A spatially distributed risk screening tool to assess climate and land use change impacts on water-related ecosystem services

Environmental Modelling & Software 83 (2016) 12e26

Contents lists available at ScienceDirect

Environmental Modelling & Software journal homepage: www.elsevier.com/locate/envsoft

A spatially distributed risk screening tool to assess climate and land use change impacts on water-related ecosystem services James E. Sample a, *, Ingrid Baber b, Rebecca Badger b a b

The James Hutton Institute, Craigiebuckler, Aberdeen, AB15 8QH, Scotland, UK Scottish Environment Protection Agency, Clearwater House, Heriot Watt Research Park, Avenue North, Riccarton, Edinburgh, EH14 4AP, Scotland, UK

a r t i c l e i n f o

a b s t r a c t

Article history: Received 2 December 2015 Received in revised form 21 March 2016 Accepted 8 May 2016

To support the implementation of the European Water Framework Directive (WFD), and as part of a tiered approach to prioritise detailed modelling, a high-level screening methodology has been developed to assess the vulnerability of water-related ecosystem services (ES) to future change. The approach incorporates a range of spatially distributed scenarios of land use and climate, which are used as inputs to a qualitative risk assessment model underpinned by expert opinion. The method makes use of widely available datasets and provides a structured way of capturing and “codifying” expert knowledge, as well as for assessing the degree of consensus between different expert groups. The range of model output reflects uncertainty in both the expert-derived assumptions and the climate & land use simulations considered. The approach has been developed in collaboration with the Scottish Environment Protection Agency (SEPA) and applied in Scotland to support the second cycle of River Basin Management Planning. Crown Copyright © 2016 Published by Elsevier Ltd. All rights reserved.

Keywords: Ecosystem services Land use Climate change Adaptation Qualitative risk assessment Water resources

Software availability

Python code and additional resources for the analysis presented are available here: https://github.com/ JamesSample/ecosystem_services_impacts.

1. Introduction Land use and water resources are closely interlinked (Howden et al., 2010). In Scotland e as in many countries e diffuse pollution from agriculture is among the main causes of failure to meet water quality targets (SEPA, 2014a, 2007), while urban expansion

Abbreviations: ES, Ecosystem Service(s); EU, European Union; FF, Future Flows; IPCC, Intergovernmental Panel on Climate Change; LCA2050, Land Capability for Agriculture 2050; LCF, Land Capability for Forestry; LCM2007, Land Cover Map 2007; LCMS1988, Land Cover Map of Scotland 1988; PET, Potential Evapotranspiration; RBMP, River Basin Management Plan(ning); SEPA, Scottish Environment Protection Agency; UKCP09, United Kingdom Climate Projections 2009; WFD, Water Framework Directive. * Corresponding author. E-mail addresses: [email protected] (J.E. Sample), ingrid.baber@sepa. org.uk (I. Baber), [email protected] (R. Badger). http://dx.doi.org/10.1016/j.envsoft.2016.05.011 1364-8152/Crown Copyright © 2016 Published by Elsevier Ltd. All rights reserved.

and afforestation may affect water quantity by changing infiltration and evapotranspiration rates. In addition, climate change is expected to lead to an intensification of the hydrological cycle (Huntington, 2006), resulting in further changes to both water quantity and quality. Meeting the food, water and energy demands of an expanding population in the face of global climate and land use change is therefore considered to be one of society's biggest challenges (Godfray et al., 2010). Across the globe, water provides a diverse range of valuable ES (Brauman et al., 2007), from drinking water provision to opportunities for recreation and the dilution of industrial discharges. Future changes to the availability and quality of water resources may therefore affect the capacity of natural systems to provide these €ter et al., 2005). This has been services (Metzger et al., 2006; Schro recognised in water policy and, under the obligations of the WFD, SEPA is required to undertake a process of River Basin Management Planning (RBMP), with the aim of maintaining or improving waterbody status while also safeguarding water-related ES. Details of the second cycle of RBMP are due for publication in 2015 (SEPA, 2014a), and to support this process there is a need to identify waterbodies where ES may be negatively affected by future change. Estimating the likely effects of land use and climate change on ES is not straightforward, as each service may respond differently to a range of factors. Physically-based, conceptual models are capable

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

of simulating some of these factors (Arnold et al., 2012; Lindstrom et al., 2010) and a variety of studies have explicitly considered the effects of future change on specific ES, for example by modelling hydropower potential (Christensen and Lettenmaier, 2007; Lehner et al., 2005) or water quality (Mehdi et al., 2015; Wilby et al., 2006) under a range of future scenarios. However, the skill of existing hydrological and water quality models is strongly dependent on the variable(s) being simulated (Gassman et al., 2007) and many models are complex and highly parameterised, making them timeconsuming to setup even for narrowly focused studies. The quality of the simulated results is also influenced by uncertainty in the model structure, parameterisation, initial conditions and input & calibration datasets (Uusitalo et al., 2015). Complex physicallybased models with large numbers of “free” parameters are especially prone to “over-fitting” the data, which makes their predictive power difficult to assess (Kirchner, 2006). For adaptation studies, a number of authors have also questioned whether the best climate scenarios currently available are good enough to be used in quantitative decision support (Frigg et al., 2013; Smith and Petersen, 2014), especially in cases where they are used to drive lengthy modelling chains to assess mitigation options. These authors argue instead for a more pragmatic approach, making use of quantitative modelling only where the numerical detail can be justified. Given time, knowledge and financial constraints, it may be impossible to obtain the measured data required to robustly parameterise complex, physically-based models (Vrana et al., 2012). Where quantitative modelling is not feasible, qualitative approaches based on expert or stakeholder opinion may be a useful way of bridging knowledge gaps and supporting high-level decision making (Heathwaite, 2003; Rowan et al., 2012), particularly if used in a hierarchical framework to identify areas worthy of more detailed (quantitative?) investigation (Volk et al., 2010). The current state-of-the-art concerning modelling with stakeholders is reviewed by Voinov et al. (2016), who state that “participatory modelling” is now one of the mainstays of environmental management and decision support. A variety of authors have also noted that the process of stakeholder elicitation can itself be beneficial, leading to better engagement and reduced levels of conflict, thereby improving the decision making process (e.g. Krueger et al., 2012). Although qualitative modelling approaches generally provide less detailed information than process-based alternatives, in many real-world applications it is sufficient for decision makers to know only the likely direction of change and an indication of magnitude (Dunn et al., 2015). In such cases, qualitative models may have the advantage of being quicker to develop and apply, so the amount of effort invested in modelling is more proportionate to the utility of the output. An additional advantage is that the assumptions and limitations of qualitative models are typically more transparent and easier to communicate, further enhancing opportunities for engagement and discussion with non-expert stakeholders, and ultimately giving the models greater credibility with users (Hall et al., 2014; Wieland and Gutzler, 2014). Key considerations when conducting a participatory modelling study include how to effectively elicit information from those taking part and how to aggregate the responses in a way that reduces bias and provides an accurate reflection of group opinion (Krueger et al., 2012; Voinov et al., 2016). One approach for dealing with bias is to use “expert calibration”, where participants are first asked to estimate some known quantities so that tendencies towards underor over-estimation can be identified and subsequently corrected. Other commonly used approaches for eliciting and aggregating group opinion include the Nominal Group (Clemen and Winkler, 1999) and Delphi (Dalkey, 1969; MacMillan and Marshall, 2006) methods, where opinions are initially gathered from each expert

