Geoscience Reference
In-Depth Information
baseline period. Weather realisations for future decades can be generated by implementing change factors
based on the probabilistic projects from the climate models. Such a weather generator is clearly suited
for driving a rainfall-runoff model to represent future conditions, including the possibility of running
multiple realisations to reflect the uncertainty in the future conditions. This has been done, for example,
by Manning
et al.
(2009).
So why would I suggest that such simulations are not fit for purpose? We have seen earlier in this topic,
particularly in Chapter 7, that there are real concerns about how well we can constrain the representation
of runoff processes in rainfall-runoff models when so many of the uncertainties in the modelling process
are epistemic, due to lack of knowledge. This is even more the case for climate models which are
known to be inadequate predictors of the baseline period in many parts of the world. In particular, for
a hydrologist, they do not adequately represent rainfalls to the extent that many hydrological studies of
the impacts of climate change need to use significant bias corrections with respect to current conditions
(see, for example, Leith and Chandler, 2010). Underpredictions of 50% of observed rainfalls are not
uncommon in upland areas, with even greater problems in parts of the Alps and Andes. These bias
corrections are then simply assumed to hold in the future. Climate models also do not adequately predict
monsoon rainfalls, el Ni no and la Ni na effects, or the North Atlantic Oscillation in the baseline period.
For hydrologists interested in floods, it is a particular concern that they do not do well in predicting
extreme rainfalls. Like our hydrological models, they are wrong and are known to be wrong (whether
that be the result of scale effects, sub-grid rain, snow and cloud parameterisations, the simplicity of land
surface parameterisations, inadequate representations of heat exchange with the oceans, anthropogenic
forcings other than greenhouse gases (Pielke
et al.
, 2009) or other causes).
We expect, of course, that with more research money devoted to climate modelling, more computer
power devoted to climate modelling at finer grid scales, better land surface parameterisations (that includes
the hydrology and hydrologists should still wince about how it is being represented!) and improved
understanding of other process representations in the models, the projections of the next generation of
climate models might well be better. But fit for purpose in assessing the impacts of climate change still
seems a long way off.
That is not to say I do not believe in climate change. I believe in climate change. I am also worried
about the possibility that the climate system, as a nonlinear dynamic system, might be subject to mode
of behaviour shifts instigated by variability that is not being predicted by the current generation of global
climate models (GCMs) (see, for example, Smith, 2000). We know that there have been rapid modal
shifts in the past, before any significant anthropogenic greenhouse gas inputs to the atmosphere (Alley
et al.
, 2003). That suggests that we should plan to adapt to the possibility of change, despite the fact that
we should be sceptical of climate model projections (see also Prudhomme
et al.
, 2010).
How, therefore? This depends on how risk averse or risk accepting we are prepared to be and that
will often be a matter of how much we are prepared to spend on an adaptation strategy. Being risk
averse generally requires more expensive measures than being risk accepting. But we can consider how
expensive the required adaptation might be for different scenarios of future change, more or less extreme,
quite independently of any climate model projections. Making such decisions does not require specifying
the probabilities of potential outcomes; there are other decision frameworks for adapting to change (e.g.
Beven, 2009, Chapter 6). In that way, it is possible to plan a response to different magnitudes of change in
terms of costs (and benefits) that might be robust with respect to future change. This does not necessarily
mean the over-design of hard infrastructure: other societal policies that permit the ability to adapt over
time might well be more cost-effective.
A general principle commonly followed in policy formulation is that the response should be propor-
tionate to the risk so that ideally we would wish to evaluate the probability associated with each magnitude
of change. UKCP09 is presented in this way, but remember that these projections are not a representation
of the odds of climate actually turning out that way - they are, rather, the probabilities of the model
