FORESCENE_D.5 by Mathieu Saurat

" # $

' %

(

),!& %

$ &0

-1- 1

! '

)

& $

! $

(% )

$ /

( 4

#$ %

" &

' ((

(()*

+ , !.

5 $ $ $

& ! ! % / )

6 (

$() & &

%$ ! & % 6 %$ ! & % 6 %$ ! &

7 7 7

( )

2-. 8 + & $ $

( )

2-. 8 + & $ $

Stefan Bringezu, Mathieu Saurat Roy Haines-Young, Alison Rollett Mats Svensson Wuppertal Institute for Climate, Environment and Energy University of Nottingham, Centre for Environmental Management Lund University, Centre for Sustainability Studies

( )

2-. 8 + & $ $

( )

2-. 8 + & $ $

Table of Content 1

INTRODUCTION

METHODOLOGY AND MODEL STRUCTURE

8 12

2.1

Bayesian modelling

2.2

Dealing with time

2.3

Dealing with uncertainties

2.4

General structure and system boundaries

2.5

Integration of the model components, linkages between the environmental topics

2.6

Data Base

2.7

Methodology for forecasting

2.8

Methodology for backcasting

ALTERNATIVE SCENARIO NARRATIVES

MODELLING EXERCISE

4.1

Example of pathway and milestone setting

4.2

Example of backcasting

4.3

Integrated alternative modelling results

DISCUSSION AND CONCLUSIONS

REFERENCES

APPENDIX

7.1

Conditional Probability Tables (CPTs)

( )

2-. 8 + & $ $

List of Figures Figure 1: Socio-industrial metabolism and Driving Response (DPSIR) framework in FORESCENE

Forces-Pressures-State-Impacts10

Figure 2: The six questions of the FORESCENE framework

Figure 3: First option for dealing with time in Bayesian networks

Figure 4: Second option for dealing with time in Bayesian networks.

Figure 5: General overview of the structure of the FORESCENE model prototype.

Figure 6: Graphical representation of the Bayesian network for the FORESCENE meta model 17 Figure 7: Forecasting method in the FORESCENE Bayesian network model prototype

Figure 8: Backcasting method in the FORESCENE Bayesian network model prototype

Figure 9: Structure of embedded sustainability strategies for the derivation of alternative sustainability scenarios 26 Figure 10: Two examples of possible pathways for the environmental indicator TMR towards the sustainability target. 33 Figure 11: Response of output variable 'TMR minerals’ to varying domestic demand for services. 36 Figure 12: Response of ouput variable 'TMR minerals’ to varying shares of services in exports. 37 Figure 13: Response of output variable 'TMR minerals’ to varying material intensity of manufacturing. 37 Figure 14: Response of ouput variable 'TMR minerals’ to two example combinations of strategies. 38 Figure 15: Alternative modelling results for TMR minerals (AS1 scenario)

Figure 16: Alternative modelling results for the foreign TMR minerals (AS1 scenario)

Figure 17: Alternative modelling results for TMR fossil fuels (AS1 scenario)

Figure 18: Alternative modelling results for the foreign TMR fossil fuels (AS1 scenario)

Figure 19: Alternative modelling results for GHG emissions from fossil fuels (AS1 scenario) 45 Figure 20: Alternative modelling results for land use change outside EU related to biofuel use in EU (AS1 scenario) 45

( )

2-. 8 + & $ $

List of Tables Table 1: Data base of the FORESCENE model prototype.....................................................20 Table 2: Alternative Sustainability Scenarios .........................................................................29 Table 3: Logistic progression of TMR as a possible pathway ................................................32 Table 4: Overview of the targets used in the alternative modelling exercise .........................39

( )

2-. 8 + & $ $

1 Introduction Since the Gothenburg summit in 2001, the implementation of the concept of sustainable development has been a core challenge for policy making in the European Union. To improve the basis for policy design, and also comply with the specific needs of ex-ante impact assessments, there is a need for a forecasting framework to develop harmonised middle and long-term baseline and alternative policy scenarios. In the context of the Sustainable Development Strategy (European Commission 2001; Council of European Union, 2006) the forecasting framework should allow to develop scenarios that can be used for strategic policy preparation to better specify and disentangle the mutual relationships between environmental, economic and social trends. To be effective, policy development and appraisal needs to understand the key driving forces and their cross-cutting linkages, which lead to increased pressure on different aspects of the environment. Measures which are designed to solve single problems often risk shifting the burden to other sectors, and they may be ineffective due to the complex interaction of environmental effects. The need for policies to be based on a cross-cutting approach has been highlighted in “The 2005 Review of the EU Sustainable Development Strategy: Initial Stocktaking and future orientations” (COM(2005)37 final), as well as in the Commission’s Communication on its Strategic Objectives 2005-2009 (COM (2005) 12). However, crosscutting driving forces which are relevant for various environmental and sustainability related problems have not yet been analyzed in a systematic policy-oriented manner. The FORESCENE project is an attempt to identify cross-cutting measures that are effective and efficient, and which have the potential to mitigate several environmental problems at the same time. As shown in Figure 1, three broad environmental themes are covered: ‘resource use and waste generation’, ‘water’, and ‘biodiversity, soils and landscape’. The aim of the project has been to capture the anthropogenic cross-cutting drivers that generate environmental pressures and the inter-linkages between them, so that an integrated forecasting framework could be constructed. It should, in turn, offer the possibility to test alternative policy strategies against some baseline. To do so, the framework used in FORESCENE was inspired and adapted from the backcasting methodology. As shown in Figure 2, the project’s framework consists of six questions that were addressed during a series of workshops. By applying the framework the project aimed to: • determine cross-cutting (cross-thematic and cross-sectoral) driving forces for environmental problems in the three targeted fields (relates to Question 1); • define essential elements of sustainable development in the different topic areas, particularly in the form of desired sustainability goals (relates to Question 2); • describe cross-sectoral measures expected to exert a multi-beneficial impact over the environmental fields considered, so that the sustainability goals identified can be achieved (relates to Question 3); and, • combine the scenario elements (driving forces, strategies, goals etc) into baseline and alternative scenarios (relates to Questions 4 to 6).

( )

2-. 8 + & $ $

The present report describes the approach taken to address Question 6 “Which alternative scenarios are possible?” shown in Figure 2 while considering strategies which either represent progress towards sustainability, or in contrast may even be worse than a “business-as-usual” development. As a result of Step 4 a model prototype based on the Bayesian network methodology was developed. It consists of interlinked sub-models featuring the environmental issues and cross-cutting drivers identified in Step 1. The model also contains target parameters reflecting the goals, and input variables which operationalize key strategies to approach them, as defined by Step 2. The model prototype can then be used for forecasting or backcasting exercises. The structure of the model prototype, the methodological approach to Bayesian modelling, forecasting and backcasting, and related issues are described in chapter 2. The sustainability strategies defined in Step 2 are used as a basis for the formulation of three alternative scenario narratives which are presented in chapter 3. The model prototype offers the possibility to quantify parts of the narratives using forecasting and backcasting techniques. Chapter 4 exemplifies the different approaches and compares alternative modelling results to the modelled trends of the ‘baseline’ defined in Step 5 of the framework. Based on the ‘prototype’ modelling results, chapter 5 finally discusses the possibilities and perspectives to further develop the framework and the model.