13

independently, but the responses are then pooled and communicated back so that participants have the opportunity to iteratively revise their estimates in the light of feedback. For the Nominal Group approach, feedback takes place face-to-face in a workshop setting, whereas for Delphi the feedback is provided remotely so that participants do not feel pressured into changing their views, for example due to a desire to conform with group norms (Ayyub, 2001). As with physically-based approaches, it is desirable that qualitative modelling studies (especially those aiming to provide decision support) include an assessment of confidence in the model results. This is particularly the case for climate adaptation studies, where there is considerable uncertainty about both the magnitude and direction of climate change (Jenkins et al., 2009). Researchers investigating impacts on ES potential typically represent climate € ter uncertainty using an ensemble of future simulations (e.g. Schro et al., 2005), although some authors have stated that the results obtained from such ensembles may be misleading if interpreted incautiously, because they rarely provide a comprehensive representation of the true uncertainty in future climate (Frigg et al., 2014). Similarly, it is important that studies incorporating stakeholder or expert opinion allow for the possibility of uncertainty or lack of consensus in the opinions expressed (Voinov et al., 2016). The Nominal Group and Delphi approaches described above aim to reduce bias by shifting group opinion towards a consensus, but in many cases the disagreement among experts may itself provide important information that should be retained (Krueger et al., 2012). O'Hagan (2012) provides an overview of formal statistical methods for dealing with uncertainties in participatory modelling and Scholten et al. (2013) demonstrate how explicit consideration of the variance in expert opinion can improve posterior inference and decision making. Alternative approaches are described by Page et al. (2012) and Vrana et al. (2012), who present examples based on fuzzy set or possibility theory. All of these techniques represent complicated mathematical models in themselves, and they may therefore be difficult to articulate effectively to non-specialist audiences. As Booker and McNamara (2004) point out, experts involved in participatory modelling usually prefer to express themselves using natural language, rather than in the terminology of mathematical uncertainty. An important challenge is therefore how to translate natural language responses into suitable inputs for subsequent modelling. Krueger et al. (2012) state that expert elicitation should make use of, “formal, systematic and transparent procedures” to capture information. Haines-Young et al. (2012) and Burkhard et al. (2012) encoded their opinions in “lookup tables”, which were used to link ES potential to historic land cover data, thereby making it possible to estimate the impacts of land use change on service provision. Haines-Young et al. conclude their methodology provides a useful “rapid assessment” tool for decision makers, complementing more detailed, process-based modelling approaches. Dunn et al. (2015) used a similar elicitation procedure to assess the possible impacts of future change on water quality. Although their study does not explicitly consider ES, their method used “reclassification matrices” (analogous to look-up tables) that were initially proposed by the authors, but subsequently refined during an expert workshop. These matrices provide a qualitative link between changes in climate and land use variables and associated changes in water quality. At larger spatial scales, several authors have chosen to explore land use and climate change impacts on ES by adopting the qualitative vulnerability assessment framework presented by the Intergovernmental Panel on Climate Change (IPCC) in the Third € ter et al. Assessment Report (IPCC, 2001; see also section 2.3). Schro (2005) and Metzger et al. (2006), for example, used this method to

14

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

assess ES potential across the whole of Europe, identifying potentially beneficial changes (e.g. increases in forest area) as well as negative ones (e.g. declining soil fertility and water availability). In the study presented here, we adopt a qualitative, participatory approach based on reclassification matrices that is similar to that employed by Dunn et al. (2015). However, because the range of ES under consideration here is broad, the number of matrices would be too great to be tackled effectively during an expert workshop. Instead, a more manageable series of structured workshop exercises has been devised, from which the necessary matrices can be generated automatically. The workshop used a simplified version of the Nominal Group approach to encourage discussion and the exercises were designed to allow participants to express themselves using natural language (i.e. “large” versus “small”) rather than mathematical terminology. We also make use of a modified version of the IPCCs vulnerability assessment framework which, when coupled to the expert-derived reclassification matrices, provides an overall risk classification for a range of ES in approximately 3000 waterbodies across Scotland. This procedure inherently includes a qualitative assessment of model confidence that incorporates diversity in both future simulations and expert responses. In summary, this paper describes work undertaken collaboratively between researchers and regulators to support the second cycle of RBMP in Scotland. The main objectives were: i. To develop a transparent and repeatable framework for capturing expert opinion on the effects of land use and climate change on water-related ES ii. To link the framework to a spatially distributed, qualitative model of ES provision iii. To apply the model under a range of different land use, climate and expert opinion scenarios in order to generate maps, at national scale, showing the modelled change (and associated confidence) in ES provision for each service in each of SEPA's WFD waterbody catchments Although the study focuses on Scotland, the key issues and policy context are similar in many other countries, especially within the EU. The methodology described is general and the datasets required are widely available, so we believe the approach could be easily transferred to other regions. 2. Methodology 2.1. Overview The methodology as a whole is illustrated in Fig. 1 and described in detail in the subsequent sections. The approach taken combines expert opinion with spatially distributed simulations of land use and climate change. To integrate these different data types, all changes were translated onto a simple qualitative scale of five classes ranging from “large decrease” to “large increase” (Table 1). “Large” changes were defined as those likely to lead to changes in ES potential of relevance to Scotland's second cycle of RBMP (e.g. changes that could affect WFD status). For convenience, integer scores from
2 to þ2 were also assigned as shorthand representations of these categories. By eliciting expert opinion as to what constitutes e.g. a “large decrease” in the context of particular ES, it is possible to develop reclassification matrices which can be used to translate changes in climate and land use into changes in service provision. Furthermore, the combined effects of climate and land use change acting together can be estimated using an additional reclassification step

Fig. 1. Key methodological stages elaborated in the main text. ES, Ecosystem service(s).

Table 1 Qualitative changes.

scoring

system

used

to

represent

Score

Description


2
1 0 þ1 þ2

Large decrease Small decrease No change Small increase Large increase

which integrates the results from each factor acting separately. Values in this reclassification matrix represent the relative influence of climate versus land use change on service provision for each ES, and were also assigned by expert opinion. One of the guiding principles in developing the methodology is that any interested group of individuals (e.g. policy makers or regulators) should be able to complete a set of relatively simple exercises to encapsulate their beliefs. An algorithm can then be used to “codify” these responses into reclassification matrices, which can be used to estimate risk levels under a range of climate and land use scenarios. In addition, by asking a variety of experts with different backgrounds to complete the exercises, the variability in the reclassification matrices can be used to indicate uncertainty (i.e. diversity of opinion) among different expert groups.

2.2. Key datasets 2.2.1. Waterbody catchments and ES data As part of Scotland's commitments under the WFD and the first cycle of RBMP, SEPA has defined catchments for approximately 3000 inland, coastal and transitional (estuarine) waterbodies

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

across the country (Fig. 2). For each of these catchments, datasets are available to describe the presence or absence of a range of water-related ES (SEPA, 2014b). The research presented here focuses on a sub-set of 12 services considered by SEPA to be most relevant to the second cycle of RBMP and for which good quality data are available (Table 2).

2.2.2. Climate scenarios Climate change is expected to alter rainfall patterns and intensities as well as evapotranspiration rates, thereby affecting water resource availability for human and agricultural use (Arnell et al., 2000; Huntington, 2006). Although there is still considerable uncertainty in the magnitude and direction of the anticipated changes at local and regional scales, the overall picture for Northwestern Europe is of warmer, drier summers, warmer, wetter winters and an increased frequency of extreme events (Burt and Ferranti, 2012; Jenkins et al., 2009). As a first approximation, the climate change effects on waterrelated ES may be assessed by considering changes in runoff (defined as the net balance between incoming rainfall and evapotranspiration). Changing runoff patterns will clearly have a direct effect on the quantity of water available for ES provision: for example, warmer winters could have implications for snow accumulation and melting processes (Chernet et al., 2013), while increases in summer potential evapotranspiration (PET) could lead to

Fig. 2. 2013 WFD status for Scotland's inland, coastal and transitional waterbody catchments.

