Climate Policy 1 (2001) 433–449
Integrated assessment of abrupt climatic changes
Michael D. Mastrandrea a,∗, Stephen H. Schneider b
a Department of Geological and Environmental Sciences, Stanford University, Stanford, CA 94305, USA
b Department of Biological Sciences, Stanford University, Stanford, CA 94305, USA
Received 9 February 2001; received in revised form 6 June 2001; accepted 25 August 2001
Abstract
One of the most controversial conclusions to emerge from many of the first generation of integrated assessment
models (IAMs) of climate policy was the perceived economic optimality of negligible near-term abatement of
greenhouse gases. Typically, such studies were conducted using smoothly varying climate change scenarios or impact
responses. Abrupt changes observed in the climatic record and documented in current models could substantially
alter the stringency of economically optimal IAM policies. Such abrupt climatic changes — or consequent impacts —
would be less foreseeable and provide less time to adapt, and thus would have far greater economic or environmental
impacts than gradual warming. We extend conventional, smooth IAM analysis by coupling a climate model capable
of one type of abrupt change to a well-established energy–economy model (DICE). We compare the DICE optimal
policy using the standard climate sub-model to our version that allows for abrupt change — and consequent enhanced
climate damage — through changes in the strength (and possible collapse) of the North Atlantic thermohaline
circulation (THC). We confirm the potential significance of abrupt climate change to economically optimal IAM
policies, thus calling into question all previous work neglecting such possibilities — at the least for the wide ranges
of relevant social and climate system parameters we consider. In addition, we obtain an emergent property of our
coupled social–natural system model: “optimal policies” that do consider abrupt changes may, under relatively low
discount rates, calculate emission control levels sufficient to avoid significant abrupt change, whereas “optimal
policies” disregarding abrupt change would not prevent this non-linear event. However, there is a threshold in
discount rate above which the present value of future damages is so low that even very large enhanced damages in
the 22nd century, when a significant abrupt change such as a THC collapse would be most likely to occur, do not
increase optimal control levels sufficiently to prevent such a collapse. Thus, any models not accounting for potential
abrupt non-linear behavior and its interaction with the discounting formulation are likely to miss an important set
of possibilities relevant to the climate policy debate. © 2001 Elsevier Science B.V. All rights reserved.
Keywords: Emergent properties of coupled socio–natural systems; Integrated assessment; Abrupt non-linear climate change;
Thermohaline circulation; Discount rate; Climate policy analysis
∗ Corresponding author. Tel.: +1-650-724-3747; fax: +1-650-725-4387.
E-mail address: mikemas@stanford.edu (M.D. Mastrandrea).
1469-3062/01/$ – see front matter © 2001 Elsevier Science B.V. All rights reserved.
PII: S 1 4 6 9 - 3 0 6 2 ( 0 1 ) 0 0 0 3 8 - 9

