RUMED Scoping Study Final Report 13 09 10

advertisement
NERC Scoping Study
Reducing Uncertainty in Models
for Environmental Decision-making
Environment Pollution and Human Health Theme
Final Report
Keith Beven, Kevin Jones, Phil Haygarth,
Rob MacKenzie, Paul McKenna, Trevor Page,
Andy Sweetman, Nigel Watson
Lancaster Environment Centre
Lancaster University, UK
September 2010
NERC Environment Pollution and Human Health Theme
Scoping Study: Reducing Uncertainty in Models for
Environmental Decision-making
Executive Summary
NERC, in collaboration with Defra, commissioned Lancaster Environment Centre to
undertake a scoping study on the subject of Reducing Uncertainty in Models for
Environmental Decision-Making, within the context of the NERC Theme on
Environment, Pollution and Human Health. This has involved three workshops in the
areas of atmosphere, land and water, and chemicals in the environment and a
“synthesis” workshop.
There is always uncertainty in decision-making. Uncertainty is only important in so far
as it affects what decision is made. It might, for example lead to the use of more
formal risk-based decision-making theory where uncertainties can be clearly
evaluated probabilistically, allowing the cost-benefit of policy options to be weighed.
It might lead to precautionary, robust or adaptive decision-making strategies where
uncertainties are large or difficult to assess, as they often will be in environmental
problems.
A realistic assessment of uncertainty is also the first step in a process of assessing
what might best be done to reduce or constrain that uncertainty by making some
improvements in the science (models) that leads to evidence (data and model
predictions) on which decisions can be based within the prevailing socioeconomic
and political context. There can be uncertainty in drivers, pressures, states, impacts
and responses (DPSIR). To move from drivers to impacts generally needs a model
(or cascade of models) and therefore needs to consider:
(a) uncertain/incomplete process representations
(b) uncertain model parameters
(c) uncertain inputs/boundary conditions
(d) missing factors
All involve knowledge (epistemic) rather than only random (aleatory) uncertainties.
One implication of knowledge uncertainties is that the best available model will not
necessarily be fit for purpose. In such cases, decision makers will not be robust in
basing decisions on model predictions. Model fitness should be assessed in relation
to the uncertainties involved, the evaluated accuracy of model output, transparency
of the modelling process, the available ”headroom” (a precautionary gap between
model predictions and a regulatory threshold), and the ability to address the
regulatory decision.
Discussions at the four workshops that formed the basis for this scoping study
revealed a number of general principles relevant to a programme of Reducing
Uncertainty in Environment Decision Making.
1. Most would be learned from the study of real-world policy-relevant problems
with the involvement of the relevant stakeholders.
2. End–to-end analyses from source to impacts on human or ecosystem health
are required for decision-making. Reduction of uncertainty in any individual
model should be seen in the context of the broader DPSIR framework.
3. Data availability from multiple sources is often a critical issue in making
research on these types of end-to-end or cross-compartment systems
effective.
The provision of exemplar “Gold standard” data sets could be
valuable in certain application areas.
4. Most such problems will involve multiple sources of uncertainty and cascades
of model components.
5. Many such problems involve multiple scales, when the type and quality of
data supporting an analysis of uncertainty might vary across scales.
6. There is the potential to link to other on-going projects funded by NERC and
others (such as the Defra Demonstration Test Catchments, Virtual
Observatory, EQUIP, Natural Hazards Risk and Uncertainty, EPSRC/NIREX)
7. Many important epistemic uncertainties require fundamental scientific
research, beyond the scope of projects relevant to a research programme
arising from this scoping study, before the knowledge can be encapsulated in
(even uncertain) models (e.g. chemical speciation parameters in complex
environments; epidemiology of mixtures of pollutants and microbes).
In addition, some specific research questions were identified that should form the
basis of a research programme.
8. More research is needed on testing model components as hypotheses while
allowing for different sources of uncertainty, including epistemic uncertainties
and the potential for some data to be non-informative in inferring robust
models that get the right result for the right reasons.
9. More research is needed on the propagation of uncertainties within end-toend cascades of model components, including impacts on human and
ecosystem health.
10. More research is needed on identifying the role of critical measurements that
will be effective in reducing uncertainty in end-to-end model cascades.
11. More research is needed on how to communicate uncertain model predictions
to decision makers, including the assumptions that underlie any quantitative
analysis.
12. More research is needed to develop a protocol that incorporates uncertainty
especially when not all uncertainties are easily treated probabilistically.
These points provide the basis for suggestions for a structured programme of future
research, including suggestions for some priority policy areas in which research on
reducing uncertainty would be valuable. There will be substantial added-value from
such a programme in the form of embedded software quality management and
model evaluation best-practices in the UK modelling community. The report is
completed by an Appendix of specific models and applications that have been
identified as possible components of such a research programme. This list is not
intended to be exclusive.
Glossary
Defra
Department of Environment , Food and Rural Affairs, UK
DPSIR
Driver, Pressures, States, Impacts, Response Concept
EPHH
Environment, Pollution and Human Health theme, NERC, UK
EPSRC
Engineering and Physical Science Research Council, UK
ESRC
Economic and Social Research Council
FERA
Food and Environment Research Agency, UK
FP7
European Union 7th Framework Programme
LWEC
Land Water and Environmental Change, NERC
MNP
Milieu- en Natuurplanbureau, Netherlands
MRC
Medical Research Council
NCR07
National Research Council, 2007, Report on Models in Environmental
Regulatory Decision Making
NERC
Natural Environment Research Council, UK
NUSAP
Numeral, Unit, Spread, Assessment, Pedigree uncertainty concepts
PM2.5
Fine particles, capable of passing through an inlet with 50% cut-off
efficiency for particles with aerodynamic diameters of 2.5 micrometers.
REACH
Registration, Evaluation, Authorisation and restriction of Chemicals,
UK
UKCP09
UK Climate Predictions 2009
UKWIR
UK Water Industry Research
Table of contents
1
Background ...................................................................................................... 1
2
What this report is (and is not) about ................................................................ 4
3
The Scoping Study Workshops ........................................................................ 7
4
Synthesis ....................................................................................................... 10
4.1
A general framework for assessing and reducing uncertainties (Responses
to Question 1) ............................................................................................ 10
4.2
Prioritization of problems (Responses to Question 2) ................................ 13
4.3
Horizon scanning (Further responses to Question 2) ................................. 14
4.4
Assessing the impacts of change (Responses to Question 3).................... 14
5
Defining a research program .......................................................................... 15
6
References..................................................................................................... 17
APPENDIX A - Workshop 1: Air ............................................................................. 21
APPENDIX B - Workshop 2: Land and Water......................................................... 34
APPENDIX C - Workshop 3: Chemicals in the Environment ................................... 48
APPENDIX D – Modelling case studies ................................................................... 58
APPENDIX E – List of participants .......................................................................... 75
NERC Environment Pollution and Human Health Theme
Scoping Study: Reducing Uncertainty in Models for
Environmental Decision-making
1
Background
NERC, in collaboration with Defra, commissioned Lancaster Environment Centre to
undertake a scoping study on the subject of Reducing Uncertainty in Models for
Environmental Decision-Making, within the context of the NERC Theme on
Environment, Pollution and Human Health. This has involved three workshops in the
areas of atmosphere, land and water, and chemicals in the environment, along with a
final “synthesis” workshop.
It is perhaps worth noting at the outset that the only mention of uncertainty or
reducing uncertainty in the NERC Strategy 2007-2011 document is in the title of the
section on the Climate System theme (improving predictions, reducing and
quantifying uncertainty).
The Environment Pollution and Health theme implicitly
recognises uncertainty by identifying “reliable predictive capability” as a key output of
the theme. The low visibility of uncertainty seems surprising now that there has been
a rather rapid change in how the possibilities and limitations of environmental
modelling are perceived over the last few years.
Not for the first time, there is an
appreciation that there are real difficulties (as discussed below) in matching
predictive capability and decision-making within the “bounded rationality” (Simon,
1978) that inevitably leaves us blind to unexpected consequences and external
shocks.
There is always, and has always been, uncertainty in decision making. Uncertainty is
only important in so far as it affects what decision is made. It might, for example lead
to the use of more formal risk-based decision making theory where uncertainties can
be clearly evaluated probabilistically, allowing the cost-benefit of policy options to be
assessed. It might lead to precautionary, robust or adaptive decision making
strategies where uncertainties are large or difficult to assess, as they often will be in
environmental problems (see, for example, Beven, 2009).
1
A realistic assessment of uncertainty is also the first step in a process of assessing
what might best be done to reduce or constrain that uncertainty by making some
improvements in the science (theory, often encapsulated in models) that leads to
evidence (data and model predictions) on which decisions can be based within the
prevailing socioeconomic and political context. In making a decision there can be
uncertainty in drivers, pressures, states, impacts and responses. Models will often
be used to inform the decision-making process since to move from drivers to impacts
generally needs a model (or cascade of models).
This study is concerned with
reducing uncertainty in the model predictions and therefore needs to consider:
(a) uncertain/incomplete1 process representations
(b) uncertain model parameters
(c) uncertain inputs/boundary conditions2
(d) missing factors
All involve knowledge (epistemic) rather than only random (aleatory) uncertainties
(see Box 1). Epistemic uncertainties are invoked when we simply do not know
enough about a knowledge system and its drivers (e.g. Papineau, 1979; Beven,
2009; 2010). The result of epistemic uncertainties is that prediction errors might have
complex (non-aleatory, non-stationary) behaviours; will not always be informative
about model adequacy; and will not be the same in prediction as in calibration (or in
transfers from site to site and application to application)(e.g. Doherty and Welter,
2010). This difference is treated at length in Models in Environmental Regulatory
Decision Making (Committee on Models in the Regulatory Decision Process,
National Research Council, 2007, hereafter referred to as NRC07). The importance
of epistemic uncertainty was also recognised in the recent review of risk and
uncertainty commissioned by the NERC Natural Hazards theme (Rougier et al.,
2009), but the non-aleatory and non-stationary nature of this uncertainty was not
brought out.
Incomplete or simplified process descriptions are often called parameterisations (see e.g.
Stensrud, 2007), but the use of a more complete small scale theory at the larger scales required
for practical applications is also effectively a parameterisation (Beven, 2009)
2 Sometimes observational inputs and boundary conditions are called input parameters
1
2
Box 1 Types of Uncertainty
There are many different ways of classifying different types of uncertainty.
At the
most fundamental level we can distinguish between those that could be reduced
given further knowledge or measurements, and those that cannot and should be
treated as random. Knowledge uncertainties, which could be reduced, are often
called epistemic uncertainties.
aleatory uncertainties.
Irreducible random uncertainties are often called
This distinction was made by Knight (1921) who called
epistemic uncertainties the real uncertainties.
Aleatory uncertainties are often described as those due to “natural variability”.
Aleatory uncertainties can be treated in the form of probabilities; epistemic
uncertainties are often treated as if they can be represented as probabilities, but this
might lead to overconfidence in uncertainty estimation if the structure of the
epistemic uncertainty is non-stationary in space or time (as it often will be). Model
structural error is an epistemic uncertainty that will generally have non-stationary
characteristics.
This very simple classification can be made more complicated to more properly
reflect real uncertainties in two ways.
The first complication is to distinguish
between those epistemic uncertainties that we already perceive as being important
(and which might indeed be reduced by further understanding or measurements),
and those that have not yet even been thought about (the unknown unknowns or
ontological uncertainties).
Little can be done about this latter type of uncertainty
since, by definition, we have (yet) not even perceived which unknown unknowns
might be important. Thus, while model predictions might inform a decision-making
process, it is necessary to be careful about treating the uncertain outcomes of
models as “truth generators” in the sense of encompassing the full range of potential
outcomes from drivers to impacts.
The second complication is to recognise that not all natural variability is simply
random. This is important when models are driven by natural boundary conditions as
assessed by limited measurements (or predictions of other models subject to
uncertainty).
The characteristics of the uncertainties associated with such
assessments may be complex and may involve epistemic as well as aleatory
components. An example is the quite different errors that might be associated with
different rainstorms falling over a catchment area. Another example is when we are
prepared to treat the species sensitivity to pollutants as a probabilistic process when
3
there is also epistemic uncertainty about the form of distribution to use, perhaps
because only a short observational record is available.
Similarly, there may be
epistemic uncertainties associated with estimates of model parameter distributions,
when it is not always clear how estimates based on measurements might relate to
the effective values required to get good model predictions at larger space and time
scales.
Despite these two complications, the distinction between aleatory uncertainties, that
can be treated as probabilities, and epistemic uncertainties, that should not, will still
hold, albeit that there may be epistemic uncertainty about the properties of aleatory
uncertainties. There is, therefore, a danger of confusion when model predictions
subject to epistemic uncertainty are presented as if they are probabilities.
An
example here is the outputs of the UKCP09 ensemble climate predictions. These
are presented as probability quantiles about potential future climate when in fact they
represent an interpolation of the probability surface of the outputs of a sparse sample
of model predictions for a particular emission scenario.
The probabilities are of the
distribution of model outputs, not of future climate, under that emission scenario.
The difference is important when there are significant (epistemic) differences
between model predictions and actual climate in the recent past. This reinforces the
point that it is important to convey to decision-makers the assumptions on which a
model uncertainty assessment is based, and to convey what a given uncertainty
metric actually means.
One implication of knowledge uncertainties is that the best available model will not
necessarily be fit for purpose (see Box 3 below). This should be assessed in relation
to the uncertainties involved. In such cases, decisions will not be robust if based
solely on model predictions.
2
What this report is (and is not) about
This scoping study is concerned with uncertainty in the prediction of models of the
environment. For the purposes of this report, and consistent with NRC07, an
environmental model is a tool for explaining and predicting scientific phenomena
where empirical observation is not available. Prediction is an important aspect of
modelling. Some models are simple one-to-one relationships with well-defined
4
functional form; others are million-line computer codes with largely unexplored
phase-spaces. We are not concerned with truth claims about “reality”. Models are
never complete because human and natural systems are always more complicated
than the model can be.
Environmental models can be repositories of scientific
knowledge, and we may value them as such, but we do not have to worry here about
how “true” our underlying theories are, only that the models should be consistent with
available observations, allowing for the relevant sources of uncertainty. Models are
then heuristically valuable simulations or projections that can be used to inform
decision making (e.g. Edwards, 2010, p352).
This report is about models of the environment (including the human interaction with