( )

2-. 8 + & $ $

Figure 1: Socio-industrial metabolism and Driving Forces-Pressures-State-ImpactsResponse (DPSIR) framework in FORESCENE

( )

2-. 8 + & $ $

Figure 2: The six questions of the FORESCENE framework

( )

2-. 8 + & $ $

2 Methodology and model structure 2.1 Bayesian modelling Bayesian networks (BNs) are an increasingly popular method applied to uncertain and complex domains such as environmental modelling and management. They can be used as or in combination with decision tools. BN models offer the possibility to incorporate knowledge of different accuracies (e.g. absence/presence of an observation vs. quantitative data from measurements) and from different sources (Marcot et al. 2001). Carefully elicited, expert knowledge can be combined with empirical data, in a mathematically coherent manner. Parameter values that come with uncertainties can be expressed as probability distributions rather than average values. The higher the uncertainty, the wider is the probability distribution. The end-point modelling results are also represented as probability distributions which prevents overconfidence in single estimated values and allows for an estimation of risks and uncertainties (Uusitalo 2007). The Bayesian network approach to modelling presents, however, three main shortcomings (Borsuk et al. 2004, Uusitalo 2007). First, BNs are acyclic graphs and therefore do not support feedback loops (Jensen 2001). Second, and closely related shortcoming, is the difficulty of modelling temporal dynamics in BNs. A possible workaround consists in using a separate network for each time slice (Jensen 2001, Uusitalo 2007). This is however often very tedious. The third difficulty associated with BN models is their limited ability to deal with continuous data when used in compiled form. In such a form, the continuous variables and parametric equations between variables need to be discretized over discrete domains chosen by the user. This implies a trade-off as the discretization can only account for rough characteristics of the original continuous distributions and relationships (Friedman and Goldszmidt 1996). The problems inherent to the discretization of continuous variables can be avoided by using the other major approach in Bayesian modelling beside Bayesian networks, namely hierarchical simulation-based modelling (Uusitalo 2007, Gelman et al. 1995). In the FORESCENE model prototype both branches of Bayesian modelling are implemented. The network approach is used, for example, with the 'biodiversity, soils and landscape' submodel that mainly consists of discrete variables. Once the submodel is compiled the BN can provide instant responses to queries such as the modification of an input parameter. The result is also directly visible in the graphical representation of the network. The simulation-based approach is used for other modules of the model prototype that contain mainly or exclusively continuous variables. The probability distributions of target parameters are estimated by generating samples from these distributions by simulation (Gelman et al. 1995, Borsuk et al. 2003). The results, however, require some processing before they can be shown. But, on the other hand, the least possible information is lost.

2.2 Dealing with time It is important to note that the time dimension is not explicitly present in a Bayesian network model. Using a collection of networks is an option to represent an evolution over time. Figure 12

( )

2-. 8 + & $ $

3 shows how time dynamics is dealt with in FORESCENE, for some parts of the model prototype (e.g. biodiversity and soils submodel). The networks of the time series differ from one another by the input values of their marginal nodes, which correspond to initial input values for given years. These inputs are calculated separately and for different scenarios. The target nodes from each network then deliver output values for each year considered, which can, in turn, be worked out into time series outside the networks. Another option consists in having the different years in one overall network (see Figure 4). This option is adapted for networks used with simulation based modelling (see previous section). It also facilitates the updating of input node values, for example for different scenarios, as there is only one file to update. However, this option is to be reserved to networks that are not too large and ramified.

Figure 3: First option for dealing with time in Bayesian networks

( )

2-. 8 + & $ $

Figure 4: Second option for dealing with time in Bayesian networks.

Note: Node types are as follows: input and control nodes (turquoise and orange), intermediary nodes (pink), output/target nodes (yellow)

2.3 Dealing with uncertainties Building a model and operating it is confronted to two main types of uncertainties: uncertainty in the causal structure and parameter uncertainty. In the present model prototype, as in most models, the first type of uncertainty is not quantified. Therefore, the real uncertainty in model predictions will be greater than that suggested by the model. In the case of Bayesian networks, the following options have been suggested for addressing uncertainty in model structure: Bayesian model averaging, learning from additional data and model testing (Borsuk et al. 2004). Considering the third option, we have compared the modelling results from the model prototype with existing data. This comparison, however, does not provide a true validation, as much of the model building was based on the same or related research that generated the data. However, in contrast to most modelling approaches, parameter uncertainty can be accounted for at a very early stage in the modelling process, rather than at the end using sensitivity analysis. When setting up input and control nodes, parameter uncertainty can be captured by probability distributions. The modelling results from the output nodes can therefore also be expressed as probability distributions that reflect the combined uncertainties from all input and control parameters, and hence uncertainty in model predictions. This probabilistic approach implies that even with a fixed criterion regarding the desired value of a given output parameter, the choice of action on the control parameters depends on the degree of confidence required by decision makers. That degree of confidence can be translated directly in terms of percentiles of the probability distributions of target parameters. In most modelling approaches, such a margin of safety in obtaining the desired outcome would be sought through conservative model assumptions. Though such assumptions can be justified, the modeller implicitly chooses a particular level of confidence in making them. Such a task, however, is a risk management decision that should better be made by decision makers. Therefore a model approach that accounts for prediction uncertainties provides an explicit basis for choosing a decision criterion that includes a margin of safety. The size of 14

( )

2-. 8 + & $ $

the margin, decided by decision makers, might be reduced by the decision makers themselves if they settle for a lower degree of confidence, or by reducing prediction uncertainty. The latter requires further data collection to reduce the uncertainty of input parameters and of relationships between parameters. This, in turn, requires further communication with stakeholders, experts and, eventually, decision makers. In that respect, the explicit dealing with uncertainty of the Bayesian approach fits well in a heuristic approach. 2.4 General structure and system boundaries The general structure of the model prototype developed in the FORESCENE project is shown in Figure 5. The geographical system boundaries are that of the EU-25 with regard to the driving forces but the impacts on most of the environmental issues are considered at the global level.

Figure 5: General overview of the structure of the FORESCENE model prototype.

Note: The components of the model considered beyond the EU-25 system borders are represented with shadows.

The central part of the figure represents the organising structure of the model, consisting of separately developed modules. In the graphical representation of a Bayesian network, variables are depicted by nodes, and a dependence between one variable and another is represented by an arrow. The absence of an arrow between any two nodes implies the conditional independence. Nodes can be either discrete (i.e. with a defined finite set of possible values called states) or continuous (i.e. can take in a value between any other two values). The ‘sub-models’ are the component parts of the model prototype, covering one of the three environmental topics, and ‘modules’ are component parts within a sub-model. The following sub-models and modules are shown in Figure 6: 15

•

Economy

•

Mineral materials

•

Fossil fuels

•

Biofuels

•

GHG emissions

•

Agricultural land use

•

Water

•

Biodiversity and soils

( )

2-. 8 + & $ $

Some modules have been or could be developed in more complex versions, in order, for example, to further differentiate and parametrise mineral, fossil fuel, greenhouse gas and crop types. This potential for refinement illustrates the strength of the Bayesian approach which allows for gradual elaboration of individual elements. The biofuel module actually also includes a simplified transport module. The biomass and agricultural land use is not fully functional at this development stage of the model prototype. Solely land use for biofuel crop production within and outside the EU is calculated from within the BN module at that stage. Results from existing land use models are used in an aggregated form outside the BNs to ensure consistency of the input data sets and of the results stemming from the functioning part of the module. Any Bayesian network can probably be best explained by starting with their end-points and proceeding in the “up-arrow” direction. When building the diagram, intermediate variables and relationships are only included if they contribute to the model’s ability to predict values of the end-point indicators, that is if they are controllable (e.g. human dependent variables), predictable or observable at the scale of the modelling exercise. If a variable does not have one of these characteristics, then it is not explicitly included, and the resulting variability becomes part of the uncertainty associated with the model (Borsuk 2004). For some modules (e.g. economy) the end-point or intermediary nodes deliver interim results that are used in the other modules. The end-point indicators of most modules (e.g. TMR minerals for the minerals module) are proxies for the assessment of the progress made towards the sustainability goals. In developing the model structure, one logically starts with the end-point nodes, which are then related to their immediate causal variables, which are then related back to their own causes, and so on, back to the drivers. The parentless nodes at the top of the diagram can either be considered marginal variables representing natural variability, or those that will be influenced by sustainability strategies. Causal relationships between the nodes are characterized either with functional equations or conditional probability tables (CPTs, see Appendix). The latter are the contingency tables of conditional probabilities stored at each node, containing the probabilities of the node given each configuration of parent values. The first option is, when possible, to be preferred because the specification of functional equations between variables reduces the variability of the results. Probabilities are introduced into the functional equations by assigning probability distributions to the equation parameters, which represent knowledge uncertainty about the parameter value from a Bayesian perspective (see section ‘Dealing with uncertainty’). 16

( )

2-. 8 + & $ $

Figure 6: Graphical representation of the Bayesian network for the FORESCENE meta model Central figure: submodels are indicated with rectangular blocks, model parameters with round nodes, and causal relationships are indicated with arrows. Surrounding insets: networks representing submodels of the main network. (Note: network images used in this figure do not correspond to their latest versions).

