National Hydrology Conference 2013
Fealy
02 - THE USE OF CLIMATE MODELS IN GUIDING THE DECISION
MAKING PROCESS: IMPLICATIONS FOR HYDROLOGICAL IMPACT
MODELLING
Rowan Fealy
Department of Geography, National University of Ireland, Maynooth
Abstract
Global climate model outputs have traditionally been, and continue to be, employed as the
key input to hydrological impact models in order to assess the likely impacts of climate
change at the catchment, river or stream level. Such ‘top-down’ approaches are fundamentally
premised on the notion that climate models produce accurate and reliable ‘forecasts’ of future
climate. This is in spite of the fact that it is widely recognised that global climate models are
subject to significant uncertainties. These uncertainties result in differences between models,
not just in terms of the projected magnitude of change, but also the direction of change,
particularly evident in important impacts-relevant variables such as precipitation. An inability
to incorporate these uncertainties into our current methodological frameworks for adaptation
decision making has rendered existing efforts at ‘climate proofing’ to be, at best, inadequate,
and at worst, dangerous. While the adaptation literature abounds with hydrological impact
studies, evidence of the uptake and subsequent implementation of the findings from such
studies by policy makers is sparse. Thus, in spite of the significant investment, both in terms
of economic expenditure and research time, a significant gap still exists between the science
and the implementation of robust adaptation policies which seek to reduce our exposure to the
risks posed by climate change.
This paper will present a brief overview the types of uncertainties that can accrue in climate
models and argues that we are unlikely to be able to reduce such uncertainties within the
decision making timeframe. Consequently, the paper will consider alternative ‘scenarioneutral’ approaches to robust decision making, which recognise that climate is only one of
many drivers of change. The paper concludes by arguing for a repositioning of climate
models within the decision making framework.
1. INTRODUCTION
The latest report from the Intergovernmental Panel on Climate Change (IPCC), the Fifth
Assessment Report (AR5) (IPCC, 2013a1), provides a comprehensive overview and
assessment of the long term changes that have occurred in the climate system over both the
instrumental and pre-instrumental records. Furthermore, the IPCC report unambiguously
states that it is extremely likely that human influences have been the dominant driver of the
observed warming, particularly since the middle of the 20th century (IPCC, 2013a). While the
1,2
A report accepted by Working Group I of the IPCC but not approved in detail.
16
National Hydrology Conference 2013
Fealy
observational evidence for changes in the global temperature record is clear, observed
changes in global precipitation over land areas are less clear, in part attributed to station
coverage but also record length. However, the lack of an identifiable trend in global
precipitation is consistent with findings from global and continental scale assessments of river
discharge and runoff, which represent catchment integrated values of precipitation, leading
the IPCC (2013a) to conclude that confidence is low for an increasing trend in global
discharge during the 20th century. In spite of the lack of significant trends in global
precipitation analysis, evidence for increases in the frequency and intensity of extreme
precipitation events at continental scales, particularly from North America, Central America
and Europe, does exist in the observational records (IPCC, 2013a).
While it is clear, from the accumulated literature reported on in the Fifth Assessment Report
(IPCC, 2013a), that the climate system is warming, and will likely continue to do so in
response to the radiative forcing associated with increasing anthropogenic greenhouses gas
emissions, the direction and magnitude of changes in moisture related variables, such as
precipitation, at sub-continental scales are somewhat less clear. This appears true for both the
observations and future model simulations of precipitation.
“There has been some improvement in the simulation of continental-scale patterns of
precipitation since the AR4. At regional scales, precipitation is not simulated as well, and
the assessment is hampered by observational uncertainties.” (IPCC, 2013b:112)
Deficiencies in a global circulation, or climate, model’s (GCM) ability to reproduce the
statistics of an observed series, particularly that of precipitation, can arise for a number of
reasons, such as, the omission of key climate process within the model and the use of
particular parameterisation schemes, but also due to the dependency of variables, such as,
precipitation, on climate processes which occur at a scale smaller than can be resolved by the
climate model. Ultimately, climate models are incomplete in their description of the real
world climate and represent a simplification of what is a very complex system, and therefore
subject to error. Consequently, any future projection of anthropogenic climate change arising
from increased concentrations of atmospheric CO2 are subject to a high degree of uncertainty
(Jones, 2000).
This uncertainty arises largely as a consequence of both aleatory (‘unknowable’ knowledge)
and epistemic, or systematic, (‘incomplete’ knowledge) uncertainties (Hulme and Carter,
1999; Oberkampf et al., 2002) (Box 1). Aleatory uncertainties are considered to be irreducible
and result from an inherent indeterminacy of the system being modelled (Hulme and Carter,
1999; Oberkampf et al., 2002). For example, future human behaviour and actions which give
rise to emissions of greenhouse gases are not ‘knowable’ in any deterministic way; therefore,
future greenhouse gas emissions or concentrations must be prescribed based on some set of
predefined assumptions (Hulme and Carter, 1999). This inherent indeterminacy means that
even if the climate modelling community could develop a ‘perfect’ climate model, model
simulations of future climate would still remain uncertain. This type of uncertainty is referred
to as model uncertainty, by the IPCC (2013a).
