climate change,

 J. Climate, 14, 198–207.Hegerl, G., and F. Zwiers (2011), Use of models in detection and attribution

of climate change,  Wiley Interdiscip. Rev. Clim. Change, 2(4), 570–591,doi:10.1002/wcc.121.

4022

Page 23: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 23/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Hegerl, G. C., et al. (1996), Detecting greenhouse gas induced climatechange with an optimal fingerprint method,  J. Climate, 9, 2281–2306.

Hegerl, G. C., et al. (2000), Optimal detection and attribution of climatechange: Sensitivity of results to climate model differences,  Clim. Dyn.,16 (10–11), 737–754.

Hegerl, G. C., et al. (2007), Understanding and attributing climate change,in  Climate Change 2007: The Physical Science Basis. Contribution of  Working Group I to the Fourth Assessment Report of the Intergov-ernmental Panel on Climate Change, edited by S. Solomon, et al.,

 pp. 663–745, Cambridge University Press, Cambridge, United Kingdom

and New York, NY, USA.Huntingford, C., P. A. Stott, M. R. Allen, and F. H. Lambert (2006),Incorporating model uncertainty into attribution of observed temperaturechange, Geophys. Res. Lett., 33, L05710, doi:10.1029/2005GL0248312.

IPCC (2007a),  Summary for policymakers, in  Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change,edited by S. Solomon, et al., 18 pp., Cambridge University Press,Cambridge, United Kingdom and New York, NY, USA.

IPCC (2007b), in Climate Change 2007: Synthesis Report. Contribution of  Working Groups I , II and III to the Fourth Assessment Report of the

 Intergovernmental Panel on Climate Change, edited by R. Pachauri andA. Reisinger, 104 pp., IPCC, Geneva, Switzerland.

Ishii, M., A. Shouji, S. Sugimoto, and T. Matsumoto (2005), Objective anal-yses of sea-surface temperature and marine meteorological variables for the 20th century using icoads and the Kobe collection,  Int. J. Climatol.,25(7), 865–879.

Jones, C. D., et al. (2011a), The HadGEM2-ES implementation of 

CMIP5 centennial simulations,   Geosci. Model Dev.,   4, 543–570,doi:10.5194/gmd-4-543-2011.Jones, G. S., and P. A. Stott (2011), Sensitivity of the attribution of near 

surface temperature warming to the choice of observational dataset,Geophys. Res. Lett., 38, L21702, doi:10.1029/2011GL049324.

Jones, G. S., P. A. Stott, and N. Christidis (2008), Human con-tribution to rapidly increasing frequency of very warm NorthernHemisphere summers,   J. Geophys. Res.,   113, D02109, doi:10.1029/2007JD008914.

Jones, G. S., N. Christdis, and P. A. Stott (2011b), Detecting the influence of fossil fuel and bio-fuel black carbon aerosols on near surface temperaturechanges, Atmos. Chem. Phys., 11, 799–816.

Jones, G. S., M. Lockwood, and P. A. Stott (2012), What influence willfuture solar activity changes over the 21st century have on projectedglobal near surface temperature changes? J. Geophys. Res., 117 , D05103,doi:10.1029/2011JD017013.

Jun, M., R. Knutti, and D. W. Nychka (2008), Spatial analysis to quantifynumerical model bias and dependence: How many climate models are

there? J. Am. Stat. Assoc., 103(483), 934–947.Karoly, D. J., and Q. Wu (2005), Detection of regional surface temperaturetrends, J. Climate, 18, 4337–4343.

Karpechko, A. Y., N. P. Gillett, G. J. Marshall, and J. A. Screen (2009),Climate impacts of the Southern Annular Mode simulated by the CMIP3models, J. Climate, 22, 3751–3768.

Kaufmann,R. K.,H. Kauppi, M. L. Mann, andJ. H. Stock (2011), Reconcil-ing anthropogenic climate change with observed temperature 1998–2008,

 Proc. Natl. Acad. Sci. U. S. A., 108(29), 11,790–11,793.Kennedy, J. J., N. A. Rayner, R. O. Smith, D. E. Parker, and M. Saunby

(2011), Reassessing biases and other uncertainties in sea-surface tem- perature observations since 1850 part 2: Biases and homogenisation, J. Geophys. Res. , 116 , D14104, doi:10.1029/2010JD015220.

Kim, S. T., and J.-Y. Yu (2012), The two types of ENSO in CMIP5 models,Geophys. Res. Lett., 39, L11704, doi:10.1029/2012GL052006.

