Memorandum submitted by Professor Keith
Beven (FL 148)
EXECUTIVE SUMMARY
This evidence refers to four issues: land management
and flood runoff generation, flood forecasting, and the impacts
of climate change on flooding, and institutional issues.
— Links between land use management
and flood runoff generation are complex and strong linkages have
only been seen at small scales. It is difficult to identify evidence
of the impacts of change on floods at larger scales given large
climatic variability. However, every opportunity should be taken
to encourage the use of small scale mitigation measures, eg. as
part of farm management plans.
— Land use planning does have a major
effect on costs of flooding. Allowing development in flood risk
areas with associated defences will not necessarily protect against
all floods, and must take account of whole catchment knock-on
effects.
— The National Flood Forecasting System
is a valuable development but there now needs to be a spread of
good practice in modern adaptive flood forecasting to all catchment
areas.
— In small catchments improvements
in forecasting rainfall inputs will be the only way of providing
forecasts with sufficient lead time for people to react.
— There is no strong evidence that
we are yet seeing the impacts of climate change rather than climate
variability. There have been periods in the past of increased
frequency of flooding. Predictions of the impacts of climate change
must take account of the inherent uncertainties in the estimates.
— The Environment Agency has not sufficiently
grasped the issue of uncertainty in both its own data and the
model predictions commissioned from consultants.
EVIDENCE
1. My name is Keith Beven. I have been Professor
of Hydrology and Fluid Dynamics at Lancaster University since
1991. I have previously worked at the Institute of Hydrology,
Wallingford (now part of the NERC Centre for Ecology and Hydrology),
and the Universities of East Anglia, Leeds, and Virginia. I have
also been a visiting Professor at the University of California,
Santa Barbara, USA, the Catholic University of Leuven, Belgium,
the Ecole Polytechnique Federale de Lausanne, Switzerland, and
Uppsala University, Sweden.
2. My work as a hydrologist has been primarily
concerned with rainfall-runoff modelling, flood frequency estimation,
flood inundation estimation, flood forecasting and estimating
uncertainty associated with model predictions. I have published
300 papers and authored or edited 7 books. I am currently the
lead investigator on the Defra FD2120 project on "Analysis
of historical data sets to look for impacts of land use and management
change on flood generation" and the theme leader for Risk
and Uncertainty in the EPSRC/Defra/EA Flood Risk Management Research
Consortium (FRMRC). I have previously contributed to the Defra
FD2114 review of land use effects on flood runoff generation and
the Defra FD2118 review of Board Scale Modelling.
My evidence refers to 4 issues: land management
and flood runoff generation, flood forecasting, and the impacts
of climate change on flooding, and institutional issues.
LAND MANAGEMENT
AND FLOOD
RUNOFF GENERATION
3. My involvement in the Defra FD2114 and
FD2120 projects has shown that given the natural variability of
the climate, it is very difficult to identify clear trends in
flood runoff generation except in very local circumstances eg
where there has been significant urbanisation or forest harvesting
on small catchments. On larger catchments, such changes are more
piecemeal and will have a smaller effect on the overall response,
making the effects of change difficult to identify given the natural
variability of the climate and the uncertainties in rainfall and
discharge datasets.[59]
Patterns of change in land management will also interact with
channel and drainage management effects. In fact, it is possible
that such changes will have different effects on runoff generation
under different conditions. Field drainage and upland gripping,
for example, will tend to increase the storage capacity of the
soil prior to a flood event (reducing runoff) but once the soils
are saturated, the drains and grips will provide faster pathways
to streams and the main channels. Blocking the grips will slow
down the time it takes runoff to get to the stream, but will mean
that the soils are more likely to saturate and provide a higher
volume of runoff. Speeding up runoff that would normally contribute
to the peak of a flood would help reduce the flood peak; slowing
runoff that would normally discharge from a catchment before a
flood peak, might increase the magnitude of a peak. The processes
are complex, and the spatial patterns of measures need to be taken
into account in catchment flood management plans (hence, as in
the Carlisle flood in 2005, the perception on the part of some
residents that flood defences in one part of the catchment had
made the flooding worse elsewhere). Scaling up local changes to
the catchment scale is a topic of research in the new FRMRC2 project,
but predicting the impacts of land use change on floods is likely
to be highly uncertain for the foreseeable future.
