CCN at Hydropredict 2010
CCN was the instigator for a special session at the Hydropredict 2010
international conference at the Czech University of Life Science, Prague, held on
September 23rd this year. The session was entitled “Uncertainty in Predicting
the Impacts of Catchment Change and its Implications for Decision Making”.
Abstracts for the papers in the session may be found below. Keith Beven gave
one of the keynote lectures (with co-authors Dave Leedal and Ruth Alcock of
Lancaster University and Neil Hunter, Caroline Keef and Rob Lamb of JBA
Consulting) on “Guidelines for Good Practice in predicting the future: the
Catchment Change Network”. The talk outlined the CCN Guidelines for Good
Practice decision framework approach to involving stakeholders in the
evaluation of risk and uncertainty in prediction, with examples of predicting
uncertain flood risk maps. A second keynote talk was given by Patrick Willems
of KU Leuven, Belgium together with Jan Staes, on “Impact of climate change on
river hydrology and ecology”. They presented an interdisciplinary approach to
evaluating the impacts of climate change, including using the daily outputs from
15 different GCMs to define low, medium and high scenarios of change (after
some statistical downscaling) in order to evaluate potential impacts on the
complexity of river ecology.
There were then a number of contributed talks, the first from Tobi Krueger
(University of East Anglia) on “Model limitations and prediction uncertainty in
the context of analytic-deliberative catchment management” in which he
outlined the experience of trying to keep multiple stakeholders involved in the
process of defining a phosphorous management strategy for two catchments in
the UK. Maria Polo from the University of Cordoba then presented a study in
uncertainty assessment of forecasting extreme values in a mountainous
catchment in Southern Spain. This was based on a weather generator calibrated
on current conditions and modified to reflect future changed conditions. A
presentation from Satish Bastiola from the National University of Ireland
looked at using multiple hydrological models within a GLUE and Bayesian Model
Averaging project in projecting the uncertainty in hydrological responses under
future conditions.
Jim Freer of Bristol University then presented an analysis of Regional Climate
Model precipitation simulations of past climate for the areas of three larger UK
catchments catchments. On many of the evaluation measures used, the RCM
predictions (without downscaling) did not match the observations particularly
well. He therefore queried how much faith we should put into their predictions
of a changed climate. Linda Speight of Newcastle University then gave an
analysis of the complexity of trying to make assessments of the changing impact
of catastrophic fluvial flooding events, including trying to allow for the
uncertainty in the breaching of flood defence structures. Finally, Keith Beven
presented a joint paper on behalf of Wouter Buytaert of Imperial College on
uncertainty in the regionalisation problem in extrapolating sets of parameter
values from gauged sites in either space or time (changed land use in future).
They argued that this should be treated as a learning process, with prior
estimates of changed parameters being refined as new measurements become
available.
The papers in this session served as a useful counterweight to many earlier
papers in the Hydropredict2010 meeting, in which the outputs from various
combinations of the available climate models ensembles were used to predict
future hydrology using rainfall-runoff models calibrated (usually optimised
without uncertainty) on historical data. While this is an obvious thing to do, and
it seems to be relatively easy to get research funding to do so because
governments are interested in getting advice about the impacts of future change,
most of these papers showed no critical assessment of the results of such a
process. There are two issues here (see also the CCN blog item). The first is
whether the GCM and hydrological models are fit for purpose. Since computing
power limits the number of GCM ensemble members, and knowledge constraints
limit the accuracy of hydrological models, it is important to distinguish between
models that are the best available and models that are fit for purpose. These are
not necessarily the same. The best available (GCM, hydrological, ecological)
models will not necessarily have the required accuracy to be fit for purpose. Jim
Freer’s talk suggested that the RCM ensemble available in UKCP09 might not be
fit for purpose in predicting effects on floods in the basins tested.
The second issue is that testing for change is a form of hypothesis testing.
Hypothesis testing should take account of the relevant uncertainties in
observations and modelling results. This was suggested nearly 20 years ago by
Wolock and Hornberger (1991) in assessing trends in hydrological data. The
lesson does not seem to have sunk in: many results were presented without any
evaluation of the uncertainties involved. The inferences being drawn about the
magnitude (and even the direction) of change might therefore be wrong. There
is a really question about whether these inferences are robust in respect of the
uncertainties in both the GVM/RCM and impact model predictions.
Reference
Wolock, D and Hornberger, G M, 1991, Hydrological effects of changes in levels
of atmospheric carbon dioxide, J. Fortecasting, 10:105-116.