Knutson, T. R., et al. (2006), Assessment of twentieth-century regionalsurface temperature trends using the GFDL CM2 coupled models,

 J. Climate, 19(9), 1624–1651.

Knutti, R. (2010), The end of model democracy?   Clim. Change,   102,395–404.Knutti, R., R. Furrer, C. Tebaldi, J. Cermak, and G. A. Meehl (2010),

Challenges in combining projections from multiple climate models, J. Climate, 23(10), 2739–2758.

Knutti, R., and J. Sedlácek (2013), Robustness and uncertainties in the newCMIP5 climate model projections, Nature Climate Change, 3, 396–373,doi:10.1038/nclimate1716.

Lean, J. L., and D. H. Rind (2008), How natural and anthropogenic influ-ences alter global and regional surface temperatures: 1889 to 2006,Geophys. Res. Lett., 35, L18701, doi:10.1029/2008GL034864.

Lockwood, M. (2008), Recent changes in solar outputs and the global meansurface temperature. iii. Analysis of contributions to global mean air surface temperature rise,  Proc. R. Soc. A, 464(2094), 1387–1404.

Mantua, N. J., and S. R. Hare (2002), The Pacific decadal oscillation, J. Oceanogr., 58, 35–44.

Masson, D., and R. Knutti (2011), Climate model genealogy, Geophys. Res. Lett., 38, L08703, doi:10.1029/2011GL046864.

Meehl, G. A., et al. (2007a), The WCRP CMIP3 multi-model dataset:A new era in climate change research,   Bull. Am. Meteorol. Soc.,   88,1383–1394.

Meehl, G. A., et al. (2007b),  Global climate projections, in  Climate Change2007: The Physical Science Basis. Contribution of Working Group I tothe Fourth Assessment Report of the Intergovernmental Panel on ClimateChange, editedby S. Solomon, et al., pp.747–845, CambridgeUniversityPress, Cambridge, United Kingdom and New York, NY, USA.

Mitchell, J. F. B., et al. (2001), Detection of climate change and attributionof causes, in   Climate Change 2001: The Scientific Basis. Contributionof Working Group I to the Third Assessment Report of the Intergov-ernmental Panel on Climate Change, edited by J. T. Houghton, et al.,

 pp. 695–738, Cambridge University Press, Cambridge, United Kingdomand New York, NY, USA.

Morice, C. P., J. J. Kennedy, N. A. Rayner, and P. D. Jones (2012),Quantifying uncertainties in global and regional temperature changeusing an ensemble of observational estimates: The HadCRUT4 dataset,

 J. Geophys. Res. , 117 , D08101, doi:10.1029/2011JD017187.Murphy, J. M., et al. (2004), Quantification of modelling uncertainties in a

large ensemble of climate change simulations, Nature, 430, 768–772. Nagashima, T. et al. (2006), Effect of carbonaceous aerosols on surface tem-

 perature in the mid twentieth century, Geophys. Res. Lett.,  33, L04702,doi:10.1029/2005GL024887.

 Nozawa, T., T. Nagashima, H. Shiogama, and S. A. Crooks (2005),Detecting natural influence on surface air temperature change inthe early twentieth century,   Geophys. Res. Lett.,   32, L20719,

doi:10.1029/2005FL023540.Overpeck, J. T., G. A. Meehl, S. Bony, and D. R. Easterling (2011), Climatedata challenges in the 21st century, Science, 331, 700–702.

Pennell, C., and T. Reichler (2011), On the effective number of climatemodels, J. Climate, 24, 2358–2366.

Press, W. H., S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery (1992), Numerical Recipes in Fortran: The Art of Scientific Computing  2nd ed.,963 pp., Cambridge University Press, Cambridge, UK.

Santer, B. D., et al. (1995), Towards the detection and attribution of ananthropogenic effect on climate, Clim. Dyn., 12, 77–100.

Santer, B. D., et al. (2007), Identification of human-induced changes inatmospheric moisture content,  Proc. Natl. Acad. Sci. U. S. A.,  104 (39),15, 248–15, 253.

Santer, B. D., et al. (2008), Consistency of modelled and observed tem- perature trends in the tropical troposphere,   Int. J. Climatol.,   28 (13),1703–1722.

Smith, D. M., et al. (2007), Improved surface temperature predictionfor the coming decade from a global climate model,   Science,   317 ,

796–799.Smith, T. M., R. W. Reynolds, T. C. Peterson, and J. Lawrimore (2008),Improvements to NOAA’s historical merged land-ocean surface temper-ature analysis (1880–2006), J. Climate, 21(10), 2283–2296.

SREX (2012),   Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. A Special Report of Working Groups I and II of the Intergovernmental Panel on Climate Change,Cambridge University Press, Cambridge, UK and New York, NY, USA.