the environment) that are or could be used in policy support and regulatory decisionmaking. It is not about environmental decision-making per se, and even less about
reducing uncertainty in environmental decision-making (but see Box 2); it is about
reducing uncertainty in models used for environmental decision-making, which is a
much narrower remit.
This report is about data as well as environmental models, but only to the extent that
environmental models are built from data (observations), are evaluated using data,
process data in order to function, and are set against data in policy-making and
regulation (NRC07; Edwards, 2010 Chapters 1 and 2; Stott and Thorne, 2010). We
will discuss data in any of these roles when they hold the key to reducing uncertainty
in environmental modelling.
The report does not consider climate change models specifically, although the
general remarks we make can and should be applied when climate model results are
used for decision-making. Because the policy stakes are high, the robustness of
climate models as evidence is a focus of much attention. This has had the effect of
embedding elements of software management best practice in climate modelling, at
least to the extent that best practice can be accommodated when dealing with
models that take months of CPU-time to run (Easterbrook and Johns, 2009). Much of
this best practice is “institutionalised” in the model inter-comparison exercises (see,
e.g., http://cmip-pcmdi.llnl.gov/index.html). Epistemic uncertainty has begun to be
addressed in the climateprediction.net project (Piani et al., 2005).
5
Box 2. Uncertainty and Making Decisions
While this scoping study is not about methodologies for decision-making in the face
of uncertainty (but rather reducing uncertainty in environmental models as an input to
the decision making process), there are important links between the way in which
models are evaluated, the communication of uncertainty, and decision-making
methodologies.
All decision-makers deal with uncertainty all the time but it is not
sufficiently appreciated that a consideration of robustness to uncertainty in potential
futures might make a difference to the decision made. It is possible to evaluate the
sensitivity of decisions to model uncertainty in a number of different ways.
Classical techniques for risk-based decision making, for example, require that all
sources of uncertainty are treated in probabilistic form so that ranking of options can
be achieved by integrating a cost function over the probabilities of predicted
outcomes (e.g. Bedford and Cooke, 2001).
This implies both completeness of the
uncertainties considered, including the cost function and a probabilistic treatment of
recognised knowledge uncertainties.
There are other methods of decision-making under uncertainty that are less
dependent on treating all uncertainties probabilistically (see Beven, 2009, Ch. 6, for a
summary). The InfoGap methodology of Ben-Haim (2006), for example, looks at the
robustness of a model-based decision in achieving defined minimum requirements to
the potential for some best estimate model to be wrong (for an environmental
application see Hine and Hall, 2010).
A decision-maker might also, in face of severe uncertainty, revert to being riskaverse or precautionary. The important point here is that deciding on a response to
model uncertainty in formulating a decision will depend on two important inputs. The
first is a realistic assessment of the uncertainty associated with a model; the second
is conveying to a decision-maker the assumptions on which that assessment is
based particularly where a decision might depend on cascades of model components
in a driver-source-pathway-impact-response system. A clear understanding of these
assumptions might guide the decision-making strategy and an appreciation for where
more scientific research is required in reducing uncertainty.
6
3
The Scoping Study Workshops
Three subject area workshops, in the areas of Atmosphere, Land and Water, and
Chemicals in the Environment, involving academics, consultants, other practitioners,
Defra, FERA, and Environment Agency staff were held in June and July 2010. Each
of these workshops was asked to address the following questions:
1. What are the most significant sources of uncertainty in the environmental
modelling area being considered?
2. Which of these are of most importance to decision makers?
3. Which of these can be most reduced?
4. What research programmes are already addressing this task?
5. Where and how should further research be focussed? Is the reduction of
uncertainty necessarily always the most appropriate action?
Summaries of the discussions in each of the three subject area workshops are
provided in Appendices A (atmosphere), B (land and water), and C (chemicals in the
environment).
The final synthesis workshop discussions are incorporated into the
main part of this report, while Appendix D gives details of models and applications
from specific policy areas that might form components of a directed research
programme.
The workshops brought out a number of common issues from the different areas. In
particular, expanding on the points listed in Section 1,
(a)
Uncertain or incomplete process representations. Workshop participants
readily pointed to structural inadequacies in models used for policy-making.
Examples are simplified chemistry in air quality models (Appendix A) and processes
controlling diffuse pollution Appendix B). Critical comments arose particularly when
policy-facing models used approaches to environmental processes that reflected
computational constraints that are (arguably) no longer valid. For example, it was
argued that modelling based on the Gaussian plume/puff solutions to fluid dispersion
7
should
now be
replaced
by high-resolution,
large-eddy-resolving,
Eulerian
simulations, and that this model-switch would be more cost-effective than pursuing
ad hoc adjustments to the plume models. A similar switch to Eulerian modelling has
been recommended to Defra by Monks et al. (2007) for the modelling of surface
ozone. The corollary of such recommendations is that there should be routine
procedures for "sunsetting" models that are no longer state-of-the-art and for building
skills in newer modelling frameworks (cf., Murrells et al., 2009).
(b) Uncertain model parameters. Many internal model parameters are "inherently"
uncertain in that the model does not characterise the environment well enough for
our best calculations of parameter values to be used. An example is the use of a
single deposition velocity for a trace gas like ozone in the atmospheric boundary
layer because the model does not distinguish between different land surfaces.
Workshop participants also emphasized repeatedly the difficulty of estimating
effective parameter values for models. One concern, particularly in the poorly mixed
environmental compartments, is that model parameters might be different from those
measured in the laboratory or at small scales in the field (e.g., Beven 2006, Butler et
al., 2008; Verver et al., 2000). Thus it is often necessary to infer effective values of
multiple (interacting) parameters on the basis of limited observational data with
consequent implications for uncertainty in prediction and for testing models as
hypotheses about system function (Beven, 2010).
(c) Uncertain inputs and boundary conditions I. Contributors to the study
commented on the frequent lack of adequate data to evaluate process
representations and model predictions. It was suggested that there was a need for
“gold standard” datasets, against which to benchmark models (Appendix A). The
atmospheric dispersion modelling community has classic datasets, such as the
Prairie Grass experiments of the 1950s, but there was a strong feeling that new
datasets, that tested more of the model parameterisations, were needed.
Our
attention was drawn to an interesting and practicable strategy for making progress
with the "model-data symbiosis" (Edwards, 2010, chapter 13) that has been put
forward in the field of combustion chemistry (Frenklach, 2007) and would be readily
adapted to many environmental science sub-disciplines. Rather than hoping for the
perfect model-evaluation dataset, this "process informatics" framework combines
organization of scientific data, shared scientific tools for analysis and processing of
these data, and engagement of all of the relevant scientific community in the data
collection and analysis in an ongoing cycle of model evaluation and testing.
8
(d)
Uncertain inputs and boundary conditions II. Workshop participants
discussed the uncertainty produced when transferring information across scales and
places. For the global fluid compartments (or at least those most actively mixing - the
atmosphere and the ocean mixed layer), the "uniqueness of place" problem should
not be as severe as for solid or dense multi-phase compartments like soil, but scaledependent parameterisations of inputs and boundary conditions are still a source of
considerable uncertainty in modelling frameworks such as atmospheric mesoscale
modelling (see, e.g., WRF Users Guide, 2010, chapter 5).
(e) Difficulties in targeting and prioritising uncertainties. It was not clear that
workshop participants, as representative of UK modelling expertise, were always
confident in defining the most significant uncertainties at different time and space
scales, suggesting that there is need for work specifically looking at this. Difficulties
in identifying the most critical uncertainties are due in part to the multiple purposes
(in research and policy-making) that most models are used for. The most critical
uncertainties then depend on the research or policy question being asked. There is a
tension between academic model-builders, who often aim to develop flexible code
capable of addressing several topics, and model-building for policy, which requires
parsimony and transparency (NRC07). The issue was considered further in the final
synthesis workshop.
(f) Missing factors. Air workshop participants chose to include the category of “wild
cards” alongside an analysis of uncertainties sorted by input-process-outputs, and
many topics were placed in the wild card category (Appendix A). All workshop
participants were clear that the most significant missing factors are the “unknown
unknowns” for which, by definition, there is no information to inform a model
uncertainty study. Since it is, by definition, impossible to programme research to find
unknown unknowns, it is important that there is a place for serendipity in model
development for environmental decision-making; research programme management
should be sufficiently agile to handle (and even encourage) the discovery of
unexpected results.
In addition to expert assessment of the importance of uncertainties in categories (a)
– (f), above, the workshops drew attention to general methods for the systematic
evaluation of uncertainties and their importance, particularly for complicated models
with very many degrees of freedom (Beven and Binley, 1992; Rabitz et al., 1999;
9
Beven, 2006; Ziehn and Tomlin,2009). We find that such methods for model
evaluation have not become embedded in the practice of the UK modelling
community for environmental decision-making (see also recommendations R4.1 –
R4.4 of Monks et al., 2007).
4
Synthesis
The final Synthesis workshop was intended to draw some common methodological
themes from the previous workshops for the practice and research needs of
uncertainty reduction in modelling for decision making. Following a summary of the
discussions in the first three workshops, three major questions were posed for
discussion:
1. How can knowledge uncertainties be identified and reduced in cost-effective
ways, in end-to-end model cascades from source to receptor and health
impacts?
2. What is the priority of exemplar problems of interest to relevant government
agencies (including horizon scanning of potential new policy drivers in
environmental pollution and health)?
3. Given the inevitability of knowledge uncertainties, how should the impacts of
future change be assessed to inform relevant decisions in different
application areas?
4.1
A general framework for assessing and reducing uncertainties
(Responses to Question 1)
The question of how to deal with and reduce knowledge uncertainties was discussed
at some length in the light of the inputs from the three previous workshops.
It was
concluded that a general framework was required that should be general to a wide
range of policy-relevant application areas.
number of components as follows:
10
Such a framework would require a
(a) A component for deciding on the importance and level of uncertainty analysis
required. All regulatory or policy decisions should recognise the potential for
uncertainties in the evidence coming from model predictions but not all
policy-relevant problems will justify extensive effort to assess the
uncertainties. A structured approach based on the NUSAP3 methodology is
used, for example by MNP in the Netherlands (see Funtowicz and Ravetz,
1990; Petersen, 2006; van der Sluijs et al., 2004).
(b) A component for testing models as fit for purpose while allowing for
uncertainties in model inputs, parameter values, and observational data.
This should include testing different models as competing hypotheses about
system response (see Box 3).
(c) A component for reducing and managing uncertainty in hypothesis testing
(d) A component for communicating uncertainty assessments (and the
associated assumptions) to the decision and policy making process
Each of these components requires a methodology to communicate effectively the
potential methods and assumptions of those methods to have acceptance and buy-in
from decision and policy makers. One strategy for such a methodology is to evaluate
parameter and input/boundary condition uncertainties within a structured decision
framework (e.g. Pascual, 2005). Decisions about models, parameters and
inputs/boundary conditions must then be made explicit, so that they can be
discussed, criticised if necessary and agreed by the stakeholders involved.
Communication of uncertainty and the assumptions that underlie how it is evaluated
to the decision maker needs to be as effective as possible (e.g. Faulkner et al.,
2007). The concept of addressing and agreeing specific decisions at different steps
in the analysis can also be useful in the communication of assumptions (see Box 2).
Such a decision-base approach extends naturally to uncertainty reduction. We do
know that the estimation of uncertainty should not be the end point of an application.
Part of the outcome in a particular application may also be to try and reduce the
uncertainty, e.g. by commissioning more research or data collection. The concepts of
3
see www.nusap.net (accessed 12.8.10)
11
the critical observation (which distinguishes between, or conditions, model
representations as a means of reducing uncertainty), model-data symbiosis, and
process informatics, are important in this respect. These concepts underpin the issue
of hypothesis testing discussed earlier, but assessing the real information content of
a data set for a particular purpose needs further research.
Box 3. Testing models as hypotheses in the face of uncertainty
A number of issues arise in testing models as hypotheses to determine whether they
might be fit for purpose. Uncertainty in the modelling process means that there is
always the possibility of accepting a poor model when it should be rejected (a form of
false positive or Type I error) or rejecting a good model when it should be accepted
(false negative or Type II error).
In statistical hypothesis testing we generally
choose to test a hypothesis only with a certain probability of being wrong (e.g. at the
5% level). It is difficult to carry this over to simulation models applied to places and
data sets that are unique in space and time with only single realisations of any
observational errors. Except in some special circumstances we cannot carry out
replicate sets of observations so any such 5% error criterion would have to be
assessed in some different way.
Consider each of the possibilities for being wrong.
We would wish to avoid Type I
errors in testing because the performance of models falsely retained for use in
prediction might lead to false inferences and poor decision making. So how to avoid
Type I errors? The primary reason why a Type I error might occur is because there
is enough uncertainty in the inputs to the model and the observational data used in
evaluation that whatever performance indicator is used to make a decision about
model acceptability, it cannot differentiate between good and poor models. This
might be because of a particular or peculiar sequence of observational errors,
including “rogue” observations.
So to avoid Type I errors, we need to be careful
about the commensurability of observed and model variables and try to ensure that
only periods of good quality data are used in formal model evaluation.
It is perhaps more important still to avoid Type II errors. If we do retain a poor model
for use in prediction (Type 1 error), then hopefully further evaluation in the future
might reveal that it gives poor predictions and therefore our choice can be later
refined as part of the learning process. But we really would not want to eliminate a
good model (i.e. make a Type II error) just because of poor-quality inputs.
12
Any approach to testing models as competing hypotheses about system response
should involve issues such as whether the driving data are fit for purpose; the real
information content of observations (and when observations might not be
informative); and the role of critical observations that could be made in differentiating
between potential models.
Note that the use of optimization or maximum likelihood estimation or Bayesian
inference will not generally make a distinction between the best available model and
models that are fit for purpose. Statistical inference is carried out conditional on an
implicit assumption that the model structure is correct.
Alternative rejectionist
approaches to model evaluation have been suggested as an alternative approach.
Both of these methodologies are summarised in Beven (2009, Ch. 4).
4.2
Prioritization of problems (Responses to Question 2)
A second aim of the synthesis workshop was concerned with the prioritization of
problems of current concern that would benefit from reduced uncertainty in the
modelling component of the decision-making process.
A number of useful
suggestions arose from the three subject workshops. These were (not in any order
of priority):