4. Small scale experiments do demonstrate
that particular land uses and management strategies do increase
flood runoff, even if the effects cannot easily be identified
at larger scales. There are also small scale experiments that
show that management strategies (eg tree planting or ploughing
across runoff pathways to increase infiltration) and local runoff
detention structures can reduce runoff. What is not clear is the
relative costs and benefits of encouraging local detention schemes
in both rural and urban developments in reducing flood risk at
larger scales. However, all opportunities to create joint benefits
to reduce runoff (eg the siting of tree planting in Farm Environment
Plans) should be taken and guidelines for good practice in reducing
runoff developed.
5. Land use planning also has a direct impact
on flood damages. The very rapid increrases in flood damages over
the last three decades have been due much more to the development
that has been allowed within flood risk areas than to any signficant
increase in the occurrence of flooding. Of course, such developments
are often only allowed when there is also some flood protection
undertaken, but local planning committees do not necessarily appreciate
the knock-on effects that protection in one location might have
on the risk of flooding downstream (and possibly outside their
Authority area). The Environment Agency can only advise planning
committees and in many cases their advice against developments
has not been heeded. In addition, it is often not properly appreciated
that even if a development is protected against the "one
in one hundred year" event, there is a 1% probability that
a bigger event will occur next year, and the year after, and the
year after that . . . In Carlisle, the plans for flood defences
that were on display at the time of the event would not have protected
the town against the flood which was bigger than the "100
year" event. Flood proofing of new developments in a way
that retains maximium storage capacity for the flood plain while
minimising the costs of damage should be considered as a recommended
strategy in considering the potential for future floods. There
may be both building regulation and insurance mechanisms for promoting
proofing rather than defense strategies.
FLOOD FORECASTING
6. The investment in the National Flood
Forecasting System (NFFS) by the Environment Agency will be an
extremely valuable foundation towards improving flood forecasts
in the future by putting all the different forecasting systems
onto a common foundation. The models that have been implemented
are, however, mostly legacy models. This has the advantage that
there is sometimes local experience of their use and limitations
in the past. It has the disadvantage that the models are mostly
not using up-to-date data assimilation and uncertainty estimation
in presenting results to decision makers for flood warning purposes.
There now needs to be a process of spreading best practice adaptive
forecasting methods across the NFFS.
7. An important concept in flood forecasting
is that of "lead time". A forecast needs to have sufficient
lead time to be useful to allow warnings to be given and people
to react. In larger catchments in the UK, lead times of 6 to 36
hours can be achieved by forecast systems using rainfall inputs,
but in smaller catchments (such as Boscastle) that might be subject
to flash floods, the only way lead times can be increased would
be by basing predictions on rainfall forecasts. The MetOffice
NIMBUS scheme for combining propagation of rainfall radar estimates
into the future with NWP rainfall estimates is adequate for detecting
the potential for high rainfalls that could cause local flooding
but not yet adequate for local forecasting with sufficient lead
time to enable a public response to warnings. This might improve
as the resolution of atmospheric NWP models reduces and new generation
radar systems come on-line, but this is still a research issue.
We have tried to test the use of high resolution NWP for flood
forecasting purposes, but because of the difficulty of running
the MetOffice Unified Model outside its normal range and computing
environment, have been able to look at only a very limited sample
of events. A more general evaluation will be constrained by the
limited numbers of sub-daily raingauge measurements in many catchments.
The point accuracy of such measurements is still needed to complement
the more general but much less accurate (especially in extreme
events) radar coverage.
IMPACTS OF
CLIMATE CHANGE
ON FLOODING
8. Floods occur rarely. They are also difficult
to measure. Under extreme conditions gauges get washed away or
overtopped and carrying out check measurements at gauging sites
can be both logistically difficult and dangerous. Thus the data
available on flood discharges is limited and of variable quality.