15

Table 2 Ecosystem services included in the assessment. ID

Ecosystem service

1 2 3 4 5 6 7 8 9 10 11 12

Hydroelectricity generation Farmed fish production Whisky production Agricultural irrigation Drinking water provision Dilution and dispersal of sewage discharges Dilution and dispersal of whisky discharges Dilution and dispersal of fish farm discharges Habitat for iconic wildlife Recreational fishing Swimming, diving, surfing Kayaking, sailing, boating

higher soil moisture deficits and an increase in irrigation demand (Brown et al., 2010). Secondarily, changing runoff patterns may also affect ES provision by affecting water quality, for example by influencing leaching processes (Barco et al., 2013) and the mobilisation and transport of particulate pollutants (Rose et al., 2012). The most comprehensive and widely used climate simulations for the UK are provided by the UK Climate Projections project (UKCP09; e.g. Jenkins et al., 2009). A variety of datasets are available, but for spatially explicit modelling the most commonly used are the outputs of the 11-member perturbed physics ensemble, based on a single regional climate model called HadRM3 (Hadley Centre, 2008). These datasets have been further refined by the Future Flows project (Prudhomme et al., 2012), which applied statistical downscaling and bias correction techniques to produce gridded daily time series of precipitation and evapotranspiration. Incorporating the 11 Future Flows simulations into our analysis makes it possible to consider climate change uncertainty as part of the risk assessment framework. However, it is worth emphasising that the Future Flows simulations are based on output from a single climate model, using greenhouse gas concentrations from the medium (A1B) emissions scenario. They therefore do not represent the full range of climate uncertainty encapsulated by UKCP09 and, in turn, UKCP09 only partially represents the true uncertainty regarding future climate (Frigg et al., 2013). The estimates of climate uncertainty presented are therefore indicative and must be interpreted with caution. To assess the magnitude of future climate change, a standard climatological baseline of 1961e1990 was compared to a future period of 2041e2070 (taken to be broadly representative of the 2050s). The 2050s were chosen to be compatible with the Scottish Government's long term “Vision for 2050”, as set out in the Scottish Land Use Strategy (Scottish Government, 2011). Note, however, that because the Future Flows datasets cover the period from 1951 to 2098, it would be possible to repeat the analysis using any baseline and future timespans within this interval. 2.2.3. Land use scenarios Water quality is very closely linked to land use and management (Weatherhead and Howden, 2009). In broad terms, different land uses are associated with different sources and degrees of pollution (Dunn et al., 2015), although other factors such as management, climate and topography also play an important role in determining pollutant losses from land to water. Changes in land use might therefore be expected to affect ES provision via changes to water quality (Dawson and Smith, 2010), although water quantity influences are also possible because some land cover types (e.g. coniferous forestry) are associated with higher evapotranspiration rates than others (Sahin and Hall, 1996). A baseline land use dataset (Fig. 3a) was developed by

16

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

reclassifying the Land Cover Map of Scotland 1988 (LCMS1988; MLURI, 1993) into six broad categories, believed to be of greatest importance from the point of view of water-related ES: arable, improved grassland, broadleaved forest, coniferous forest, seminatural and urban. A single future scenario (Fig. 3b) was also developed to represent one possible realisation of land use during the 2050s, with the focus primarily on arable expansion and afforestation, as outlined by the Scottish Government's Land Use Strategy (Scottish Government, 2010). Urban expansion was not included as, according to the Land Cover Map 2007 (LCM2007; Morton et al., 2011), built-up areas cover just 1.8% of the country, compared to 10% and 15% respectively for arable agriculture and forestry. Scotland's population has been relatively stable over the past 40 years, rising from 5.24 million in 1974 to 5.35 million in 2014 (National Records of Scotland, 2015). Future trends are poorly constrained, but the population is expected to continue to increase slowly, perhaps by around 8% by 2037 (National Records of Scotland, 2015). These changes are not insignificant and may have implications for ES delivery, but the projected expansion is nevertheless small compared to the potential changes in agriculture and forestry highlighted by the Land Use Strategy. Expansion of arable agriculture was constrained using the Land Capability for Agriculture 2050 dataset (LCA2050) of Brown et al. (2008), which identifies areas of the country likely to become more suitable for cropping as a result of climate change. All prime land in LCA2050 was assumed to become arable by the 2050s,

resulting in a 36% increase in agricultural area. For forestry, the future scenario was based on data from the Woodland Expansion Advisory Group (Towers et al., 2011) and the Land Capability for Forestry (LCF) map (Bibby et al., 1988), which were combined to identify areas for possible woodland expansion. The Scottish Government has previously expressed an aspiration for 25% total forest cover and the future scenario assumes that this is achieved by the 2050s. In order to distribute new woodland spatially, classifications in the LCF dataset were converted sequentially, starting by planting trees in the most favourable areas and then including less favourable land classes until the 25% target was reached. Areas of land not converted either to arable or to forestry in the future scenario were assigned the same land use class as on LCM2007, which is the most up to date national scale dataset of land use for Scotland. The future scenario represents the maximum possible extent of both arable agriculture and forestry by the middle of the 21st century. In particular, the area of improved grassland (and therefore of mixed farming) in the scenario is implausibly reduced, because much of the existing improved grassland is considered to be suitable for arable farming in the future, while much of the remainder is capable of supporting trees and is therefore converted to woodland. Although unrealistic, the scenario is nevertheless useful as it likely reflects the upper limit of possible land use change impacts on Scotland's water resources. It therefore provides a sensitivity test, in the sense that any ES not affected by this scenario is unlikely to be vulnerable to more modest changes.

Fig. 3. (a) Baseline land use. (b) Future land use scenario.

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

2.3. Vulnerability assessment framework

“small” changes, but not “large” ones (Fig. 5).

To assess the impact of changes in climate and land use on water-based ES, a risk classification methodology was adapted from the vulnerability assessment framework proposed by the IPCC (2001). This defines vulnerability as a function of sensitivity, adaptive capacity and exposure which, in the context of the work presented here, can be expressed as follows:

2.4. Capturing expert opinion

 Exposure is the magnitude of land use or climate change that takes place  Sensitivity describes how much an ES changes for a given amount of exposure  Adaptive capacity describes the ability of aspects of an ES to change in order to mitigate negative effects. For example, farmers may choose to grow drought-resistant crops with lower irrigation demand, or run-of-river hydropower developers may fit several small turbines in place of a single large one, thereby increasing the range of flows over which a plant can operate. By combining these concepts with the qualitative classification scheme in Table 1, it is possible to visualise vulnerability as shown in Fig. 4. The exposure axis is divided into a qualitative scale running from
2 to þ2; the sensitivity and adaptive capacity axes are each divided into three classes: low, medium and high. For each “exposure variable”, each ES can be assigned to a position on the sensitivity and adaptive capacity axes. This provides a method for translating changes in runoff and land use into changes in ES provision, the key assumption being that services with low sensitivity and/or high adaptive capacity will be buffered against changes driven by external factors such as climate and land use. In the example given in Fig. 4, the high adaptive capacity of “ES 2” results in smaller changes to service provision than for “ES 1”. Adaptation is only necessary when changes negatively affect ES provision. For example, if “ES 2” on Fig. 4 represented drinking water provision, a decrease in the resource might be considered bad, whereas an increase could be good. In this situation, the adaptive capacity can be used to buffer declining resources, but need not also buffer increasing resources. The response scores might therefore read [
1, 0, 0, þ1, þ2] instead of [
1, 0, 0, 0, þ1]. The amount of “buffering” possible is determined by each service's position on the sensitivity e adaptive capacity axes. ES occupying cells in the lower-right of the grid are assumed to have no ability to buffer change, whereas those in the upper left corner are capable of buffering even “large” impacts. Services along the diagonal are assumed to have enough buffering capacity to mitigate

Fig. 4. Overall vulnerability expressed as a combination of sensitivity, adaptive capacity and exposure. ES, ecosystem service; L, low; M, medium; H, high.

17

A workshop was held in Edinburgh on March 16th, 2015. Attendees included water resources and ES experts, environmental economists, specialists in adaptation & sustainability and senior policy officers, many of whom are actively involved in developing Scotland's second RBMP. Prior to the event, several participants provided comments and guidance on the approach being proposed. This allowed iterative refinement of the methodology and ensured core aspects of the process had support from key workshop participants on the day. The eight participants were divided into two equal groups and each group was asked to complete six exercises (Table 3), designed to capture opinions in a structured way. The exercises are described in detail in the following sections, illustrated by example responses from one of the groups. Handouts for the exercises are also provided as supplementary material. Exercises 1, 2 and 3 make it possible to assess the climatechange-only impacts on each ES, whereas exercises 1, 4 and 5 permit consideration of land-use-change-only effects. Exercise 6 makes it possible to combine these two separate results in order to estimate the joint effects of climate and land use change acting together. Differences between the results from each group were used to give some indication of the degree of consensus provided by expert opinion. With just two groups it is not possible to characterise this uncertainty in detail, but with additional resources the analysis could be made more robust by repeating the workshop with a larger number of groups and participants. 2.4.1. Exercise 1: adaptive capacity e sensitivity grids for land use and climate change This exercise determines the degree to which each ES can mitigate impacts due to climate and land use change (i.e. the “buffering capacity” e see Fig. 5). Workshop participants were provided with two blank sensitivity e adaptive capacity grids. On the first they were asked to assign each of the 12 ES (Table 2) to a position on the grid assuming the exposure variable was “change in runoff”. This task was then repeated for the second grid, but with the exposure variable switched to “change in land use”. Illustrative choices are shown in Fig. 6. 2.4.2. Exercise 2: key times of year This exercise determines which sub-sets of the climate change data are used to estimate future changes. By default, average annual changes in runoff are considered between the baseline and future

Fig. 5. “Buffering” scores based on sensitivity and adaptive capacity.