Diffuse and point pathogen sources in catchment areas

Cycle 2 requirements of the Water Framework Directive

Radionuclide impacts on health

Urban air quality (particularly modelling of PM2.5)

Bio-aerosol source and dispersion characteristics, particularly with respect to
composting

Contaminated Land Hazards

REACH (with possible industrial backing)

Environmental fate of pharmaceutical products

Pathogen risks from insects
These issues link to other research councils / bodies, including ESRC / MRC /
EPSRC (sustainable water, sustainable urban environments), LWEC, UKWIR, and
13
FP7. It was also recognized that a number of these issues involve data sets from
multiple sources which are not all easily or readily available. Other NERC projects,
such as the prototype Virtual Observatory are starting to address data availability.
Following the Synthesis workshop, some specific examples of models that might be
used as demonstrators for the development of techniques in reducing uncertainty in
environmental models from different policy areas have been identified4.
Full details
are given in Appendix D.
4.3
Horizon scanning (Further responses to Question 2)
In addition, looking further ahead, a number of issues were identified as being
potentially of great interest to policy making in relation to environment, pollution and
health in the future. These were (not in any order of priority):

Requirements for real-time prediction with uncertainty (to facilitate discounting
of quality exceedances)

Predicting and controlling occurrences of Crohn’s disease

Impacts of dinoflagellates and shellfish on health

Increasing campylobacter infections – uncertainties in source tracking based
on different methods and models

Moving from threshold levels to exposure-reduction regulatory frameworks
We recognize that there are also important issues in predicting and inducing
behavioural socio-economic changes in policy setting (e.g. increased rainwater
harvesting; changes in agricultural practices; pharmaceutical innovations or changes
in prescribing; changes in eating habits) but we consider these to be outside the
remit of this scoping study concerned with environmental modelling.
4.4
Assessing the impacts of change (Responses to Question 3)
The importance of prognosis (rather than diagnosis) to decision makers was noted in
our workshops; it was also discussed at length by the US Committee on Models in
the Regulatory Decision Process in 2007 (NRC07) under the heading “extrapolation”.
Extrapolation or prognosis often requires the assessment of the impacts of future
Discipline-specific summaries of models used in policy-making have been produced elsewhere
e.g., Monks, 2007; AQEG, 2009.
4
14
change and, as such, is greatly subject to epistemic uncertainty. Evaluating model
predictions focuses attention on the theoretical basis for the model structure.
The nature of the relevant change will be application-area-specific but may include
climate change, land management change, and societal/behavioural change. All of
the types of future change have been important in past studies, for example in the
Land Use Futures: making the most of land in the 21st century5. Uncertainties in the
predictions of such changes (such as in the UKCP09 climate change projections6)
will have a potential effect on policy and decision making in a number of areas.
There is also overlap here with existing programmes, such as the NERC EQUIP
consortium project (although this does not include a major human health
component).
Predicting the impacts of future change is predicated on understanding and
assessing past changes and model hindcasts of past change. This is critical to the
evaluation of models as hypotheses about system responses. There was some
discussion in workshops about whether the historical data available are fit for this
purpose. In many cases it was felt that data were lacking in quantity, quality and
availability (e.g. information about farm practices, prescriptions issued, chronic health
impacts …). It was felt that in future it might be useful to extend the NERC Virtual
Observatory concepts to other types of data.
5
Defining a research program
Discussions at the four workshops that formed the basis for this scoping study
resulting in a number of general guidelines relevant to a research programme for
Reducing Uncertainty in Models for Environment Decision Making.
1. Most would be learned from the study of real-world policy-relevant problems
with the involvement of the relevant stakeholders.
2. End–to-end analyses from source to impacts on human or ecosystem health
are required for decision-making. Reduction of uncertainty in any individual
model should be seen in the context of the broader DPSIR framework.
5
6
See http://www.foresight.gov.uk/OurWork/ActiveProjects/LandUse/LandUse.asp
See http://ukcp09.defra.gov.uk/
15
3. Data availability is often a critical issue in model evaluation. The provision of
exemplar “Gold standard” data sets could be valuable in certain application
areas.
4. Most such problems will involve multiple sources of uncertainty and cascades
of model components.
5. Many such problems involve multiple scales, when the type and quality of
data supporting an analysis of uncertainty might vary across scales.
6. There is the potential to link to other on-going projects funded by NERC and
others (such as the Demonstration Test Catchments, Virtual Observatory,
EQUIP, Natural Hazards Risk and Uncertainty, EPSRC/NIREX).
7. Many important epistemic uncertainties require fundamental scientific
research, beyond the scope of this scoping study, before the knowledge can
be encapsulated in (even uncertain) models (e.g. chemical speciation
parameters in complex environments; epidemiology of mixtures of pollutants
and microbes).
In addition, some specific research questions were identified that should form the
basis of a research programme.
8. More research is needed on testing model components as hypotheses while
allowing for different sources of uncertainty, including epistemic uncertainties
and the potential for some data to be non-informative in inferring robust
models that get the right result for the right reasons.
9. More research is needed on the propagation of uncertainties within end-toend cascades of model components, including impacts on human and
ecosystem health.
10. More research is needed on identifying the role of critical measurements that
will be effective in reducing uncertainty in end-to-end model cascades.
11. More research is needed on how to communicate uncertain model predictions
to decision makers, including the assumptions that underlie any quantitative
analysis.
12. More research is needed on how to develop a protocol to incorporate
uncertainty into decisions, especially when not all uncertainties are easily
treated probabilistically in a risk-based decision making framework (e.g. a
generalization of headroom concepts or the “reasonable worse case”
approach of the ECB Technical Guidance Document on Risk Assessment).
16
These guidelines include methodological development and method development
(applications to particular practical problems).
The two are not easily separated.
Although we would expect there to be some generality or commonality in the
methodologies for dealing with cascades of model components, epistemic
uncertainties, or hypothesis testing of model components across application areas,
the details will inevitably be application-specific. This suggests that the research
programme should be structured around a small number of interesting (to policy
makers) applications with a meta-project facilitating exchange of ideas, methods and
best practice across application areas. If software quality management and
systematic model evaluation is insisted upon within such a research programme, we
are confident that substantial added-value will accrue from the embedding of best
practice across the UK environmental modelling community.
We would suggest that, in developing projects for such a programme, any in-scope
project would have to address explicitly:

Their assumptions about different sources of uncertainty, including epistemic
uncertainties, embedded within explicit software quality assurance

The development of methodologies for hypothesis testing of model
components as fit for purpose (i.e. systematic model evaluation).

The design of critical experiments for hypothesis testing and reducing
uncertainty