We cannot control the occurrence of floods in time, and improving
the quality of measurements would be expensive. The consequence
is that we should expect that any predictions about flood levels
and discharges (for example, of the 1 in 100 year or 0.01 annual
probability of exceedence event) will be associated with significant
uncertainty. We should also remember that statistically there
is a finite (if small) probability that the next event will be
bigger than the design flood for defences (as noted above in respect
of Carlisle). A larger event does not necessarily indicate a change,
but can arise within the natural variability of flood events.
9. It is convenient to analyse the limited
data that we do have statistically but as well as the uncertainty
about flood magnitudes, there will also be uncertainty about the
form of the distribution of events, and how the magnitude of events
might be changing over time with changes in catchment characteristics
(eg continuing urbanisation), land and channel management strategies
(eg the recent publicity over the reduction in the EA maintenance
programme in small channels) and the potential effects of climate
change. The statistical analysis assumes stationarity, we suspect
that the distribution of events might well be changing over time.
10. Analysis of such changes is highly constrained
by the quantity and quality of the data available and the variability
of the natural process. Thus, as with the land use effects, it
is very difficult to distinguish any changes in flood characteristics
by an analysis of the available data. There is some evidence that
in the last 10-15 years, the number of flood events exceeding
different threshold discharges has been increasing as a result
of increasing winter rainfalls, particularly in the west of the
country. However, there does seem to have been other periods of
increased frequency of flooding in the past[60].
There is no strong evidence therefore to conclude, yet, that we
are seeing the effects of climate change rather than climate variability.
We should therefore be very wary of drawing strong conclusions
based on model estimates of changing flood frequencies based on
estimates of change from GCM models. We know that such models
have difficulty in predicting current extremes adequately (one
of the reasons for moving to ensembles of models in UKCIP08) and
should expect their predictions of future conditions to be associated
with significant uncertainty. Such uncertainties are not necessarily
only statistical in nature, but making realistic estimates of
the uncertainties might change the nature of decisions that are
based on the model predictions.
11. This clearly does not mean that, as
more data are collected in the future, we might not see detectable
changes in flood frequency, only that those changes are and will
be difficult to distinguish from natural variability. Some allowance
should be made for the potential for future change but the current
EA guidance to allow for 20% above the current estimate of the
100 year discharge needs revisiting to allow for the uncertainties
in making such an estimate. The 20% figure is acting effectively
as a precautionary value. Given the uncertainties involved in
estimating flood exceedence probabilities, a precautionary rather
than model prediction approach, might well be an appropriate but
will require a re-evaluation of the decision making framework
and planning guidance.
INSTITUTIONAL ISSUES
12. Since the formation of the Environment
Agency there has been a significant loss of experienced staff
to consultancies. In general, the preparation of flood risk maps,
and catchment flood management plans are contracted out by the
EA to consultants. Reorganisation has also meant that some understanding
of local conditions in particular catchments has been lost (and
indeed, the recent reorganisation means that in some cases, responsibilities
for some catchment areas are now split between different sections
within the Agency). There appear to be few mechanisms in place
to record this local understanding when expertise is lost due
to retirements, transfers or loss to consultancies. While it is
entirely appropriate that the EA should make use of the expertise
available in consultancies, it is not clear that there is sufficient
expertise to allow critical assessment of the reports produced
by consultants which often depend on complex and uncertain model
calculations and uncertain data. The Agency has not fully grasped
the uncertainty issue (or how to constrain it): it does not ask
consultants to estimate uncertainties associated with model predictions;
it does not assess the uncertainties associated with its own data
sets. It could be more pro-active in this respect. There will
be a cost involved in taking account of uncertainties in decision
making, but the robustness of decisions might be improved.
Prof Keith Beven
November 2007
59 Defra, 2007, FD2120: Analysis of historical data
sets to look for impacts of land use and management change on
flood generation, Interim Report. Back
60
See Beven, K J (1993), Riverine flooding in a warmer Britain,
The Geographical Journal, 159, 157-161. Back
|