( )

2-. 8 + & $ $

2.5 Integration of the model components, linkages between the environmental topics FORESCENE aimed to focus on the interdependencies between the three environmental topics ‘resource use and waste’, ‘water’, and ‘biodiversity and soils’. At the present stage of development not all interdependencies considered are explicitly modelled in the Bayesian networks. Some linkages, such as those between agricultural land use and biodiversity, are accounted for but are partly exogenously determined. For example, because the biomass and land use module is not fully functional at the present point (it is limited to biofuels), the inputs ‘agricultural land use in EU’ or ‘intensity of agriculture in EU’ are determined exogenously based on results from existing models and studies and injected into the model, after ensuring consistency with results from the other BN modules. The in-built modular structure of the model prototype implies that attention needs to be paid to the linkages between the modules, also within a sub-model (corresponding to one environmental topic). The input nodes of one given module consist of its own marginal nodes and of output or intermediary nodes from another module. Often, the purpose of a marginal node in one module is to operationalize a linkage between this module and a ‘parent’ module that influences it. For example, water intensity coefficients (marginal/input nodes of the water module) operationalize the linkages with the mineral materials and fossil energy modules by associating water abstraction to material use. Therefore, input data sets generated exogenously (such as water coefficients) and fed into the marginal nodes need to be consistent across all modules as well as with regard to modelling results coming as input data from other modules.

2.6 Data Base Data mining and literature assessment, conducted in order to ‘populate’ the model prototype, served the following purposes: provide a basis of empirical data and expert statements from which relationships between nodes could be characterised and quantified (whether parameters in functional equations, or probabilities in CPTs); account for parameter uncertainty with probability distributions when different studies publish diverging data; compile time series of input data to be fed into the marginal nodes of the model prototype. Assumptions and modelling results from existing business-as-usual forecasts were considered in FORESCENE to build input data sets for the baseline scenario. Knowledge from different sources could be integrated into one data set while acknowledging their diversity through the use of probability distributions rather than point average values. To each module in the model corresponds a spreadsheet containing time series (in five year time slices) for the input nodes. Spreadsheets are linked in a way to ensure consistency of input data across the modules (see also section 3.7 ‘Methodology for forecasting’ below). An overview of the data base of the FORESCENE model prototype is shown in Table 1.

( )

2-. 8 + & $ $

Table 1: Data base of the FORESCENE model prototype Module

Data sources

Economy

Eurostat, GINFORS model, EEA

Mineral materials

WI data base, ETC/RWM NAMEA, MOSUS

Fossil fuels

WI data base, Eurostat, EEA, DG TREN

Biofuels

WI data base, Eurostat, EEA, DG-ENER

GHG emissions

ETC/AAC, IPCC, UNEP/IPSRM, DG TREN

Agricultural land use

FAO, FAPRI, EEA, UNEP/IPSRM, Eurostat

Water

Aquastat, Eurostat

Biodiversity and soils

EEA, Eurostat, FAO, DG AGRI, and diverse specialised literature sources

2.7 Methodology for forecasting This section explains the forecasting technique used in FORESCENE. Figure 7 gives an overview of the different steps involved. The upper, middle and lower parts of the figure summarize the process into three phases: Construction of the network, parameterization and characterization of the linkages Construction of input data sets for each module Model runs, representation and analysis of the results. The upper and lower parts of the figure are further divided into two boxes. On the left-hand side the technical description concerns network modelling, while on the right-hand side simulation-based modelling is described. The former option is reserved to BNs with discrete variables (e.g. ‘Agricultural management’ is ‘green’ or ‘brown’) or continuous variables that can be discretized without increasing the variability of the results (e.g. ‘pollutant load’ with known thresholds). The latter method regards BNs with continuous variables linked by functional equations, which would be much less accurate if discretized. Both approaches start with a graphical representation of the model structure (upper part of Figure 7). In FORESCENE, a commercial software called Netica1 was used. Parameters deemed necessary to be considered can stem from participatory processes as in the Steps 1 and 2 of the FORESCENE framework. The quantitative relationships characterizing the conditional distributions between the model components need then to be established. The degree of belief and uncertainty underlying the functional relationships between the variables is acknowledged through probabilities in CPTs or equations with parameters expressed as probability distributions.

www.norsys.com

( )

2-. 8 + & $ $

Once the linkages are characterized, the marginal nodes (without parent nodes) are updated with their probability distributions (discrete or continuous). Necessary information are read for each time slice, for each marginal node of each module by a vba-macro from a common spreadsheet and written into the text-file (i.e. non graphic) version of the BN. When the model is “populated”, one can re-open its graphical version and run the forecast. With discrete nodes and a compiled network (lower part, left-hand side of Figure 7) the modelled probability distributions of target nodes over their discrete states (e.g. ‘Biodiversity status’ is declining, stable or improving) can be read directly in Netica. Each time slice is modelled in a separate BN. If an input node is modified, the influence of the change immediately appears. For simulation-based modelling, sample cases should be generated (Netica has a Monte Carlo-like function for it) in a sufficient number to cover the extent of the probability distributions of the target nodes. Then, using the open source softwares for numeric calculation Scilab2 or for statistics R3, these distributions can be visualised and their characteristics calculated. With the simulation-based Bayesian modelling, results are more than single point values: the predictive precision, given model uncertainties, of policy relevant variables is quantified.

www.scilab.org

www.r-project.org

( )

2-. 8 + & $ $

Figure 7: Forecasting method in the FORESCENE Bayesian network model prototype

( )

2-. 8 + & $ $

2.8 Methodology for backcasting This section presents the backcasting technique used in FORESCENE. Figure 12 gives an overview, summarized in three steps (upper, middle and lower parts of the figure), as follows: Construction of the network, parameterization and characterization of the linkages (same as in the forecasting case) Defining quantitative targets for each time step Model runs, representation and exploitation of the results. As in the previous section, the lower part of the figure is divided into two boxes: Bayesian network modelling on the left-hand side, and simulation-based modelling on the right-hand side. The first step (establishing the BN) is identical to the forecasting process, and the same BN can actually be used for both. Then, quantitative desired goals and milestones (i.e. intermediate goals for some time steps up to the horizon considered) used in the scenario narratives are assigned to the corresponding target nodes at the corresponding time slices (middle part of Figure 12, left). A path (linear, S-shaped, other) is defined from today’s values for the target nodes being considered to the values set as ‘desirable goals’, via the intermediate milestones (middle part of Figure 12, right). This way, each target node of interest has a goal value set for it at each time step. Backcasting with Bayesian network modelling (i.e. for compiled BNs of discrete or discretized variables) consists in switching target nodes onto the state corresponding to the desired goal (e.g. ‘Biodiversity status’ set to ‘favourable’). The program then updates the whole BN using Bayesian inference, which concretely consists in re-calculating the probability distributions of all nodes (backwards up to the input nodes), after a particular desired state was imposed to a target node. One can then see on the network the new “backcasted” configuration of input nodes. In the illustration of Figure 12 (lower part, left), the discrete probability distribution of the input node “Agrienvironmental support” changed when the status of the output node “Status of terrestrial biodiversity” was forced to ‘favourable’. The probability than more than half of the rural development budget has to be spent on agri-environmental support increased by about 15% up to almost 80%, strongly suggesting the necessity to increase the budgetary effort in that area. So as to know exactly which share of the rural development budget should be spent on agrienvironmental support, it is a question that is not accessible to the level of detail of the model prototype in its current form. In the case of simulation-based modelling, sample cases (see previous section) of the target nodes are generated at all time steps with different sets of input values for the input nodes representing key strategic drivers that can be influenced in some way. Then, using a graphical or analytical method, suitable sets of input nodes’ values are determined that drive the values of the target nodes along the chosen path towards the desired goals. The graphical backcasting method in simulation-based modelling is shown in Figure 12 (lower part, right). Drawing, firstly, a horizontal line at the desired goal value for the target parameter (represented on the y axis) and observing where it intersects the curve of predictions, and, secondly, drawing a vertical line from this intersection to the horizontal axis, suggests the suitable value for the input (driving) parameters (represented on the x axis). However, there are in fact several curves of 23

( )

2-. 8 + & $ $

predictions (percentile curves) and the choice of the intersection of the line drawn horizontaly with one or the other percentile curve depends on the degree of confidence one wishes for the desired goal to be met, given the uncertainty in model predictions. If a higher degree of confidence is required, then the intersection with a curve of higher percentile must be used to determine the necessary value for the driving input parameter (Borsuk et al. 2003). In a traditional model using only median predictions (or, equivalently, model predictions that do not account for uncertainty), modelling results would appear as point values, not probability distributions. In such a case the modelled curve would in fact be the 50% percentile curve and there would de facto be 50% confidence that the desired goal will be met if the input parameter has the value shown by the graphical backcasting method. A positive aspect in the simulation-based Bayesian modelling approach is, therefore, that the model assesses the uncertainty associated with the predictions. It provides an explicit basis for choosing sustainability strategies (i.e. modified input parameters compared to baseline) which include a margin of safety. The acceptable size of the margin depends on the risk tolerance of decision makers who have to make a choice based on the predictive uncertainty in the Bayesian model (Borsuk et al. 2003). Finally, this backcasting method is probably well suited for a heuristic approach, involving consultations with decision makers and stakeholders. The â&#x20AC;&#x153;backcastedâ&#x20AC;? sets of inputs will certainly not be the panacea that shows the way towards sustainability. Nevertheless, based on the model results it can be discussed whether such a development is feasible. Considering the whole model, the method should also highlight the potentially negative side-effects of the modelled sustainability path on other parts of the model, indirectly connected to the drivers considered. In case of disagreement, the process can be iterated, alternative targets or pathways can be considered. Some controversial scenario elements set aside in the previous steps of the Framework (e.g. in Questions 2 and 3) could surface again. It is also necessary to include this reflection in the decision process as a whole by considering what modelling or consulting activities are required next to determine who should do what and when.