17
National Hydrology Conference 2013
Fealy
Epistemic or systematic uncertainties arise primarily from a lack of complete knowledge of
the system. Epistemic uncertainties are considered to be reducible, as our understanding or
knowledge of the particular system or environment increases. For example, the envelope of
possible values of the sensitivity of the climate system may be narrowed as our understanding
of the key climate processes improves. Conversely, it could also be that case that with
additional research we could also find that we did not previously include a particular process,
which may result in the climate sensitivity envelope increasing. Parameterisations schemes, or
empirical approximations employed by climate models to represent processes that are not
dynamically resolved, also fall into this category of uncertainties. Epistemic uncertainties
contribute to model spread, represented by a range of responses in a model ensemble (Figure
1).
Typical sources of uncertainty
1. Problems with data
a) Missing components or errors in the data
b) “Noise” in the data associated with biased or incomplete observations
c) Random sampling error and biases (non-representativeness) in a sample
2. Problems with models
a) Known processes but unknown functional relationships or errors in the structure of the model
b) Known structure but unknown or erroneous values of some important parameters
c) Known historical data and model structure, but reasons to believe parameters or model structure will
change over time
d) Uncertainty regarding the predictability (e.g., chaotic or stochastic behaviour) of the system or effect
e) Uncertainties introduced by approximation techniques used to solve a set of equations that
characterize the model.
3. Other sources of uncertainty
a) Ambiguously defined concepts and terminology
b) Inappropriate spatial/temporal units
c) Inappropriateness of/lack of confidence in underlying assumptions
d) Uncertainty due to projections of human behaviour (e.g., future consumption patterns, or
technological change), which is distinct from uncertainty due to “natural” sources (e.g., climate
sensitivity, chaos)
Box 1. Typical sources of uncertainty encountered in the modelling framework (from Moss and
Schneider, 2000).
In a significant development, reflected in the Fifth Assessment Report (IPCC, 2013a), the
climate modelling community have moved away from employing socio-economic based
‘storylines’ of emissions (SRES)(e.g. Nakicenovic, 2000), each making different assumptions
about future greenhouse gas emissions, to representative concentration pathways (RCPs),
which define a specific emissions trajectory and subsequently, its radiative forcing. The RCPs
are not dependent on any particular set of social, economic or technological developments.
Unlike the earlier suite of SRES scenarios, the RCPs explicitly include the potential for
introducing climate policies, which seek to mitigate future climate change. While any
particular representative concentration pathway may never be realised (denoted by the use of
‘representative’), and hence no associated probabilities can be attached to individual RCPs
(IPCC, 2013a), they do provide an essential tool for exploring potential future changes in the
climate system arising from anthropogenic activities.
18
National Hydrology Conference 2013
Fealy
However, due to the non-linear dynamics of the (model) climate system, differences in model
structure, representation of physical and dynamical processes and parameterisation schemes,
even when forced with the same emissions scenario, climate models will always simulate a
range of future outcomes (Figure 1; 5-95% ranges). The spread in model outcomes is most
evident at the sub-continental and regional scales. For example, differences occur between
GCMs not just in the magnitude, but also direction, of projected precipitation changes. The
lack of an ‘optimal solution’ has presented significant challenges for decision makers who are
seeking to develop tailored, cost effective, adaptation strategies at scales appropriate to their
sector.
Figure 1. Time series of simulated global mean annual surface temperature anomalies, relative to
1986-2005, from the CMIP5 experiments. The multi-model mean for each RCP is indicated by the
respective solid line and the 5-95% range across the individual models is indicated by the shading.
Numbers indicate the number of models contributing to the different concentration driven
experiments. For each RCP, a range of plausible values are simulated. (Source: IPCC, 2013a)
While individual climate scenarios represent a plausible future outcome, their usefulness is
ultimately limited in that it they have no degree of probability attached (Jones, 2000).
Furthermore, it could also be argued that by virtue of ‘knowing’ the likely impacts associated
with a particular emissions trajectory, decision makers may decide on more, or less, stringent
controls on emissions, thus resulting in a particular pathway becoming less, rather than more,
likely.
2. APPROACHES TO HANDLING MODEL UNCERTAINTY
In response to these acknowledged limitations, a number of approaches have been developed
which seek to cater for issues associated with model spread, such as adoption of a ‘best guess’
framework or taking the ensemble mean or median value from a range of scenarios. However,
this practice, which may ultimately result in the suppression of crucial uncertainties should be
considered as ‘dangerous’ as any subsequent policy decisions may only reflect a partial
assessment of the risk involved (Parry et al., 1996; Risbey, 1998; Hulme and Carter, 1999;
New et al., 2007). Consequently, such top-down or ‘predict and provide’ approaches (Figure
2), which place a strong emphasis on having ‘reliable’ model ‘predictions’, are not considered
particularly useful for subsequent use in sensitivity or risk analysis, due to an inability to
attach probabilities or likelihoods to the selected climate scenario(s). Without a clear
19
National Hydrology Conference 2013
Fealy
statement on the uncertainties that have, or have not, been incorporated into the research,
decision makers need to exercise extreme caution as any subsequent decisions may not
encompass the full range of associated risks. Such policy decisions can give rise to
maladaptation (over- or under- adaptation).