The development of methodologies for communication of uncertainties and
assumptions to decision makers
Appendix D lists the details of models and application areas identified during the
scoping study, that might be useful in developing a research programme in this area.
Because of the time scale of the study and the limited participation in the workshops,
we recognize that this list will be incomplete, but it serves, nonetheless, to sketch out
the likely range of models and modelling issues to be addressed in a future research
programme. All those areas identified have significant policy implications.
6
References
AQEG (Air Quality Expert Group), 2007, Ozone in the United Kingdom, available at
http://www.defra.gov.uk/environment/airquality/publications/ozone/pdf/aqegozone-report.pdf.
Bedford, T & Cooke, R, 2001, Probabilistic Risk Analysis: Foundations and Methods,
17
Cambridge University Press.
Ben-Haim, Y, 2006, Info-Gap Decision Theory, 2nd Edition, Academic Press:
Amsterdam.
Beven, K J, 2006, A manifesto for the equifinality thesis, J. Hydrology, 320, 18-36.
Beven, K J, 2009, Environmental Modelling – An Uncertain Future?, Routledge:
London
Beven, K J, 2010, Preferential flows and travel time distributions: defining adequate
hypothesis tests for hydrological process models, Hydrol. Process. 24: 15371547.
Beven, K.J. and A.M. Binley (1992), The future of distributed models: model
calibration and uncertainty prediction, Hydrological Processes, 6, 279-298.
Butler, T., Taraborrelli, D., Brühl, C., Fischer, H., Harder, H., Martinez, M., Williams,
J., Lawrence, M., and Lelieveld, J., 2008, Improved simulation of isoprene
oxidation chemistry with the ECHAM5/MESSy chemistry-climate model:
lessons from the GABRIEL airborne field campaign, Atmos. Chem. Phys., 8:
4529–4546.
Committee on Models in the Regulatory Decision Process, National Research
Council, 2007, Models in Environmental Regulatory Decision Making,
National Academies Press, Washington D.C., available as PDF at
http://www.nap.edu/catalog/11972.html
Doherty, J and Welter, D, 2010, A short exploration of structural noise, Water
Resources Research, VOL. 46, W05525,doi:10.1029/2009WR008377
Easterbrook, S.M., and T.C. Johns, 2009, Engineering the Software for
Understanding Climate Change, Computing in Science & Eng., 11: 64-74.
Edwards, P.N., 2010, A Vast Machine, MIT Press, Massachusetts, USA.
Faulkner, H, Parker, D, Green, C, Beven, K, 2007, Developing a translational
discourse to communicate uncertainty in flood risk between science and the
practitioner, Ambio, 16(7), 692-703
Frenklach, M., 2007, Transforming data into knowledge—Process Informatics for
combustion chemistry, Proc. Combustion Institute, 31: 125–140.
Funtowicz, S.O. and J.R. Ravetz, 1990, Uncertainty and Quality in Science for
Policy, Kluwer
Hine, D and J. W. Hall, 2010, Information gap analysis of flood model uncertainties
and regional frequency analysis Water Resour. Res., 46, W01514,
doi:10.1029/2008WR007620, 2010
Knight, F H (1921) Risk, Uncertainty and Profit, Houghton-Mifflin Co., reprinted
University of Chicago Press, 1971.
Murrells, T., S. Cooke, A. Kent, S. Grice, A. Fraser, C. Allen, R.G. Derwent, M.
Jenkin, A. Rickard, M. Pilling, M. Holland, S. Utembe, 2009, Modelling of
18
Tropospheric Ozone - Project Summary Report: 2007-2009, Report
AEAT/ENV/R/2899 to The Department for Environment, Food and Rural
Affairs, Welsh Assembly Government, the Scottish Executive and the
Department of the Environment for Northern Ireland, available at
http://www.airquality.co.uk/reports/cat05/1003151144_ED48749_Final_Repor
t_tropospheric_ozone_AQ0704.pdf, last accessed 1 September 2010.
Monks, P.S., R.S. Blake and P. Borrell, 2007, Review of tools for modelling
tropospheric ozone formation and assessing impacts on human health &
ecosystems. A report to collate, evaluate and summarise information on tools
for modelling ozone (O3) formation and assessing impacts on human health
and ecosystems prepared for the United Kingdom Department for
Environment, Food and Rural Affairs.
Papineau, D., 1979, Theory and Meaning, Clarendon Press, Oxford, UK
Pascual, P., 2005, Wresting Environmental Decisions From an Uncertain World, ELR
News & Analysis, 35, 10539-10549, Environmental Law Institute, Washington
D.C.
Petersen, A.C., 2006, Simulating Nature: a philosophical study of computersimulation
uncertainties and their role in climate science and policy advice, Het Spinhuis,
Netherlands.
Piani , C., Frame D.J., Stainforth D.A., Allen M.R. , 2005, Constraints on climate
change from a multi-thousand member ensemble of simulations, Geophys.
Res. Lett., 32: L23825, doi: 10.1029/2005GL024452.
Rabitz, H., Aliş,Ö. F., Shorter, J., Shim, K., 1999, Efficient input-output model
representations. Computer Physics Communications, 117: 11–20.
Rougier, J et al., 2009, Scoping study on the Analysis, Propagation and
communication of Probability, Uncertainty and Risk' (SAPPUR)., unpublished
NERC Scoping Study, Natural Hazards Theme (see
http://www.bris.ac.uk/brisk/sappur).
Simon, H.A., 1978, Rational Decision making in Business Organizations, Nobel
Memorial Lecture, available at http ://nobelprize.org
/nobel_prizes/economics/laureates/1978/simon-lecture.html. [last accessed 8
August 2010]
Stensrud, D.J., 2007, Parameterization schemes : keys to understanding numerical
weather prediction models, Cambridge University Press, Cambridge, U.K.,
pp459.
Stott, P.A., and P.W. Thorne, 2010, How best to log local temperatures?, Nature,
465: 158-159.
van der Sluijs, J.P., P.H.M. Janssen, A.C. Petersen, P. Kloprogge, J.S. Risbey, W.
Tuinstra, and J.R. Ravetz, 2004, RIVM/MNP Guidance for Uncertainty
Assessment and Communication: Tool Catalogue for Uncertainty
Assessment. Report No. NWSE-2004-37. Utrecht University, The
Netherlands [online]. Available: http://www.nusap.net/downloads
/toolcatalogue.pdf [last accessed 6 August
19
Verver, G.H.L., H. van Dop, and A.A.M. Holtslag, 2000, Turbulent mixing and the
chemical breakdown of isoprene in the atmospheric boundary layer, J.
Geophys. Res.-Atmospheres, 105(D3): 3983-4002.
WRF Users Guide, 2010, User’s Guide for the Advanced Research WRF (ARW)
Modeling System Version 2.2, available at
http://www.mmm.ucar.edu/wrf/users/docs/user_guide/contents.html, last
accessed 1 September 2010.
Ziehn, T., and A.S. Tomlin, 2009, GUI-HDMR - A Software Tool for Global Sensitivity
Analysis of Complex Models. Environmental Modelling and Software, 24:775–
785
20
APPENDIX A - Workshop 1: Air
Monday 14 June 2010
Lancaster Environment Centre, Lancaster University
Attendees
Charles Chemel
University of Hertfordshire
Roy Harrison
NERC
Nick Hewitt
Lancaster Environment Centre
Rob Kinnersley
Environment Agency
Rob MacKenzie
Lancaster Environment Centre
Paul McKenna
Lancaster Environment Centre
Sarah Metcalfe
Nottingham University
Trevor Page
Lancaster Environment Centre
Ron Smith
CEH
Andy Sweetman
Lancaster Environment Centre
Alison Tomlin
University of Leeds
Roger Timmis
Environment Agency
Session 1: Welcome & individual introductions
To begin Roy Harrison introduced the workshop in the context of the NERC’s
Environment, Pollution and Human Health Theme. This was followed by Rob
MacKenzie, the workshop convenor, who gave a general overview of the use of
atmospheric models in decision-making. All attendees were invited to present
examples of their experience of encountering uncertainty within a policy or regulatory
framework.
Roy Harrison
Theme Action Plan Report to SISB by October 2010. Funding decision by May 2011.
Bid ca. £3M from NERC + £2M from DEFRA, but could push for more. Bob Watson’s
move to DEFRA has increased the profile of social science in environmental
decision-making, so there is the possibility of ESRC buy-in if case is made.
Need major science advances, not minor tinkering. Must fit to NERC strategy.
Model should be in use by agency or wanted for use by agencies (there was some
discussion about what to do about operational models that were “structurally
unsound”). Non-operational models are within scope if used by decision-makers.
Diagnosis of uncertainty must be tied to specific models.
21
Proposals must have a willing agency partner
Generic/philosophical programme has been set up by John Rees in the Natural
Hazards Theme
Rob MacKenzie
Described the time-space domains covered by “operational” models – fit to timespace of problems. Slides showed absence of models at smallest scales, but slide
was missing the roadside model DMRB.
Joint programme between NERC and Met Office should make it easier for academics
to influence MO model structure.
There is a new model comparison exercise being carried out by DEFRA
The AQEG report 2009 was introduced. AQEG suggest a strategy for reducing
uncertainty in models. Scoping study must be cognisant of this.
Alison Tomlin
Propagate errors through simulations – provides error bars on output
Sensitivity studies show which params drive uncertainty
Uncertainty due to model structure, input params, and model parameters (e.g. rate
coeffs)
Street-canyon modelling (MISCAM) example for York. Roadside concs very sensitive
to wind input.
AT has made an uncertainty software suite available online: GUI-HDMR
(http://www.engineering.leeds.ac.uk/erri/people/tomlin/GUI-HDMR.shtml).
Charles Chemel
Work for EA – CREMO – see website
Contribution of a given point source to regional air quality and deposition
Need to get right answer for right reasons – dynamic evaluation looks at model
sensitivity – needs diagnostic evaluation, which is very difficult (i.e. what is driving
model response to a change in input?).
Eulerian model predicts very different impacts compared to dispersion models of
point sources for regional AQ.
Rob Kinnersley
Reintroduced source-transport-receptor-impact-exposure paradigm. E.g.: ammonia
from chicken and pigs. The screening process is catching too many farms which turn
out not to have a problem. Factor of 2 error is common when screening model
22
compared to observations: probably dominated by oversimplification of source terms
as well as uncertainties in wind/meteorology.
Are CFD models mature and fast enough now to provide better operational models
for dispersion?
Are critical loads the right metric for ecosystems? How do we model human health
impacts?
Data assimilation is beginning to become important for hazardous release mapping.
Roy confirmed that studies on optimal sampling could be within scope of the study.
Roger Timmis
Heathrow example: the proposals were accompanied by modelling which was
criticised as having poor model study structure and poor auditability. Uncertainty in
future traffic, chemistry and meteorology will all have to be considered in a
transparent and reproducible way (i.e., not just choosing best possible future).
RT asked us to note cultural differences in modelling for decision-making: USA
stipulate models to be used and even inputs (e.g. a standard set of random
numbers), whilst UK prefer to see problems addressed using best available
technology.
Sarah Metcalfe
Use of GLUE for AQ modelling: HARM atmospheric model. Sensitivity studies for
critical loads: most sensitive to emissions. Which parameters should be assessed?
Are parameters driving uncertainties stable through space and time?
How reliable are measurements?
How to represent uncertainties in standard damage metrics?
Communicating uncertainty to regulators is a persistent problem.
Ron Smith
This team use lots of different models to aid decision-makers.
Modelling of landscape impacts: from EMEP to small-scale ammonia models. More
detailed models can lead to more uncertainty because of the increase in the number
of unconstrained parameters.
Persistent problems (over 20 years!) getting mean SO2 modelled over UK. Models
are routinely in error by factor 2-4.
Shortage of measurements to assess models
Should match scales of model and regulation
There is a need to educate policy-makers – social-science role
23
The fringes of towns are important for the public-perception of environmental quality.
Nick Hewitt
NH’s work spans the space and time domain illustrated in the opening slides.
Often don’t understand emissions for a given scenario/environment, as well as
transport and impacts
Session 2: Sources of uncertainty
Discussion on sources of uncertainty in atmospheric models used for environmental
decision-making was structured under the headings of i) inputs; ii) processes; iii)
outputs and iv) others/wildcards. These were added to separate flipcharts and
grouped where possible.
Inputs
 Types of uncertainty
o Parametric
o Physical inputs (meteorology, background data)
o Behaviour - emissions
o Model resolution

Does background matter? Yes, but we need to test a model’s ability to
predict increment above background. Uncertainty is inherent in background
concentrations

Boundary & initial conditions

Emissions
o Now & future
o Quantifying emission rates – species specific, suitable temporal
resolution, suitable spatial resolution
o Emissions inventory still a key uncertainty, especially for fugitive
emissions – how can back-calculation from monitoring reduce
uncertainty, if at all?
o If you have only £1 to spend to improve model predictions spend it on
emissions.

Met/climate data

Met data that can be generalized – turbulent structure as a function of land
surface for instance. Issue is “what uncertainty is introduced by using generic
24
met data?”, is this acceptable? Would the resource for site-specific
meteorology deliver sufficient improvement to justify?

Expert knowledge on parameters can vary

Explore statistical techniques for getting best estimates of input parameters
& ranges & constraints

Closing the loop (iterative). Identify key inputs (using sensitivities) to improve
our knowledge of them

Don’t ignore scenarios & futures
Process
 Process representation - model structure
o How to assess is the right processes are in the model (need targets
from measurement)
o Allowing for missing processes
o Effective parameters - scaling rules

Multi-scale solution – One atmosphere

Model ensembles – how should we treat these?

Near-source dispersion modelling options

One specific – peri-urban AQ

Relationship between generic models & site specific models

Role of models of different complexity – comprehensive versus “cut-down”,
knowing when to revert back from cut-down version.

Explore consistency & difference between models

Empirical models & monitoring versus compendious modelling

Importance of short-term processes on long-term targets/standards
o Can short time-scale processes affect long term averages? Yes, if
compliance if based on meeting an upper percentile rather than mean.
Non-linearity if also important.

Sensitivity is scale-dependent
o What value are small scale experiments for real world atmospheric
modelling processes?

Chemistry & missing
o Chemistry of ozone & PM precursors.
Outputs
 Not all uncertainty is source – transport – receptor, background often not well
understood
25

Sometimes we cannot express inputs probabilistically meaning we cannot
assess predictive uncertainty, we don’t know enough

Focus on outcome
o Validity/reliability of the regulatory/policy decision
o Most important thing for decision makers is – Is the regulatory/policy
decision fit for purpose?
o Importance of uncertainty for decisions depends on available
headroom
(within
an
ambient
air
quality
standard),
actual
“importance” is inversely related to “headroom”
o Decision usually relate to some “protective” standard e.g. human
health, this standard will have temporal & spatial scales. Most
important uncertainty depends on these scales within the decision
making standard

Reducing uncertainty in real-time models – optimum number and placement
of monitors (context – air quality in major incidents)

How to communicate risk – model output needs to be expressed as
probability of harm

Propagation of errors in modelling

Impact modelling – specific – impact of air pollution on protected habitats e.g.
are critical loads appropriate? (lacking dynamic time component)

Some outputs are more useful than others in terms of constraining models

How can I improve the discriminatory power of screening tools so I don’t
waste time on detailed modelling of low-risk sites?

Data
o Available datasets? What suitable target datasets do we have or not
have? E.g. tracers, smog chamber, wind tunnel, etc.
o “Gold standard” datasets for model verification e.g. DAPPLE tracer
outputs
o Targetted experiments – smog chambers, wind tunnels, field studies
o Collecting data for process-based evaluation
o Improving John Stedman’s maps

Linking model outputs & environmental targets/standards

Rules of thumb
Others/Wildcards
 Most significant uncertainty
26
o Depends on time/space scale of environmental impact
o Usually emissions is most important (ties in with comments under
“inputs”), but it may not always be the case that emissions are more
important

We want reductions in uncertainty that are effective at improving decisions

We want targeted reductions i.e. focus on uncertainties that effect decisions

Model tests
o Intercomparison
o Science within parameterisations
o Comparisons with observations
o Do tests reveal uncertainty?

Conditional selection of observations for model testing – so that tests can
focus on particular parameterised processes and improve them e.g. plume
knock-down, penetration of boundary layer

Better exploitation of existing/archived observed data to test model & reveal
uncertainties e.g. by developing conditionally selected observation pools for
particular processes or impact

How generalisable is uncertainty analysis?

Peaks – focus on identifying processes & uncertainty behind peaks e.g. in a
time series

Try to define uncertainties at different scales: Street, Urban, Regional,
Transboundary, organise in Source-Pathway-Target-Structure