( )

2-. 8 + & $ $

Figure 8: Backcasting method in the FORESCENE Bayesian network model prototype

( )

2-. 8 + & $ $

3 Alternative scenario narratives The first three questions of the FORESCENE framework have provided elements of scenarios and a system to organise them. As a result, we can define a broad set of sustainability goals and their relationships to the various strategies that might achieve them. General and more specific strategies can be combined into an embedded structure (see Figure 9). This can be used as a basis for the formulation of narratives and for the modelling. The narratives combine different strategies with varying strength within a broader context. They define alternative potential pathways to the future which can be modelled based on the more specific quantifiable parameters, and compared with the ‘baseline’ of FORESCENE. Figure 9 represents the ‘alternative scenario’ structure in terms of a set of hierarchical relationships and strategies. The elements in the diagram have been arranged so that they become increasingly more specific moving downwards (i.e. potentially measurable). The arrangement from left to right reflects the degree of consensus expressed within the groups consulted; those elements to the right of the diagram covered potential developments that were more controversial such encouraging new lifestyles that gave less emphasis to consumption as an element of well-being through, say, as finding satisfaction in alternatives to paid employment. Figure 9: Structure of embedded sustainability strategies for the derivation of alternative sustainability scenarios

There was broad agreement amongst those consulted that most strategies for achieving sustainable development had to include an improvement in the environmental performance of consumption and production systems. All of the elements in Figure 9 below this high level strategy are consistent with this broad aim. Some, such as reduction in GHG and waste emissions are more obviously so, while others are linked more indirectly, such as the localisation of markets. The latter would reflect, for example, changing lifestyles and the development of local food markets and employment systems that reduce the need for transport of people and goods. The framework shown in Figure 9 should be read in terms of there being a family of scenarios nested under the broad theme of improving the efficiency of consumption and production systems. 26

( )

2-. 8 + & $ $

Using the combinations of the lower, more specific elements of the framework shown in Figure 9, a series of alternative scenarios can be ‘backcast’ from the strategies identified. Although those consulted agreed on a similar, general desired future direction, a range of potential ‘futures’ might be imagined on the basis of different combinations of the elements being emphasised or downplayed. Moreover individual, sectoral strategies might be implemented at different speeds, or be achieved with different levels of success by the end of the scenario period. Therefore, in an attempt to cover a range of the possible outcomes, three contrasting, alternative trajectories have been developed (see Table 2). Two of them (‘Commitment to Change’, and ’Muddling Through’) reflect the same broad set of ambitions derived from the workshops, and differ only in terms of the degree to which the more specific desired sustainable outcomes identified in the lower parts of Figure 9 have been attained. The third is a more pessimistic scenario in which most of the objectives have not been realised and performance across most of the areas identified in Figure 9 does not even match the baseline. Table 2 expands each of the three scenarios in terms of the specific elements identified in the lower part of Figure 9. In developing the scenarios the aim has been to move from the general trends identified through to the specification of the states which particular nodes in the FORESCENE model would require for particular time periods. Where possible we have attempted to build the alternative scenarios around the same set of nodes used for the baseline analysis. However, since some of the alternatives include the implementation of new measures (such as a Soil Framework Directive) there are also some structural differences between baseline and alternative scenarios. The first scenario describes what might be termed an ideal outcome. Under the ‘commitment to change’ the foundations needed to achieve more sustainable outcomes are put in place early and to greater effect. It is imagined that there are marked improvements in the efficiencies of using energy, materials and water, and an expansion in renewable energy production. The rate of change is moderate to high and significant reductions have been achieved by 2030, with the rate of investment in R & D being mostly over 3%. The level of greenhouse gas emissions has fallen by 20% by 2020, and the 80% cut by 2050 has been attained. There is reducing pressure from the expansion of urban land and intensive agriculture from around the same time. Improvements in production of second generation fuel crops and other renewable energy technolgies means that pressure from fuel crop expansion is minimised. The area for food crops is adequate to meet needs, and so EU consumption does not lead to pressure at global scales. Regulation measures, including strong cross-compliance have been implemented, and farming and forestry management practices are predominantly green as early as 2020. The second scenario is less promising than the first, in terms of achieving concrete sustainable outcomes. In this case while there are achievements, they are patchy. This scenario therefore represents a future that is characterised essentially as ‘muddling through’. As with the first scenario, improvements in the efficiencies of using energy materials and water are made, and there is expansion in renewable energy production. However, the rate of change is only moderate compared to the first scenario. The rate of investment in R & D is greater than in the first scenario, at 2-3%. However, the patchy nature of the outcome is indicated by greenhouse gas emissions continuing to reduce only slowly. The 2020 target is not met until 2035, and only a 65% reduction is achieved by 2050. There is on-going pressure from the expansion of urban land and intensive agriculture. As with the first scenario strong regulation measures and better management practices are in place. 27

( )

2-. 8 + & $ $

The final scenario is the one which sets out a pessimistic future, characterised as â&#x20AC;&#x2DC;failing to deliverâ&#x20AC;&#x2122;. Thus even by 2050 the transition to sustainability has not really been achieved. Increased consumption of resource intensive goods has meant that there has been less progress towards higher energy and resource efficiency. There is some expansion in renewable energy production, but the rate of change has been slow. The rate of investment in R & D has been no more than 2%. Greenhouse gas emissions have increased and there is still pressure from the expansion of urban land and intensive agriculture. Lack of improvements in efficiency of production of fuel crops means that area for food crops is insufficient and that EU consumption exerts pressure on land at Global scales. Nevertheless on the positive side regulation measures, including strong cross-compliance have been implemented, and farming and forestry management practices are mostly green.

GHS reduction targets for 2020 and 2050 met.

Habitat and WFD goals only achieved by 2025.

Green management practices in forestry and agriculture widespread.

Diversity of agricultural production is high and from 2030, and there is the EU approache s selfsufficiency in most key food and energy products.

Shift from paid employment to alternatives

Strong soil framework directive implementted by 2025.

Localisation of markets

Shift to precaution and prevention, and internalisation of damage costs

Rate of urban expansion and growth in area of intensive agriculture slackens by 2030. Improvements in production of fuel crops means that area for food crops is adequate and that EU consumption does not lead to pressure at Global scales.

Improving status of terrestrial and aquatic semi-natural habitats

No marked expansion of renewable energy sources and reduced pressure on area of land devoted to food cops from biofuels.

Reduced pressure from urban sprawl, intensive agriculture and expansion of transport infrastructure

High rates of investment in R&D (>3%) means that increase in nutrient transfer efficiencies are good so that there has been improvement in water quality.

Improved planning and decision making

Significant improvements in the efficiency of material use and waste minimisation. Water abstraction rates are stable or declining.

Expansion of renewable energy sources

AS1, Commitment to change: Under this scenario there is clear commitment to the goals of sustainability. There are marked improvements in the efficiencies of using energy, materials and water, and there is expansion in renewable energy production. The rate of change is moderate to high and significant improvements have mostly been achieved by 2030. The rate of investment in R& D has been over 3%. Greenhouse gas emissions have reduced significantly by 2030 and there is reducing pressure from the expansion of urban land and intensive agriculture from 2030. Regulation measures, including strong cross-compliance have been implemented, and farming and forestry management practices are predominantly green as early as 2020.

Reduced pollution loads

Alternative scenario

Improving energy, material and water efficiencies

Table 2: Alternative Sustainability Scenarios

Demand for consumer goods has increased at 2000 rates until 2025, but returns to 2000 levels by 2050.

GHS reduction targets for 2020 and 2050 not fully met.