Abstracts from the special session
Guidelines for good practice in flood risk mapping: The catchment
change network
K. J. BEVEN1, D. LEEDAL1, R. E. ALCOCK1, N. HUNTER2, C. KEEF2 & R.
LAMB2
1 Lancaster Environment Centre, Lancaster University, Lancaster LA1 4YQ, UK
2 JBA Consultants, Broughton Hall, Skipton, North Yorkshire, BD23 3AE, UK
k.beven@lancaster.ac.uk
Abstract The assessment and mapping of flood risk involves many different sources of
uncertainty. Many of these sources of uncertainty involve epistemic uncertainties that are not
necessarily easy to represent statistically. This can create problems for communication
between analyst and users when uncertain flood risk maps are being prepared. It is suggested
that one way of dealing with this problem is to define Guidelines for Good Practice in the
form of a set of decisions that must be agreed and recorded for later evaluation and review.
The Catchment Change Network (CCN) is a knowledge transfer project, funded by the UK
NERC, that aims to bring academic research and practitioners together to produce guidelines
for good practice for uncertainty estimation in predicting the future in the areas of flood risk,
water quality and water scarcity all of which involve important epistemic uncertainties. The
paper will set out the background to developing Guidelines for flood risk mapping and give
an application to a site in Yorkshire, UK.
Impact of climate change on river hydrology and ecology: case study
for interdisciplinary policy oriented research
PATRICK WILLEMS1, JAN STAES2 & PATRICK MEIRE2
1 Katholieke Universiteit Leuven, Hydraulics Division & Leuven Sustainable Earth (LSUE) research centre,
Kasteelpark Arenberg 40, BE-3001 Leuven, Belgium
Patrick.Willems@bwk.kuleuven.be
2 Universiteit Antwerpen, Ecosystem Management Research Group, Universiteitsplein 1c, BE-2610 Wilrijk,
Belgium
Abstract The impact of climate change on river hydrology and ecology is a subject that
receives increasing attention and has strong implication for hydrological, ecological,
economic and social policy. Because climate change affects such wide variety of disciplines,
pursuing research in this field requires an interdisciplinary approach. There is a need to
simultaneously understand and project the climate change, and to project and effectively deal
with its impacts on the present and future aquatic ecosystem, which presents a great challenge
to the global research community. This paper summarizes the experiences obtained by a
Belgian research project in which key experts from climatological, hydrological and
ecological research communities, as well as water managers and policy makers, are brought
together, in order to improve the decision making regarding the impact of climate change on
aquatic and floodplain ecosystems. The project focuses on a case study “Grote Nete”, which
allow us to adapt these relevant issues, while focusing on the combined information from
climate projections. Interdisciplinary linkages and feedback mechanisms in the interfacing
between climatology and hydrology, between hydrology and ecology and with water
managers and policy makers are discussed, as well as the current knowledge and interfacing
gaps and related challenges.
Model limitations and prediction uncertainty in the context of
analytic-deliberative catchment management: acceptance by
stakeholders and their role in improving model predictions
T. Krueger (1), A. Inman (2), K. Hiscock (1), L. E. D. Smith (2)
(1) School of Environmental Sciences, University of East Anglia, Norwich, NR4 7TJ, UK (t.krueger@uea.ac.uk,
phone +44 (0)1603 592041)
(2) Centre for Development, Environment and Policy, University of London (SOAS), Wye, TN25 5AH, UK
Abstract. It has increasingly been realised that the protection of water resources requires a ‘twintrack’ approach of scientific research and deliberative stakeholder engagement at the catchment scale
(Smith and Porter, 2009, DOI 10.1007/s10113-009-0102-z). Computer models can be used to facilitate
this process, but the prediction uncertainty resulting from the simplifications and stationarity
assumptions inherent in models and the paucity of data to parameterise, drive and test them should be
accounted for. There is little data, however, on how well stakeholders accept uncertainties and how
trust in models can develop. In this paper, we analyse both of these processes using case studies of
diffuse water pollution in the Broads in Norfolk and the upper Tamar catchment in southwest England.
It will be shown how gaps in process knowledge and data limitations led us to develop simple nutrient
export coefficient models, and how we quantified their uncertainties. The main thrust of the paper will
then report how stakeholder engagement with the modelling approach was facilitated, how stakeholder
expectations evolved, to what extent prediction uncertainties were accepted, and how ownership of and
contribution to the modelling approach developed. It will be discussed how models may be advanced
by incorporating local stakeholder knowledge and scientific expertise explicitly, and how this
knowledge can best be elicited and utilised. Our modelling approach presents exciting opportunities for
mutual learning, which can enhance both the societal responsibility of individuals and organisations as
well as the scientific enquiry.
Uncertainty assessment for long-term forecasting of extreme values
in streamflows due to catchment changes in a Mediterranean
mountainous watershed in Southern Spain
M.J. POLO
Fluvial Dynamics and Hydrology Research Group, Universidad de Córdoba, Área de Ingeniería Hidráulica,
Edificio Leonardo da Vinci. Campus de Rabanales. 14071 Córdoba,Spain.
e-mail: mjpolo@uco.es
M.A. LOSADA
Environmental Flow Dynamics Research Group, Universidad de Granada, CEAMA. Av. del Mediterráneo s/n.