Figure 2. Traditional top down or ‘predict and provide’ approach to climate change impact
assessment for adaptation (Shine and Desmond, 2011). This approach requires ‘reliable’ climate
model predictions of future climate in order for the resultant adaption to be successful.
For example, future sea level rise estimates lie in the range from 0.26 to 0.82 m for 20812100, relative to 1986-2005 (IPCC, 2013a). A sea wall built on the basis of defending coastal
infrastructure against a projected future mean model sea level rise of 0.55 m (ensemble
mean), while economically viable, may underestimate the uncertainties and be ineffectual
against a sea level rise of greater than 0.55 m. Conversely, a sea wall built to defend against a
future projected sea level rise of 0.82 m, while significantly reducing the potential risks
associated, may be very costly to justify on the basis of uncertainty alone. Under-adaptation
also has potential additional risks associated with it, particularly where hard infrastructure,
such as a sea wall, is advocated as the adaptation response, as it can lead to the safe
development paradox. This paradox results from a perception of reduced risk due to the
presence of hard infrastructure, which in turn may lead to development taking place that
would otherwise have not, in the absence of the infrastructure. If the infrastructure fails, the
potential exposure to risk is therefore greatly increased. Catastrophic failure of the surge
protection levee system in New Orleans during Hurricane Katrina, in 2005, is a case in point;
the levee system had previously facilitated ‘safe’ development in the surrounding low-lying
regions, thereby increasing the exposure to risk if/when the levees failed.
Giorgi and Mearns (2002), recognising differences in model ‘skill’, demonstrated a procedure
for calculating model average, uncertainty range and collective reliability of a range of
regional climate projections from ensembles of different AOGCM simulations. The
Reliability Ensemble Averaging (REA) method weights GCMs based on individual model
performance and criteria for model convergence. This procedure acknowledges that models
have different levels of skill associated with modelling different aspect of the climate system
and weights models accordingly. However, the use of model convergence as a criterion for
20
National Hydrology Conference 2013
Fealy
model reliability has been subject to critique, as similarities, or a lack of independence (e.g.
model code sharing, similar parameterisations schemes) in model structure may result in two
or more models producing similar outputs and therefore ascribed a higher weight than a truly
independent model. A model might also demonstrate apparent skill, solely due to error
cancellation, rather than reflecting any real skill.
An additional weakness of relying on such top-down approaches have tended to dismiss the
possibility of local adaptation (‘dumb farmer’ hypothesis) or only assume an arbitrary level of
adaptation (‘clairvoyant farmer’) (Dessai and Hulme, 2003). Fundamentally, the predict and
provide approach requires climate models to provide precise, reliable and accurate
‘predictions’ of future climate, ultimately a key limiting factor in developing robust
adaptation (Dessai et al., 2009).
In spite of this recognised limitation, the ‘predict and provide’ approach has predominated the
scientific literature; yet substantive evidence which demonstrates the translation of the
resultant scientific outputs into meaningful adaption is much less prevalent (Wilby and
Dessai, 2010; Wilby et al., 2009). As an alternative, sensitivity analyses have been employed
to assess the sensitivity of a system to incremental changes in climate (e.g. ±10%, ±20%,
±30% change in precipitation; ±1oC, ±2oC, ±3oC change in temperature) and constitute a
bottom-up approach to informing climate adaptation policy (Dessai and Hulme, 2003). In
order to test the sensitivity of a system, sensitivity analysis try to account for simultaneous
changes in a number of variables and can also take into account uncertainty in inputs (Katz,
2002). An additional advantage of this approach is that the imposed, incremental, changes
may be informed by the output from a climate model or climate models.
A number of authors argue that due to the uncertainties associated with climate change
projections, that it is better to adapt to present day, or recent historic, climate variability as
this represents a good proxy for near term climate change (Dessai and Hulme, 2003). In
support of this, Burton et al. (2002) advocate the need for policy development to start, at least
in the initial phases, with an assessment of vulnerability to present day variability and
extremes. Analogue approaches, where adaptation measures are assessed against past climate,
have appeared as an alternative approach in impacts assessments.
However, such bottom-up approaches are not without their opponents. Sensitivity analysis
can be used to generate response surfaces from which risk thresholds can be identified, such
as ‘dangerous’ climate change, however, they may not necessarily produce consistent or
plausible scenarios of future changes (Jones and Mearns, 2003). While a significant weakness
of the analogue approach assumes that the past climate encompasses the full range of
variability that is likely to occur in the future.