---

434
M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
1. Introduction: context of the use of integrated assessment models for climate policy analysis
Integrated assessment models (IAMs) have become increasingly important tools in interdisciplinary
climate policy analysis. As instruments of analysis, IAMs allow study of the behavior of coupled systems
— which may have emergent properties that are not revealed by simulations with each disciplinary
sub-model alone. However, many such early-generation models are limited by restrictive assumptions
(Schneider, 1997) relative to the nature of the complex systems they attempt to simulate. First generation
IAMs applied to the climate problem calculate optimal emission control rates — often via the imposition
of a tax on carbon — given some standard set of population, technology and energy systems scenarios.
The impacts of a climate change scenario on the economy are usually treated via a climate damage
function, through which a given amount of climate change — some degree of global warming — is
translated into a consumption loss, often a percentage reduction in gross domestic product (GDP). This
loss of consumption leads to a loss of utility — usually the quantity to be maximized in IAMs. Most
of these first IAMs consistently produced economically optimal solutions with very modest near-term
mitigation.
Such coupled models typically employ simple climate sub-models that produce only smoothly-varying
climate change scenarios. The translation of gradual warming into climate damage omits a significant
source of potential climate damage — abrupt non-linear climate changes. If climate changes are smooth
then the capacity of society to adapt is higher. Changes that are gradual and more foreseeable lead to
lower damages compared to very rapid changes such as abrupt unanticipated events or “surprises”. Many
large-scale environmental systems have been shown to possess multiple equilibria and the potential
to switch rapidly between them, as well as hysteresis behavior (Higgins et al., 2001). One example
is the multiple stable equilibria of the thermohaline circulation (THC) of the North Atlantic, creating
the possibility of rapid and irreversible (in the near-term) ocean circulation changes, with significant
detrimental effects. Because of the time lag between emissions and impacts, and the abrupt nature of
such an event, seemingly “optimal” near-term policy behavior that does not account for the possibility
of abrupt changes could precondition a catastrophic collapse of this ocean current system by the 22nd
century. The simple climate models used in first generation IAMs are not equipped to represent such
changes, nor are most economy models coupled to them 1 . This study addresses these deficiencies, and
sketches some preliminary policy implications.
These same themes are present in the study of decision making under uncertainty. Working Group III
of the IPCC TAR (IPCC, 2001c) presents the results of seven models from an Energy Modeling Forum
project (Manne, 1995) on hedging strategies for low-probability, high-consequence climatic scenarios
under uncertainty resolved (i.e. the uncertainty is replaced with knowledge) in 2020. These models
employ cost–benefit analysis and show little abatement until after uncertainty about the occurrence of high
damages is resolved — low near-term abatement. However, when abrupt non-linear changes are involved,
a wait-and-see approach may lead toward an undesirable outcome: by the time the threatening outcome
is clearly identified, it may then be extremely expensive or even impossible to avoid. Working Group III
also describes research (Ha-Duong et al., 1999) that supports such concerns by showing decoupling from
1 For example, these models often assume perfect foresight of all future time periods. We introduce a wide range of increased
damages due to abrupt climatic change within this formulation, noting that damages will be greater without such artificial fore-
sight. Taking this one step further, elimination of perfect foresight may increase optimal control rates by enhancing precautionary
savings behavior (Kuntz-Duriseti, 2001; Lippman and McCall, 1981; Blanchard and Mankiw, 1988).

---

M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
435
current emissions trends if the interplay between the inertia of the economic system and uncertainty over
the required magnitude of emissions abatement is considered, a very similar situation.
We demonstrate the potential importance of abrupt non-linear climate damages with a modeling exercise
coupling a simple IAM to a simple climate–ocean model capable of representing one form of abrupt
non-linear climate change — weakening or collapse of THC. Owing to this simplicity, our quantitative
results are only meant to be illustrative of principles we believe should be taken into consideration by the
policy community.
2. The DICE model
William Nordhaus’ dynamic integrated climate and economy (DICE) model (Nordhaus, 1992) has been
a widely used IAM because of its relative simplicity and transparency. It couples a simple globally and
seasonally-averaged two-box climate model (Schneider and Thompson, 1981) with an economic model
of similar complexity. The coupled climate–economy system is solved as a simple optimal growth model
that maximizes discounted utility from consumption in all considered time periods with perfect foresight
(a traditional social welfare function). The model determines the optimal forecast for future emissions
reductions by balancing the costs of reducing emissions with the costs of climate change, represented by
a climate damage function.
Since its creation, many papers have demonstrated the sensitivity of the DICE model results to changing
structural assumptions (Schultz and Kasting, 1997; Roughgarden and Schneider, 1999). This paper relaxes
a simplification of most conventional IAMs via the addition of a sub-model that includes abrupt, non-linear
climate changes. The DICE model makes no attempt to incorporate certain non-linear behaviors found
in more complex general circulation models (GCMs) (Manabe and Stouffer, 1999) or observed in nature
(Broecker, 1997). The original DICE climate model is capable only of smooth temperature changes —
no discontinuities in the slope of the time-evolving changes — given a smooth CO2 increase scenario.
However, when forced by certain smooth emissions scenarios, GCMs can display abrupt non-linear
responses, in particular the weakening or collapse of the thermohaline circulation (THC) of the North
Atlantic Ocean that warms northern Europe by as much as 10◦C (Broecker, 1997). Such abrupt non-linear
changes could rapidly change the rate of temperature increase in as little as a decade, could cause regional
surface temperatures to alter much more rapidly than global averages, or could even produce a regional
temperature decrease while the rest of the world warms (Schneider and Thompson, 2000) 2 . Several
simulations suggest that this circulation is sensitive both to the level at which greenhouse gases are
stabilized in the future and also to the rate of increase of such concentrations in the atmosphere (Stocker
and Schmittner, 1997).
This paper explores the implications of incorporating abrupt non-linear climate behavior associated
with reduction and/or collapse of THC into the DICE model by creating the Enhanced DICE model,
or E-DICE. Only a few previous studies have considered rate-dependent damages from global warming
(Peck and Teisberg, 1994; Toth et al., 1997), the possibility of abrupt climate change (Toth et al., 1997),
and specific abrupt climate change damages from ocean circulation changes (Keller et al., 2000a,b).
2 Paleoclimatic data suggest that THC collapse has occurred many times, particularly in glacial periods when episodes of
icebergs discharged into the North Atlantic provided a cap of low salinity, low density melt-water that facilitates the formation
of sea ice, reduces the strength of the sinking of surface water and in turn, inhibits the Gulf Stream and its extension into
northwestern Europe (Broecker et al., 1985; Seidov and Maslin, 1999).