18

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26 Table 3 Summary of the six workshop exercises used to gather expert opinion. Exercise

Title

1 2 3 4 5 6

Develop adaptive capacity e sensitivity grids for land use and climate change Identify key times of year Decide on nature of change (good, bad or neutral?) Develop land use hierarchy Assign land use weightings Assign climate and land use weightings

Fig. 6. Example of ecosystem services assigned to sensitivity e adaptive capacity grids assuming (a) exposure ¼ change in runoff and (b) exposure ¼ change in land use.

periods. However if, for example, workshop participants prioritised spring and summer instead, the algorithm would estimate percentage changes based upon these months only. Each group was asked to select the key season(s) of primary relevance to each ES. For example, in the case of agricultural irrigation it might be reasonable to choose spring and summer, as these are the main growing seasons as well as the times of year when water is most likely to be scarce. Illustrative responses are shown in Fig. 7. 2.4.3. Exercise 3: nature of change Adaptation need only take place to mitigate undesirable changes and not potentially beneficial ones. This exercise required participants to decide whether increases and decreases in runoff might be considered good, neutral or bad for each ES. The results of this exercise were used to decide whether to apply the available “buffering capacity” (from exercise 1) to all changes, or just to changes in one particular direction. Illustrative choices are shown in Fig. 7. 2.4.4. Exercise 4: land use hierarchy Participants were asked to rank each of the six land use classes (section 2.2.3) according to their relative effects on water quality and quantity. The template provided (Fig. 8) requires the land categories to be divided into three levels. This is done in order to be compatible with the qualitative classification scheme in Table 1. For example, based on the choices illustrated in Fig. 8, a change in land use from arable to coniferous forestry would result in a two-step improvement in

water quality (þ2; a “large increase”), but also a one-step decrease in water quantity (
1; a “small decrease”). By restricting the template to three levels, the maximum change that can take place between any two land use classes is ±2, which is compatible with the largest change allowed by the qualitative scale. Note that there is no restriction placed on the number of land use classes assigned to each level and that levels can be left empty if desired. 2.4.5. Exercise 5: land use weightings If a particular change in land use results in a þ2 change in water quality and a
1 change in water quantity (as in the example from Fig. 8 mentioned above), the overall land use change impact can be estimated by summing these scores. This assumes that water quantity and quality are equally important for each ES, which is not true in many cases: hydropower potential, for example, is entirely determined by water quantity (regardless of quality), whereas drinking water provision could be significantly affected by both. For this reason, it is necessary to assign relative weights to the quantity and quality scores from exercise 4. The overall land use effect is then calculated as a weighted sum:

SLU ¼ wqual Cqual þ wquant Cquant

(1)

Where SLU is the overall land use score; wqual is the weight assigned to water quality; Cqual is the quality change score from Fig. 8 due to switching from one land use to another; and wquant and Cquant are the equivalents for water quantity. In exercise 5, participants were asked to suggest values for the weightings, wqual and wquant, normalised such that they sum to 1.

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

19

Fig. 7. Example results from exercise 2 (key seasons) and exercise 3 (nature of change).

Fig. 8. Example results from exercise 4.

Example responses are shown in Table 4. Note that the [Quality, Quantity] scores assigned for hydropower are [0, 1], whereas for drinking water they are [0.5, 0.5], which is consistent with the discussion above. 2.4.6. Exercise 6: climate and land use weightings In the final step, participants were asked to weight the relative influences of land use versus climate change. This is very similar to the previous exercise, except here the focus is on comparing the effects of climate versus land use on each ES, rather than comparing the water quality versus quantity effects associated with land use change alone. Example responses are shown in Table 5.

2.5. Classification matrices The choices made by each expert group were encoded in a set of reclassification matrices, which were used to spatially distribute the workshop results. 2.5.1. Climate only The first step towards developing reclassification matrices for climate is to estimate runoff change. This was done by considering UKCP09 change quantiles to separate “large” changes from “small” ones, following the approach of Towers et al. (2012). The runoff matrix is illustrated in Fig. 9 and shows, for example,

Table 4 Example weightings to reflect whether the land use change influence for a particular ES is due to changes in water quantity, water quality or a mixture of the two. ID

Ecosystem service

Quality weight

Quantity weight

1 2 3 4 5 6 7 8 9 10 11 12

Hydroelectricity generation Farmed fish production Whisky production Agricultural irrigation Drinking water provision Dilution and dispersal of sewage discharges Dilution and dispersal of whisky discharges Dilution and dispersal of fish farm discharges Habitat for iconic wildlife Recreational fishing Swimming, diving, surfing Kayaking, sailing, boating

0 1 0.33 0.33 0.5 0.5 0.5 0.5 0.66 0.66 1 0.66

1 0 0.66 0.66 0.5 0.5 0.5 0.5 0.33 0.33 0 0.33

20

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

Table 5 Example weightings to reflect the relative influence of climate and land use change. ID

Ecosystem service

Land use weight

Climate weight

1 2 3 4 5 6 7 8 9 10 11 12

Hydro electricity Farmed fish Whisky Agricultural irrigation Drinking water Dilution and dispersal of sewage discharges Dilution and dispersal of discharges from whisky Dilution and dispersal of fish farm discharges Habitat for iconic wildlife Recreational fishing Swimming, diving, surfing Kayaking, sailing, boating

0.33 0.66 0.33 0.33 0.5 0.5 0.5 0.5 0.66 0.66 0.66 0.5

0.66 0.33 0.66 0.66 0.5 0.5 0.5 0.5 0.33 0.33 0.33 0.5

that decreases in precipitation of more than 15% coupled with increases in potential evapotranspiration (PET) of more than 15% are predicted to cause “large decreases” (
2) in runoff. Note that precipitation is assumed to dominate PET, such that a >15% increase in PET is not sufficient to cancel a >15% increase in precipitation; instead, the net effect is a “small increase” (þ1) in runoff. These assumptions are broadly consistent with the climate of Scotland, but could be easily modified for applications elsewhere. Climate impacts on each ES are initially estimated on a grid-cellby-grid-cell basis, and are only later aggregated to catchment level (see Section 2.6.1). Fig. 10 shows a schematic of the steps performed by the risk classification algorithm in order to determine the overall climate change impact for each grid cell. 2.5.2. Land use only The results from exercise 4 (Fig. 8) can be used to develop transition matrices for water quality and water quantity. An example for water quality based on the choices illustrated in Fig. 8 is shown in Fig. 11 (the equivalent for water quantity is not shown for brevity). Note that the matrix is symmetric. These matrices are used to provide the change scores, Cqual and Cquant , in Eq. (1), while the weights wqual and wquant are taken from Table 4. A schematic of the steps used by the risk classification algorithm to determine overall impact in each grid cell due to land use change alone is shown in Fig. 12. 2.5.3. Climate and land use combined To estimate the impacts of climate and land use acting together, the scores in each grid cell from processing climate (CClim) and land use (CLU) separately were combined in a final weighting step:

STotal ¼ wLU CLU þ wClim CClim

(2)

Where the weights, wLU and wClim , are taken from Table 5. This provides an overall score, still in the range from
2 to þ2, representing the effect in each grid cell of climate and land use change acting simultaneously. This approach is simplistic, as it assumes the

Precipitation Change (%)

Runoff Matrix PET Change (%) < -15 -15 to -5 -5 to +5 +5 to +15 > +15

<15 -1 -1 -2 -2 -2

-15 to -5

-5 to +5

+5 to +15

0 -1 -1 -2 -2

1 1 0 -1 -1

2 2 1 1 0

> +15 2 2 2 1 1

Fig. 9. Runoff matrix (after Towers et al., 2012). PET, potential evapotranspiration.