Habitat and WFD goals s only achieved by 2025.

Green management practices in forestry and agriculture widespread.

Intensive agriculture and specialisation is still widespread, but diversity of agriculture increases after 2030, and there is greater EU self sufficiency in food and energy production.

Shift from paid employment to alternatives

Strong soil framework directive implementted by 2025.

Localisation of markets

Rate of urban expansion and growth in area of intensive agriculture slackens by 2030. Expansion of non-food crop area means that EU need for food cops leads to some pressure at Global scales.

! Shift to precaution and prevention, and internalisation of damage costs

Marked expansion in area of land devoted to bio-fuels.

Improved planning and decision making

Moderate rates of investment in R&D means that increase in nutrient transfer efficiencies are good so that there has been some improvement in water quality.

% Improving status of terrestrial and aquatic semi-natural habitats

Reduced pressure from urban sprawl, intensive agriculture and expansion of transport infrastructure

Moderate improvement in the efficiency of material use and waste minimisation. Water abstraction rates are stable.

2-. 8 + & $ $

Expansion of renewable energy sources

AS2, Muddling through: Under this scenario some of the sustainability goals identified are achieved in the medium term, but success is patchy and modest. In contrast to AS1, there have been improvements in the efficiencies of using energy materials and water are made, and there is expansion in renewable energy production. The rate of change is moderate and significant improvements have been achieved by 2030. The rate of investment in R& D has been between 2-3%. However, greenhouse gas emissions have continued to increase slowly and there is still pressure from the expansion of urban land and intensive agriculture. On the positive side regulation measures, including strong cross-compliance have been implemented, and farming and forestry management practices are mostly green.

( )

Reduced pollution loads

Alternative scenario

Improving energy, material and water efficiencies

Demand for consumer goods has continued to grow at 2000 rates, but slackens by 2040.

Sgnificant reductions in GHG emissions not achieved.

Habitat and WFD goals s only achieved by 2050.

Strong soil framework directive only implementted by 2050.

Green management practices in forestry and agriculture widespread.

Intensive agriculture and specialisation is still widespread and there is some tendency to shift problems outside EU in order to remedy detrimental environme ntal effects â&#x20AC;&#x2DC;at homeâ&#x20AC;&#x2122;.

Shift from paid employment to alternatives

Rate of urban expansion and growth in area of intensive agriculture does not slacken. Lack of improvements in efficiency of production of fuel crops means that area for food crops is insufficient and that EU consumption exerts pressure on land at Global scales.

Localisation of markets

Marked expansion in area of land devoted to bio-fuels.

! Shift to precaution and prevention, and internalisation of damage costs

Low rates of investment in R&D means that increase in nutrient transfer efficiencies are modest, meaning that improvements in water quality are slow.

Improved planning and decision making

Reduced pressure from urban sprawl, intensive agriculture and expansion of transport infrastructure

Modest improvement in the efficiency of material use and waste minimisation. Water abstraction is increasing or stable at best compared to 2000 levels.

% Improving status of terrestrial and aquatic semi-natural habitats

Expansion of renewable energy sources

AS3, Failing to deliver: Under this scenario the transition to sustainability has been unsuccessful or weak. Increased consumption of resource intensive goods has meant that there has been less progress towards higher energy and resource efficiency. There is some expansion in renewable energy production, but the rate of change is slow and significant improvements are not achieved until 2050. The rate of investment in R& D has been no more than 2%. Greenhouse gas emissions have increased and there is still pressure from the expansion of urban land and intensive agriculture. On the positive side regulation measures, including strong cross-compliance have been implemented, and farming and forestry management practices are mostly green, but these are token gestures.

2-. 8 + & $ $

Reduced pollution loads

Alternative scenario

( )

Improving energy, material and water efficiencies

Demand for consumer goods has continued to grow at 2000 rates.

4 Modelling exercise 4.1 Example of pathway and milestone setting From the participatory processes undertaken at previous stages of the project, it appears that one criterion deemed necessary for sustainability in the EU is to adjust resource use in the economy towards a more sustainable level. Using total material requirements (TMR) as indicator, the sustainability target has been defined in quantitative terms as an 80% reduction of the TMR of EU25 in 2050 in comparison to its level in 2000. According to the methodological description in section 2.8 and Figure 8, once the sustainability goal for the 2050 time-horizon is defined, the next step consists in breaking it down to intermediary targets, for milestone years. There are theoretically an infinite number of possible pathways from 2000 TMR level down to an 80% reduction in 2050. One could be to ‘do nothing’, for example, until 2040 (i.e. follow the business-as-usual scenario) and then suddenly decide to (or be forced to) invest heavily in ways to reduce TMR to reach the target. In practice, it is unlikely, however, that such an approach would be successful – and so alternative pathways can be considered. Figure 10 shows two examples of how TMR could evolve from its actual level toward the sustainability target in 2050. The dashed curve represents a linear reduction of TMR from 2010 onwards. This simple, but easy to visualise, pathway was implicitely recommended by the experts involved in the different participatory processes. They mentioned a 20% TMR reduction by the year 2020 as a suitable interim target on the way to the sustainability target of 80% reduction. It corresponds to the linear pathway drawn in Figure 10. The other curve (plain line) shows a pattern that may be more realistic. The S-shaped (or logistic) curve is a classic empirical description for technology life cycles (Grübler 1998). It shows how new technologies successively go through embryonic, growth and maturity (or saturation) stages. This representation may be more realistic because reaching the sustainability goal for TMR is expected to require innovations in the production of goods in order to drastically increase material productivity beyond the observed inherent increase. The curve presented in Figure 10 is of the form A / (1 + e-λ(t-t0)) with A the asymptotical value (the sustainability target of 80% reduction) and λ a shape parameter set to 0.3. The inflection point t0 (where half of the way towards the goal is achieved) occurs in the year 2030. Table 3 shows the progression of TMR reduction as expected with a S-shaped pathway. In such a case, half of the reduction effort occurs in only ten years between 2025 and 2035.

Table 3: Logistic progression of TMR as a possible pathway 2000 2010

2015

2020

2025

2030

2035

2040

2045

2050

0.9%

3.8%

14.6%

40%

65.4%

76.2%

79%

80%

However, this representation also shows only very few improvement until 2020. The curve only starts kicking off around 2022-2023. That is, this example S-shaped pathway lacks a clear, ambitious though reachable target for the medium term (around 2020). For this reason, the

( )

2-. 8 + & $ $

somewhat simpler linear pathway may be more appropriate when there is reason to assume that certain measures are implemented which take effect also in medium term reducing the absolute resource use of the economy. Once the form of a probable pathway is agreed upon and quantified, it can be used with the FORESCENE model prototype in backcasting mode. The method is described in section 2.8. The aim is, for each time-slice, to determine a suitable combination of input parameters (driving and control factors) that could allow for the output variable to reach the alternative values (compared to the baseline scenario) set by the chosen pathway. Next section shows an example of backcasting for the year 2050 (i.e. the end-point of the sustainability pathway), using the graphical method presented in Figure 8.

Figure 10: Two examples of possible pathways for the environmental indicator TMR towards the sustainability target.

4.2 Example of backcasting To provide a concrete example, the backcasting method explained in section 2.8 is applied to the combination of the economy and ‘non renewable non energetic materials’ (NRNE, or also ‘mineral materials’) modules of the model prototype. The output variable of interest is ’TMR minerals’ (total material requirements) for the EU-25. The economy module is a parent module of the NRNE module. Final demand, distinguished between domestic demand and exports, is an end-point of the economy module and drives direct material input (DMI) of the European economy. The linkage is 33

( )

