understanding the social cost of carbon: a technical ... · models x 5 socioeconomic scenarios x...

Steven RoseEnergy & Environmental Analysis Research Group, EPRI

U.S. Energy AssociationDecember 8, 2014

Understanding the Social Cost of Carbon: A Technical Assessment

2© 2014 Electric Power Research Institute, Inc. All rights reserved.

The Social Cost of Carbon (SCC)

$37 per metric ton of CO2

U.S. Government’s “central” SCC estimate of the global societal damages from a metric ton of today’s CO2 emissions


$0

$20

$40

$60

$80

$100

$120

$140

$160

$180

$200

2010 2020 2030 2040 2050

2007

$ /

CO2

US Government SCC Values

3% (95th

percentile)

US Government Social Costs of Carbon by Discount Rate

2.5%

3%

5%

Source: Developed from USG (2010) and USG (2013)

“Central” values

$37

Solid = USG (2013) estimates

Dashed = USG (2010) estimates

2015


Why Should We Care?

• SCC is an estimate of the damages to society from CO2

• SCC is in use broadly in USG rulemakings (going back to 2008) –states and others using as well

• For foreseeable future, CO2 (all GHGs) will be regulated under the Clean Air Act and efficiency policies

• USG legally obligated to value CO2 (9th Circuit)

• USG SCC values recently updated in 2013 (significantly higher)– Revised SCC estimates attracting a great deal of attention– Variety of issues being raised – legal, process, & technical

• Two key technical challenges– Robustness – establishing confidence in SCC estimates– Application – using estimates properly

• General lack of technical information and understanding



$37 per metric ton of CO2

U.S. Government’s “central” SCC estimate of the global societal damages from a metric ton of today’s CO2 emissions

What does this mean?


Trying to Better Understand the SCC

• Currently difficult to interpret and evaluate the SCCs

• EPRI has undertaken an initiative aimed at better understanding and advancing methods– Developing detailed understanding of modeling– Evaluating alternatives

Team: Steven Rose, Delavane Turner, Geoffrey Blanford, John Bistline, Francisco de la Chesnaye, Tom Wilson



Definition: The net present value of global climate change impacts from one additional net global tonne of carbon dioxideemitted to the atmosphere at a particular point in time

SCC in 2020 is the discounted value of the additional impacts from the marginal

emissions increase in 2020

2000 2300

2000 2300

2000 2300

2000 2300 2000 2300

Population

Income

Emissions (CO2, etc.) Temperature Climate damages

CO2 pulse

Socioeconomics

2000 2300 2000 2300 2000 2300

2000 2300

Dashed = after CO2 pulse


Types of Impacts Being Monetized

• Health• Agriculture• Forestry• Sea level• Water resources• Energy consumption

(space cooling & heating)• Migration• Hurricanes• Ecosystems• Catastrophic

Impact types included and formulations vary by model

Based on sector specific impacts studies in the

literature


0

5

10

15

20

25

30

35

40$(

3)$2

5 $5

2 $7

9 $1

06

$134

$1

61

$188

$2

15

$243

$2

70

$297

$3

25

$352

$3

79

$406

$4

34

$461

$4

88

$515

$5

43

$570

$5

97

$625

$6

52

$679

Num

ber o

f est

imat

es

1995$ per metric ton CO2 for 1995 emissions

Histogram of SCC Estimates in the Literature

(Derived from Tol 2008)

Note: not a distribution

Vast Range of SCC Estimates Have Been Produced

Developed from Tol (2008) meta analysis of SCC estimates

Range reflects differences in models, assumptions, and

impacts included (and not apples-to-apples)

-$3 to $655/tCO2


$0

$20

$40

$60

$80

$100

$120

$140

$160

$180

$200

2010 2020 2030 2040 2050

2007

$ /

CO2

US Government SCC Values

3% (95th

percentile)

US Government Social Costs of Carbon by Discount Rate

2.5%

3%

5%

Source: Developed from USG (2010) and USG (2013)