Murphy et al. (2004) suggests that only multi-model ensembles, sampling as wide a range of
model uncertainties as possible, can reliably show the spread of possible regional changes, a
finding confirmed by Lopez et al. (2009) and others. While this technique can produce a wide
range of scenarios, which are useful for examining the range in projected climate response at
the regional scale, the resultant scenarios are considered as being equally plausible and have
no associated likelihood of occurrence. For example, if in the case of a climate projection of
21
National Hydrology Conference 2013
Fealy
precipitation, a variable which is inherently difficult to simulate reliably, two models produce
scenarios with similar magnitude changes but opposite in sign (Giorgi and Francisco, 2000),
i.e. model A simulates a 10% increase and model B a 10% decrease in regional precipitation;
can a decision maker then assume that there is going to be 0% change (ensemble mean) in
precipitation? Adaptation measures required for a 10% increase in precipitation (improved
flood defences) are likely to be significantly different to those required for a decrease in
precipitation (such as, additional reservoir capacity). Knowledge about the potential future
range of changes is a far more powerful tool, particularly where a wide range of futures are
considered. While such inter model differences may, or may not, be reduced through
increased scientific understanding, the quantification, and subsequent communication, of
uncertainties is considered more desirable than assuming a perfect model in adaptation
development.
Increasingly, the incorporation of probabilities in climate change impact assessments is
becoming more wide spread. As researchers move from employing single trajectory, topdown approaches towards the use of multiple scenarios from multiple GCMs in climate
impact assessments, characterising uncertainty in the associated scenarios has become
increasingly feasible. From a policy development perspective, the identification and
communication of uncertainties, and their ranges, may have a useful application in strategic
decision making (Burton et al., 2002).
Irrespective of the approach adopted, transparency in method is considered crucial, with the
onus on the individual researcher to explicitly state the approach adopted and the assumptions
made to represent uncertainty (Hulme and Carter, 1999: Moss and Schneider, 2000; Dessai
and Hulme, 2003).
3. CLIMATE INFORMATION AT SCALES APPROPRIATE TO DECISION
MAKERS
Due to computational limitations, the typical spatial resolution of many AOGCMs
(atmosphere-ocean global circulation/climate models) is currently in the order of greater than
100 square kilometres (E.g. model horizontal resolution T63 ~180km; T159 ~125km; T106
~110km). While this has been demonstrated as adequate to capture low frequency, largescale, variations in the climate system (e.g. Stephenson and Pavan, 2003), many important
processes occur at much smaller spatial scales, such as, those processes associated with
convective cloud formation and precipitation, and thus are too fine to be resolved in the
dynamic modelling process.
As a consequence, a number of techniques have been developed to ‘downscale’ coarse
resolution GCM output to finer spatial and temporal scales for use in decision making.
Dynamical regional climate models (RCMs) and empirical statistical downscaling (SD) are
the primary means by which regional- or local- scale information is derived from a parent
GCM(s). However, the incorporation of an additional downscaling ‘layer’ to generate high
resolution scenarios will act to both propagate and contribute to uncertainty within the
modelling framework (e.g. Kahn et al., 2006; Rowell 2006; Hingray et al., 2007; Gachon and
22
National Hydrology Conference 2013
Fealy
Dibike, 2007; Dibike et al., 2008) resulting in significant regional variations between
downscaled model projections, even when forced with the same GCM and emissions scenario
(Haylock et al., 2006).
Figure 3. Projected changes in the mean ensemble precipitation (%) from the ENSEMBLES regional
climate modelling project for 2016–2035, relative to 1986–2005, for DJF (left) and JJA (right)
derived from 10 GCM-RCM combinations for the SRES A1B scenario. The stippling indicates where
80% of the models agree in the sign of the change (IPCC, 2013 after Rajczak et al. (2013).
In spite of these shortcomings, employing either dynamical or statistical downscaling has
been considered to ‘add value’ to climate projections, when compared to GCM output at the
grid scale (e.g. Katz, 2002; Rowell, 2006; Fealy and Sweeney, 2007; Feser et al., 2011).
However, Dessai et al. (2009) caution against confusing accuracy with precision; while higher
resolution climate projections may represent higher precision than the parent GCM, this
should not be confused with increased accuracy of projection. In addition, the notion of added
value has been questioned by a number of authors (e.g. Castro et al., 2005; Rockel et al.,
2008; Pielke Sr. and Wilby, 2012). However, Katz (2002) argues that some downscaling
techniques have the potential to be useful, even if they do not offer an improvement over the
GCM employed, through enabling uncertainty analysis to be undertaken (E.g. Wilby and
Harris, 2006; Fowler et al., 2007; Hashmi et al., 2009) (Figure 3).
Fealy (2010; 2013), in an analysis of uncertainty at the regional/local scale for Ireland,
applied an approach to develop probabilistic projections for Ireland on the basis of previously
available statistically downscaled scenarios (Fealy and Sweeney, 2007; 2008a; 2008b). The
methodology employed by Fealy (2010; 2013) was adapted from Hulme and Carter (1999),
Jones (2000) and New and Hulme (2000) and applied to two impacts relevant climate
variables, that of seasonal mean temperature (oC) and precipitation change (%), for a selection
of 17 GCMs. The methodology had previously been applied directly to climate model (both
global and regional) output, but was refined by Fealy (2010) for application to statistically
downscaled data, for a selection of synoptic stations in Ireland.
23
National Hydrology Conference 2013
Fealy
Figure 4. Seasonal probably distributions for precipitation change (%) for Claremorris synoptic
station (River Suck), for the period 2070-2099, relative to 1961-1990. Based on Fealy (2013).