Stainforth, D. A., et al. (2005), Uncertainty in predictions of the cli-mate response to rising levels of greenhouse gases,   Nature,   433,403–406.

Stone, D. A., M. R. Allen, and P. A. Stott (2007), A multimodel update onthe detection and attribution of global surface warming,  J. Climate,  20,517–530.

Stott, P. A., and G. S. Jones (2012), Observed 21st century temperaturesfurther constrain decadal predictions of future warming,  Atmos. Sci. Lett.,13(3), 151–156.

Stott, P. A., and S. F. B. Tett (1998), Scale-dependent detection of climate

change, J. Climate, 11(12), 3282–3294.Stott, P. A., et al. (2000), External control of 20th century temperature bynatural and anthropogenic forcing,  Science, 290(5499), 2133–2137.

Stott, P. A., M. R. Allen, and G. S. Jones (2003), Estimating signal ampli-tudes in optimal fingerprinting, part II: Application to general circulationmodels,   Clim. Dyn.,   21   (5–6), 493–500, doi:10.1007/s00382-003-0314-8.

Stott, P. A., et al. (2006a), Observational constraints on pastattributable warming and predictions of future global warming,

 J. Climate, 19(13), 3055–3069.Stott, P. A.,et al. (2006b), Transient climate simulations with the HadGEM1

climate model: Causes of past warming and future climate change, J. Climate, 19, 2763–2782.

Stott, P. A., et al. (2010), Detection and attribution of climate change:A regional perspective,   Wiley Interdiscip. Rev. Clim. Change,   1 (2),192–211.

4023

Page 24: Jones Et Al 2013 Model Attribution

7/18/2019 Jones Et Al 2013 Model Attribution

http://slidepdf.com/reader/full/jones-et-al-2013-model-attribution 24/24

JONES ET AL.: ATTRIBUTION OF TEMPERATURES WITH CMIP5

Stouffer, R. J., G. Hegerl, and T. S. (2000), A comparison of surface air tem- perature variability in three 1000-yr coupled ocean-atmosphere modelintergrations, J. Climate, 13, 514–537.

Taylor, K. E. (2001), Summarizing multiple aspects of model perfor-mance in a single diagram,   J. Geophys. Res.,   106   (D7), 7183–7192,doi:10.1029/2000JD900719.

Taylor, K. E., R. J. Stouffer, and G. A. Meehl (2012), An overview of CMIP5 and the experiment design,   Bull. Am. Meteorol. Soc.,   93 (4),485–498, doi:10.1175/BAMS-D-11-00094.1.

Tett, S. F. B., et al. (2002), Estimation of natural and anthropogenic contri-

 butions to 20th century temperature change,  J. Geophys. Res., 107 (D16),4306, doi:10.1029/2000JD000028.Thompson, D. W. J., J. J. Kennedy, J. M. Wallace, and P. D. Jones (2008), A

large discontinuity in the mid-twentieth century in observed global-meansurface temperature, Nature, 453(7195), 646–649.

Trenberth, K. E., et al. (2007), Observations: Surface and atmospheric cli-mate change, in   Climate Change 2007: The Physical Science Basis.Contribution of Working Group I to the Fourth Assessment Report of   the Intergovernmental Panel on Climate Change, edited by S. Solomon,

et al., pp. 235–336, Cambridge University Press, Cambridge, UnitedKingdom and New York, NY, USA.

van Vuuren, D. P., et al. (2011), The representative concentration pathways:An overview, Clim. Change, 109(1–2, SI), 5–31.

Verbeek, J. (1997), Wind stress and SST variability in the North Atlanticarea: Observations and five coupled GCMs in concert,   Mon. Weather 

 Rev., 125, 942–957.von Storch, H., and F. Zwiers (2013), Testing ensembles of climate change

scenarios for “statistical significance”,  Clim. Change,   117 (1–2), 1–9,doi:10.1007/s10584-012-0551-0.

Weigel, A. P., R. Knutti, M. A. Liniger, and C. Appenzeller (2010), Risksof model weighting in multimodel climate projections,  J. Climate,  23,4175–4191.

Wild, M. (2012), Enlightening global dimming and brightening,  Bull. Am. Meteorol. Soc., 93(1), 27–37.

Zhang, X., F. W. Zwiers, and P. A. Stott (2006), Multimodel multisignalclimate change detection at regional scale,  J. Climate, 19, 4294–4307.

Zhang, X., et al. (2007), Detection of human influences on twentieth-century precipitation trends, Nature, 448, 461–465.