“Central” values

$37

Solid = USG (2013) estimates

Dashed = USG (2010) estimates

2015

Each point (e.g., $37) result of significant

aggregation


USG SCC Approach

Feature Detail

Multiple SCC models 3 models – DICE, FUND, PAGE

Standardized uncertainties - 5 reference socioeconomic and emissions scenarios (each extended from 2100 to 2300) - 1 distribution for the climate sensitivity parameter

Model specific parametric uncertainties In FUND and PAGE climate and damage components

Standardized discounting 3 constant discount rates – 2.5%, 3%, and 5%

Thousands of SCC results 150,000 SCC estimates for a given discount rate and year (3 models x 5 socioeconomic scenarios x 10,000 runs each)

Aggregation of results - Average of 150,000 results for each discount rate and year - “3% (95th percentile)” value is 95th percentile from distribution of 150,000 results with 3% discounting

USG estimates are the result of significant aggregation – over time, world regions, impact categories, many scenarios, & models.

Making sense of the estimates requires delving into these details.


Our Study’s Assessment Approach

• Examine the inner workings of the models to elucidate the key drivers and assess the main elements

• Specifically, learn about and assess the raw modeling and results – i.e., undiscounted and disaggregated to underlying facets

• 4 separate technical assessments– 3 assessments of modeling causal chain components

• Reviewing modeling, programming component, running diagnostics, comparing

• Exploring many perspectives

– 1 overall assessment of the USG experimental design


2000 2300

2000 2300

2000 2300

2000 2300 2000 2300

Population

Income


CO2 pulse

Socioeconomics

2000 2300 2000 2300 2000 2300

2000 2300

Technical Assessment by Causal Component

Elucidating and assessing each component

Component 1 Component 2 Component 3



Study Objective

• Improved understanding of SCC modeling and estimates that informs public discussion, future SCC modeling and application, and future climate research


Key Questions

• How do the models behave, and are they different? • What drives differences? • Are differences useful information?• Are there alternative uncertainties to consider?• Are there additional uncertainties to consider?• Are the estimates robust (insensitive to alternatives)?• Are there opportunities to improve the overall USG SCC

approach?


2000 2300

2000 2300

2000 2300

2000 2300 2000 2300

Population

Income


CO2 pulse

Socioeconomics

2000 2300 2000 2300 2000 2300

2000 2300

Socioeconomics & Emissions Component Assessment


Component 1



USG’s Standardized Socioeconomic & Emissions Inputs into SCC Models

0

20

40

60

80

100

120

140

160

2000

2050

2100

2150

2200

2250

2300

Foss

il &

Indu

stria

l CO

2(G

tCO

2)

0

500

1000

1500

2000

2500

2000

2050

2100

2150

2200

2250

2300

GDP

(tril

lion

2005

US$

)0

2

4

6

8

10

12

2000

2050

2100

2150

2200

2250

2300

Popu

latio

n (b

illio

n)

USG1USG2USG3USG4USG5 (550 avg)

-3

-2

-1

0

1

2

3

4

5

2000

2050

2100

2150

2200

2250

2300La

nd N

et C

O2

(GtC

O2)

0.0

0.5

1.0

1.5

2.0

2.5

2000

2050

2100

2150

2200

2250

2300

Non

-CO

2Ky

oto

Forc

ing (

W/m

2 )

We observe inconsistencies in

implementation across models

Global population Global GDP Global FF&I CO2

Global land CO2 Non-CO2 forcing


0

20

40

60

80

100

120

140

2000

2010

2020

2030

2040

2050

2060

2070

2080

2090

2100

Foss

il &

Indu

stri

al C

O2

(GtC

O2) EMF-22 baselines

USG1USG2USG3USG4USG5 (550 avg)

Socioeconomic and Emissions Uncertainty

050

100150200250300350400450

2000

2010

2020

2030

2040

2050

2060

2070

2080

2090

2100

Glob

al G

DP (t

rillio

n 20

05 U

S$)

0

2

4

6

8

10

12

2000

2010

2020

2030

2040

2050

2060

2070

2080

2090

2100

Glob

al p

opul

atio

n (b

illio

n)Very different socioeconomic structures

USG1 low emissions from high econ growth

USG2 high emissions from low econ growth

Baseline global fossil & industrial CO2

Baseline global GDP

Baseline global population

USG2USG2

USG1

USG1

USG2USG1

Uncertainty in socioeconomic structure not considered


Is Socioeconomic Structure Important?