Background

Averaging times

Unknown unknowns!
Session 3: Future research – scale analysis
Participants were invited to write ideas for future research to reduce uncertainty on
Post-It notes and place them upon axes of “Importance” and “Achievability”. Four
sets of these axes were provided for research at atmospheric different scales: i)
street; ii) urban; iii) regional; and iv) trans-boundary.
27
Highly important
Highly achievability
Low importance
Low achievability
Low importance
Highly achievability
Importance
Highly important
Low achievability
Achievability
Street
High Importance / High Do-ability
• Representation of in-street flows
• Very near source - set of guidance on correct use of CFD, etc. in
environment at street-scale
• Dense sensor networks for high temporal & spatial resolution data gathering
(£££)
• Sub-grid scale emissions
• Communicating risk of harm
• Optimised monitor placements to verify model and calculate better source
terms e.g. in major incidences
• Primary NO2 fraction for traffic fleet
• Background [NO2], [O3]
High Importance / Low Do-ability
• Short-time scale modelling (e.g. emergency response) including inverse
models
• Real time wind measurements relevant to target sites
• Source attributes from ambient data for incidents (e.g. fires). Also use of
attributed estimates to predict impacts elsewhere
28
• Representation of traffic demand / instantaneous emissions for free / mixed /
saturated flows
Low Importance / High Do-ability
• Building geometries - effect on disposition
• Turbulent / chemistry interactions
• Tools for assessing uncertainty at this scale. How good are they?
Low Importance / Low Do-ability
• Non traffic emissions. Are they likely to grow?
• Actual, real-life vehicle emissions
Urban
Hi Importance / High Do-ability
• Communicating risk of harm
• Dense sensor networks for high temporal
and spatial resolution data
gathering
• Generalised rules for definition of urban canopy (incorporated into ABMStype & MESO-models)
• Sub-grid scale emissions
• Categorisation of urban roughness - turbulence flow profiles
• Reaction rates for VOC / NOx / O3 chemistry
• Particulates size distributed composition (measure to constrain model)
High Importance / Low Do-ability
• Detection of signal to noise: Can we confirm from (noisy) ambient data if
controls (eg low emission zones) are working?
• Inverse modelling of source terms + use to check if emission controls are
working
• Model resolution
• Non-exhaust
• Estimation of boundary layer heights
Low Importance / High Do-ability
• Methods for quantifying or incorporating segregation. Chemistry / mixing
29
• Tools for assessing uncertainty at this scale. How good are they?
• Improved temporal resolution of emissions
• Urban heat islands - effect on stability
• Numerical scheme
Low Importance / Low Do-ability
• Actual, real life vehicle emissions
• Urban / industrial - relationship between MET data at nearest station & at
site of interest
• Estimation of background
Regional
Hi Importance / High Do-ability
• Communicating risk of harm
• Isoprene chemistry
• Observations to define landscape variation
• Nesting sub-grid scale modelling
• More monitoring to allow genuinely regional assessment of uncertainties
• Deposition parameter
High Importance / Low Do-ability
• MET / AQ / climate feedbacks
• Process model for BVOC’s emissions
• Non-exhaust
• Model resolution
Low Importance / High Do-ability
• Boundary layer height - model, or measurement network? (exploit airport
ceileometer network?)
• More complete emissions inventories (especially agricultural sources)
• Tools for assessing uncertainty at this scale. How good are they?
• LRT of dust / primary particulates
Low Importance / Low Do-ability
• Transport UT - LS
30
• More co location of measurements
• Estimation of background
Transboundary +
Hi Importance / High Do-ability
• Isoprene chemistry
• Sub-grid process
• Communicating risk of harm
• Global dimming and brightness due to pollution increase / decrease
• Deposition parameter
High Importance / Low Do-ability
• Earth-system feedbacks
• MET / AQ / climate feedbacks
• Process model for BVOC’s emissions
• Incorporating frontal transport (& chemistry)
• Transport UT - LS
• More integration with hemispheric / global models to feed into regional
models
Low Importance / High Do-ability
• Tools for assessing uncertainty at this scale. How good are they?
• Effects
of
climate
change
on;
Emissions,
(+impacts/exposure)
Low Importance / Low Do-ability
[purposely left blank]
Session 4: The Ideal Future Research Programme
Future research programme in 2 years’ time.
This session focused on questions to Roy for clarification:
31
Chemistry,
Removal
Can proposals be cross media? Interaction of compartments is natural, but no strong
pressure to build cross-compartment research. Need for govt partner might make it
difficult to develop generic proposals. May need to work with Chief Scientists’ group
to help generic proposals find a home. May have workshops to help learning across
projects.
Does programme have to be very tightly-knit? No, can be scatter-gun so long as
each project is well-defined.
Measurements and monitoring: focused on optimised measurements for reducing
uncertainty – campaign and monitoring both important. Find critical measurements.
Use the term “field observations” rather than “monitoring”. Latter day equivalent of
Prairie grass experiment could be best use of field obs. Are data already in
existence? Perhaps there is room for a project to make UK data as easily accessible
as USA data.
Inverse modelling may be important in finding a role for monitoring.
Lab studies: should focus on key parameters under controlled conditions
Allow data mining, but insist on adding value to databases
Consider replacement of existing regulatory models with more physically realistic
representations
Horizon-scanning: Call should include work on decisions that are coming at us on 5year timescale – e.g. modelling PM2.5, future ozone climate; exposure reduction
paradigm.
Modelling potential solutions – decision-makers will probably keep proposers in line.
For example, policy-relevant study of impact of conversion to electric vehicles.
The call could include proposals for developing a protocol for handling uncertainty in
decision-making.
Focus of the scoping study moved a little during the day: Roy acknowledged that
projects that could impact on a range of DEFRA models (rather than one) would be
in-scope.

A mixture of project types (individual and cross-boundary)

Must have a practitioner partner (could present difficulties for generic
projects)

Workshops to help practitioners to organise their thoughts, develop a protocol

Monitoring / field observations / campaigns

Gold standard dataset to test models

Data mining and using existing data
32

Relating questions to influences / decisions coming down the line

Clear line of sight to decisions being made

How bold are we thinking?
33
APPENDIX B - Workshop 2: Land and Water
Tuesday 29 June 2010
Lancaster Environment Centre, Lancaster University
Attendees
Ruth Alcock
Lancaster Environment Centre
Keith Beven
Lancaster Environment Centre
Julian Dawson
Macaulay Land Use Research Institute
Alan Godfree
United Utilities
Richard Gooday
ADAS
Phil Haygarth
Lancaster Environment Centre
Ed Henderson
National Nuclear Laboratory UK
Ian Jones
CEH
Paul McKenna
Lancaster Environment Centre
Trevor Page
Lancaster Environment Centre
Roger Pickup
School of Health and Medicine,
Kate Snow
United Utilities
Roger Timmis
Environment Agency
Jim Walker
Environment Agency
Tim Morley
Know Innovation
Session 1: Welcome & individual introductions
Phil Haygarth welcomed participants and introduced Tim Morley as the workshop
facilitator. They presented the agenda for the day and gave a general outline of the
scoping study’s history, drivers, aims and objectives incorporating a brief
presentation provided by Roy Harrison at the first seminar. The participants were
then invited to introduce themselves.
Trevor Page – Research Associate, Lancaster Environment Centre. Recent
research interests include diffuse pollution from agriculture.
Richard Gooday – ADAS. Research interests include modelling of diffuse pollution.
Ian Jones – Centre for Ecology & Hydrology. Physical limnologist.
34
Ruth Alcock – Coordinator of Catchment Change Network, a NERC-funded
Knowledge Transfer Network and Editor of Environment International.
Ed Henderson – Senior Technologist (Hydrology), Contaminated Land and Waste
Assessment Team National Nuclear Laboratory.
Jim Walker – Principal Research Scientist - Flooding and Communities, Research,
Monitoring and Innovation, Evidence Directorate, Environment Agency
Keith Beven – Distinguished Professor Lancaster Environment Centre. Research
interests in hydrological modelling and approaches to modelling uncertainty.
Kate Snow – United Utilities, Catchment Policy Manager.
Alan Godfree – United Utilities, Principal Public Health Scientist. Environmental and
public health interests, including drinking water safety.
Julian Dawson – Macaulay Land Use Research Institute. Environmental
biogeochemist, interests include integrated management of diffuse pollution.
Roger Timmis – Environment Agency, Research Expert - Air, Land and Water.
Roger Pickup – School of Health and Medicine, Lancaster University. Microbiologist
with interest in a number of pathogens in the environment.
Phil Haygarth – Co-Director of Centre for Sustainable Water Management,
Lancaster University.
Session 2: Defining the boundaries of the workshop
A general consensus was reached that the workshop should focus upon land and
water interactions from catchment watershed to the near coastal waters. The
workshop will focus upon pollutants within this remit and their impact on human
health, but also upon ecosystems.
Participants were invited to make suggestions of specific topics they would see
included in the discussion
35

Pathogens, specifically cryptosporidium and cyanobacteria (linking to
eutrophication)

Flooding. Participants agreed that the impact of pollutants carried by flood
water, rather than the flood waters themselves, should be within scope.

Irrigation, resulting in food contamination.

Bathing waters

Groundwater. For example, predicting the effects of leaks to groundwater
whether in the specific case of nuclear waste disposal (low level repositories)
or more general groundwater pollution.

Legacy issues. For example, groundwater pollutant plumes which remain
after the pollutant source has been removed; flood sediments re-depositing
pollutants on the floodplain (examples given of dioxins in Bolsover and
positive effects of flooding deposits in New Orleans leading to reduced blood
Pb levels due to capping of contaminated land); Chernobyl fall-out in Northern
British Isles; mine waters.

Sustainable Urban Drainage – though it was commented that the built
environment might be considered within the remit of EPSRC rather than
NERC, it was pointed out that proposals that could require cross-research
council collaboration should not be dismissed.

Socio-economic factors – especially upon predictions of future scenarios.

High magnitude – low frequency events – are often of most importance
and can be most difficult to manage or prepare for.
A contrast was drawn between this and the first (Atmosphere) workshop in that there
is less modelling commonality in the Land-Water field, with this in mind the group
developed a framework for the subsequent discussions which involved addressing
uncertainties in different environmental compartments. Roger Timmis provided an
extract of an Environment Agency report which structured the factors to consider in
the release of pollutants into the environment through a number of categories, each
with a menu of choices. The figure below shows an example of this for waterbourne
releases with each factor having an associated “menu” of choices.
36
From Environmental Analysis Co-operative (1999), Emissions and your licence to
operate. A guide for assessing releases to the environment, ISBN 0 88295 4239.
The participants agreed that with was a good framework for structuring the
discussion of sources of uncertainty.
Sessions 3 and 4: Sources of uncertainty
Participants were invited to consider sources of uncertainty in Land-Water
applications and write them on Post-It notes, these were then placed upon axes of
“Levels of uncertainty” and “Importance to decision makers”. Seven sets of these
axes were provided for a sequence of environmental compartments, as defined by
those present:
i)
substance
ii)
source
iii)
mobilisation
iv)
delivery
v)
transformation
vi)
exposure
vii)
health impacts
Roger Pickup gave an example of mycobacterium paratuberculosis and described
how it mapped onto the compartments. Other named pollutants (e.g. cyanobacteria)
or groups of pollutant type (e.g. pharmaceuticals) were mapped through the
compartments.
37
High uncertainty
Highly important
Low uncertainty
Low importance
Low uncertainty
Highly important
Uncertainty
High uncertainty
Low importance
Importance to decision makers
38
Substance
Drugs & Medicine
Veterinary Medicines
High
uncertainty
Emerging pathogens, or yet to
emerge
Emerging pollutants (e.g. nano
particles)
Personal Care Products
Organic Compounds Solvents
(TCE, etc) Hydrocarbons (eg
MTBE, BTEX)
Substances mobilised by floods
to floodplains (e.g. heavy
metals, other industrial
pollution)
Pollutants in urban surfaces
waters
Trace metals
Pesticides (organic
contaminants)
Low
uncertainty
Sediment & associates
Radio-isotopes
Enteric viruses
Low importance to decision
makers
Pollutant mixtures with
antagonistic effects
Particle associated substances
Cyanobacteria
Phosphorus
Nitrogen
Cryptosporidium
Viruses
Known bacteria
Protozoa
High importance to decision makers
39
Sources
Pathogen quantification
High
uncertainty
Diffuse source
Terrorism
Drugs/medicines
Animals-food
Sewage sludge
Cyanobacteria
Nutrients - internal
- diffuse
– point
Contaminated sediments
Industrial storage on floodplain
Pollutants picked up by flashy
urban runoff
Crypto – other species
Crypto - wild mammals
Crypto - humans (septic tank,
sewers, WWTW)
Firewater (e.g. Buncefield)
Petrochemicals, deposition,
point sources
Land use (cropping/ poaching
stream banks)
Societal aspects of
sources/release rates
(consumption, activity,
behaviour) linked to policy
Animal
Human
Environment
Low
uncertainty
Leaks to ground disposals
Industrial contamination
Animal?
Crypto – farm animals
Sewage
Point sources
Phosphorus fertilizers
Feed concentrates
Detergents
High importance to decision
makers
Low importance to decision
makers
40
Mobilisation
Control by rainfall substrate
High
uncertainty
Crypto – point sources –
WWTW, septic tanks
Phosphorus mobilisation by
septic tanks
Drugs & medicines – water &
particles
Flashy urban runoff
Run-off from land receiving
organic wastes
Suspended sediments
Dissolved phases
Crypto – surface runoff from
grazing land
Crypto –muck/slurry spreading
Drugs & medicines – water
-particles
Soil erosion
Cyanobacteria – Nutrients –
Internal
Diffuse
Point
Hydrological conditions
Raindrop impact
Substances (re)introduced from
environmental storage latency
reservoirs after time delay
Pathogen route/quantification
Flood discharges (velocity,
location)
Low
uncertainty
Solubilisation
Snowpack storage – mobilised
in spring thaw
Reservoir turn over
Immediate to groundwater or
release from unsaturated zone
Low importance to decision
makers
Direct discharge WWTW to
water body
Phosphorus – particle
detached solubilisation – from
point sources
High importance to decision
makers
41
Delivery
Control by - hydro-driver
- substrate
-sinks
High
uncertainty
Drugs & medicines
Phosphorus delivery from
septic tanks
Groundwater movement &
transport
Flashy urban runoff
Drainage
Flood routing, overbank flows &
deposition
Cyanobacteria – Nutrients –
Internal
Diffuse
Point
Coastal transport (currents)
Pathogen delivery
Phosphorus delivery of solutes
and particles
High flows
Sedimentation
Water column transport
Crypto – through soil filtration
Low
uncertainty
Rivers to bathing areas
Attenuation strategy = negative
delivery
Low importance to decision
makers
High importance to decision
makers
42
Transformation
Can increase or decrease “harm” to receptors (health) e.g. radionuclide decay – decrease in harm;
organic material with metals attached, degradation releases bound metals.
Transformation of all P to algal
input in receiving water
Complex – controlled by
“reactivity” – chemical,
biological
High
uncertainty
Crypto – in reservoir/stream
degradation
Mechanism known –
environmental rate unknown
Geochemical processes &
take-up by crops
Cyanobacteria – nutrients,
climate, in-lake interation
Survival T90?
Adsorption
Deposition & remobilisation in
stream
Degradation (irganic)
Immobilization
Re-suspension
Bioaccumulation
Drugs & medicines –
biologically active
Geochemical reactions
Low
uncertainty
Little transformation in urban
runoff
Low importance to decision
makers
High importance to decision
makers
43
Exposure
High
uncertainty
Water food - drugs &
medicines, endocrine
disruption, synergistic impacts
Crypto – hand to mouth from
contaminated land / farm visits
Through flood plain food
sources
Crypto – bathing waters
Exposure via food
Water quality – food (aquatic &
on land)
Direct impact on people &
property
Flood inputs to reservoirs
Bathing
Phosphorus / eutrophication –
exposure to cyanobacteria
through swimming, bathing in
lakes, sea and rivers
Exp. Pathogens
Dose varies – host, pathogen
Particles (estuaries)
abstractions from groundwater
for drinking water, beach
springs, food (fish, seaweed)
Drinking
Low
uncertainty
Inhalation of toxin (red tides)
Recreational
Recreational drinking water
Crypto – drinking water
Low importance to decision
makers
High importance to decision
makers
44
Health impact
High
uncertainty
What’s on the sediment?
Health impacts of P &
cyanobacteria to humans –
rashes, skin disorders, vomiting
Health impact varies – host,
pathogen
Drugs & medicines, endocrine
disruption, synergistic impacts
Dose-response ingestion
pathway
Pathogenicity
Direct & indirect (Human &
ecosystem)
Controlled releases monitored
Potential exposed group
studies
Effects of low doses?
Dose-response
Crypto can kill – immune
compromised
Low
uncertainty
Low importance to decision
makers
Cyanobacteria – potentially
fatal
High importance to decision
makers
45
Specific models or types of models which could be considered (this list was not
intended to be comprehensive):