18006 Granada, Spain
e-mail: mlosada@ugr.es
Abstract. The long-term forecasting of extreme values in streamflows is usually made from the return
period analysis of the available annual maxima floods at given sections along the river. Whereas
precipitation series usually extend up to 60-100 years and relatively numerous meteorological stations
are spread over the territory, discharge measurements are not so abundant and, when available, the data
series duration is very often not long enough to provide us with significant extension for accurate
results. Moreover, when prediction of the effects of catchment changes (i.e. soil uses or extreme events
such as fires, earthquakes, landslides…) needs to be made for planning purposes, the use of physical
distributed models is usually the only possibility, provided that current calibration is available and
assuming default contrasted values for the model parameters under the new conditions. To overcome
the lack of both long-time series and experimental and field data under the new sceneries of analysis,
this work presents a methodology to forecast long-term extreme discharge values and applies it to a
mountainous non-regulated basin in Southern Spain, where a significant relationship between annual
precipitation and annual maximum daily discharge was observed. Firstly, under the assumption that
precipitation occurrence is the main uncertainty source in the observed discharge variability, a
forecasting model has been built from Monte Carlo simulation of the annual precipitation in 40 yrinterval, its relationship with daily discharge and daily-instantaneous discharge transformation. The
derived probability functions for extreme instantaneous discharge at the study site provide us with a
stochastic forecasting for flood management and planning under the current conditions in the
catchment, and a more accurate approach than the return period deterministic estimation from the
available 12 yr-interval discharge data. Secondly, a proposed scenery of soil use change in the
catchment is included by a hydrological distributed model and the relationship between annual
precipitation and annual maximum daily discharged is corrected from the results of the simulation of
the historical time series available; the forecasting model described above is finally run on the new
functions, its results quantifying the associated uncertainty due to precipitation occurrence. Risk
analysis for flooding areas, and many other derived variables, can be easily drawn from such results.
Evaluation of the role of hydrological modelling uncertainties in
climate change impact assessments: a case study including basins
located in Ireland
Satish Bastola, Conor Murphy and John Sweeney
ICARUS, Dept of Geography, National University of Ireland, Maynooth, Ireland
Satish.Bastola@nuim.ie
Abstract. Assessment of the impact of climate change on water resources are subjected to a range of
uncertainties; associated with estimates of future emissions of greenhouse gases, uncertainties about
the response of the climate system to these changes at global and local scales, and uncertainties
associated with hydrological modelling. The uncertainties associated with hydrological models have
traditionally been given less attention in impact assessments until relatively recently. In order to
examine the role of hydrological model uncertainty (parameter and structural uncertainty) in climate
change impact studies and include a full consideration of impact model uncertainty two approaches
were used; Generalized Likelihood Uncertainty Estimation (GLUE) and Bayesian Model averaging
(BMA). Six sets of statistically downscaled climate scenarios derived from three global circulation
models and two emission scenarios, namely SRES A2 and B2, and four conceptual hydrological
models were used within the GLUE framework to define the uncertainty envelop for future estimation
of stream flow. The investigation on four Irish catchments shows that the role of hydrological model
uncertainty is remarkably high therefore it should be routinely considered in impact studies. The
average width of the prediction interval arising from uncertainties associated with parameterization of
hydrological modelling is nearly 40 % of the average flow and it increased nearly to 70% of the
average flow when different model structures are included in order to address the uncertainty in model
structure. Similarly, the overall uncertainty envelop increased nearly to 140% of the average flow for
the three periods, namely 2020s, 2050s and 2070s. The BMA was also used and compared with the
uncertainty envelope derived from the GLUE method. For the implementation of BMA, a total of 24
predictions obtained from six climate scenarios and four hydrological models obtained from GLUE
method were post processed using BMA. In BMA, the probability density function from each model at
any given time is modelled by a gamma distribution with heteroscedastic variance. The weight and
variance parameter of the BMA were estimated from the period 1970-1990 and subsequently used to
produce the probabilistic estimation of stream flow for future periods. The prediction intervals
estimated from both BMA and GLUE are comparable. The median prediction for three selected time
periods and four basins shows progressive increase of winter discharge when moving from 2020s to
2070s and progressive decrease in summer discharge from 2020s to 2070s.
Climate change ensembles driving modelling cascades with
uncertainty analysis to quantify the resultant impacts on flood risk.
Is this currently feasible?
Freer, J.E., F. Wetterhall, He, H. Cloke, Y.. Pappenberger, M. Wilson, G.
McGregor
1. Geography, King's College London, London, United Kingdom.
2. School of Geographical Sciences, University of Bristol, Bristol, United
Kingdom.