• Defines the relationship between society & emissions

• Implications for both climate AND damage results

– Socioe structure Emissions Climate change

– Socioe structure Societal size & composition climate vulnerability & adaptation

Sensitivity of damages explicitly evaluated in damage component assessment

Temp CO2 Conc Total Income

Per Capita Income

PopSize/Comp Other

DICE X XFUND X X X X X XPAGE X X X

Drivers of climate damages in the models


Socioeconomic/Emissions Component Assessment Key Observations• Inconsistencies to address

– Implementation of standardized socioe/emissions inputs– Socioe/emissions extensions to 2300

• Additional uncertainties to consider – Range of socioe/emissions, socioe structure, 2300 extensions

• Some futures not equally likely and shouldn’t be weighted as such

• Average “policy” socioeconomic & emissions scenario is problematic


2000 2300

2000 2300

2000 2300

2000 2300 2000 2300

Population

Income


CO2 pulse

Socioeconomics

2000 2300 2000 2300 2000 2300

2000 2300

Climate Modeling Component Assessment


Component 2



Climate Modeling Structure

• Structural differences across the three models in all characteristics

• Different model specific parametric uncertainties considered across models

Climate Modeling Structural CharacteristicsAtmospheric concentrations

CO2

Non-CO2 KyotoNon-CO2 non-Kyoto

Radiative forcingCO2

Non-CO2 KyotoNon-CO2 non-Kyoto

Global mean temperatureRegional temperaturesClimate feedbackTime steps


We Isolate the Component & Run Diagnostics

0

20

40

60

80

100

120

140

160

2000 2050 2100 2150 2200 2250 2300

billi

on m

etric

tons

CO

2

USG2

USG5

2 Scenarios

USG2 = higher emissions scenario

USG5 = lower emissions scenario

Diagnostic Scenarios: We standardize the set of emissions & radiative forcing inputs (CO2 & non-CO2) and run deterministic and probabilistic scenarios

E.g., standardized total CO2 projection input


Projected Global Temperatures

0

1

2

3

4

5

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

deg

C ab

ove

pre-

indu

stria

l

DICE

FUND

PAGE

USG2

USG5

For the same emissions

scenario, 1˚C variation

Driven by modeling differences in

carbon cycle, non-CO2 forcing, forcing

to temperature translation, and

climate sensitivity responsiveness


Projected Incremental Temperatures (for a 1 billion tC pulse in 2020)

0

0.0005

0.001

0.0015

0.002

0.0025

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

deg

C

DICE

FUND

PAGE

USG5

USG2

Driven by differences in incremental emissions

implementation and climate modeling

For the same incremental emissions scenario, significant differences in the

incremental temperature change over time

Always higher response off lower emissions scenario


Sensitivity of Climate Responses

Temperature to emissions Most – DICE, Least - FUND

Temperature to climate sensitivity Most – PAGE, Least – FUND

0

0.001

0.002

0.003

0.004

0.005

0.006

DICE FUND PAGE DICE FUND PAGE

USG2 USG5

Chan

ge in

tem

pera

ture

(Deg

C)

1.5

3.0

4.5

6.0

Climatesensitivity

Incremental change in global temperature in 2100 varying climate sensitivity

The climate models are not equally sensitive to alternative assumptions

PAGE most sensitive


0

1

2

3

4

5

6

2000 2050 2100

degr

ees C

FUND - USG2

0

1

2

3

4

5

6

2000 2050 2100

degr

ees C

1%

5%

25%

50%

75%

95%

99%

Mean

Det

PAGE - USG2

0

1

2

3

4

5

6

2000 2050 2100

degr

ees C

Model Specific Uncertainty – Temperature (e.g., USG2 with climate sensitivity 3˚C)