---

436
M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
While Keller et al. address the incorporation of ocean circulation changes, as do we, our approach differs
in that we couple a physical climate–ocean model to the optimal growth DICE model, whereas Keller
et al. tabulate data from a more complex climate–ocean model, but do not directly couple the economy
model to a physical model. Our approach allows us to find salient emergent properties of the coupled
system, since internal feedback processes can interact between our sub-models using an iterative coupling
technique.
3. The E-DICE model
The E-DICE model we develop extends the DICE model to incorporate potential damages from changes
in THC overturning rate by modifying the original DICE climate damage function. We add a term ε to the
exponent of the damage function, creating a more non-linear, “hockey stick”-like function (see Appendix
A). Fig. 1 compares the DICE damage function with a representative E-DICE damage function. Since
this “top down” damage function formulation is not based on more mechanistic “bottom up” analyses,
we use a range of plausible assumptions to test the sensitivity of this IAM to an array of abrupt events and
display its results compared to standard results using a smooth climate change model without the added
parameter for abrupt damages.
The E-DICE model determines values for ε by exchanging information with an enhanced climate model
that simulates THC. This model, the simple climate demonstrator (SCD) (Schneider and Thompson,
2000), was developed as an extension of an original two-box climate model (Schneider and Thomp-
son, 1981) (the conventional climate model in DICE). The SCD model retains suitable simplicity for
use in integrated assessment while demonstrating the major features of global warming simulations per-
formed by recent GCMs — specifically the response of THC to anthropogenic CO2 emissions. The
present circulation strength is thought to be (with considerable observational uncertainty) roughly 20
Fig. 1. Comparison of DICE and E-DICE damage functions. The E-DICE damage function here corresponds to an
of 0.98, a
mid- to high-range value. The temperature profile used to generate the DICE damage curve is that of the original DICE optimal
solution (3% PRTP). E-DICE was run with a maximum THC enhanced damage of 10% GWP, a rate of time preference of 1.8%,
an initial THC overturning rate of 15 Sv (1 Sv = 1 million m3/s), and a climate sensitivity of 4.5◦C for an equivalent doubling
of CO2.

---

M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
437
Sverdrups (Sv), where 1 Sv = 1 million m3/s. It is well-known from many modeling studies that the
strength of the overturning circulation initially decreases with increasing CO2, and for small and slowly
developing emissions scenarios the strength recovers, whereas for large and/or rapid forcing the over-
turning may cease entirely — the so-called thermohaline catastrophe. SCD reproduces this established
behavior.
Details of the behavior of SCD and the steps in an E-DICE/SCD run are outlined in Appendix A.
4. Results with Nordhaus discounting
The output of the canonical DICE model produces optimal carbon taxes for the future. Here, that
analysis is expanded to include enhanced damages from THC reduction: the E-DICE/SCD runs. Fig. 2
compares the optimal carbon tax forecasts for the E-DICE/SCD model with the optimal forecasts from the
original DICE results containing no enhanced damages (see Appendix A for information on methods).
Discount rates, carbon cycle formulation, and other parameters are all left as in DICE to isolate the
enhanced damage effects we single out. This graph indicates that incorporating THC damages always
increases the optimal carbon taxes in the E-DICE model compared to the original DICE model. It also
shows that while the lower enhanced damage estimates (1–10% gross world product (GWP) were there
to be a full collapse of the THC) lead to noticeable changes in the optimal carbon tax forecast, a very
Fig. 2. E-DICE optimal carbon taxes for selected maximum THC enhanced damages compared to original DICE. Optimal
carbon taxes increase as THC enhanced damage estimates increase. Relatively sizable enhanced damages are required before a
significant increase in carbon tax occurs. Final atmospheric CO2 concentrations (in model year 2350) are shown for each run in
parts per million by volume (ppmv).