Source Apportionment – Faecal Indicator Organisms/Pathogens

Simple source model – Input = Activity x Release Rate

Soil Erosion – Universal Soil Loss Equation

Sorption-Desorption models (Langmuir)

Rainfall-flood models

Flood Estimation Handbook (FEH) models

Lake models e.g. PROTECH

Geochemical models – transformation from solid to dissolved

Societal behavioural model

Epidemiological models

Toxicological models

Dermatological models

Risk-based assessment of individual pollutants

CFD

Phosphorus Indicator tool

SCIMAP

INCA P

PSYCHIC

PEDAL

MOPS

MODFLOW
Sessions 5 & 6: Future Research Priorities
The scoping study should learn from work being carried out by projects such as the
Virtual Observatory and the Demonstration Test Catchments (Eden, Wensome and
Hampshire Avon). Where possible the Environment Agency’s experiences relating to
the Water Framework Directive should be used. United Utilities are part of an £11M
project examining land management practices in agricultural environments and their
impact upon pollution, they were also involved in SCAMP (Sustainable Catchment
Management Programme) which has now ended but could still be used as a good
source of data for reducing uncertainty in some areas of modelling (e.g. data relating
to water colour).
46
In addition to the factors listed in Session 4, and most important those in the top right
hand corner of the axes some other priorities for future research were identified:

Critical observations – identifying data which can address specific sources of
uncertainty.

Developing a protocol to deal with uncertainty, perhaps akin to guidelines for
best practice developed by the Flood Risk Management Research
Consortium. The site specific nature of the land-water theme presents
particular problems when trying to develop a generic approach to handling
uncertainty.

Pollutant mixtures – potentially with synergistic or antagonistic effects

Knowledge uncertainty – leading to scenario analysis

The concept of headroom for regulators, dealing with uncertainty with
background levels and permitted standards.

