2.6.1. Treatments of Uncertainties in Previous IPCC Assessments
The IPCC function is to assess the state of our understanding and to judge
the confidence with which we can make projections of climate change and its
impacts. These tentative projections will aid policymakers in deciding on actions
to mitigate or adapt to anthropogenic climate change, which will need to be
re-assessed on a regular basis. It is recognized that many remaining uncertainties
need to be reduced in each of (many) disciplines, which is why IPCC projections
and scenarios are often expressed with upper and lower limits. These ranges
are based on the collective judgment of the IPCC authors and the reviewers of
each chapter, but it may be appropriate in the future to draw on formal methods
from the discipline of decision analysis to achieve more consistency in setting
criteria for high and low range limits (McBean et al., 1996; see Raiffa,
1968, for an introduction to decision analysis).
Although the SAR on impacts, adaptation, and mitigation (IPCC, 1996b) explicitly
links potentially serious climate change with mitigation and adaptation assessment
in its Technical Summary, the body of the report is restricted mostly to describing
sensitivity and vulnerability assessments (see also Carter et al., 1994).
Although this methodology is appropriate for testing sensitivity and vulnerability
of systems, it is poorly suited for planning or policy purposes. IAMs available
to SAR authors (e.g., Weyant et al., 1996) generate outcomes that are
plausible but typically contain no information on the likelihood of outcomes
or much information on confidence in estimates of outcomes, how each result
fits into broader ranges of uncertainty, or what the ranges of uncertainty may
be for each outcome (see Chapter 1 and Section
2.4 for further discussions of integrated assessment issues). However, several
studies since the SAR do use probability distributions (e.g., Morgan and Dowlatabadi,
1996, and citations in Schneider, 1997).
IPCC Working Group I (WGI) in its contribution to the SAR (IPCC, 1996a) uses
two different methods or techniques to estimate climate change: scenarios and
projections. A scenario is a description of a plausible future without estimation
of its likelihood (e.g. the individual IS92a-f emission scenarios or climate
scenarios generated by GCMs in which a single emission path is used). Scenarios
may contain several sources of uncertainty but generally do not acknowledge
them explicitly
Careful reading of the SAR WGI Technical Summary (IPCC, 1996a) reveals that
the term projection is used in two senses:
- A single trajectory over time produced from one or more scenarios (e.g.,
projected global temperature using the IS92a emissions scenario with a climate
sensitivity of 2.5°C).
- A range of projections expressed at a particular time in the future, incorporating
one or more sources of uncertainty (e.g., projected global warming of 0.8-3.5°C
by 2100, based on IS92a-f emission scenarios and a climate sensitivity of
1.5-4.5°C at 2xCO2).
Projections are used instead of predictions to emphasize that
they do not represent attempts to forecast the most likely evolution of climate
in the future, only possible evolutions (IPCC, 1996a, Section F.1). In
the SAR, projection and scenario are used to describe possible
future states, with projections used mainly in terms of climate change and sea-level
rise. This usage defines climate projection as a single trajectory of
a subset of scenarios. When used as input into impact assessments, the same
climate projections commonly are referred to as climate scenarios.
 Figure
2-1: Schematic depiction of the relationship between "well-calibrated"
scenarios, the wider range of "judged" uncertainty that might
be elicited through decision analytic techniques, and the "full"
range of uncertainty, which is drawn wider to represent overconfidence in
human judgments. M1 to M4 represent scenarios produced by four models (e.g.,
globally averaged temperature increases from an equilibrium response to
doubled CO2 concentrations). This lies within a "full"
range of uncertainty that is not fully identified, much less directly quantified
by existing theoretical or empirical evidence (modified from Jones, 2000). |
Projected ranges are constructed from two or more
scenarios in which one or more sources of uncertainty may be acknowledged. Examples
include projections of atmospheric CO2 derived from the IS92a-f emission
scenarios (IPCC, 1996a), global temperature ranges (IPCC, 1996a), and regional
temperature ranges (CSIRO, 1996). A range of projections will always be more
likely to encompass what actually will transpire than a single scenario. Although
projected ranges are more likely to occur than single scenarios, they are not
full-fledged forecasts because they incorporate only part of the total uncertainty
space. The relationship between scenarios and projected ranges as treated in
the SAR is shown schematically in Figure 2-1.
A projected range is a quantifiable range of uncertainty situated within a
population of possible futures that cannot be fully identified (termed "knowable"
and "unknowable" uncertainties by Morgan and Henrion, 1990). The limits
of this total range of uncertainty are unknown but may be estimated subjectively
(e.g., Morgan and Keith, 1995). Given the finding in the cognitive psychology
literature that experts define subjective probability distributions too narrowly
because of overconfidence (see Section 2.6.5.3), the inner
range represents the "well-calibrated" range of uncertainty. Thus,
the wider range of uncertainty represents a "judged" range of uncertainty,
based on expert judgments—which may not encompass the full range of uncertainty
given the possibility of cognitive biases such as overconfidence. Although the
general point remains that there is always a much wider uncertainty range than
the envelope developed by sets of existing model runs, it also is true that
there is no distinct line between "knowable" and "unknowable"
uncertainties; instead, it is a continuum. The actual situation depends on how
well our knowledge (and lack thereof) has been integrated into assessment models.
Moreover, new information—particularly empirical data, if judged reliable
and comprehensive—eventually may narrow the range of uncertainty to well
inside the well-calibrated range by falsifying certain outlier values.
If the full range of uncertainty in Figure 2-1 were known,
the probability of a particular outcome could be expressed as a forecast (provided
we can state the probability). Although there are significant sources of uncertainty
that cannot yet be quantified, decision analytic elicitation procedures (Section
2.4) can estimate the full range of uncertainties and conditional probabilities
(see Section 2.5.5 for an assessment of the state of the
science concerning human judgment). Conditional probabilities may be calculated
within a projected range even though the probability of the range itself remains
unknown.
Continues on next page