Fig. 10. Schematic of the procedure for estimating climate change impacts. Yellow boxes are intermediate processing steps and red boxes are gridded outputs. ES, ecosystem service; ET, evapotranspiration; Pptn, precipitation. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Water quality Baseline land use Arable Improved grass Coniferous forest Broadleaved forest Semi-natural Urban

Arable 0 0 -2

Imp. grass 0 0 -2

Future Land use Conif. B-leaf. forest forest +2 +2 +2 +2 0 0

Seminat. +2 +2 0

Urban +1 +1 -1

-2

-2

0

0

0

-1

-2 -1

-2 -1

0 +1

0 +1

0 +1

-1 0

Fig. 11. Transition matrix for water quality based on the information in Fig. 8. A similar matrix must also be constructed for water quantity.

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

21

upstream area because, for the ES under consideration, changes taking place in the upper catchment will affect water bodies downstream.

2.6.2. Offshore and transitional waterbodies The climate and land use scenarios represent changes taking place on land, but many of the ES under consideration are associated with coastal and offshore waterbodies (Fig. 2). Modelling potential impacts in these locations requires identifying which inland catchments influence which offshore waterbodies. This has been achieved using the simple rule base illustrated in Fig. 13, which shows four inland catchments (I1 to I4), four transitional catchments (T1 to T4) and one offshore catchment (O1). Transitional catchments are assumed to be predominantly influenced by the inland catchments that drain directly to them: T1 is primarily influenced by I1; T2 by I2 and I3; and T4 by I4. Transitional catchments such as T3 that have no direct link to an inland catchment are assumed to be influenced by any neighbouring transitional catchments (T2 and T4). T3 is therefore assumed to be affected by climate and land use changes taking place in I2, I3 and I4. Changes in offshore catchments are assumed to reflect changes across all the adjoining transitional catchments. For example, the change in O1 is determined by the combined changes across T1, T2, T3 and T4, which is the same as aggregating the scores for I1, I2 I3 and I4. This is a highly simplistic way of linking onshore scenarios to offshore changes in ES provision. In reality, the effects offshore will be determined by a complex interplay between coastal geomorphology, ocean tides & currents and prevailing weather conditions. Developing a more sophisticated model to incorporate these details was considered to be beyond scope, however, and for a qualitative risk assessment the simpler approach was deemed preferable.

Fig. 12. Schematic of the procedure for estimating land use change impacts only. Yellow boxes are intermediate processing steps and red boxes are model outputs. LU, land use; ES, ecosystem service. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

climate and land use change effects on ES provision can be combined as a linear sum. Other functional forms were considered but dismissed as introducing unnecessary complexity. Because the approach uses a qualitative scale with limited range (i.e. “small decrease”, “large increase” etc.), most monotonic functions would likely give the same end result, and for small changes in climate and land use the linear approach should provide a reasonable approximation to most plausible alternatives. 2.6. Data aggregation 2.6.1. Spatial data aggregation The procedure described so far involves calculating change scores on a 1 km by 1 km grid across the whole of Scotland. In order to make these results relevant to the second cycle of RBMP, it was necessary to summarise the gridded data to produce aggregated scores for each waterbody catchment. This was achieved by calculating the modal (most common) score within each catchment, including the upstream area. It is important to include the

Fig. 13. Illustrative linkages between inland catchments (I1 to I4), transitional waterbodies (T1 to T4) and offshore waterbodies (O1).

22

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

2.6.3. Representing uncertainty Uncertainty is included in the risk model by considering multiple plausible climate simulations and different versions of expert opinion. No uncertainty estimates are currently available for land use change, although if other future land use simulations were available the method could easily be extended to include them. In this example, the methodology generates 552 maps in total. For each of the 12 ES there are:  22 maps of climate change impacts (11 climate simulations times two sets of expert opinion)  2 maps of land use change impacts (one future land use simulation times two sets of expert opinion)  22 maps of climate and land use change impacts combined (11 climate simulations times one future land use simulation times two sets of expert opinion) Each of these sets has been summarised by calculating simple statistics for each waterbody consisting of (i) the median impact score and (ii) the range of scores (maximum minus minimum), which gives an estimate of the uncertainty in the output classification. The uncertainty score is a value between zero and four, with zero corresponding to high confidence (i.e. all the simulations produced the same impact score) and four indicating low confidence, because some model combinations resulted in “large decreases” (
2) in service potential whereas others predicted “large increases” (þ2). Presenting maps of the median impact and confidence side by side makes it possible to evaluate the predicted risk

to a waterbody in context with the reliability of that prediction (e.g. Fig. 14). 2.7. Incorporating data on existing ecosystem services The method described in Sections 2.1e2.6 assumes all ES are present in all waterbody catchments. In reality this is not the case: as described in section 2.2.1, SEPA have already collected data indicating the presence or absence of ES in each catchment. In the final stage of processing, the SEPA datasets were therefore used to mask the model output, such that impact scores are only presented for waterbody catchments where ES are believed to be present. 3. Results The full suite of 36 maps (12 ES times 3 future scenarios: land use only, climate only and land use and climate combined) is provided as supplementary information. Here we discuss some illustrative examples to highlight key features and guide interpretation of the results. Fig. 14 shows an example of the output illustrating combined climate and land use change effects on the potential for agricultural irrigation. Overall, under the scenarios considered, the model makes medium to high confidence predictions of a decline in the ability of natural water systems to meet future irrigation demands across large swathes of the country. Spring and summer were chosen as the most important times for this ES (in exercise 2) and, during these periods, the majority of

Fig. 14. Model results for agricultural irrigation potential under combined land use and climate change.

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

future climate simulations indicate substantial decreases in runoff. This means the climate signal represented on these maps is fairly consistent and generally associated with declining irrigation potential. Additionally, in the southwest of the country there are currently large areas of improved grassland that are converted to arable cropping as part of the future land use scenario. In exercise 4, both workshop groups assumed this change would be associated with an increase in water demand, and in exercise 5 both assigned higher weights to water quantity than to water quality. These decisions, combined with the general reduction in summer runoff linked to climate change, lead to “large decreases” in the potential for irrigation in parts of the southwest e.g. Dumfries and Galloway. A second example is given in Fig. 15, which shows the combined effects of land use and climate change on the potential for whisky production. The model makes high confidence predictions that whisky production potential will decrease by a small amount in many parts of the country, as well as making lower confidence predictions of large decreases for some waterbody catchments. In the workshop exercises, both expert groups made very similar choices for this ES: the same buffering score was assigned in exercise 1; summer was selected as the most important season in exercise 2 and the same weights were assigned in exercises 5 and 6. Because all the climate simulations predict decreases in runoff during the summer months, the modelled climate change impacts are generally consistent and are associated with small decreases in whisky production potential in most areas. In addition, those catchments where the simulated land use switches from improved grass to arable are predicted to experience further reductions in potential, leading to “large decreases” when combined with climate change effects. Note however that the areas with larger modelled changes are also associated with higher uncertainty, because in exercise 4 one group of experts thought improved grass and arable crops used similar amounts of water, whereas the other assumed that any conversion to arable cropping would lead to reductions in the amount of water available to the whisky industry. Table 6 summarises the impacts of combined land use and climate change for each ES. Note that due to the spatial variability of the output, it is difficult to give an accurate representation in a table such as this e a much better overview can be obtained by consulting the maps themselves (see supplementary information). The model makes medium to high confidence predictions of “no change” for a variety of ES, including: farmed fish production; habitats for iconic wildlife; recreational fishing; swimming, diving & surfing; and kayaking, sailing & boating. This is perhaps not surprising as many of these services are associated with either large inland lochs or coastal water bodies, which are likely to be less sensitive than smaller catchments to changes in water resources. Substantial decreases in overall service potential are modelled for whisky production and agricultural irrigation (see above). Smaller but still consistent decreases are also predicted for drinking water provision and the ability of natural water courses to dilute discharges from waste water treatment plants and the whisky industry. Run-of-river hydropower potential is perhaps the most spatially variable ES of all, with some sites towards the west coast predicted to experience small increases in power generation potential, while others further east experience either no change or small decreases. 4. Discussion The methodology provides a way of exploring the possible implications of scenarios of land use and climate change for a range of water-related ES. Although still complex, the approach is nevertheless much simpler than alternatives such as building quantitative conceptual or process-based models for each ES. Indeed,