2-. 8 + & $ $

operationalised through the share of services vs. that of manufactured goods in both domestic demand and exports associated with the corresponding material intensity of services production vs. that of manufacturing. The input (driving) parameters considered in this backcasting exercise are the share of services in domestic demand and exports, and material productivity in the manufacturing sector. The output variable targeted is TMR minerals. The target is to reach a reduction by 20% in 2020 and 80% in 2050, compared to 2005 level. In this example, the backcasting approach is applied with the model prototype for the year 2050. The Bayesian modelling method involved in this backcasting example is that of simulation-based modelling (see sections 2.1 and 2.8). The sub-model is simulated for different values of the input variable share of services in domestic demand, share of services in exports and productivity improvement in manufacturing (besides the baseline inherent improvement in material productivity), respectively. In each case the other parameters are kept as in the baseline scenario. For each case, sample files are generated and then plotted. The graphical method shown in Figure 8 allows to find combinations of input parameters that would drive the TMR of the EU-25 to the 80% reduction target. Figure 11, Figure 12 and Figure 13 present the modelling results for TMR minerals in function of the variations of the input parameters share of services in domestic demand, share of services in exports and productivity improvement in manufacturing, respectively. For each graph, the first value on the left of the x-axis shows the baseline value of the input parameter for the year 2050. The different curves represent percentiles of the modelled distributions for the output variable TMR minerals. It reflects the overall modelling uncertainty carried through uncertainty on parameters such as GDP growth, coefficients of indirect material flows associated with imports etc (see also section 2.3 â&#x20AC;&#x153;Dealing with uncertaintiesâ&#x20AC;?). Figure 11 shows the predictions for TMR minerals of EU-25 when demand for services equals respectively 65% to 100% (in 5% steps) of domestic demand. The first value (65%) is that of the baseline scenario. As expected, TMR decreases when the share of services increases. Material intensity of services is lower than that of manufactured goods (direct and indirect material intensities are accounted for in both cases). The predictive uncertainty tends to slightly decrease in absolute terms with higher values of demand for services. In relative terms, it remains a similar share of the median value. Similar patterns can be observed when the share of services in exports increase or the material productivity of goods production increases (Figure 12 and Figure 13, respectively). The target set for TMR minerals for the year 2050 is a 80% reduction compared to 2005 level. The purpose of the simulation and the three graphs is to provide the data necessary to look for an adequate combination of the three drivers that would stir TMR towards its target. In 2005, TMR minerals of EU-25 amounted to about 8.5 billion tonnes. A reduction by 80% would mean that, in 2050, TMR should amount to around 1.7 billion tonnes. Compared to the mean value of TMR in that year in the baseline scenario (about 17 billion tonnes), TMR should be reduced by about 15.3 billion tonnes in 2050. For this example calculation, one may assume that the three parameters whose influences are shown in Figure 11, Figure 12 and Figure 13, respectively, contribute approximately equally to the TMR reduction. It means that each should contribute to a reduction of about 5 billion tonnes TMR.

( )

2-. 8 + & $ $

The suitable values for the three parameters can be graphically determined from the plot of modelling results shown in the three figures. For each graph, we draw a horizontal line about 5 billion tonnes below the median value of the baseline scenario (first value plotted on the left). Then, we observe where this line intersects the curve of predictions, and, subsequently, draw a vertical line from this intersection to the horizontal axis. The point where the vertical line then crosses the x-axis suggests the suitable value for each of the input parameters. The graphical backcasting method is represented in Figure 11, Figure 12 and Figure 13 by thicker double arrows. For this example, we have represented the intersection of the horizontal arrows with the median curves (which correspond to the mean curves too because the probability distributions are normal). The results from this example graphical backcasting method are therefore the same that would come out with a model using only mean point values for its parametersâ&#x20AC;&#x2122; values (or, equivalently, a model that does not account for uncertainty). There is here 50% confidence that the criterion will be met (here the criterion was that each parameter achieves independently an absolute reduction of 5 billion tonnes TMR). However, given that in the FORESCENE model prototype the uncertainty in model predictions is quantified, the choice of suitable values for the three parameters can depend on the degree of confidence required by decision makers. If a higher degree of confidence than the 50% examplified in Figure 11, Figure 12 and Figure 13 is required, then the intersection of the horizontal arrow with a curve of a higher percentile (e.g. 75% curve) must be used to determine the suitable value for each of the input parameters. Finally, Figure 14 shows probability distributions simulated for TMR minerals corresponding to two alternative combinations of the three input parameters. The mean value, the first and third quartiles of the probability distribution are reported below the x-axis (in tonnes). The upper graph uses the parameter values graphically determined in Figure 11, Figure 12 and Figure 13 (using intersection with the 50% percentile curve). However, it appears that the effect of the three alternative parameters combined is lower than the sum of the separate effects. The mean TMR value is 3.2 billion tonnes while the target was 1.7 billion tonnes. It is due to the fact that increasing the share of demand for services partly off sets material productivity increases in goods production, because the latter see their share in demand decreasing. Therefore, a second, more ambitious, alternative combination of the three parameters (with a 70% increase in material productivity of goods, while the other two parameters remain as in the first combination) is used as input to model TMR minerals of EU-25 in 2050. This time the first quartile of the distribution equals the target (1.7 billion tonnes), which means that, given the parameter uncertainties, there is about 25% probability that, with this combination of input parameters, TMR minerals will be as high or lower than the 1.7 billion tonnes target in 2050. Considering the values displayed for the fourth quartile of TMR minerals in 2050 in Figure 14 (lower graph), it appears that, given the parameter uncertainties, there is 75% probability that TMR minerals will decrease by at least 75% in 2050 (2.1 billion tonnes for the fourth quartile represent a 75% reduction from 8.5 billion tonnes TMR in 2005). To better grasp what this alternative setting of input parameters means, it can be noted that a 70% increase in manufacturing material productivity, on top of the baseline, represents an overall yearly productivity increase of 3.9%. It means about 2.3 times the baseline yearly increase (1.7%). These results constitute a first example of the possibilities offered by the FORESCENE model prototype when used for backcasting. The procedure described above should then be repeated for the other time slice in order to build the complete path (i.e. time series) of input parameters leading 35

( )

2-. 8 + & $ $

to the sustainability targets. This work could also be coupled with consultations of decision makers, stakeholders and experts in order to improve the plausibility of the assumptions and thus the relevance of the results. The structure of the model itself, as well as the data base used for the parameters can be altered following such consultations, therefore qualifying the process for a heuristic approach.

Figure 11: Response of output variable 'TMR mineralsâ&#x20AC;&#x2122; to varying domestic demand for services.

Note: Lines represent percentiles of predictive distributions, as indicated in legend. Arrows correspond to graphical method of selecting strategies, as described in the text.

( )

2-. 8 + & $ $

Figure 12: Response of ouput variable 'TMR mineralsâ&#x20AC;&#x2122; to varying shares of services in exports.

Figure 13: Response of output variable 'TMR mineralsâ&#x20AC;&#x2122; to varying material intensity of manufacturing.

Note to Figure 12and Figure 13: Lines represent percentiles of predictive distributions, as indicated in legend. Arrows correspond to graphical method of selecting strategies, as described in the text. 37

( )

2-. 8 + & $ $

Figure 14: Response of ouput variable 'TMR mineralsâ&#x20AC;&#x2122; to two example combinations of strategies.

( )

2-. 8 + & $ $

4.3 Integrated alternative modelling results Chapter 3 presented scenario narratives, i.e. plausible stories about a vision of the future and the pathway to reach it. This chapter shows modelling results obtained from using the model prototype to quantify parts of the narratives. The modelling exercise focuses here on the first alternative scenario (AS1) ‘commitment to change’. The goals set by this narrative are the highest among the three alternative scenarios and, as such, a strong development of the input parameters is probably needed to reach them. Table 4 shows the targets that are used below for the modelling exercise. They are attributed to the three submodels, corresponding to the three environmental fields considered. The aim of the exercise is to quantify alternative developments of key input parameters that, when fed into the model prototype, drive the output variables towards the targets. Both forecasting and backcasting modelling techniques described in sections 2.7 and 2.8 were implemented. Forecasting was essentially used in simulation-based modelling as ‘trial and error’. Backcasting was used with the discrete sub-model ‘biodiversity and soils’ following the network modelling approach (see Figure 8). It was also used to some extent in simulation-based modelling, following the example given in the previous section.

Table 4: Overview of the targets used in the alternative modelling exercise Submodels Resource use and waste generation

Targets .