Influence other bodies (e.g. NIREX) to promote addressing uncertainty in
their research programme
47
APPENDIX C - Workshop 3: Chemicals in the Environment
Monday 5 July 2010
Lancaster Environment Centre, Lancaster University
Attendees
Richard Glass
FERA
Todd Gouin
Unilever
Crispin Halsall
Lancaster Environment Centre
Helinor Johnston
Defra
Kevin Jones
Lancaster Environment Centre
Marc Kennedy
FERA
Steve Lofts
CEH
Paul McKenna
Lancaster Environment Centre
Camilla Pease
Environment Agency
Frances Pollitt
Health Protection Agency
Andy Sweetman
Lancaster Environment Centre
Paul Whitehouse
Environment Agency
Tim Morley
Know Innovation
Session 1: Welcome & individual introductions
Andy Sweetman gave a brief presentation outlining the structure of the Scoping
study and the two previous workshops that had taken place (Atmosphere and LandWater).
He then described how the use of numerical models for simulating
environmental processes and predicting the impact of change is often limited by the
uncertainties inherent in these models and the data they use. Addressing these
modelling uncertainties is of particular importance to environmental decision makers,
who need to determine where best to direct resources to improve confidence, by
reducing these uncertainties, mitigating their effects, or managing in a way that “lives
with” uncertainty that is impracticable to reduce further.
Each participant was then invited to provide a brief introduction.
Marc Kennedy – FERA statistician. Interests include uncertainty in complex
computer models, dietary exposure assessments, operator exposure.
48
Richard
Glass
–
FERA.
Interests
include
pesticide
exposure
(including
environmental, worker and bystander), cumulative and aggregate exposures.
Paul Whitehouse – Evidence Directorate, Environment Agency. Interests include
environmental standards for flora and fauna, Water Framework Directive,
environmental protection versus the cost of appliance, mixtures, chemical speciation.
Camilla Pease - Environment Agency, toxicologist. Works closely with HSE, FSA,
HPA, previously with Unilever.
Helinor Johnston – Defra. Chemicals and nanotechnology. Interested in POPs and
emissions sources.
Kevin Jones - Lancaster Environment Centre, Director is Centre for Chemicals
Management. Interests span all environmental compartments up to toxicology.
Todd Gouin – Unilever. Environmental fate modeller. Concerned with risk
assessment of their products.
Frances Pollitt - Health Protection Agency. Mammalian toxicologist. Interests
include drinking water and contaminated land. Works with Committee on Toxicology
and Committee on Carcinogenicity the latter of which has recently looked into
mixtures.
Steve Lofts – Centre for Ecology and Hydrology. Environmental chemist with
interests in bioavailability and chemical speciation of metals.
Andy Sweetman - Lancaster Environment Centre. Interests include use of
environmental fate and exposure models for chemicals including POPs and dioxins.
Has worked closely with Defra.
Crispin Halsall - Lancaster Environment Centre. Interests include fate of persistent
chemicals in sensitive environments such as the polar regions.
49
Session 2: Defining the boundaries of the workshop
Andy Sweetman presented a slide (below) providing a simple summary of the
environmental processes and compartments to be considered when modelling
chemical fate and exposure.
Sources of uncertainty in chemical fate and exposure models
The topic of chemical mixtures was highlighted as one which the study must
include due to the high levels of uncertainties involved in predicting the effects
(synergistic or antagonistic) of pollutants in combination.
Temporal and spatial scales are important elements of the study, with the spatial
scale varying from local, through regional to global and temporal scale varying from
minutes to centuries.
The science of communicating uncertainty and risk should also be within the remit
of the study, as should the concept of living with uncertainty.
A number of conceptual points were raised including: whether the decision-making
process is wholly integrated within the modelling process; whether regulation based
upon the most susceptible receptors is the best methodology; and if probabilistic
methods are perceived as being too complex for practitioners. Where regulation
could currently be seen to be over-cautious, is the cost of over-regulation being
properly addressed?
Sessions 3 and 4: Sources of uncertainty
It was agreed amongst the participants that, due to the extreme breadth of the
Chemicals remit, it would be usual to focus discussions on a number of carefully
chosen scenarios. The following five were settled upon:
1. River spillage/discharge
2. Bystander exposure to pesticides
3. Contaminated land impact on wildlife and human health
4. Global POPs
5. Nanomaterials
50
Participants were invited to consider sources of uncertainty in these scenarios and
write them on Post-It notes, these were then placed upon axes of “Levels of
uncertainty” and “Importance to decision makers”. A consensus was reached that
three sets of these axes should provide a sequence of environmental compartments,
these were:
source
ix)
pathway
x)
receptor
High uncertainty
Low importance
High uncertainty
Highly important
Low uncertainty
Low importance
Low uncertainty
Highly important
Uncertainty
viii)
Importance to decision makers
Scenario 1. River spillage/discharge
Where river discharges are consented there is usually a need for modelling before
consent is given, in contrast accidental releases may not allow the time for sitespecific modelling to take place.
There is often great uncertainty or sometimes complete lack of knowledge of some
physical and chemical properties of substances, such as partitioning coefficients.
Data about the impact of the substance on human health can often be lacking or
uncertain, especially for chronic rather than acute effects.
51
The effect of atypical environmental factors (e.g. frozen river) may be uncertain,
along with information about the quantity, form and phase of the chemical in the
environment and its bioavailability, persistence or subsequent degradation.
The long-term impact of regulatory actions upon the environment may also be a
source of uncertainty, for example a decision to cease water abstraction at a point
might have long-term impacts upon other parts of the catchment.
Scenario 2. Bystander exposure to pesticide
Exposure can occur via two main routes:
1. During spraying via liquid drift from fields over a short period of time. Dermal
exposure, assumptions made about proportion of body exposed to the
environment, deposition, hand to mouth transfer.
2. After spraying, vapour from volatile components over a period of up to 2
weeks. Exposure via inhalation.
Data can be limited on chronic effects to humans, uncertainties exist in extrapolating
between animal and human impacts and some impacts cannot be observed in
animals e.g. headaches and depression.
Spray drift models may be good but uncertainties may result from emissions e.g.
particle size distributions, concentrations, quantity. Also windspeed data can be a
source of great uncertainty, especially for short distances and over short time scales.
The spatial extent of measurable impacts is also a source of uncertainty.
Regulation is often based upon the identification of sensitive receptors (e.g. elderly,
children), this then leads to conservatism in regulation and may result in an
imbalance between the benefits of regulation and the costs of implementation. There
are also issues of bystander (resident) exposure being more politically sensitive than
that of operator exposure due to the influencing concept of personal choice.
Scenario 3. Contaminated land impact on wildlife and human health
This scenario involves a complex exposure route involving land, air, water and
various forms of ingestion/exposure. Models tend to work on worst case scenario,
e.g. a person who eats only produce grown on an allotment built on contaminated
land. Models are time aggregated over lifetime, often leading to multiplication of
conservativeness. The impact of background concentrations is important and can be
52
a source of uncertainty, as can the lack of data on bioavailability. The SPOSH
(Significant Possibility of Significant Harm) approach is relevant to Defra regulation
on contaminated land. The CLEA model was discussed with its focus on worst case
scenarios and deterministic approach being criticised, this was seen as a potential
case study for future work to reduce uncertainty.
http://www.defra.gov.uk/environment/quality/land/contaminated/documents/circular01
-2006.pdf
Scenario 4. Global POPs
POPs present special challenges in environmental modelling due to their long range
transport in the atmosphere and their accumulation in the food chain. There are
newer chemicals which have only partial international bans. Uncertainties exist in the
quantity and location of sources, both primary and secondary. Modelling questions
persist over pathways taken, compartment partitioning and chemical fate in the food
chain. Models are used to predict the impacts of bans, national implementation plans
and climate change. Human health impacts can be difficult to model due to subtle
toxicology and epidemiology. The effect of mixtures is also difficult to predict.
Scenario 5. Nanomaterials
There are many knowledge gaps in the modelling of nanomaterials in the
environment. The approach taken tends to be one of Life Cycle Analysis (LCA) and
is applied to a vast range of materials, often without enough data to apply to models.
Questions exist over the applicability of existing models for well known chemicals to
new nanomaterials which might have structural similarities (e.g. asbestos and carbon
nanotubes). Should these materials be considered as physical or chemical hazards?
They are sometimes used to deliver drugs or pesticides which then brings into
question the issues of synergistic effects, they may also interact with environmental
matter e.g. dissolved organic matter. Some of the materials may undergo changes in
physical form. They tend to be difficult to monitor in the environment. Big
uncertainties may exist in the sources terms for models, source characterisation is
important.
The summary below groups the points raised into the four areas indicated in the
figure above.
Source
53
High Importance / High Uncertainty
•
Volatilisation (amount)
•
Formulation effects on volatilisation
•
Quantity
•
Source characterisation - form, quantity. e.g. nano
•
Quantity (emission) - multiple sprays
High Importance / Low Uncertainty
•
POPs. Global POP sources: Where, When, How much?
•
Analytical measurement uncertainty
Low Importance / High Uncertainty
•
Emission factors. Activities / TGD. Process cat. Env release cat.
•
Test substance relevance in toxicological studies VS reality
•
Spatial emission patterns. primary vs secondary emission sources
Low Importance / Low Uncertainty
•
Changes in physical form (in the environment)
•
What is the phase? Pure chemical, mixture, powder, etc.
Pathway
High Importance / High Uncertainty
•
Cannot measure exposure (nanomaterials)
•
Nano-particle physical form - how to handle in models?
•
Unknown / unexpected pathways
•
Chemical degradation
•
Physical - chemical property data
•
Sensitivity to interaction with environment
•
Extreme consumer
•
Other source / pathway receptors
High Importance / Low Uncertainty
•
Bioavailability (local water conditions)
•
What is the receiving medium? (air/soil/water?)
•
BREAM (spray) - ADMS (dispersion)
•
“Lag time” between source area and remote area. in chemical concs.
54
•
Available dilution. Site specific
•
Medium transport. wind speed / wind direction - existing models OK
Low Importance / High Uncertainty
•
purposely left blank. no responses
Low Importance / Low Uncertainty
•
CLEA model is used in UK - Universally - refine parameters?
•
Assumptions made in models. E.g. CLEA model are worst-case scenario
Receptor
High Importance / High Uncertainty
•
Toxicity to wildlife
•
What is the receptor?
•
Unknown / Unexpected receptors
•
How toxic is the chemical? QSAR receptor (acute)
•
Exposure to nanomaterials - quantity? - form (physical)??
•
Extrapolation of toxicity findings from test species to humans (esp. children)
•
SPOSH 2A - Legalisation!! what constitutes significant harm?
•
Sensitivity of receptor
•
The dose in man / x received that causes effects. What Tox studies are
there? could be none
•
Ingestion. Bioaccessibility to Bioavailability
•
Contaminated land scenarios - assumptions of worse case / realism?
•
Has systemic exposure occurred via ingestion / inhalation / skin absorption.
Often assume 100%. Data to refine?
•
Human variations in physiology and effects susceptibility
•
Is test substance in toxicological study same as characterised in site
sampling?
•
Short term exposure effects - re. chronic vs conservatism in considering
lifetime exposure. How to take into account time effects?
•
Foodchain models - who eats what? - how much metabolism + degradation?
•
Total exposure dose (from lots of sources) to Human receptor
55
High Importance / Low Uncertainty
•
Is there a background exposure & what is the added increment?
Low Importance / High Uncertainty
•
purposely left blank / no responses
Low Importance / Low Uncertainty
•
Can’t measure all possible endpoints in animals (e.g. headaches, depression,
non-specific symptoms)
Sessions 5 & 6: Future Research Priorities
It was apparent from sessions 3 and 4 that the receptor compartment is the area in
which most factors were identified as being important to decision makers and
sources of high levels of uncertainty. Many of these factors pertain to the
uncertainties
in
toxicology
and
epidemiology,
sometimes
resulting
from
extrapolation of human data from test to animals. Potentially significant reductions in
uncertainty could result from collection of critical datasets, though this data
collection work might need to be incentivised. Data on chronic health impacts are
particularly lacking for many substances, but this requires a commitment to long term
monitoring.
A recurrent topic during the discussions was the importance of chemical speciation,
this is a source of uncertainty due to a lack of knowledge on effects relating to
different chemical species. Similarly, the effects of different chemical mixtures was
identified as a significant uncertainty source.
The communication/presentation of uncertainty to legislators (and the public) is an
important area needing future research.
Although many models may possess common features, properties or even submodules, it is not possible to develop one unified model but it may be possible to
develop a common framework for assessing the effects of uncertainty for any given
model, with a possible aim of identifying the areas upon which further research
should be focussed to most effectively reduce uncertainty, this could be combined
with an approach which balances the costs of implementation with the beneficial
impacts of regulation. Such a framework could be applied initally to commonly
56
accepted models such as CLEA, EUSES, BREAM, MCRA and CRÈME as
exemplars.
Reference should be made to the EU Technical Guidance Document on Risk
Assessment
http://ecb.jrc.ec.europa.eu/documents/TECHNICAL_GUIDANCE_DOCUMENT/EDITI
ON_2/tgdpart2_2ed.pdf
57
APPENDIX D – Modelling case studies
Case study 1
Contaminated Land Exposure Assessment
Model:
CLEA
Developer:
Environment Agency
Reference:
Defra (2002): Contaminated land exposure assessment model
(CLEA): Technical basis and algorithms. UK Department for
Environment, Food and Rural Affairs, London, report CLR10
Use:
Contaminated Land Exposure Assessment. The model is used
for assessing exposures levels resulting from direct and indirect
contact with contaminated soil and for defining soil guideline
values. Limited coverage of exposure situations and pathways
and conservative nature restrict its use as a comprehensive
exposure assessment tool.
Commissioned by: Environment Agency
Users:
Local authorities, consultants
Availability:
Free to download from
http://www.environmentagency.gov.uk/clea
Priority uncertainty source(s):
Uncertainties in aggregating average daily exposures to get
longer term exposures.
Improved data on exposure frequency.
Scenario uncertainty.
High levels of uncertainty in some soil–soil solution partition
coefficients.
Uncertainty in significance of pica behaviour in children as an
input.
Lack of consideration of soil chemistry in determining soil-plant
uptake factors of metals and metalloids (e.g. cadmium, mercury,
arsenic). Not considering this introduces potentially significant
uncertainties into predicted plant matter concentrations of these
chemicals.
Call user partner:
Environment Agency - Evidence Directorate (Air, Land & Water),
the directorate is expected to reach a decision on the future
support for CLEA in October 2010.
58
Case study 2
Chemical risk assessment
Model:
European Union System for the Evaluation of Substances
(EUSES)
Developer:
TSA Group Delft bv
Reference:
Vermeire T. G., Jager D. T., Bussian B., Devillers J., den Haan
K., Hansen B., Lundberg I., Niessen H., Robertson S., Tyle H.,
van der Zandt P. T. J., European Union System for the
Evaluation of Substances (EUSES). Principles and structure,
Chemosphere, Volume 34, Issue 8, April 1997, Pages 18231836, ISSN 0045-6535, DOI: 10.1016/S0045-6535(97)00017-9.
Use:
Chemical risk assessment. Decision-support instrument, which
enables users to carry out rapid and efficient tiered assessments
of the general risks posed by substances to man and the
environment. The exposure assessment, effects assessment
and risk characterisation are carried out for environmental
populations as well as for human beings, including workers,
consumers and man exposed through the environment.
Commissioned by: Co-ordinated effort of EU Member States, the European
Commission and the European Chemical Industry.
Users:
Government authorities, research institutes (e.g. CEH) and
chemical industry, consultants
Availability:
Free to download from http://ecb.jrc.ec.europa.eu/euses/
Priority uncertainty source(s):
EUSES estimates Predicted Environmental Concentrations of
chemicals which are then assessed for risk by comparison with a
Predicted No Effect Concentration derived from toxicity data.
There are key uncertainties associated with this framework:
1. Chemicals are assessed singly – there is no provision for
taking the possible effects of mixtures into account and little
understanding of the degree of uncertainty that this introduces
into the calculation of overall risk.
2. EUSES models the concentrations of a contaminant in
environmental compartments (soil, water, sediment) assuming a
steady state situation (e.g. losses from soil equal inputs). For
many metals and metalloids, the assumption that steady state is
suitable may be flawed, since these chemicals accumulate
slowly in the environment and are removed from a given
59
compartment at a much slower rate than would be the case
under steady state conditions. There is thus considerable
uncertainty associated with the EUSES modelling of such
chemicals.
3. Level III fugacity model to distribute chemicals in a six
compartment world.
4. Exposure routes for human/biota receptors.
Call user partners: Centre for Ecology & Hydrology (CEH) - Biogeochemistry
Programme
Defra - Chemicals and Nanotechnologies Division
Environment Agency - Evidence Directorate
60
Case study 3
Atmospheric dispersion modelling
Model:
ADMS-Urban
Developer:
CERC Environmental Software and Services
Reference:
D.J. Carruthers, R.J. Holroyd, J.C.R. Hunt, W.S. Weng, A.G.