3. European Centre for Medium-Range Weather Forecasts, Reading, United
Kingdom.
4. Geography, University of the West Indies, St Augustine, Trinidad and
Tobago.
5. School of Geography, Geology and Environmental Science, University of
Auckland, Auckland, New Zealand.
Abstract. Quantifying the impacts of climate change on our water resources inevitably involves
multiple model predictions, where the resultant uncertainties are difficult (if not impossible) to
quantify. The question therefore remains how do we develop effective methodologies that are 'fit for
purpose' to inform policy makers and feed into integrated catchment management solutions, enabling
us to mitigate and plan for climate change? We suggest there is an urgent need to place more
importance on understanding our predictive limitations especially when dealing with extremes (such as
changes in flood risk) our current capability may not be 'fit for purpose' in the above context.
This study evaluates the impacts on flood inundation of a changing climate within a probabilistic
framework. Global and Regional Climate Models (GCM/RCM) are used to drive catchment
hydrological and river hydraulic models within a cascaded framework. We consider three analytical
steps that must be taken into account in order to successfully provide probabilistic projections of flood
inundation: (i) downscaling of the driving variables from the GCM/RCM due to the scale mismatch
between the model types; (ii) bias correction due to GCM/RCM deficiencies in modelling consistent
seasonal and inter-annual climate variability and (iii) uncertainties in the cascade of models.
A probabilistic climatic-hydrologic-hydraulic cascade flood prediction system was set up to identify
different sources of uncertainties and understand the way in which they propagate. Critical assessment
regarding the ability of RCM model simulations to drive such a model cascade is particularly
highlighted, proving a basis for discussion on our current ability to quantify climate impacts of extreme
behaviour by such methodologies. Flood events can be caused by different mechanisms, such as highintense rainfall events over a short period or intermediate precipitation over a longer time period and
the spatial pattern of precipitation. It is therefore important that these mechanisms are sufficiently
reproduced by RCMs in climate impact assessments. The study area was a meso-scale Upper Severn
catchment located in the Midlands Region of England. Multiple RCM predictions were obtained from
the ENSEMBLES project (16 ensembles) as well as the HadRM3 members of the UKCP09 (11
ensembles).
A multi-site approach to risk assessment for the insurance industry
L. SPEIGHT1, J. HALL1, C. KILSBY1 & P. KERSHAW2
1 School of Civil Engineering and Geosciences, Cassie Building, Newcastle University, Newcastle upon Tyne, NE1
7RU, UK
linda.speight@ncl.ac.uk
Abstract Using a case study of one insurance company’s exposure from static caravans we propose a
methodology for flood risk assessment at multiple sites nested within a national framework. Following
a source-pathway-receptor approach, a mixture of statistical and physically based methods is used in a
systems based model which incorporates the most important random processes associated with flood
damage. Within the system meteorological inputs are modelled statistically using a conditional
dependence model; the water level, floodplain inundation and damage calculations are deterministic;
and the impact of flood defence failure is considered probabilistically. The methodology explicitly
couples spatial dependencies between variables affecting flood risk at national and local scales. The
output is greater understanding of risk, and the associated uncertainties, which can be used to inform
decision making.
Regionalisation as a learning process
Wouter Buytaert(1) and Keith Beven(2)
(1) Civil and Environmental Engineering, Imperial College London (w.buytaert@imperial.ac.uk)
(2) Lancaster Environment Centre, Lancaster University, UK, Geocentrum, Uppsala University, Sweden, and
ECHO/ISTE, EPFL, Lausanne, Switzerland
Abstract. Geographical migration of hydrological model structures, commonly known as
regionalisation, is a highly uncertain process. Regionalisation relies on the hypothesis that
calibrated parameter sets from a donor catchment can be useful to predict discharge of an
ungauged catchment. However, since every catchment is unique, model parameters need to be
adapted for differences between a calibration and a prediction catchment, either by
transformation or further selection. This process is inherently uncertain: model parameters, and
therefore the required changes, do not exactly represent quantities that we can measure or
calculate. Therefore, we outline an approach to learn about how model parameters should be
transformed between a gauged and an ungauged catchment. The approach consists of an
iterative process, in which a model structure is applied successively to gauged catchments. After
each step, parameter behaviour is evaluated as a function of catchment properties and
intercatchment similarities. The method is illustrated with an application of a customized version
of TOPMODEL to a set of catchments in the Ecuadorian Andes. First, parameter sets are
generated for a donor catchment. This model ensemble is then used to predict the discharge of
the other catchments, after applying a stochastic parameter transformation to account for the
uncertainty in the model migration. The parameter transformation is then evaluated and
improved before further application. The case study shows that accurate predictions can be
made for predicted basins. At the same time, knowledge is gained about model behaviour and
potential model limitations.