FUND PAGEDICE

Models considering significantly different uncertainty – PAGE substantially more


0.0000

0.0005

0.0010

0.0015

0.0020

0.0025

0.0030

2000 2050 2100

degr

ees

C

DICE - USG2

-0.0005

0.0000

0.0005

0.0010

0.0015

0.0020

0.0025

0.0030

2000 2050 2100

degr

ees C

FUND - USG2

-0.0005

0.0000

0.0005

0.0010

0.0015

0.0020

0.0025

0.0030

2000 2050 2100

degr

ees C

1%

5%

25%

50%

75%

95%

99%

Mean

Det

PAGE - USG2

Model Specific Uncertainty – Incremental Temp (e.g., USG2 with climate sensitivity 3 degC)

Models considering significantly different uncertainty – PAGE substantially more

FUND PAGEDICE


0

0.0005

0.001

0.0015

0.002

0.0025

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

deg

C

DICE

FUND

PAGE

USG vs Other Modeling – Incremental Temp(for a 1 billion tC pulse in 2020)

RCP3PD

RCP8.5

RCP8.5 = higher emissions scenario

RCP3PD = lower emissions scenario


0

0.0005

0.001

0.0015

0.002

0.0025

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

deg

C

DICE

FUND

PAGE

MAGICC (Default)MAGICC (Hadley)

USG vs Other Modeling – Incremental Temp(for a 1 billion tC pulse in 2020)

Notably different responses from more sophisticated climate modeling

RCP3PD

RCP8.5


USG vs Other Modeling – Temp Uncertainty(e.g., RCP8.5)

0

1

2

3

4

5

6

7

8

2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100

Deg C

abov

e pre

-indu

stria

l

MAGICC

FUND

PAGE

DICE

Medians

Running climate components

probabilistically (including climate

sensitivity uncertainty)

Dashed = 17th-83rd

percentiles

Notably different probabilistic result from more sophisticated climate & probabilistic modeling


Climate Component Assessment Key Observations• Significant differences in climate responses across models

(to 2100 and 2300, in total and incremental climate)

• Important climate component structural differences

• Implementation inconsistencies affecting results

• Models representing & sampling different uncertainty spaces

• Alternative climate modeling produces different results

• Additional uncertainties to consider– Alternative climate modeling, alternative parametric uncertainty,

alternative climate sensitivity distribution assumption


2000 2300

2000 2300

2000 2300

2000 2300 2000 2300

Population

Income


CO2 pulse

Socioeconomics

2000 2300 2000 2300 2000 2300

2000 2300

Climate Damages Component Assessment


Component 3



Damage Modeling Structure

DICE FUND PAGERegions 1 global region 16 regions 8 regions

Sectors 2 sectors

Sea Level Rise,Aggregate non-SLR

14 sectors

Sea Level Rise, Agriculture, Forests, Heating, Cooling, Water Resources, Tropical Storms, Extratropical Storms, Biodiversity, Cardiovascular Respiratory, Vector Borne Diseases, Morbidity, Diarrhea, Migration

4 sectors

Sea Level Rise, Economic, Non-economic, Discontinuity

Damage drivers

Temperature, total income

Temperature (global & regional), CO2conc, ocean temp, population (size, composition), income (total, per capita), technological change

Regional temperature, income (total, per capita), regional damages scaled off EU

Damage specifications

Quadratic functions Various functional forms Power functions

Adaptation Implicit (within calibrated net responses)

Mostly implicit, explicit for agriculture & SLR, increased resiliency with per capita income