Seasonal mean precipitation change (%)
River Catchment (Station)
Blackwater (Cork Airport)
Boyne (Mullingar)
Moy (Belmullet)
Suck (Claremorris)
17 GCMs (A2)
Season
DJF
MAM
JJA
SON
DJF
MAM
JJA
SON
DJF
MAM
JJA
SON
DJF
MAM
JJA
SON
DJF
MAM
JJA
SON
Min.
-0.7
-14.6
-27.2
-22.2
2.3
-30.3
-29.4
0.3
-3.0
-22.4
-15.9
-9.8
2.3
-30.3
-29.4
0.3
-2.1
-16.0
-44.9
-15.7
Q1
1.9
-5.7
-13.5
-10.9
10.4
-12.9
-15.4
2.7
-0.3
-9.3
-8.2
-4.5
10.4
-12.9
-15.4
2.7
11.5
-7.0
-35.6
-5.4
Med.
4.2
-2.4
-10.1
-8.1
12.4
-6.8
-12.6
4.4
1.4
-4.5
-6.6
-2.9
12.4
-6.8
-12.6
4.4
16.1
0.3
-19.7
-0.2
Q3
6.6
1.0
-6.9
-5.5
14.7
-0.7
-10.0
6.1
3.1
0.3
-5.1
-1.5
14.7
-0.7
-10.0
6.1
21.2
14.4
-7.8
5.9
Max.
15.0
7.1
-1.3
-1.1
26.9
8.3
-2.3
13.2
8.1
7.9
-1.2
0
26.9
8.3
-2.3
13.2
29.4
25.6
10.3
15.6
Table 1. Seasonal mean change in precipitation (%) for Cork Airport (River Blackwater), Mullingar
(River Boyne), Belmullet (River Moy), Claremorries (River Suck) and the grid box representing
Ireland from the 17 GCMs employed by Fealy (2010) for the period 2070-2099, relative to 19611990. Values for minimum (Min.), maximum (Max.), median (Med.) and quartiles (Q1-1st quartile;
Q3-3rd Quartile) are also shown. Small difference occur between the values in Figure 4 and those
outlined for Claremorris in the above table which occur due to slightly different treatments of the
data. (Table source: Bastola et al., 2012, based on data from Fealy, 2010).
24
National Hydrology Conference 2013
Fealy
While the projected mean changes in temperature and precipitation, based on the probabilistic
approach of Fealy (2010; 2013), were found to be comparable to the ensemble mean derived
from the statistically downscaled data, the probability distribution functions indicated a wide
range in the distribution of the projected changes. Projections of temperature were found to be
consistent in the direction of change, while varying in magnitude; however, results for
precipitation were found to vary in both direction and magnitude in particular seasons. In
spite of the probabilistic approach employed, for some seasons, differences in direction, or the
sign, of precipitation change, were modelled with equal probability (Figure 4). Table 1
illustrates the range in seasonal values for simulated changes in precipitation (%) for the
period 2070-2099, relative to 1961-1990, for a selection of hydrologically relevant synoptic
stations.
Rather than being an impediment to developing adaption options, the development and
subsequent use of probabilistic scenarios provides a valuable assessment of variables/seasons
where the associated uncertainties may require alternative policy options to be more
rigorously assessed. For example, uncertainties associated with projected precipitation
changes at all stations during the spring months by the 2080s, which result in both increased
and decreased precipitation being simulated with equal likelihood, is highlighted as a case in
point. From a policy perspective, these findings are particularly relevant for sectors dependent
on water supply and availability that seek to develop robust adaptation options. Under the
traditional ‘predict and provide’ approach to impact assessment, such uncertainty may be
viewed as a justification to adopt a ‘wait and see’ approach to adaptation on the basis of not
having an optimal solution, something which may be fundamentally unachievable; while
alternative approaches may readily accommodate such uncertain climate information, with the
ultimate aim of developing robust adaptation options which are insensitive to uncertainties.
Critically, the approach has the potential to highlight seasons/locations where climate
information simply cannot address the needs of the policy community (e.g. seasons/locations
where an equal likelihood of both positive and negative changes are suggested).
Ultimately, a combination of approaches is required, which combine probabilistic based
climate scenarios, in addition to other non-climate related pressures such as demographic
structure etc., and sensitivity analysis to more comprehensively asses the vulnerability or
resilience of a particular sector, community or infrastructure to future, unpredictable, changes
in the climate system. Based on a vulnerability assessment, the ‘local’ level of resilience to a
particular change in climate can potentially be determined and through the subsequent
incorporation of probabilistic scenarios, suitable adaptation measures may be assessed.
4. A WAY FORWARD?
More recently, Prudhomme et al. (2010) and Wilby and Dessai (2010) propose alternative
scenario neutral approaches to adaptation which address the sensitivity of adaptation options
or pathways to a range of plausible, but uncertain, future climates. Importantly, Wilby and
Dessai (2010) also include the potential of non-climatic pressures in influencing a systems
response or vulnerability. The scenario neutral approach offers a significant methodological
advancement over traditional top-down and bottom-up approaches, through the potential to
25
National Hydrology Conference 2013
Fealy
incorporate probabilistic climate scenarios with existing knowledge of the sensitivity of the
system under study, in developing robust or ‘no regrets’ adaptation (Figure 5).