---

438
M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
large damage estimate for THC collapse (25% GWP) is required before a sizable change in near-term
control rates appears 3 .
Final (model year 2355) CO2 concentrations shown on the graph similarly reflect sizable changes only
with large enhanced damages. Only then are significant controls on emissions implemented — which are
needed if atmospheric concentrations are to be significantly reduced relative to the case without enhanced
damages. All the simulations produce more than a tripling of CO2 in the distant future because they are
based on conventional economy, technology and population assumptions. These are certainly only one
set of social scenarios that are possible, but in this study we concentrate on the abrupt climate change
issue, and do not argue the plausibility or desirability of such conventional scenarios (see IPCC, 2000 for
alternative scenarios or Schneider, 2001 for discussion of the likelihood of future climate changes that
might be inferred from such scenarios).
5. The discount rate
A prime reason for the more than tripling of CO2 concentrations in most DICE runs is the discount rate
of the DICE model. Discounting in DICE is governed by two terms, growth in income in combination
with declining marginal utility of income, and a pure rate of time preference (PRTP). Nordhaus sets
the PRTP equal to 3% per year so as to obtain an effective discount rate equal to 6% per year (Toth,
1995; see Appendix A). Because of geophysical processes, changes in THC overturning lag by many
decades behind the CO2 emissions that cause them. Thus, damages from THC changes are pushed far
into the future, making the present value of the enhanced damages in our model particularly sensitive to
discounting. Different but equally plausible PRTP can be chosen for the model (Arrow, 1996). Howarth
suggests setting the discount rate equal to the return on risk-free assets: 0.4% per year (Howarth, 2001).
To explore the sensitivity to alternative discounting assumptions, we employ a range of PRTPs: 1.5, 3%
(original DICE default), and 4%. An elimination of time preference (PRTP = 0) that treats all time
periods equally is also considered (Azar and Sterner, 1996; Rabl, 1996).
Another alternative to changing the classic discount rate is “hyperbolic discounting”. A growing litera-
ture in the psychological and economic communities hypothesizes that the discount rate placed on future
projects declines as a project moves further into the future (Ainslie, 1992; Heal, 1997). In other words, a
person’s preferences vary greatly when asked whether a project should be initiated next year or the year
afterward, but vary to a far lesser extent for a project initiated 50 years from now or fifty one. According
to interpretations of this data, in the near future (e.g. 5 years) the preferred discount rate is very high.
However, as the time horizon extends, the preferred discount rate declines significantly. Laboratory and
field studies support such findings (Ainslie, 1992).
Tables 1–3 present optimal carbon taxes for the coupled E-DICE/SCD model runs, compared to the
uncoupled DICE model, for this range of discount rates and approaches. The original DICE model uses a
3 Keller et al. (2000b) describe scenarios in which a low control rate can be “optimal” even with a THC collapse because,
given their chosen damage functions and discount rate, it is cheaper to allow a collapse than to reduce emissions to preserve
THC. In some of our model runs, THC collapses because its preservation is not “optimally” efficient, for example, when climate
sensitivity is high, initial circulation strength is low, and the maximum damage estimate is low. Keller et al. use an enhanced
damage estimate of 1.5%, comparable to our low enhanced damages estimates, and thus have not broadly explored the range of
possible values for such significant parameters.