23

simplicity is one of the key strengths of the method: because the whole process can be communicated easily, stakeholders can objectively evaluate strengths and weaknesses before deciding how much faith to place in the output. This is often not possible for more sophisticated “black box” approaches, which may provide a lot of numerical detail, but which are frequently so complicated that the quality of the output is difficult to assess e even for modelling specialists (Jakeman et al., 2006; Kirchner, 2006). A second advantage of the method is its flexibility. As noted by Fürst et al. (2010), successful decision support requires a high level of stakeholder participation, which can only be achieved when models are communicated effectively and adapted to users’ needs. The core assumptions required for this analysis can be explained transparently and are parameterised in such a way that they can be changed according to user preferences. It would be straightforward, for example, to modify the runoff matrix, or to consider a different set of land use categories or future simulations. Additionally, the relative simplicity of the algorithm means that, unlike quantitative modelling approaches at national scale, the analysis can be re-run repeatedly to allow users to explore the implications of different sets of assumptions. One obvious disadvantage of the qualitative approach is that the results only represent relative risk, because they rely on expert opinion to distinguish e.g. “large” versus “small” changes. Different experts may have different ideas of what these terms mean, leading to what Kuhnert et al. (2010) refer to as “linguistic uncertainty”. During our workshop, “large” changes were defined as those likely to lead to changes in ES potential of relevance to Scotland's second cycle of RBMP but, given the varied backgrounds of our experts and the wide range of factors under consideration, this definition is still subject to interpretation. Within each workshop group, the experts were encouraged to discuss these issues, but previous work (e.g. Knol et al., 2010) has shown that group discussions may be dominated by particular sub-groups or vocal individuals, leading to biases and/or overemphasis of consensus. A possible improvement would have been to use the Delphi method for elicitation (Section 1), but this would involve each expert separately completing the exercises and then iteratively discussing each individual's responses with the others. This is resource-intensive and would likely have created problems of “expert fatigue” as described, for example, by Page et al. (2012). To some extent, our methodology mitigates issues of overemphasis of consensus and cognitive bias by dividing the experts into more than one group. The within-group discussions help participants to develop a common understanding of the problem and have the benefit of reducing linguistic uncertainty, but the separation between groups prevents any one individual or sub-group from dominating the overall response. With just two groups in our case study it is not possible to characterise expert-related uncertainty in detail, but a larger study with more groups would have the potential to reduce the ambiguity associated with interpreting the exercise(s), without artificially reducing genuine uncertainty due to e.g. overemphasis of consensus. A second source of bias is the workshop exercises themselves, which limited the range of possible responses and enforced a highly conceptualised characterisation of the ways in which climate and land use change might affect ES. For example, the choice of key seasons in exercise 2 ignores the possible effects of water storage, which may be important given that some climate simulations for Scotland show decreases in summer precipitation being compensated by increases during the autumn and winter (Jenkins et al., 2009). If autumn and winter rainfall can be stored, either in the ground or in reservoirs, it is possible the effects of decreasing summer precipitation on ES provision might be less than indicated by our analysis (e.g. on Fig. 14). On the other hand, in the context of

24

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

Fig. 15. Model results for whisky production potential under combined land use and climate change.

Table 6 Summary of results for each ES. ID

Ecosystem service

Combined effects of land use and climate change Modelled impact

Modelled confidence

1 2 3 4 5 6 7 8 9 10 11 12

Hydroelectricity generation Farmed fish production Whisky production Agricultural irrigation Drinking water provision Dilution and dispersal of sewage discharges Dilution and dispersal of whisky discharges Dilution and dispersal of fish farm discharges Habitat for iconic wildlife Recreational fishing Swimming, diving, surfing Kayaking, sailing, boating

Small increase to small decrease; mostly no change No change Decrease (mostly small, but occasionally large) Widespread decrease (occasionally large) Widespread small decrease Widespread small decrease Small decrease No change or small decrease No change No change No change No change

Medium High Medium-High Medium-High Medium-High Medium-High High Medium Medium High Medium-High Medium

agricultural irrigation, many farmers in Scotland currently abstract only small amounts of water during the summer, as the incoming rainfall delivers sufficient water directly to the crops. Even if groundwater recharge during the winter maintains summer base flows, it is likely that significant changes to the irrigation regime would be necessary to deliver this water to the crops. Similarly, storage capacities in most Scottish reservoirs have been designed with historic conditions in mind, so drinking water or hydropower reservoirs may need expanding to make effective use of any seasonal increases in rainfall (Sample et al., 2015). In this sense, maps such as Fig. 14 may still provide a reasonable approximation of

changes in ES provision (despite the simplicity of the approach), because the costs of making effective use of stored water are often high. This may explain why none of the experts assigned “high” adaptive capacity to any of the ES in exercise 1 (Fig. 6a). Dealing with the bias imposed by the structure of the exercises is not straightforward, but these issues are fundamentally comparable to the subjective decisions made when constructing processbased, quantitative models. As Krueger et al. (2012) point out, informal expert (i.e. modeller) opinion enters all stages of the modelling process via choices of model structure, parameterisation, initialisation, discretisation, calibration etc. This is the case even for

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

studies not typically considered to be informed by expert judgement. At present, perhaps the most effective way of dealing with such subjectivities in both qualitative and quantitative studies is to make the key modelling assumptions transparent and explicit wherever possible, so they can be challenged and modified as necessary. Involving experts and stakeholders in the model development process, as done here, is another way to ensure that assumptions and limitations will be more acceptable to end users (Ayyub, 2001). From the point of view of decision support, a further practical limitation of the methodology is the issue of “scale compression”, which becomes apparent when comparing maps of climate or land use change impacts individually with those showing the combined effects of climate and land use change acting together. At each stage in the modelling process, the approach adopted was to “renormalise” the impact scores to keep them within the range
2 to þ2 (Table 1). This is consistent with the philosophy of a qualitative screening assessment, but it can produce outputs that are confusing at first glance. For example, in cases where land use or climate change individually lead to “large” decreases (
2), the combined effect (which could nominally be taken to equal
4) has been renormalised back to
2. The result is that “large” changes on the maps for land use or climate change individually are not of the same magnitude as “large” changes on the combined maps showing land use and climate change acting together. This is believed to be preferable to the alternative of allowing the impact scale to expand after each matrix combination, which would result in a finely differentiated metric giving the impression of containing more quantitative information than is actually present. The downside, however, is that care must be taken when comparing maps showing different scenario combinations. 5. Conclusion This paper presents a systematic and repeatable framework for capturing and “codifying” expert opinion regarding the potential for land use and climate change to affect water-related ecosystem services. The approach has been demonstrated in Scotland where a workshop was held in which groups of experts assessed the likely responses of various ecosystem services to future change. Results from the workshop have been used to drive a national scale qualitative impact model for the 2050s. The output consists of estimates of the most likely impact on each ecosystem service in more than 3000 Scottish sub-catchments, together with an indication of confidence based on the consistency of results across a range of future scenarios. The model makes medium to high confidence predictions of significant reductions in the ability of natural systems to supply water for agricultural irrigation and whisky production. Reductions are also predicted for drinking water provision and the capacity to dilute waste water and whisky production discharges. In contrast, the results include medium to high confidence predictions of approximately no change for the following ecosystem services: farmed fish production; habitats for iconic wildlife; recreational fishing; swimming, diving & surfing; and kayaking, sailing & boating. The results are spatially variable and are best interpreted by considering maps showing the median impact and range of uncertainty in each waterbody catchment. All the maps are available in the supplementary information section. The methodology provides a high-level evaluation encompassing a range of ecosystem services at large (e.g. national) spatial scales. Although less detailed than some alternative approaches, the framework is comparatively simple and transparent and nevertheless produces output that is operationally relevant to