Fundamentally, the scenario neutral approaches as espoused by both Prudhomme et al. (2010)
and Wilby and Dessai (2010) argues for the repositioning of climate change information from
the start of the climate risk assessment process to further down the risk assessment chain. This
approach shifts the requirement for accurate and reliable predictions (i.e. most likely or
probable outcome), and arguably an impossible constraint, to one where a range of (uncertain)
future projections (i.e. plausible or possible outcomes) are considered for use in assessing the
robustness of, rather than developing, different adaptation options or pathways.
Figure 5. A conceptual framework for a scenario neutral approach to developing robust adaptation
(Wilby and Dessai, 2010)
Critically, the scientific imperative to further our understanding of the dynamics of the
climate system, which seeks to reduce epistemic or systematic uncertainties in climate
simulations, remains, but it no longer acts to constrain or supersede the development of robust
adaption, a societal imperative. A key benefit to separating the scientific from the societal
imperative is that scientific developments (e.g. increased understanding of the systems under
study, new scenarios or improved models) can be readily incorporated into the scenarioneutral approach. Additionally, the incorporation of probabilistic based climate distributions
into the scenario neutral approach offers the potential to transition from a wholly
deterministic based approach to decision making to one which recognises the inherent, and
potentially irreducible, uncertainties, required for robust adaptation.
5. DISCUSSION
Kass and Raftery (1995) suggest that ‘any approach that selects a single model and then
makes inference conditionally on that model ignores the uncertainty involved in model
26
National Hydrology Conference 2013
Fealy
selection, which can be a big part of overall uncertainty,’ and ‘this leads to underestimation of
the uncertainty about quantities of interest, sometimes to a dramatic extent’ (Kass and Raftery
1995:784). Yet in spite of this early acknowledgement, the climate modelling and impacts
community continued to produce and employ single trajectory climate scenarios for use in
impact assessments which sought to inform policy making (‘predict and provide’). While
there was a valid historical reason for such, arising from the limited number of centres who
were undertaking global climate modelling due to the computational resources required and
associated expense of running such model simulations, the implications for the policy
community were significant. Hulme and Carter (1999) consider the practice of employing a
limited number of climate scenarios as ‘dangerous’, as such an approach only reflects a partial
assessment of the associated risk involved.
While a number of techniques have developed in order to account for model differences, due
to emissions scenarios and GCMs; an inability to produce probabilistic based projections in
the past proved a limiting factor in enabling the potential risk of climate change impacts in
key sectors to be quantified and potentially hindered the subsequent development of suitable
policy responses to reduce or mitigate such impacts. While the development of probabilistic
based approaches represents a positive step forward, particularly from the use of the single
trajectory approach common in the past, they only represent a sampling of the model spread
space and do not capture the total uncertainty space (events that we may not have considered;
changes in non-climatic factors). Critically, information derived from probabilistic based
climate assessments is not independent of the methodology employed (e.g. New et al., 2007);
therefore the risk of maladaptation remains. In addition, the contribution of full end-to-end
probabilistic based climate impact assessments to the decision making process remains
largely untested with the exception of one or two peer reviewed studies (Wilby et al., 2009).
Probabilistic based projections applied in a ‘predict and provide’ framework, would also
suggest that adaptation is not possible in locations where divergent model outcomes are
simulated. If however, we reposition the use of climate models, so that rather than being the
point of departure in the adaptation process (requiring ‘reliable’ and ‘accurate’ predictions) to
one where uncertainty is recognised as being inherent, and additional non-climatic factors are
considered in a holistic way, as in the scenario-neutral approach as advocated by Wilby and
Dessai (2010) and others, then meaningful progress can potentially be made towards
achieving robust adaptation options which are insensitive to uncertainties in future changes,
climatic or otherwise.
6. ACKNOWLEDGEMENTS
Portions of this paper are reproduced from Fealy, R. (2013), Progress in Physical Geography, 37(2),
pp. 178-205. doi: 10.1177/0309133312462935 and Fealy, R. (2010), STRIVE Report Series No. 48,
Environmental Protection Agency, Johnstown Castle, pp1-66.
27
National Hydrology Conference 2013
Fealy
7. REFERENCES
Bastola, S., Murphy, C. Fealy, R. (2012) Generating probabilistic estimates of hydrological response
using a weather generator and probabilistic climate change scenarios. Hydrological Processes, 26,
2307–2321.
Burton, I., Huq, S., Lim, B., Pilifosova, O. and Schipper, E.L. (2002) From impacts assessment to
adaptation priorities: the shaping of adaptation policy. Climate Policy, 2, 145–159.
Castro CL, Pielke RA, Sr, and Leoncini G (2005) Dynamical downscaling: Assessment of value
retained and added using the Regional Atmospheric Modeling System (RAMS). Journal of
Geophysical Research, 110: D05108.
Dessai, S. and Hulme, M. (2003) Does climate policy need probabilities? Tyndall Centre Working
Paper No. 34.