Reduce TMR by 80% in 2050

Reduce foreign TMR

Reduce GHG emissions by 80% in 2050

Net import of land should not increase

Water supply and water abstraction should be balanced Overall biodiversity status: favourable

Terrestrial biodiversity status: favourable

Aquatic biodiversity status: favourable

Soil carbon: high

Soil erosion: low

Soil quality: high

Water Biodiversity, landscape and soils

Some of the targets presented in Table 4 needed to be somewhat modified, as explained below, when it seemed that the corresponding values modelled for the input parameters were not plausible. This adaptative process would certainly fit in a consultation cycle with decision makers and stakeholders, where the model (structure, parameters, data base) could be iteratively improved in order to increase the plausibility and applicability of the modelling results. For this alternative modelling exercise economic drivers were kept equal to the baseline. Essentially, GDP per capita was assumed to grow on average 2% per year until 2030 and 1.5% 39

( )

2-. 8 + & $ $

afterwards. The aim of the modelling was indeed to look for suitable values for other input parameters to drive key ouput variables towards sustainability targets in a growing economy. In the light of the ongoing economic crisis the assumptions regarding economic parameters are probably unrealistic in the short-term. Until 2050 and on average, however, a growing economy is still plausible, although with increasing uncertainty. If it had been assumed that the economy were to shrink, the impact on the output variables would have been a decrease in the absolute consumption of resources and, hence, also in the related environmental impacts. In general, a lower GDP growth enhances the chances for an absolute decoupling from resource use and impacts. However, a detailed econometric module is not part of the model prototype. It is therefore not clear how reduced expenses in R & D and other investments would impact key strategies such as ‘fostering material productivity’. Directly influenced by the economy module are the mineral materials, fossil fuel and biofuel modules. These in turn are ‘parents’ for the greenhouse gas emissions, water use, land use, and biodiversity and soils modules. The input variables considered for quantification in the mineral materials module are material productivity and the share of services in final demand for their direct influence on mineral materials consumption, and indirectly on energy demand and water use. The model suggests that if the yearly material productivity increase in manufacturing were to double (from 1.7% per year in the baseline), associated to a rise of the service economy with as much as 80% of the EU’s final demand consisting of services in 2050 (from 52% in 2005), then the reduction targets for TMR would be in sight. The material efficiency of service production shall also increase but no big jump is expected (contrary to the industrial sector) because the service sector will indirectly require resources (e.g. real estate services relying on maintenance of constructions). Figure 15 shows the development of the TMR minerals of EU-25 under these input conditions. TMR minerals (mean value) could have decreased by 24% in 2020 and 70% in 2050, compared to baseline. This differs somewhat slightly from the targets set in Table 4 (20% and 80% decrease in 2020 and 2050, repectively). Towards the end of the time horizon especially, the expansion of the service sector levers off the productivity improvement of the manufacturing sector, whose share diminishes. The model shows, however, that the original targets could be met with a 2.3 to 2.5 fold (instead of twofold) increase in yearly material productivity growth and a further expansion of the service sector in final demand (see also backcasting exercise in previous section). Though these developments are not excluded, it was decided for the sake of plausibility to revise the 2050 target for TMR minerals and stick to the first modelling results listed above. Figure 16 shows the foreign part of TMR minerals in the alternative scenario AS1. Starting from 2015, it decreases in absolute terms. Shifting environmental burden outside Europe is therefore reduced. The slight growth in the years up to 2015 can be explained by the fact that expected growing imports and indirect material flows associated with them compensate material productivity improvement for that period. In relative terms, however, the ratio foreign vs. domestic TMR still increases. It is due to the fact that, under the alternative assumptions, the domestic part of TMR decreases together with the foreign part.

( )

2-. 8 + & $ $

Figure 15: Alternative modelling results for TMR minerals (AS1 scenario)

Figure 16: Alternative modelling results for the foreign TMR minerals (AS1 scenario)

( )

2-. 8 + & $ $

The input variables considered for quantification in the fossil fuel and biofuel modules are energy productivity, the share of renewables, and specifically the share of biofuels, for their importance in fossil fuel consumption and land use change, particularly outside the EU. Modelling results show that with energy productivity increasing by 2.7% yearly (against 1.6% in the baseline), and the share of renewables increasing from 6.8% in 2005 to about 50% in 2050 (against 12% in 2030 and then constant in the baseline), the part of TMR associated with fossil fuel use in the EU-25 would decrease by 70% in 2050 (Figure 17). As in the case of TMR minerals, the original sustainability target has been revised to keep modelling results for the input parameters plausible. There is, however, still room for further parameterisation of the model prototype (e.g. after consultation with decision makers and stakeholders). An extended plausible set of input parameters could be established that would lead TMR fossil fuels towards the original sustainability targets. Similarly to the modelling results for TMR minerals, the foreign part of TMR fossil fuels is expected to decrease in absolute terms after 2015 (Figure 18). It means that the absolute direct and indirect, used and unused amount of primary resources extracted outside Europe to satisfy its need for fossil fuels decreases. The environmental impact of the EU on the rest of the world is therefore reduced, compared to baseline. Directly correlated to decreased fossil fuel consumption, Figure 19 shows modelling results for greenhouse gas emissions in alternative scenario AS1. With the set of input parameters (energy efficiency, share of renewables) described above, greenhouse gas emissions from fossil fuels would decrease by 30% in 2020 and slightly over 70% in 2050. In accordance with the revisions made for the previous targets, the last 10% needed to reach the desired sustainability goal (80% reduction) are missing in this configuration of input parameters. Additional knowledge and data regarding the development of the fossil fuel mix and the possibilities for the decarbonization or carbon recycling of fossil fuels could help close the gap in the modelling results. The increased share of renewables in the energy mix suggested by the model prototype to mitigate issues of TMR fossil fuels and greenhouse gas emissions need to be accompanied by a strategy aiming at consistently developing the renewable mix. In particular, the expansion of renewables should not come at the cost of land use change and associated greenhouse emissions and biodiversity loss outside the EU. Therefore, Figure 20 shows modelling results from the transport, biofuel and land use module group when biofuel use in road transport is capped at 5% of the transport fuel mix (against up to 9.8% in the baseline). This configuration would avoid about half of the land use change occuring outside Europe in the baseline. It is a conservative value because it is assumed also that at least half of the biofuels will be imported. However, with a lower biofuel and overall fuel demand than in the baseline, increased fuel efficiency and production of second generation biofuels in Europe, it seems plausible that the share of imports can be drastically reduced through the blend cap. Existing studies have demonstrated that such perspectives are not unrealistic (EEA 2006, EEA 2007, UBA 2008). Regarding future work, the agricultural land use module could be further developed into a functional and consistent entity coupling land use for both the production of food and non food biomass, as the focus so far has been on biofuels. As a consequence of reduced material and energy consumption in the alternative scenario, withdrawal of water for industrial and cooling purposes decreases. The number of households is expected to increase (though population will slightly decrease) but improved water efficiency off sets this trend. All in all the water balance (i.e. difference between water supply from precipitations, recycling and desalination and water abstraction) improves at the EU level. This result, however, 42

( )

2-. 8 + & $ $

hides disparities between EU regions in the context of climate change, especially between the north, where higher precipitations are expected, and the south, where more frequent drought episodes are probable. At the present stage of development, the model prototype does not account for geographical disparities with EU. The water module could, however, without important modifications be disaggregated into regional versions, provided that region specific data are available. If data are not directly available, the Bayesian approach would allow the elicitation of expert judgements to be used instead. The network constituting the biodiversity and soils module is rather complex and made of discrete variables or continuous variables that have been discretized. This combination tends to buffer the influence of upstream drivers on the output parameters as has been observed when forecasting the baseline scenario. Reciprocally, when using the backcasting technique as described in Figure 8, the signal sent upstream to the input nodes of the module when setting the target nodes to their desired states seems to get diluted. The target nodes in question are ‘overall biodiversity status’, ‘terrestrial biodiversity status’, ‘aquatic biodiversity status’, ‘soil carbon’ and ‘soil erosion’. Their target status are set to ‘favourable’ for the three biodiversity nodes, ‘high’, ‘low’ and ‘high’ for the three soil nodes, carbon content, erosion and soil quality, respectively. For the reasons mentioned above, more model testing and calibrating would be needed if one were to draw conclusive results from this backcasting exercise. However, input variables such as ‘cross-compliance’, ‘farming management’, ‘intensity of R & D’, and introduction of a ‘soil framework directive’ tend to increase the probability of a “greener” outcome. It suggests that investing efforts into increasing crosscompliance, developping green agricultural and forest management, fostering R & D and implementing a strong protection of soils through a policy instruments such as a ‘soil framework directive’ would help drive biodiversity and soil quality in Europe towards the sustainability goals set for them, and increase the probability of these targets to be met.

