Memorandum by Professor Richard S J Tol,
Hamburg, Vrije and Carnegie Mellon Universities
ABSTRACT
Probabilistic scenarios cannot be avoided. If
one is mainly interested in quantitative results, this is obvious.
A number is meaningless without a confidence interval. If one
is mainly interested in qualitative results, probabilistic analysis
is not necessarily called for, or so it seems. However, an insight
is meaningless if it is not robust. Alternative scenarios are
needed to test robustness. Alternative scenarios should span the
range of not implausible futures. That range can only be derived
from probabilistic scenarios. Besides the question in its title,
this paper investigates whether the SRES scenarios span the range
of not implausible futures.
The SRES scenarios were severely criticised
by Castles and Henderson. That critique focused on the use of
exchange rates. The choice of exchange rates, however, does not
matter much (relative to the other uncertainties) for carbon dioxide
concentrations and hence for climate change. On the other hand,
the choice of exchange rates does matter for assumed development
pathways, and hence vulnerability to climate change, and for the
distribution of carbon dioxide emissions, and hence the distribution
of mitigation costs and responsibilities. The choice of the discount
rate matters so much, primarily because of the convergence of
per capita incomes and emission intensities assumed in the SRES
scenarios.
The SRES scenarios were built with models that
were originally designed for the analysis of energy policies.
Such models use scenarios, but here they were used to build scenarios.
Using and building scenarios are different things. Also, for building
emissions scenarios, more knowledge is required than knowledge
of energy systems. Furthermore, the models used were calibrated
to data sets with a relatively short time span. Because of funding
constraints, validating the models against longer time series
was never a priority.
I collected long term data on population, per
capita income, energy use, and carbon dioxide emissions from energy
use. I plotted these data together with the four alternative projections
according to the IMAGE 2.2 model. I used the data to estimate
the Kaya identity in differential form. I extrapolated the model,
and used the forecast error to calculate the relative probabilities
of the four SRES scenarios. The following results emerge.
The population scenarios are largely in accordance
with history. It is peculiar that the A1 and B1 scenarios have
the same populations, even though their economies are very different.
The per capita income scenarios for developed
countries are largely in accordance with history. For developing
countries, this is not the case. China's economy, for instance,
has been stagnant if not declining for most of the last five centuries.
Only the last two decades saw rapid economic growth in China.
All four scenarios continue the pattern of most recent times.
Rapid economic growth is also foreseen, in all scenarios, for
other developing countries. For Africa, this is a clear break
with the past. The four scenarios foresee rapid convergence of
incomes across the world in the current century, even though the
past two centuries witnesses income divergence.
The projections of energy intensities only partly
conform with history. The fastest decreases of energy intensities
in the scenarios are not faster than was observed in the past.
However, the scenario foresee decreases only, even though energy
intensities have increased as well in the past. Energy intensities
across the world converge in all scenarios, not faster than the
maximum observed rate, but always faster than the minimum observed
rate.
The projections of emission intensities for
individual regions span the range of observed past behaviour.
All scenarios foresee further convergence of emission intensities
first—in line with past observations—but divergence
later. The scenarios all show the same qualitative pattern of
convergence, and diverge only minimally quantitatively.
The above pattern suggests that the SRES modellers
know a lot about the supply side of the energy system, but less
about the demand for energy. Their knowledge of economic development
is lacking. Their demographic expertise is sound, but strangely
separated. My personal knowledge of the SRES modellers confirms
this assessment.
The relative probabilities of the four alternative
SRES scenarios confirms this picture. The scenarios for the period
2000-50 for populations each have probabilities of over 10 per
cent; for emission intensities, the A2 and B1 scenario are most
likely, but the other two scenario have more than a 0.1 per cent
chance. For the scenarios for per capita incomes and energy intensities,
the situation is different. Here, the A2 scenario is by far the
most likely, and the other three scenarios are extremely unlikely.
For the period 2050-2100, a similar picture
emerges, albeit less extreme. The Kaya trend projection method
used here is, however, less suitable for projecting this far into
the future.
If one applies the same trend projection method
directly on emissions, the relative probabilities SRES scenarios
are all acceptable (that is, not lower than 10 per cent chance)
for the period 2000-50. This suggests that the errors in the underlying
scenarios largely cancel each other out.
The following conclusions can be drawn. The
SRES scenarios are not equally likely. The A2 is by far the most
realistic. The SRES scenarios do not span the range of plausible
futures. The range of emissions can be somewhat wider, and the
range of underlying development can be much wider. The SRES scenarios
do not accord with past trends. On the one hand, this makes for
interesting scenarios. On the other hand, it is odd that all SRES
scenarios break with past trends at the same time, and that this
trend break is sometimes at the point where data end and scenarios
start.
The SRES scenarios are therefore useful as emissions
scenarios. When used in climate models, they more or less span
the range of not implausible futures. The SRES scenarios are less
useful for climate change impact analysis, at least for those
studies in which impacts do not only depend on climate change
but also on vulnerability and hence development. The SRES scenarios
are less useful for emission abatement studies, at least for those
studies that use regional models and are interested in the distribution
of mitigation costs and responsibilities.
15 January 2005
| 