Exogenous adaptation policy

Structural differences across models

Different model specific parametric uncertainties


We Isolate the Component & Run Diagnostics

0123456789

2000

2050

2100

2150

2200

2250

2300

degr

ees C

abo

ve p

re-in

dust

rial

USG2USG5

Global Mean Temperature

0

500

1000

1500

2000

2500

2000

2050

2100

2150

2200

2250

2300

ppm

CO2 Concentration

0

2

4

6

8

10

12

2000

2050

2100

2150

2200

2250

2300

billi

on

Global Population

0

200

400

600

800

1000

1200

1400

1600

2000

2050

2100

2150

2200

2250

2300

trill

ion

$200

5

Global GDP

Diagnostic Scenarios: We standardize climate & socioeconomic inputs and run deterministic and probabilistic scenarios

Regional population & GDP

Standardized input projections2 Scenarios

USG2 = higher temp scenario (~4 deg C by

2100)

USG5 = lower temp scenario (~2 deg C by

2100)


Projected Total Global Damages

Driven by differences in damage modeling

structure & parameterization

For the same temperature & socioeconomic scenario, ~3x variation in damages

For the same temp &

socioeconscenario, net damages to net benefits


Projected Incremental Global Damages(standardized climate signal from a 1 billion tC pulse in 2020)

Driven by differences in damage modeling

structure & parameterization

For the same incremental temperature

change scenario, ~4x variation in

incremental damages


Damage Responsiveness to Temperature –Global Damages

Total climate damages as a function of global temperature

Figures with a USG2 reference socioeconomic condition

DICE and PAGE damages more responsive to warming

2100 w/ GDP ~$270T

2050 w/ GDP ~$120T


Damage Responsiveness to Income –Global Damages

FUND with increasing benefits with income at lower warming levels. DICE & PAGE more responsive to income at higher warming levels.

Total climate damages as a function of non-OECD income

Figures with a USG2 reference climate condition

2100 w/ 4˚C2050 w/ 1.8˚C


PAGE

FUN

D

2050 USG2

DICEALL

Damage Responsiveness to Temperature –Sectoral Damages

Cooling

AgricultureHeating

Water resourcesSLR

Non-economic

Discontinuity

Economic

SLR

2050 sectoral climate damages as a function of global temperature

Non-SLR

Particular sectors driving results

Sectoral damages differ in sign (damage or benefit) and responsiveness to warming


PAG

EFU

ND

2050 USG2

Damage Responsiveness to Temperature –Regional Damages

China

EU

US

Regional damages not equally responsive to warming. FUND with net benefits for some regions. PAGE with net damages for all regions. FUND

modeling individual regions. PAGE scaling damages off EU damages.

2050 regional climate damages as a function of global temperature


Key Parts of Incremental Damages (USG2 scenario to 2300)

Model specific features dominate incremental damages

210021002100

China cooling

ROW agriculture

China agriculture

SLR

Non-SLR Non-econ

Economic

SLR

ROW cooling

Discontinuity


Model Specific Uncertainty – Incremental Global Damages (e.g., USG2 temperature & socioeconomics)

Models considering significantly different damage uncertainty – FUND modeling significantly more than PAGE

and DICE, but PAGE has higher mean

FUND has broader

distribution with larger tails

Means


Damage Specification Literature SourcesModel

(version) Damage type Study Basis Links to SCC models

DICE (2010)

Aggregate non-SLR IPCC (2007b), Tol (2009)1 Calibration DICE, FUND, PAGE

SLR coastal impacts Undocumented FUND (v3.8) Agriculture

Kane et al. (1992), Reilly et al. (1994), Morita et al. (1994), Fischer et al. (1996), Tsigas et al. (1996)

Calibration

Tol (2002b) Income elasticity Forestry Perez-Garcia et al. (1995), Sohngen et

al. (2001) Calibration Tol (2002b) Income elasticity Energy Downing et al. (1995), (1996) Calibration Hodgson and Miller (1995) Income elasticity Water resources Downing et al. (1995, 1996) Calibration Downing et al. (1995, 1996) Income elasticity

Coastal impacts Hoozemans et al. (1993), Bijlsma et al. (1995), Leatherman and Nicholls (1995), Nicholls and Leatherman (1995), Brander et al. (2006)