( )

2-. 8 + & $ $

Figure 17: Alternative modelling results for TMR fossil fuels (AS1 scenario)

Figure 18: Alternative modelling results for the foreign TMR fossil fuels (AS1 scenario)

( )

2-. 8 + & $ $

Figure 19: Alternative modelling results for GHG emissions from fossil fuels (AS1 scenario)

Figure 20: Alternative modelling results for land use change outside EU related to biofuel use in EU (AS1 scenario)

( )

2-. 8 + & $ $

5 Discussion and conclusions The modelling techniques (forecasting and backcasting in network and simulation-based modelling) presented in this report helped us quantify, in a plausible way, the necessary development of key driving parameters in order to achieve the sustainability targets. Strategies such as enhanced resource productivity in combination with the capping of certain resource use (e.g. biofuel imports and related crop land) were shown as potentially effective to reach a more sustainable future. The more ambitious of the three scenario narratives elaborated in the project (‘commitment to change’) will require the most stringent implementation of those strategies. Taken into the broader perspective of the project’s framework, the modelling results would need to be further discussed with policy makers and stakeholders. On the one hand, the results can provide orientation for decision making. On the other hand, articulation of further demand for decision support will allow to further develop the modelling activities, revise assumptions, further extend the model and combine it with existing models and data bases. The sustainability targets selected for the modelling exercise reach further than it seems at first sight. The ambitious goals set for the reduction of Europe’s total material requirement (TMR) translates not only into reduced use of primary natural resources but also into reduced waste generation within and outside the EU (a consequence of reducing the material throughput of the socio-industrial system). This also reduces landscape disturbances due to mining and associated and subsequent material flows down to final disposal of consumer products. Reducing TMR also implies the mitigation of shifting environmental problems to other regions of the world through imports of raw materials or material intensive products. The targets set for the reduction of greenhouse gas emissions are also expected to contribute to the mitigation of climate change as well as related problems. Issues such as water supply in Southern Europe, or biodiversity will be influenced in a sustainable manner if those ambitious goals are met with strategies that do not induce a shifting of the problem onto other resources or world regions. Therefore, in this case, the targets for greenhouse gas emissions are met under the condition to halt the conversion of natural land into cropland for the purpose of cultivating bio-fuel crops (as the conversion adversely affects biodiversity and the carbon balance of biofuels). Some of the changes assumed to be a precondition for a more sustainable development, such as the strong shift towards a service oriented economy, need to be discussed with regard to their relevance, plausibility and feasibility. An assessment according to those criteria will probably depend on the general frame that one is willing to adopt. If the underlying assumption is that of an economic growth which is still considerably based on manufacturing as today, so that for instance a growing car production for export would still be needed for our economy to grow, then the results would have little chance of looking acceptable. At that point one could look back at the diverse elements collected along the different steps of the FORESCENE Framework and remember that the somewhat fuzzy ideas of ‘changing paradigm’ or ‘redefining wealth and happiness away from material possessions’ came up more than once in the different workshops. One could then imagine that in a system where economic growth is sustained by growing production of education, music, art, cultural services, in all their forms and at all ages, as well as growing provision of elder care, rather than by increased goods production or increased transport, this set of modelling results could eventually make sense. Last but not least, the FORESCENE approach (framework and model) can be applied in different contexts where certain interlinked problems shall be overcome. It can be used to quantify the 46

( )

2-. 8 + & $ $

conditions needed to reach a normative desirable future. It can be further developed towards a heuristic learning tool to support stakeholder and expert involvement in the policy preparation phase. The level of details can be further expanded through combination with specific sectoral models.

( )

2-. 8 + & $ $

6 References Borsuk, M. E., C. A. Stow and K. H. Reckhow. 2003. Integrated approach to total maximum daily load development for Neuse river estuary using Bayesian probability network model (Neu-BERN). Journal of Water Resources Planning and Management 129(4): 271-282. Borsuk, M. E. 2004. Predictive assessment of fish health and fish kills in the Neuse river estuary using elicited expert judgment. Human and Ecological Risk Assessment 10: 415-434. Borsuk, M. E., C. A. Stow and K. H. Reckhow. 2004. A Bayesian network of eutrophication models for synthesis, prediction, and uncertainty analysis. Ecological Modelling 173: 219-239. Cain, J. 2001. Planning improvements in natural resources management – Guidelines for using Bayesian networks to support the planning and management of development programmes in the water sector and beyond. Centre for Ecology & Hydrology: UK. Commission of the European Communities. 2001. A Sustainable Europe for a Better World: A European Union Strategy for Sustainable Development. Brussels, 15.5.2001, COM(2001)264 final. Commission of the European Communities. 2005a. Commission’s Communication on its Strategic Objectives 2005-2009 (COM (2005) 12). Commission of the European Communities. 2005b. The 2005 Review of the EU Sustainable Development Strategy (COM(2005)37 final). Council of the European Union. 2006. Review of the EU Sustainable Development Strategy (EU SDS) − Renewed Strategy. 10907/06 EEA (European Environment Agency). 2006. How much bioenergy can Europe produce without harming the environment? Copenhagen: EEA Report No 7/2006. EEA. 2007. Estimating the environmentally compatible bioenergy potential from agriculture. Copenhagen: EEA Technical Report No 12/2007. Fargione,J., J. Hill, D. Tilman, S. Polasky and P. Hawthorne. 2008. Land clearing and the biofuel carbon debt. Science 319(5867): 1235-1238. Friedman, N. and M. Goldszmidt. 1996. Discretizing continuous attributes while learning Bayesian networks. In: Proceedings of teh 13th International Conference on Machine Learning (ICML). Morgan Kaufmann Publishers, San Francisco, CA, pp. 157-165. Gelman, A., J. B. Carlin, H. S. Stern, and D. B. Rubin. 1995. Bayesian data analysis. In: Texts in Statistical Science. Chapman & Hall. Grübler, A. 1998. Technology and global change. Cambridge: Press Syndicate of the University of Cambridge. Jensen, F. V. 2001. Bayesian networks and decision graphs. Springer-Verlag: New-York. Marcot, B. G., R. S. Holthausen, M. G. Raphael, M. Rowland and M. Wisdom. 2001. Using Bayesian Belief Networks to evaluate fish and wildlife population viability under land management alternatives from an environmental impact assessment. Forest Ecol. Manage. 153, 29-42. UBA (Umwelt Bundesamt). 2008. Nachhaltige Flächennutzung und nachwachsende Rohstoffe. [Sustainable land use and regenerative natural resources.] Berlin: UBA, Wuppertal Institute, IUSE, IFEU. 48

( )

2-. 8 + & $ $

Uusitalo, L. 2007. Advantages and challenges of Bayesian networks in environmental modelling. Ecological Modelling 203: 312-318.

( )

2-. 8 + & $ $

7 Appendix 7.1 Conditional Probability Tables (CPTs) Cain (2001) defines a Bayesian Network as a ‘graphical tool for building decision support systems to help make decisions under uncertain conditions’. The key phrase to focus on in this definition is ‘uncertain conditions’. As Cain points out, BNs were originally developed to allow the impact of uncertainty in management to be accounted for. Using the tools decision makers could balance the desirability of an outcome against the chance that the management option selected might fail. The representation of a system in terms of a set of relationships that have probabilities associated with them is at the heart of the Bayesian approach. An example of a simple BN is shown in Figure 1. If we think about the sorts of things that might influence agricultural yield, for Figure 1: A simple BN (after Cain, 2001) example, then these might include water supply and fertilizer applications. The amount of water applied to the crop might, in turn, be influenced by such factors as soil type and the level of irrigation. Figure 1 shows this diagrammatically. The key variables (that is the things that can change, such as yield, fertilizer applications and soil type etc.) are shown as a set of boxes or nodes. The relationships between the variables are shown as a set of arrows. These arrows simply set out the linkages between the variables; they show what influences what. The arrows describe what we think the causal relationships are within the system; notice that the arrows have a direction to express this idea of causality. In Figure 1 each of the nodes are shown as being able to take various states. Thus yield can be ‘good’ or ‘poor’, or soil type can be ‘sandy’ or ‘clay’. In the diagram, the nodes are represented by a special type of box, called a ‘belief bar’, which we can use to express the probability that the variable (node) is in a particular state, and how this might influence the other nodes to which it is linked. At the moment no real probabilities have been assigned, and all the nodes show a 50:50 chance of being in a particular state. What the BN allows us to do is to assign probabilities to the states for the different nodes to describe how we think the world actually works. The way we do this is to build Conditional Probability Tables (CPT) for each node.

( )

2-. 8 + & $ $

Table 1: CPT for the Crop water Application Node

Table 1 shows the CPT for the node â&#x20AC;&#x2DC;Crop Water Applicationâ&#x20AC;&#x2122; that is part of the network described in Figure 1. Technically CPT is the contingency table of conditional probabilities4 stored at each node, containing the probabilities of the node given each configuration of parent values. Clearly when building a BN we do not have to confine ourselves to variables that have only two states. The software can handle a number of discrete states or even continuous values.

The conditional probability of an event is the probability of the event occurring under certain given conditions.