Dessai, S., Hulme, M., Lempert, R. and Pielke R. (2009) Climate prediction: a limit to adaptation, in
Adger N, Lorenzoni I, O’Brien K (eds) Adapting to Climate Change: Thresholds, Values,
Governance. Cambridge University Press: Cambridge, UK.
Dibike, Y.B., Gachon, P., St-Hilaire, A., Ouarda, T.B.M.J. and Nguyen, V.T.-V. (2008) Uncertainty
analysis of statistically downscaled temperature and precipitation regimes in Northern Canada.
Theoretical and Applied Climatology, 91(1-4): 149-170.
Fealy, R. (2010) An Assessment of Uncertainties in Climate Modelling at the Regional Scale: The
Development of Probabilistic Based Climate Scenarios for Ireland, STRIVE Report Series No. 48,
Environmental Protection Agency, Co. Wexford, pp 1-66. ISBN 978-1-84095-345-9.
Fealy, R. (2013) Deriving probabilistic based climate scenarios using pattern scaling and statistically
downscaled data: A case study application from Ireland, Progress in Physical Geography, 37(2),
pp. 178-205.
Fealy, R. and Sweeney, J. (2007) Statistical downscaling of precipitation for a selection of sites in
Ireland employing a generalised linear modelling approach, International Journal of Climatology,
27(15): 2083-2094.
Fealy, R. and Sweeney, J. (2008a) Statistical downscaling of temperature, radiation and potential
evapotranspiration to produce a multiple GCM ensemble mean for a selection of sites in Ireland.
Irish Geography, 41(1): 1-27.
Fealy, R. and Sweeney, J. (2008b) “Climate Scenarios for Ireland”, in Sweeney et al., (2008) Climate
Change–Refining the Impacts for Ireland, Environmental Protection Agency, Ireland.
Feser F, Rockel B, von Storch H, et al. (2011) Regional climate models add value to global model
data: A review and selected examples. Bulletin of the American Meteorological Society 92: 1181–
1192.
Fowler HJ, Blenkinsop S, and Tebaldi C (2007) Linking climate change modelling to impacts studies:
Recent advances in downscaling techniques for hydrological modelling. International Journal of
Climatology 27(12): 1547–1578.
Gachon P. and Dibike Y. B. (2007) Temperature change signals in northern Canada: Convergence of
statistical downscaling results using two driving GCMs, International Journal of Climatology,
27(12): 1623-1641.
Giorgi, F. and Francisco, R. (2000) Evaluating uncertainties in the prediction of regional climate
change. Geophysical Research Letters, 27: 1295-1298.
Giorgi, F. and Mearns, L.O. (2002) Calculation of average, uncertainty range, and reliability of
regional climate changes from AOGCM simulations via the “Reliability Ensemble Averaging”
(REA) Method, Journal of Climate, 15: 1141-1158.
Hashmi MZ, Shamseldin AY, and Melville BW (2009) Statistical downscaling of precipitation: Stateof-the art and application of bayesian multi-model approach for uncertainty assessment. Hydrology
and Earth Systems Sciences Discussions, 6: 6535–6579.
Haylock MR, Cawley GC, Harpham C, et al. (2006) Downscaling heavy precipitation over the United
Kingdom: A comparison of dynamical and statistical methods and their future scenarios.
International Journal of Climatology 26(10): 1397–1415.
28
National Hydrology Conference 2013
Fealy
Hingray, B., Mouhous, N., Mezghani, A., Bogner, K., Schaefli, B. and Musy, A. (2007) Accounting
for global-mean warming and scaling uncertainties in climate change impact studies: application to
a regulated lake system. Hydrology and Earth System Sciences, 11 (3): 1207-1226.
Hulme, M. and Carter, T.R. (1999) Representing uncertainty in climate change scenarios and impact
studies, in: Representing uncertainty in climate change scenarios and impact studies (Proc.
ECLAT-2 Helsinki Workshop, 14-16 April, 1999 (Eds. T. Carter, M. Hulme and D. Viner).
Climatic Research Unit, Norwich, UK, 128 pp
Hulme, M. and Carter, T.R. (2000) The changing climate of Europe. Chapter 3 in: Parry, M.L. (ed.)
Assessment of the Potential Effects of Climate Change in Europe. Report of the ACACIA
Concerted Action, November 2000, UEA, Norwick, UK, 350 pp.
IPCC (2013a) Final draft of Working Group I Contribution to the IPCC Fifth Assessment Report
Climate Change 2013: The Physical Science Basis. [Accessed 18 October, 2013 www.ipcc.ch].
IPCC (2013b) Final draft of Working Group I Contribution to the IPCC Fifth Assessment Report
Climate Change 2013: The Physical Science Basis Summary for Policymakers. [Accessed 18
October, 2013 www.ipcc.ch].
Jones, R.N. (2000) Analysing the risk of climate change using an irrigation demand model, Climate
Research, 14: 89-100.
Jones, R.N. and Mearns, L.O. (2003) Assessing Future Climate Risks, in Lim, B., Carter, I. and al
Huq, S. (eds). Adaptation Policy Framework, Technical Paper 5, United Nations Development
Programme, New York.