---

M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
439
Table 1
Optimal carbon taxes (US$/t C), 2005
Model:
Uncoupled
E-DICE/SCD 1% maximum
E-DICE/SCD 5% maximum
PRTP a:
DICE
damage enhancement
damage enhancement
0%
101.01
118.06
210.20
1.5%
17.06
19.14
27.61
3%
5.43
5.99
8.17
4%
2.90
3.19
4.29
Hyperbolic
67.39
78.54
137.80
a PRTP = pure rate of time preference.
Table 2
Optimal carbon taxes (US$/t C), 2055
Model:
Uncoupled
E-DICE/SCD 1% maximum
E-DICE/SCD 5% maximum
PRTP a:
DICE
damage enhancement
damage enhancement
0%
178.37
211.81
395.14
1.5%
38.79
44.36
67.90
3%
15.04
16.97
24.83
4%
9.14
10.28
14.83
Hyperbolic
153.24
180.98
331.87
a PRTP = pure rate of time preference.
PRTP of 3%, and the tables include the results of uncoupled DICE runs using other PRTPs or hyperbolic
discounting.
The numerical results in the tables confirm that our results are particularly sensitive to the choice of
PRTP, which affects both the magnitude of the carbon taxes and also the magnitude of the change in carbon
tax due to THC-enhanced damages. As is well-known, the lower the PRTP, the higher the present value of
distant future climatic damages, and thus, the larger the increase an incorporation of THC damages causes
in the optimal carbon tax. It is interesting that the E-DICE results appear more sensitive to the PRTP
Table 3
Optimal carbon taxes (US$/t C), 2105
Model:
Uncoupled
E-DICE/SCD 1% maximum
E-DICE/SCD 5% maximum
PRTP a:
DICE
damage enhancement
damage enhancement
0%
231.92
278.59
540.85
1.5%
54.47
63.30
101.78
3%
21.73
24.89
38.36
4%
13.35
15.76
22.89
Hyperbolic
214.77
256.81
483.71
a PRTP = pure rate of time preference.

---

440
M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
Fig. 3. “Cliff diagram” of equilibrium THC overturning varying PRTP and climate sensitivity. As PRTP increases, the climate
sensitivity threshold at which collapse of the THC occurs decreases (with a central value of initial overturning rate of 20 Sv). This
is because higher discount factors lead to lower emissions control rates, and thus allow a greater risk of climate change sufficient
to trigger abrupt non-linear climatic responses (the “threshold” value of climate sensitivity above which a collapse occurs should
not be taken as quantatively fixed, a possible impression from this figure. This diagram holds many other parameters relevant
to that threshold in the middle of their ranges, and thus collapse thresholds for each PRTP could be either higher or lower than
those in this figure if other relevant parameters were simultaneously co-varied).
than to varied physical parameters such as climate sensitivity (IPCC, 2001a). Fig. 3 is a “cliff diagram”
showing the equilibrium THC overturning for different combinations of climate sensitivity and PRTP
values. As the PRTP decreases, circulation is preserved for disproportionately higher climate sensitivities
since the lower PRTP allows larger emissions reductions and thus it takes a higher climate sensitivity to
reach the “cliff”. Conversely, the higher the discounting factor, the lower the climate sensitivity needed
to experience the overturning collapse.
Employing hyperbolic discounting yields optimal carbon taxes almost as high as the 0% PRTP case,
despite the fact that for the first few decades of the simulation the hyperbolic formulation actually has
a higher discounting effect than conventional exponential discounting. However, more than a century
hence, when the THC-enhanced damages are largest, the hyperbolic discount factor is smallest. Un-
der hyperbolic discounting, these large future damages affect near-term optimal carbon taxes signif-
icantly and outweigh the reduction due to the high discounting of the first half of the 21st century.
If hyperbolic discounting is a better empirical expression of the PRTP than conventional exponen-
tial discounting, the present value of long term highly damaging events will lead to much stronger
“optimal” near-term emissions reductions. Of course, as discussed earlier, the “correct” choice of dis-
count rate (or discounting formulation) is as much a value laden preference as it is a technical issue
about economic growth rates over time, opportunity costs or other factors that are part of the debate
(Arrow, 1996). Our IAM, which incorporates rapid climatic changes in the distant future created by