25

regulators. Furthermore, the data requirements are small compared to process-based modelling, and the amount of time invested is more proportionate to the practical utility of the output, given our present state of system knowledge. We believe the method complements more sophisticated (e.g. quantitative, physically-based) modelling approaches by providing a high-level screening assessment that can be used to prioritise more detailed river basin management planning. Acknowledgements This research was funded by the Scottish Government's Rural and Environment Science and Analytical Services and the Centre of Expertise for Waters (CREW). The authors would like to thank the workshop participants for their input, as well as Dr. Kit Macleod and Dr. Leah Jackson-Blake for their helpful comments and discussion. We are also grateful to Dr. Tobias Krueger and two anonymous reviewers for their constructive comments on the draft manuscript. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.envsoft.2016.05.011. References Arnell, N.W., Grabs, W., Mimikou, M., Schumann, A., Reynard, N.S., Kaczmarek, Z., Oancea, V., Buchtele, J., Kasparek, L., Blazoka, S., Starosolszky, O., Kuzin, A., 2000. Effect of Climate Change on Hydrological Regimes and Water Resources in Europe: Summary Report of Projects EV5V-CT93-0293 and EV5V-CT94-0114 (Southampton, UK). Arnold, J.G., Moriasi, D.N., Gassman, P., Abbaspour, K.C., White, M.J., Srinivasan, R., Santhi, C., Harmel, R.D., van Griensven, A., van Liew, M.W., Kannan, N., Jha, M., 2012. SWAT: model use, calibration, and validation. Trans. ASABE 55, 1491e1508. Ayyub, B., 2001. Elicitation of Expert Opinions for Uncertainty and Risks. CRC Press. Barco, J., Gunawan, S., Hogue, T.S., 2013. Seasonal controls on stream chemical export across diverse coastal watersheds in the USA. Hydrol. Process 27, 1440e1453. http://dx.doi.org/10.1002/hyp.9294. Bibby, J.S., Heslop, R.E.F., Hartnup, R., 1988. Land Capability for Forestry in Britain (Aberdeen). Booker, J.M., McNamara, L.A., 2004. Solving black box computation problems using expert knowledge theory and methods. Reliab. Eng. Syst. Saf. 85, 331e340. http://dx.doi.org/10.1016/j.ress.2004.03.021. Brauman, K.A., Daily, G.C., Duarte, T.K., Mooney, H.A., 2007. The nature and value of ecosystem services: an overview highlighting hydrologic services. Annu. Rev. Environ. Resour. 32, 67e98. http://dx.doi.org/10.1146/ annurev.energy.32.031306.102758. Brown, I., Towers, W., Rivington, M., Black, H., 2008. Influence of climate change on agricultural land-use potential: adapting and updating the land capability system for Scotland. Clim. Res. 37, 43e57. http://dx.doi.org/10.3354/cr00753. Brown, I., Poggio, L., Gimona, A., Castellazzi, M., 2010. Climate change, drought risk and land capability for agriculture: implications for land use in Scotland. Reg. Environ. Chang. 11, 503e518. http://dx.doi.org/10.1007/s10113-010-0163-z. Burkhard, B., Kroll, F., Nedkov, S., Müller, F., 2012. Mapping ecosystem service supply, demand and budgets. Ecol. Indic. 21, 17e29. http://dx.doi.org/10.1016/ j.ecolind.2011.06.019. Burt, T.P., Ferranti, E.J.S., 2012. Changing patterns of heavy rainfall in upland areas: a case study from northern England. Int. J. Climatol. 32, 518e532. http:// dx.doi.org/10.1002/joc.2287. Chernet, H.H., Alfredsen, K., Killingtveit, Å., 2013. The impacts of climate change on a Norwegian high-head hydropower system. J. Water Clim. Chang. 4, 17. http:// dx.doi.org/10.2166/wcc.2013.042. Christensen, N.S., Lettenmaier, D.P., 2007. A multimodel ensemble approach to assessment of climate change impacts on the hydrology and water resources of the Colorado River Basin. Hydrol. Earth Syst. Sci. 11, 1417e1434. http:// dx.doi.org/10.5194/hess-11-1417-2007. Clemen, R.T., Winkler, R.L., 1999. Combining probability distributions from experts in risk analysis. Risk Anal. 19, 187e203. http://dx.doi.org/10.1111/j.15396924.1999.tb00399.x. Dalkey, N., 1969. The Delphi Method: an Experimental Study of Group Opinion. Dawson, J.J., Smith, P., 2010. Integrative management to mitigate diffuse pollution in multi-functional landscapes. Curr. Opin. Environ. Sustain 2, 375e382. http:// dx.doi.org/10.1016/j.cosust.2010.09.005. Dunn, S.M., Towers, W., Dawson, J.J.C., Sample, J., McDonald, J., 2015. A pragmatic

26

J.E. Sample et al. / Environmental Modelling & Software 83 (2016) 12e26

methodology for horizon scanning of water quality linked to future climate and land use scenarios. Land use policy 44, 131e144. http://dx.doi.org/10.1016/ j.landusepol.2014.12.007. Frigg, R., Smith, L.A., Stainforth, D., 2013. The myopia of imperfect climate models: the case of UKCP09. Philos. Sci. 80, 886e897. Frigg, R., Bradley, S., Du, H., Smith, L.A., 2014. Laplace's demon and the adventures of his apprentices. Philos. Sci. 81, 31e59. Fürst, C., Volk, M., Makeschin, F., 2010. Squaring the circle? Combining models, indicators, experts and end-users in integrated land-use management support tools. Environ. Manage 46, 829e833. http://dx.doi.org/10.1007/s00267-0109574-3. Gassman, P., Reyes, M.R., Green, C.H., Arnold, J.G., 2007. The Soil and Water Assessment Tool: historical development, applications and future research directions. Trans. ASABE 50, 1211e1250. Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., Toulmin, C., 2010. Food security: the challenge of feeding 9 billion people. Science 327, 812e818. http://dx.doi.org/ 10.1126/science.1185383. Scottish Government, 2010. Draft Land Use Strategy Environmental Report. Scottish Government, 2011. Getting the Best from Our Land: A Land Use Strategy for Scotland. Hadley Centre, 2008. Met Office Hadley Centre Regional Climate Model (HadRM3PPE) Data. Haines-Young, R., Potschin, M., Kienast, F., 2012. Indicators of ecosystem service potential at European scales: mapping marginal changes and trade-offs. Ecol. Indic. 21, 39e53. http://dx.doi.org/10.1016/j.ecolind.2011.09.004. Hall, D.M., Lazarus, E.D., Swannack, T.M., 2014. Strategies for communicating systems models. Environ. Model. Softw. 55, 70e76. http://dx.doi.org/10.1016/ j.envsoft.2014.01.007. Heathwaite, A.L., 2003. Making process-based knowledge useable at the operational level: a framework for modelling diffuse pollution from agricultural land. Environ. Model. Softw. 18, 753e760. http://dx.doi.org/10.1016/S1364-8152(03) 00077-X. Howden, N.J.K., Burt, T.P., Worrall, F., Whelan, M.J., Bieroza, M., 2010. Nitrate concentrations and fluxes in the River Thames over 140 years (1868-2008): are increases irreversible? Hydrol. Process 24, 2657e2662. http://dx.doi.org/ 10.1002/hyp.7835. Huntington, T.G., 2006. Evidence for intensification of the global water cycle: review and synthesis. J. Hydrol. 319, 83e95. http://dx.doi.org/10.1016/ j.jhydrol.2005.07.003. IPCC, 2001. Climate Change 2001: Impacts, Adaptation and Vulnerability, Contribution of Working Group II to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press. Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in development and evaluation of environmental models. Environ. Model. Softw. 21, 602e614. http://dx.doi.org/10.1016/j.envsoft.2006.01.004. Jenkins, G.J., Murphy, J.M., Sexton, D.M.H., Lowe, J.A., Jones, P., Kilsby, C.G., 2009. UK Climate Projections: Briefing Report. Exeter. Kirchner, J.W., 2006. Getting the right answers for the right reasons: linking measurements, analyses, and models to advance the science of hydrology. Water Resour. Res. 42, 1e5. http://dx.doi.org/10.1029/2005WR004362. Knol, A.B., Slottje, P., van der Sluijs, J.P., Lebret, E., 2010. The use of expert elicitation in environmental health impact assessment: a seven step procedure. Environ. Health 9, 19. http://dx.doi.org/10.1186/1476-069X-9-19. Krueger, T., Page, T., Hubacek, K., Smith, L., Hiscock, K., 2012. The role of expert opinion in environmental modelling. Environ. Model. Softw. 36, 4e18. http:// dx.doi.org/10.1016/j.envsoft.2012.01.011. Kuhnert, P.M., Martin, T.G., Griffiths, S.P., 2010. A guide to eliciting and using expert knowledge in Bayesian ecological models. Ecol. Lett. 13, 900e914. http:// dx.doi.org/10.1111/j.1461-0248.2010.01477.x. Lehner, B., Czisch, G., Vassolo, S., 2005. The impact of global change on the hydropower potential of Europe: a model-based analysis. Energy Policy 33, 839e855. http://dx.doi.org/10.1016/j.enpol.2003.10.018. Lindstrom, G., Pers, C., Rosberg, J., Stromqvist, J., Arheimer, B., 2010. Development and Testing of the HYPE (Hydrological Predictions for the Environment) Water Quality Model for Different Spatial Scales. MacMillan, D.C., Marshall, K., 2006. The Delphi process? an expert-based approach to ecological modelling in data-poor environments. Anim. Conserv. 9, 11e19. http://dx.doi.org/10.1111/j.1469-1795.2005.00001.x. Mehdi, B., Lehner, B., Gombault, C., Michaud, A., Beaudin, I., Sottile, M.-F., Blondlot, A., 2015. Simulated impacts of climate change and agricultural land use change on surface water quality with and without adaptation management strategies. Agric. Ecosyst. Environ. 213, 47e60. http://dx.doi.org/10.1016/ j.agee.2015.07.019. €ter, D., 2006. Metzger, M.J., Rounsevell, M.D.A., Acosta-Michlik, L., Leemans, R., Schro The vulnerability of ecosystem services to land use change. Agric. Ecosyst. Environ. 114, 69e85. http://dx.doi.org/10.1016/j.agee.2005.11.025. MLURI, 1993. The Land Cover of Scotland: Final Report (Aberdeen). Morton, D., Rowland, C., Wood, C., Meek, L., Marston, C., Smith, G., Wadsworth, R., Simpson, I.C., 2011. Final Report for LCM2007 e the New UK Land Cover Map (CS Technical Report 11/07.