Calibration

Diarrhoea WHO Global Burden of Disease (2000)2 Calibration

WHO Global Burden of Disease (2000) Income elasticity Vector-borne diseases Martin and Lefebvre (1995), Martens et

al. (1995, 1997), Morita et al. (1995) Calibration Link and Tol (2004) Income elasticity Cardiovascular and

respiratory mortality Martens (1998) Calibration

Storms CRED EM-DAT database,3 WMO (2006) Calibration

Toya and Skidmore (2007) Income elasticity Ecosystems Pearce and Moran, (1994), Tol (2002a) Calibration

PAGE (2009)

SLR Anthoff et al. (2006)4 Calibration & income elasticity FUND

Economic Warren et al. (2006)5 Calibration DICE, FUND, PAGE

Noneconomic Warren et al. (2006) Calibration DICE, FUND, PAGE

Discontinuity Lenton et al. (2008), Nichols et al., (2008), Anthoff et al. (2006), Nordhaus (1994)6

Calibration DICE, FUND

Adaptation costs Parry et al. (2009) Calibration

Damage formulations based

on older climate impacts literature,

with some formulations based on those from the

other models


Damage Component Assessment Key Observations• Significant differences in damage responses across models

(to 2100 and 2300, in total and incremental damages)

• Important damage component structural differences

• Model specific features driving results, e.g., – DICE – damages increase quadratically– FUND – agricultural CO2 fertilization, cooling demand, China

damages, adaptation– PAGE – regional scaling, non-economic damages, discontinuity

damages, adaptation

• Models representing & sampling different uncertainty spaces

• Additional uncertainty to consider – 2013 revisions


Overall Experimental Design Assessment

USG SCC experimental design features1. Multiple models2. Standardized uncertainties3. Model specific parametric uncertainties4. Standardized discounting5. Tens of thousands of SCC estimates6. Aggregation of estimates into USG SCC values


Experimental Design Issues• Significant structural differences across models – Do they reflect

differences in expert opinion? e.g.,– Socioeconomic/emissions – different sets of emissions and radiative forcing– Climate – carbon cycle, climate sensitivity, feedbacks, uncertainty– Damages – unique model specific factors that dominate results

• Consideration of uncertainty – we find reasonable alternative specifications, additional uncertainties, and artificial variation

• Comparability and independence of model results – Inconsistencies across modeling (implementation, structural, uncertainties) & inter-model relationships raise questions about statistical comparability which is required for averaging

• Robustness of overall results – Current results may not be robust (i.e., insensitive to reasonable alternatives) given our observations regarding (1) model sensitivity, and (2) existence of alternative assumptions and modeling

• Experimental design challenges – Issues with the overall design, in particular the multi-model approach (with consistency and comparability issues)


Recommendations

1. Internal review of the modeling – to evaluate differences, improve comparability and uncertainty representation, and enhance robustness

2. Revisit experimental design – worth revisiting given challenges of multi-model approach (e.g., chose best approach for each component)

3. Evaluate robustness – useful given sensitivity of models and issues with uncertainty considered. Will increase confidence in results.

4. Peer review the approach and models – USG SCC approach is novel and peer review would be valuable. Model review also practical given regulatory use.

5. Provide additional documentation and justification – will facilitate communications & interpretation, and increase public confidence

6. Provide application guidance – SCC application also an issue, guidance on proper application needed


Concluding Remarks

• The social cost of carbon is important

• However, the USG estimates are difficult to interpret and assess

• Better understanding of the modeling and what the estimates represent is needed– To inform public discussion,– Improve SCC modeling and application, and – Facilitate climate research broadly – impacts analyses in

general, climate science, climate economics, & integrated assessment


New StudyUnderstanding the Social Cost of Carbon: A Technical Assessment

http://epri.co/3002004657

(full report & ES downloads)

Objective: inform public social cost of carbon (SCC) discussion, future SCC modeling and use, and climate research broadly


Thank You!

Questions/comments:

Steven Rose

[email protected]

202-293-6183

understanding the social cost of carbon: a technical ... · models x 5 socioeconomic scenarios x...

Documents