---

M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
441
choices made and implemented over the 21st century, highlights the policy consequences of such value
choices.
6. Emergent properties of coupling physical and social sub-models
One of the main purposes of the use of integrated assessment models is to characterize a range of
possible events and test the sensitivity of outcomes like THC weakening to a variety of plausible assumed
parameter values or baseline assumptions. Such analysis is also important because it can reveal emergent
properties of the studied system — in this case the coupled social–natural system of climate, ocean, and
economy. In this spirit, we summarize our results by constructing a broad (but not full) range of scenarios
with our E-DICE model, based on picking moderately high and low estimates for several parameters
simultaneously, so that we cascade our uncertainties on the “high” and “low” sides for illustrative purposes
(Fig. 4). We do not claim these to be defined range limits, as we have not conducted a formal procedure to
estimate the subjective probabilities of these high and low cases, let alone to investigate possible outlier
events beyond the range limits we explore here (e.g. see discussions in Moss and Schneider (2000);
Schneider (2001)). The THC profiles corresponding to DICE runs are the results of uncoupled SCD
model runs using the optimal CO2 concentrations from DICE.
For our “lower bound” case we allow the cascade of low climate sensitivity (1.5◦C for 2 × CO2) with
high initial THC overturning rate (25 Sv), low enhanced damages (1% additional GWP loss from a total
collapse), and baseline DICE PRTP (3%). The high initial overturning rate and low climate sensitivity
decrease the likelihood of a THC collapse (Schneider and Thompson, 2000). Fig. 4 shows (filled square
curve) that the reduction in THC over the 200-year period of the simulation is not trivial (from 25 to
about 15 Sv), but at 2200 the THC is still far away from an abrupt collapse. Carbon taxes for this case
are shown in the lower panel of the figure, also with filled square (optimal taxes for this DICE run start at
about US$ 4/t C and end at about US$ 24/t C). The unfilled square curves show the results of E-DICE for
these same parameter choices. Note that the enhanced E-DICE damages are sufficient to slightly increase
the optimal carbon tax relative to the DICE run without THC damages. The increased carbon tax lowers
emissions slightly but in this “lower bound” case the small decline produces only a trivial reduction to
the THC weakening.
The triangle curves represent an intermediate case, with 20 Sv initial overturning rate, 5% enhanced
maximum damages, 3◦C climate sensitivity for CO2 doubling and a 2% PRTP. Note that the higher climate
sensitivity produces a much greater reduction in THC strength, and thus much larger damages in E-DICE
relative to DICE. Thus, the optimal carbon taxes for E-DICE are substantially larger than those for DICE
— and likewise the mitigation of THC weakening is more noticeable, but still not very sizable.
Our “upper bound” case (filled circles for DICE and open circles for E-DICE) use a 15 Sv initial THC
overturning rate, a 4.5◦C climate sensitivity for CO2 doubling, 10% maximum enhanced damages and a
1.5% PRTP. The high impact case is the most interesting because it shows why abrupt non-linear changes
can make such a large difference to integrated assessment conclusions. Note that the original DICE run
shows a rapid decline in THC almost immediately after 2000, and there is an “abrupt” collapse of the
THC — “abrupt” meaning over a decade or so — around 2100. The rapidly decreasing profile of THC
— filled circle curve on the upper panel of Fig. 4 — shows that enhanced damages will be much larger
than for our lower and intermediate cases since the percent change in THC is very large and because the
changes occur earlier, and thus, will have a much higher present value, even with exponential discounting.

---

442
M.D. Mastrandrea, S.H. Schneider / Climate Policy 1 (2001) 433–449
Fig. 4. Optimal carbon taxes for E-DICE and DICE with a range of parameter values, and the THC overturning profiles associated
with those scenarios. The curves delineate a range of solutions, and illustrate the sensitivity of the E-DICE model to the PRTP.
As the PRTP decreases, the increase in carbon tax grows when THC enhanced damages are included. The circle curves illustrate
an emergent property of this coupled socio–natural system. Under the optimal DICE run (filled circle curve), THC collapses
abruptly. When the potential THC damages are incorporated, the increase in optimal carbon tax reduces emissions sufficiently to
prevent the collapse (open circle curve). However, when PRTP is increased from 1.5 to 1.8% or greater, the discounted present
value of the enhanced damages is insufficient to lower emissions below the threshold that causes abrupt THC collapse.
As a result, the carbon taxes for the E-DICE case (open circle curve in lower panel of Fig. 4) are much
larger than for the DICE case in which the abrupt THC shutoff occurs. Note that this shutoff in THC
occurred in conventional DICE despite the carbon tax seen in the lower panel with filled circles — starting
about $14/t C and rising to about US$ 68/t C in 2200. That tax — based on a damage function that was

---

Document Outline

• Integrated assessment of abrupt climatic changes
  • Introduction: context of the use of integrated assessment models for climate policy analysis
  • The DICE model
  • The E-DICE model
  • Results with Nordhaus discounting
  • The discount rate
  • Emergent properties of coupling physical and social sub-models
  • Conclusions
  • Acknowledgements
  • DICE and E-DICE
  • Simple climate demonstrator (SCD)
  • E-DICE/SCD
  • Discounting
• Model improvements needed
  • References

---