National Records of Scotland, 2015. Scotland's Population: the Registrar General's Annual Review of Demographic Trends (2014). O'Hagan, A., 2012. Probabilistic uncertainty specification: overview, elaboration techniques and their application to a mechanistic model of carbon flux. Environ. Model. Softw. 36, 35e48. http://dx.doi.org/10.1016/j.envsoft.2011.03.003. Page, T., Heathwaite, A.L., Thompson, L.J., Pope, L., Willows, R., 2012. Eliciting fuzzy distributions from experts for ranking conceptual risk model components. Environ. Model. Softw. 36, 19e34. http://dx.doi.org/10.1016/ j.envsoft.2011.03.001. Prudhomme, C., Haxton, T., Crooks, S., Jackson, C., Barkwith, A., Williamson, J., Kelvin, J., Mackay, J., Wang, L., Young, A., Watts, G., 2012. Future Flows Hydrology: an ensemble of daily river flow and monthly groundwater levels for use for climate change impact assessment across Great Britain. Earth Syst. Sci. Data Discuss. 5, 1159e1178. http://dx.doi.org/10.5194/essdd-5-1159-2012. Rose, N.L., Yang, H., Turner, S.D., Simpson, G.L., 2012. An assessment of the mechanisms for the transfer of lead and mercury from atmospherically contaminated organic soils to lake sediments with particular reference to Scotland, UK. Geochim. Cosmochim. Acta 82, 113e135. http://dx.doi.org/10.1016/ j.gca.2010.12.026. Rowan, J.S., Greig, S.J., Armstrong, C.T., Smith, D.C., Tierney, D., 2012. Development of a classification and decision-support tool for assessing lake hydromorphology. Environ. Model. Softw. 36, 86e98. http://dx.doi.org/10.1016/ j.envsoft.2011.09.006. Sahin, V., Hall, M.J., 1996. The effects of afforestation and deforestation on water yields. J. Hydrol. 178, 293e309. http://dx.doi.org/10.1016/0022-1694(95)028250. Sample, J.E., Duncan, N., Ferguson, M., Cooksley, S., 2015. Scotland's hydropower: current capacity, future potential and the possible impacts of climate change. Renew. Sustain. Energy Rev. 52, 111e122. http://dx.doi.org/10.1016/ j.rser.2015.07.071. Scholten, L., Scheidegger, A., Reichert, P., Maurer, M., 2013. Combining expert knowledge and local data for improved service life modeling of water supply networks. Environ. Model. Softw. 42, 1e16. http://dx.doi.org/10.1016/ j.envsoft.2012.11.013. €ter, D., Cramer, W., Leemans, R., Prentice, I.C., Araújo, M.B., Arnell, N.W., Schro Bondeau, A., Bugmann, H., Carter, T.R., Gracia, C.A., de la Vega-Leinert, A.C., €€ Erhard, M., Ewert, F., Glendining, M., House, J.I., Kankaanpa a, S., Klein, R.J.T., Lavorel, S., Lindner, M., Metzger, M.J., Meyer, J., Mitchell, T.D., Reginster, I., , S., Sitch, S., Smith, B., Smith, J., Smith, P., Sykes, M.T., Rounsevell, M., Sabate Thonicke, K., Thuiller, W., Tuck, G., Zaehle, S., Zierl, B., 2005. Ecosystem service supply and vulnerability to global change in Europe. Science 310, 1333e1337. http://dx.doi.org/10.1126/science.1115233. SEPA, 2007. Significant Water Management Issues in the Scotland River Basin District. SEPA, 2014a. Current Condition and Challenges for the Future: Scotland River Basin District. SEPA, 2014b. SEPA River Basin Management Planning: Benefits of the Water Environment [WWW Document]. URL. http://www.environment.scotland.gov.uk/ get-interactive/discover-data/. Smith, L.A., Petersen, A.C., 2014. Variations on reliability: connecting climate predictions to climate policy. In: Boumans, M., Hon, G., Petersen, A.C. (Eds.), Error and Uncertainty in Scientific Practice. Pickering and Chatto, pp. 137e156. Towers, W., Sing, L., Ray, D., Crow, P., 2011. Woodland expansion GIS Project: Final Report. Woodland Expansion Advisory Group. Towers, W., Dunn, S.M., Dawson, J.C.C., Sample, J., 2012. The Potential Risks to Water Quality from Diffuse Pollution Driven by Future Land Use and Climate Change. Centre of Expertise for Water. http://www.crew.ac.uk/sites/www.crew.ac.uk/ files/publications/Full report_climate change and water.pdf. Uusitalo, L., Lehikoinen, A., Helle, I., Myrberg, K., 2015. An overview of methods to evaluate uncertainty of deterministic models in decision support. Environ. Model. Softw. 63, 24e31. http://dx.doi.org/10.1016/j.envsoft.2014.09.017. Voinov, A., Kolagani, N., McCall, M.K., Glynn, P.D., Kragt, M.E., Ostermann, F.O., Pierce, S.A., Ramu, P., 2016. Modelling with stakeholders e next generation. Environ. Model. Softw. 77, 196e220. http://dx.doi.org/10.1016/ j.envsoft.2015.11.016. €ller, M., Wurbs, D., 2010. A pragmatic approach for soil erosion risk Volk, M., Mo assessment within policy hierarchies. Land use policy 27, 997e1009. http:// dx.doi.org/10.1016/j.landusepol.2009.12.011. Vrana, I., Vaní cek, J., Kov ar, P., Bro zek, J., Aly, S., 2012. A group agreement-based approach for decision making in environmental issues. Environ. Model. Softw. 36, 99e110. http://dx.doi.org/10.1016/j.envsoft.2011.12.007. Weatherhead, E.K., Howden, N.J.K., 2009. The relationship between land use and surface water resources in the UK. Land use policy 26, S243eS250. http:// dx.doi.org/10.1016/j.landusepol.2009.08.007. Wieland, R., Gutzler, C., 2014. Environmental impact assessment based on dynamic fuzzy simulation. Environ. Model. Softw. 55, 235e241. http://dx.doi.org/ 10.1016/j.envsoft.2014.02.001. Wilby, R.L., Whitehead, P.G., Wade, A.J., Butterfield, D., Davis, R.J., Watts, G., 2006. Integrated modelling of climate change impacts on water resources and quality in a lowland catchment: river Kennet, UK. J. Hydrol. 330, 204e220. http:// dx.doi.org/10.1016/j.jhydrol.2006.04.033.