Geoscience Reference
In-Depth Information
can have an important effect on runoff generation but it has proven very difficult to upscale such in-
formation to larger scales (though see Jackson
et al.
(2008) for a recent attempt to do so). There have
been many studies that report predictions of the hydrological impacts of catchment change but these
have been almost entirely based on fitting new parameters with the hindsight that change is known to
have occurred or speculative scenarios of future change without later verification (and mostly without
allowing for uncertainty in the modelling process). Uncertainties in basic hydrological data can make it
difficult to identify change (Brath
et al.
, 2006). In fact, the experience in groundwater modelling, where
there have been a number of such
post-audit
verification studies, is not too promising in this respect.
Most predictions of groundwater responses have been shown to be wrong (Konikow and Bredehoeft,
1992). The result has been a continuing discussion about the possibility of validation of groundwater
models, much of which is also relevant for rainfall-runoff modelling (see, for example, the work of
Oreskes
et al.
, 1994).
8.9 Changing Risk: Climate Change
There are literally hundreds of papers that have been published in peer reviewed journals that have applied
rainfall-runoff models to the problems of assessing the impacts of climate change on stream discharges.
Some of this work is summarised in the latest hydrological impacts review of the Intergovernmental
Panel on Climate Change (IPCC) (Parry
et al.
, 2007). Such work has attracted significant amounts of
funding, driven by the requirements of policy makers who need to respond to the potential impacts of
future changes. I will not review that body of work here, as I believe that it is not generally (as yet) fit
for purpose (see also Hall, 2007; Wilby
et al.
, 2008; Beven, 2011).
For it to be fit for purpose, it should reflect the uncertainties in the future scenarios for catchment
and climate change and the uncertainties in representing rainfall-runoff processes. Very few of the
published studies do this in an adequate way, in part because they are constrained by the limitations of
the various components in the uncertainty cascade. As noted in Section 8.8, there are very important
limitations in the representation of the impacts of land management and other catchment changes on
runoff processes. These are probably dominated, however, by the limitations of predicting future climate
inputs. The science of climate prediction has progressed dramatically in the last decade. There are now
finer resolution coupled ocean-atmosphere models at the global scale; there are high-resolution regional
climate models nested within these global models; and there are ensembles of model predictions from
both multiple implementations of models in different research centres and “stochastic physics” ensembles
for single models.
As an example, the UK Climate Impacts Programme now provides, in its UKCP09 product, a prob-
abilistic interpretation of potential future changes under different IPCC emission scenarios, with maps
of the 10%, 50% and 90% projections for various variables. These are based on 11 fine-resolution dy-
namic regional climate model runs and 300 coarser resolution decadal regional model runs that have
been merged and bias-corrected relative to a 1961 to 1990 baseline period using a Bayesian Gaussian
emulator method to provide spatial posterior distributions for predicted variables on a 25 km grid for the
whole of the UK (see http://ukcp09.defra.gov.uk/). These projections have been used directly as inputs
to hydrological models (e.g. Bell
et al.
(2007a) used the G2G model of Section 6.2). These probabilistic
projections are given a qualitative interpretation. It is expected that the change is
very unlikely
to be
less than the mapped 10% projection and
very unlikely
to be greater than the mapped 90% projection;
while the 50% projection can be taken as a current best estimate of the projected changes. That is the best
current assessment available, but making direct use of these probabilities is to treat them as if the model
was correct and the range of potential outcomes was complete. This is not the case (see also Hall, 2007).
UKCP09 also provides a weather generator that can be used to generate weather sequences for any
5 km grid in the country. The weather generator has been fitted to interpolated weather data for the