Kass, R.E. and Raftery, A.E. (1995) Bayes factors. Journal of American Statistical Association, 90:
773–795.
Katz, R.W. (2002) Techniques for estimating uncertainty in climate change scenarios and impact
studies. Climate Research, 20: 167-185.
Khan MS, Coulibaly P, and Dibike Y (2006) Uncertainty analysis of statistical downscaling methods
using Canadian Global Climate Model predictors. Hydrological Processes 20: 3085–3104.
Lopez, A., Fung, F., New, M., Watts, G., Weston, A. and Wilby, R.L. (2009) From climate model
ensembles to climate change impacts: A case study of water resource management in the South
West of England. Water Resources Research, 45, W08419.
Moss, R.H. and Schneider, S.H. (2000) “Uncertainties in the IPCC TAR: Recommendations to Lead
Authors for More Consistent Assessment and Reporting", in Pachauri R., T. Taniguchi, and K.
Tanaka (eds.), Guidance Papers on the Cross Cutting Issues of the Third Assessment Report of the
IPCC, Geneva, Switzerland: World Meteorological Organization, 33-51.
Murphy, J., Sexton, D., Barnett, D., Jones, G., Webb, M., Collins, M. and Stainforth, D. (2004)
Quantification of modelling uncertainties in a large ensemble of climate change simulations,
Nature, 430: 768-772.
Nakicenovic, N., Alcamo, J., Davis, G., de Vries, B., Fenhann, J., Gaffin, S., Gregory, K., Gr, A.,
Yong Jung, T., Kram, T., La Rovere, E.L., Michaelis, L., Mori, S., Morita, T., Pepper, W., Pitcher,
H., Price, L., Riahi, K., Roehrl, A., Rogner, H-H., Sankovski, A., Schlesinger, M., Shukla, P.,
Smith, S., Swart, R., van Rooijen, S., Victor, N. and Dadi Z. (2000) Special Report on Emissions
Scenarios: A Special Report of Working Group III of the Intergovernmental Panel on Climate
Change, Cambridge University Press, Cambridge, U.K., pp 599.
New, M. and Hulme, M. (2000) Representing uncertainty in climate change scenarios: a Monte-Carlo
approach, Integrated Assessment, 1:203-213.
New, M., Lopez, A., Dessai, S. and Wilby, R. (2007) Challenges in using probabilistic climate change
information for impact assessments: an example from the water sector. Philosophical Transactions
of the Royal Society A , 365: 2117-2131.
Oberkampf, W.L., DeLand, S.M., Rutherford, B.M., Diegert, K.V. and Alvin, K.F. (2002) Error and
uncertainty in modelling and simulation, Reliability Engineering and System Safety, 75: 333-357.
Parry, M.L., Carter, T.R. and Hulme, M. (1996) What is a dangerous climate change? Global
Environmental Change, 6: 1-6.
Pielke RA Sr, and Wilby RL (2012) Regional climate downscaling – what’s the point? EOS 93(5): 52–
53.
29
National Hydrology Conference 2013
Fealy
Prudhomme, C.,Wilby, R. L., Crooks, S., Kay, A. L. & Reynard, N. S. (2010) Scenario-neutral
approach to climate change impact studies: application to flood risk. Journal of Hydrology, 390,
198–209.
Rajczak, J., P. Pall and C. Schär (2013): Projections of extreme precipitation events in regional climate
simulations for Europe and the Alpine Region, Journal of Geophysical Research, 118 (9), 36103626, doi:10.1002/jgrd.50297
Risbey, J.S. (1998) Sensitivities of water supply planning decisions to streamflow and climate scenario
uncertainties. Water Policy, 1: 321-340.
Rockel B, Castro CL, Pielke RA Sr, et al. (2008) Dynamical downscaling: Assessment of model
system dependent retained and added variability for two different regional climate models. Journal
of Geophysical Research 113: D21107.
Rowell, D. P. (2006) A Demonstration of the Uncertainty in Projections of UK Climate Change
Resulting from Regional Model Formulation. Climatic Change, 79 (3): 243-257.
Shine, T. and Desmond, M. (2013) Ireland Adapts to Climate Change, Climate Change Research
Programme (CCRP), Report 9, Environmental Protection Agency, Johnstown Castle, Co.Wexford,
Ireland, pp 1-14.
Stephenson DB and Pavan V (2003) The North Atlantic Oscillation in coupled climate models: A
CMIP1 evaluation. Climate Dynamics 20: 381–399.
Wilby RL and Dessai S (2010) Robust adaptation to climate change. Weather 65: 180–185.
Wilby, R. L., Troni, J., Biot, Y., Tedd, L., Hewitson, B.C., Smith, D. M. and Sutton, R.T. (2009) A
review of climate risk information for adaptation and development planning. International Journal
of Climatology, 29:1193-1215.
Wilby, R.L. and Harris, I. (2006) A framework for assessing uncertainties in climate change impacts:
low flow scenarios for the River Thames. Water Resources Research, 42 (2).
Wilby, R.L.and Dessai, S. (2010) Robust adaptation to climate change. Weather, 65(7), 180-185.
30