Robins, D.D. Apsley, D.J. Thompson, F.B. Smith, UK-ADMS: A
new approach to modelling dispersion in the earth's atmospheric
boundary layer, Journal of Wind Engineering and Industrial
Aerodynamics, Volume 52, May 1994, Pages 139-153, ISSN
0167-6105, DOI: 10.1016/0167-6105(94)90044-2.
Use:
Modelling the dispersion of buoyant or neutrally buoyant
gaseous and particulate emissions to the atmosphere, used as a
tool for tackling air pollution problems in cities and towns. It can
be
used
to
examine
emissions
from
6000
sources
simultaneously. Emissions can be of any duration. It is used to
assess current and future air quality with respect to the air
quality standards such as the EU Air Quality Directive, UK
NAQS. The effects of plume rise, wet and dry deposition, radioactive decay, hilly and variable roughness terrain, coastal
regions and large buildings are allowed for. Output includes
mean values, variances and percentiles of air concentration,
dosage (time integrated concentration raised to some power),
and deposition to the ground. For emissions of radio-active
isotopes estimates can be made of the ground level gammaradiation dose rate beneath the plume. A user interacts with the
model via a system of menus based on Microsoft Windows so
that it is easy to use.
Commissioned by: 12 public bodies including HMIP
Users:
Local authorities, Defra (project EPG 1/3/176), Environmental
Consultants, Environment Agency.
Availability:
Sold commercially by CERC
Priority uncertainty source(s):
Modelling of fluctuations in concentration for time scales less
than one hour. Modelling during low wind speeds. Accuracy of
Gaussian
dispersion
in
conditions
of
complicated
and
heterogeneous terrain (dispersion modelling vs. Computational
fluid dynamics modelling).
61
Call user partners: Environment Agency - Air Quality Modelling and Assessment
Unit (AQMAU)
CERC (Developers of ADMS software)
62
Case study 4
Surface Air Quality
Model:
Ozone Source-Receptor Model (OSRM)
OSRM is a Lagrangian trajectory model of tropospheric
photochemistry. The model runs on externally generated 4-day
back trajectories (generated by the UK Met Office NAME model,
for example) ending on a 10 x 10 km grid covering the UK. The
model ingests NAEI and EMEP emissions. Chemical conversion
is calculated using the STOCHEM chemistry scheme (circa 70
chemical species and 180 reactions).
Reference:
Hayman, G.D., J. Abbott, C. Thomson, T. Bush, A. Kent, RG
Derwent, ME Jenkin, MJ Pilling, A., Rickard and L. Whitehead,
(2006a) “Modelling of Tropospheric Ozone”. Final Report
(AEAT/ENV/R/2100 Issue 1) produced for the Department for
Environment, Food and Rural Affairs and the Devolved
Administrations on Contract EPG 1/3/200.
Developer:
AEA Technology & Environment
Use:
UK Air Quality policy applications
Commissioned by: Defra
Users:
Defra, UK Devolved Authorities, AEA Technology & Environment
Availability:
Proprietary
Priority uncertainty source(s):
(a)
uncertain/incomplete
process
representations:
(i)
boundary layer dynamics (incomplete); (ii) volatile organic
compound chemistry (incomplete)
(b)
uncertain model parameters: (i) rate coefficients for
lumped
parameters;
(ii)
photolysis
rates;
(iii)
deposition
velocities; (iv) temporal emission factors
(c) uncertain inputs/boundary conditions: (i) emissions,
particularly the
speciation
of
volatile
organic
compound
emissions (anthropogenic and biogenic) and sub-diurnal time
variation; (ii) uncertainties in trajectories, particularly when using
forecast trajectories or climate model output; (iii) chemical
boundary conditions
(d) missing factors: (i) 3D stirring and mixing, including the
effect of mixing on chemistry; (ii) convection
Call user partner:
Defra - Air Quality and Industrial Pollution (AQIP) Division
63
Case study 5
Pesticide bystander and residential exposure assessment
Model:
Bystander
and
Residential
Exposure
Assessment
Model
(BREAM)
Developer:
The Arable Group (TAG)
Reference:
M.C. Butler Ellis, A.G. Lane, C.M. O'Sullivan, P.C.H. Miller, C.R.
Glass, Bystander exposure to pesticide spray drift: New data for
model development and validation, Biosystems Engineering, In
Press, Corrected Proof, Available online 30 June 2010, ISSN
1537-5110, DOI: 10.1016/j.biosystemseng.2010.05.017.
Use:
Bystander and Residential Exposure Assessment Model
Developed as a replacement to current empirical exposure
assessment approaches. The model is suitable for regulatory
use for bystander exposure to pesticides resulting from spray
drift from typical arable boom sprayers.
Commissioned by: Pesticide
industry,
regulators
(e.g.
Pesticide
Regulation
Directorate)
Users:
Defra
Availability:
Contact Defra/TAG
Priority uncertainty source(s):
Process descriptions: address the assumptions made about
dermal exposure and ingestion;
Inputs: improving data on emissions
Outcomes: need for more epidemiological data on chronic
effects on humans; research into improving the translation of
results of tests on animals to humans; extrapolation of data to
model life time exposures.
Call user partner:
The Food and Environment Research Agency (FERA) - Human
and Environmental Exposure division.
64
Case study 6
Groundwater modelling: exposure to radionuclides
Model:
TRAnsport and Fluid Flow Including geoChemistry (TRAFFIC)
Developer:
BNFL
Reference:
Watts, L., Janin, S., and McGarry, R., A methodology for
estimating predictive uncertainty in groundwater contaminant
modelling
using
the
hydrogeochemical
transport
code,
TRAFFIC, Calibration and Reliability in Groundwater Modelling,
(Proceedings of the ModelCARE 96 Conference held at Golden,
Colorado, September 1996). IAHS Publ. no. 237, 1996. 571,
1996.
Model:
MODFLOW
Developer:
USGS
Reference:
Harbaugh, A. W., and McDonald, M. D. “User’s documents for
MODFLOW-96, an update to the U.S. Geological Survey
modular finite difference ground-water flow model,” U.S.
Geological Survey Open File Report 96-485, 1996.
Availability:
Download from
http://water.usgs.gov/nrp/gwsoftware/modflow2000/modflow2000.html
Model:
Finite Element subsurface FLOW system (FEFLOW)
Developer:
WASY GmbH
Reference:
Trefry, M.G.; Muffels, C., FEFLOW: a finite-element ground
water flow and transport modeling tool, Ground Water 45 (5):
525–528, 2007.
Availability:
Download and obtain licence at http://www.feflow.info
Use:
Modelling groundwater flow.
Users:
Sellafield Ltd, NNL, Consultants, e.g. Serco Ltd.
Priority uncertainty source(s):
A range of models are used in the prediction of transport in
groundwater and a number of common uncertainties require
addressing, these include source chemistry and release
mechanisms, gas generation and migration, dual/double porosity
effects, geosphere geochemistry (including sorption) and
fracture-matrix interaction.
Call user partners: Sellafield Ltd Land Quality Team
65
National Nuclear Laboratory (NNL) - Contaminated Land and
Waste Assessment Team
Low Level Waste Repository (LLWR) Ltd
66
Case study 7
WFD classifying good ecological status
Model:
WinBUGS (Bayesian analysis Using Gibbs Sampling)
Developer:
Cambridge University & Imperial College
Reference:
Spiegelhalter, D. J.; Thomas, A.; Best, N.; Gilks, W. R. BUGS
0.5: Bayesian Inference Using Gibbs Sampling Manual; Medical
Research Council Biostatistics Unit, Institute of Public Health:
Cambridge, UK, 1996.
Use:
Flexible software for the Bayesian analysis of complex statistical
models using Markov chain Monte Carlo (MCMC) methods. The
Water Framework Directive (WFD) requires water bodies to be
classified in five quality classes (high, good, moderate, poor,
bad) for ecological status.
The criteria for high, good and
moderate ecological status described in the WFD need to be
made operational because they will be used to set the practical
quality targets for surface water management.
Commissioned by: MRC
Users:
Environment Agency
Availability:
Free to download from
http://www.mrc-bsu.cam.ac.uk/bugs/welcome.shtml
Priority uncertainty source(s):
The Environment Agency uses the WinBUGS Bayesian
approach applied to fisheries data to generate classifications of
ecological status and is interested in having further work carried
out in this area. The Agency is willing to provide access to the
code and datasets, along with induction in their use.
Call user partner:
Environment Agency – Evidence Directorate Risk & Forecasting
and Air, Land & Water.
67
Case study 8
Pollution Climate Mapping
Model:
PCM
Developer:
AEA Technology
Reference:
Grice, S., Cooke, S.L., Stedman, J.R., Bush, T.J., Vincent, K.J.,
Hann, M., Abbot, J. and Kent, A.J. (2008) UK air quality
modelling for annual reporting 2007 on ambient air quality
assessment under Council Directives 96/62/EC, 1999/30/EC and
2000/69/EC. Didcot, AEA (Report AEAT/ENV/R/2656).
Use:
Compliance assessment for reporting against EU ambient limit
values. Baseline projections for evaluation of future compliance
and scenario analysis to assess policy options. This is a
spreadsheet model and closely calibrated to monitored data.
The scope to improve model uncertainty is probably more limited
than for other models but it is a very important policy tool.
Commissioned by: Defra
Users:
AEA Technology under contract to Defra
Availability:
Not freely available but Defra owns the model and is happy to
make it available – investment would probably be required for
other organisations to successfully learn to use the model
Priority uncertainty source(s):
Consultation with contractors required.
Projected roadside concentrations of PM2.5
Call user partner:
Defra – Science and Evidence Team – Atmosphere and Local
Environment Programme
68
Case study 9
Air Quality Forecasting
Model:
Community Multiscale Air Quality Modeling System (CMAQ)
Developer:
USEPA
Reference:
Byun and Schere, 2006 D. Byun and K.L. Schere, Review of the
governing equations, computational algorithms, and other
components of the Models-3 Community Multiscale Air Quality
(CMAQ) modeling system, Applied Mechanics Reviews 59
(2006), pp. 51–77.
Use:
Air Quality forecasting by Defra. Also looking into developing
other uses.
Commissioned by: USEPA
Users:
AEA Technology under contract to Defra.
Univ of Hertfordshire under contract to EA
King’s College, London
Power generators
Availability:
Freely available. Model code is open source and available from
http://www.cmascenter.org/index.cfm
Priority uncertainty source(s):
Consultation with contractors required.
Call user partner:
Defra – Science and Evidence Team – Atmosphere and Local
Environment Programme
69
Case study 10
Emission reduction scenario modelling
Model:
UK Integrated Assessment Model (UKIAM)
Developer:
Imperial College
Reference:
Oxley, T., H. ApSimon, A. Dore, M. Sutton, J. Hall, E. Heywood,
T. Gonzales del Campo and R. Warren. 2004. The UK
Integrated Assessment Model, UKIAM: A National Scale
Approach to the Analysis of Strategies for Abatement of
Atmospheric Pollutants Under the Convention on Long-Range
Transboundary Air Pollution. Integrated Assessment. 4(4):236249.
Use:
Integrated assessment modelling tool used to evaluate and
optimise the impacts of different emission reduction scenarios.
Uses source-receptor matrices from FRAME. The FRAME
model is used by Defra for National scale estimation of sulphur
and nitrogen deposition for use in calculation of exceedance of
critical loads.
Commissioned by: Defra
Users:
Imperial College under contract from Defra
Availability:
Not freely available but Defra has paid for development
Priority uncertainty source(s):
Consultation with contractors required.
Call user partner:
Defra – Science and Evidence Team – Atmosphere and Local
Environment Programme
70
Case study 11
Atmospheric deposition modelling
Model:
EMEP4UK
Developer:
CEH / Norwegian Meteorological Institute
Reference:
Vieno, M., Dore, A. J., Wind, P., Di Marco, C., Nemitz, E.,
Phillips, G., Tarrason, L., and Sutton, M. A.: Application of the
EMEP Unified Model to the UK with a Horizontal Resolution of
5×5 km2, Atmospheric Ammonia – Detecting Emission Changes
and Environmental Impacts, edited by: Sutton, M. A., Reis, S.,
and Baker, S. M. H., Springer, Dordrecht, The Netherlands, 30
367–372, 2009.
Use:
Being developed for national scale assessment of deposition of
acidifying and eutrophying pollutants. Ozone concentrations also
available.
Commissioned by: Defra
Users:
CEH/University of Edinburgh under contract to Defra
Availability:
The EMEP4UK model is not freely available but Defra has paid
for
development.
The
EMEP
Unified
model
(v3.0,
http://www.emep.int/OpenSource/index.html) and WRF model
which are large components of the model are open source
models.
Priority uncertainty source(s):
Consultation with contractors required.
Call user partner:
Defra – Science and Evidence Team – Atmosphere and Local
Environment Programme
71
Case study 12
Meteorological modelling
Model:
Numerical
Atmospheric
dispersion
Modelling
Environment
(NAME III)
Developer:
UK Met Office
Reference:
Jones A.R., Thomson D.J., Hort M. and Devenish B., 'The U.K.
Met Office's next-generation atmospheric dispersion model,
NAME III', in Borrego C. and Norman A.-L. (Eds) Air Pollution
Modeling and its Application XVII (Proceedings of the 27th
NATO/CCMS International Technical Meeting on Air Pollution
Modelling and its Application), Springer, pp. 580-589, 2007.
Availability:
Available for external research use under licence
Model:
Met Office Unified Model (MetUM)
Air Quality in the Unified Model (AQUM)
Developer:
UK Met Office
Reference:
Cullen, M. J. P. 1993 The unified forecast/climate model.
Meteorol. Mag., 122, 81–94
Availability:
Available for external research use under licence
Model:
United Kingdom Chemistry and Aerosols (UKCA) Model
Developer:
UK Met Office
Commissioned by: Defra, DECC, NERC
Reference:
Evaluation of the new UKCA climate-composition model – Part
1: The stratosphere: Morgenstern, O., Braesicke, P., O'Connor,
F. M., Bushell, A. C., Johnson, C. E., Osprey, S. M., and Pyle, J.
A., Geosci. Model Dev., 2, 43-57, 2009.
Availability:
Available as a patch to the external Unified Model from vn6.6
Use:
Models are used to support research to examine the
assumptions made and scenarios put forward towards revised
legislation
Users:
UK Met Office, Defra, Researchers.
Priority uncertainty source(s):
Consultation with contractors required.
Call user partner:
Defra – Science and Evidence Team – Atmosphere and Local
Environment Programme
72
Case study 13
Management of Chemicals in River Basins
Model:
Geography-referenced Regional Exposure Assessment Tool for
European Rovers (Great-er)
Developer:
EU Chemical Industry, CEFIC Long range Research Initiative
Reference:
T. Feijtel, G. Boeije, M. Matthies, A. Young, G. Morris, C.
Gandolfi, B. Hansen, K. Fox, M. Holt, V. Koch, R. Schroder, G.
Cassani,
D.
Schowanek,
J.
Rosenblom,
H.
Niessen,
Development of a geography-referenced regional exposure
assessment tool for European rivers - great-er contribution to
great-er #1, Chemosphere, Volume 34, Issue 11, June 1997,
Pages 2351-2373, ISSN 0045-6535, DOI: 10.1016/S00456535(97)00048-9.
Use:
GIS based fate model providing chemical concentrations and
water quality along a river. In the model, rivers are segmented
into stretches, surrounding waste water drainage areas are
segmented into geographic units, which are linked to river
discharge points and hence to river stretches. Rivers are
classified into main rivers (which are modelled in detail) and
small surface waters (which are considered as a part of the
waste water pathway system). Spatial scale is determined by the
resolution of the geographical segmentation and by the
classification of the rivers. Segmentation of the drainage area
allows efficient geo-referenced modelling of emissions, waste
water transport and treatment, and processes in small surface
waters.
Commissioned by: CEFIC LRI
Users:
EU Chemical Industry, Regulators (EA, HSE, Defra).
Availability:
Available from http://www.great-er.org/pages/home.cfm
Priority uncertainty source(s):
Uncertainty around loss processes for most pharmaceuticals
e.g. the removal during sewage treatment, in-stream die-away
rates, photodegradation, sorption to sediments, etc.
Awaiting further clarification on specific areas for research
sought from users.
Call user partner:
Environment Agency (as Great-er project sponsor)
73
Case study 14
Review of historical case studies
A considered review could be carried out using historical case studies where users
have encountered model uncertainties which have compounded the environmental
decision-making process, or opportunities which have allowed testing of existing
models and their uncertainties.
Suggested examples could include, but should not be limited to:
i)
2010 Eyjafjallajökull eruption – modelling of volcanic ash plume
ii)
Heathrow third runway – environmental impact report
iii)
Buncefield oil storage depot disaster 2005 – modelling of fall-out
iv)
UK Nirex Ltd investigations of a potential radioactive waste repository at
Sellafield – water migration in rock
v)
Ozone hole - Total Ozone Portable Spectrometer (TOPS) conflict with
BAS data.
vi)
9/11 effect on aircraft induced contrails
74
APPENDIX E – List of participants
Project team
Keith Beven
Phil Haygarth
Kevin Jones
Rob MacKenzie
Paul McKenna
Tim Morley
Trevor Page
Andy Sweetman
Convenor of Workshop 4
Co-convenor of Workshop 2
Co-convenor of Workshop 3
Convenor of Workshop 1
Project Co-ordinator
Workshop facilitator
Co-convenor of Workshop 2
Co-convenor of Workshop 3
Workshop attendees
Ruth Alcock
Keith Beven
Charles Chemel
Julian Dawson
Richard Glass
Alan Godfree
Richard Gooday
Todd Gouin
Crispin Halsall
Roy Harrison
Phil Haygarth
Ed Henderson
Nick Hewitt
Helinor Johnston
Ian Jones
Kevin Jones
Dave Kay
Marc Kennedy
Rob Kinnersley
Steve Lofts
Rob MacKenzie
Paul McKenna
Sarah Metcalfe
Tim Morley
Gordon Nichols
Trevor Page
Camilla Pease
Roger Pickup
Frances Pollitt
Ron Smith
Kate Snow
Andy Sweetman
Workshops
Lancaster Environment Centre
2
Lancaster Environment Centre
2, 4
University of Hertfordshire
1
Macaulay Land Use Research Institute
2
Food and Environment Research Agency
3
United Utilities
2
ADAS
2
Unilever
3
Lancaster Environment Centre
3
Natural Environment Research Council
1
Lancaster Environment Centre
2
National Nuclear Laboratory UK
2
Lancaster Environment Centre
1
Department for Environment, Food and Rural Affairs
3
Centre for Ecology and Hydrology
2
Lancaster Environment Centre
3
Aberystwyth University
4
Food and Environment Research Agency
3
Environment Agency
1
Centre for Ecology and Hydrology
3
Lancaster Environment Centre
1, 4
Lancaster Environment Centre
1, 2, 3, 4
Nottingham University
1
Know Innovation
1, 2, 3, 4
Health Protection Agency
4
Lancaster Environment Centre
1, 2, 4
Environment Agency
3
School of Health and Medicine, Lancaster University
2
Health Protection Agency
3
Centre for Ecology and Hydrology
1
United Utilities
2
Lancaster Environment Centre
1, 3
75
Roger Timmis
Alison Tomlin
Roger Timmis
Jim Walker
Paul Whitehouse
Environment Agency
University of Leeds
Environment Agency
Environment Agency
Environment Agency
1, 2
1
1
2
3, 4
Telephone interviewees
Tom Coles
Peter Costigan
David Devaney
John Goudie
David Lee
Eunice Lord
Claire McCamphil
Dan McGonigle
Caron Montgomery
Steve Nelson
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
ADAS
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Department for Environment, Food and Rural Affairs
Additional report consultants
Alex Bond
David Carruthers
Bernard Fisher
Sarah Honour
Jo Jeffries
Mark Willans
Robert Willows
Quintessa Limited Scientific & Mathematical Consultancy
CERC
Environment Agency
Department for Environment, Food and Rural Affairs
Environment Agency
National Nuclear Laboratory
Environment Agency
76
Download