knowledge elicitation training and dissemination report€¦ · knowledge elicitation tools (knets)...

Deliverable 3.5.8 – Tisza River Basin Case Study

Knowledge Elicitation Training and Dissemination Report

Facilitators: Sukaina Bharwani (SEI Oxford, UK), Dagmar Haase (UFZ, Germany). KnETs software developer: Michael Fischer, University of Kent, UK Report of the NeWater project – New Approaches to Adaptive Water Management under Uncertainty www.newater.info

http://www.newater.info/

This working document serves as base for the case study and work blocks to plan research and stakeholder activities. It is part of the next 18-month implementation plan and will be updated regularly.

Title Knowledge Elicitation Training Workshop Purpose This document describes the purpose, workshop activities and lessons learnt Filename Minutes TfT Tisza March 2008.doc Authors Dagmar Haase (UFZ), Sukaina Bharwani (SEI) Document history Current version Draft version in general estimated as being useful by PICP Contact [email protected]

Date 11.08.2008 Status Draft, version 0.1 Target readership NeWater WB and WP leaders General readership NeWater GA & stakeholders from NeWater CS Correct reference Not to be referred to yet, project internal only

This part of the Deliverable 3.5.8. has been submitted as a regular paper in a Special Issue on Participation for the Journal Ecology & Society.

mailto:[email protected]

Content of the manuscript

Policy summary 4

1. Introduction 6

1.1 Tacit knowledge 7

1.2 Objectives 8

2. The Case Study 8

3. Methods 10

3.1 Designing the game for the Tisza case 10

3.2 Applying KnETs in the field 13

4. Results 14

4.1 Creation of decision-making heuristics 14

4.2 Verification of the decision heuristics and elicitation of tacit knowledge 16

4.3 Validation phase 17

5. Discussion 19

6. Conclusions 20

Acknowledgements 20

References 20

Policy Summary Identifying decision criteria and analysing human behaviour is fundamental for understanding decision-making processes and actions that steer river basin management and water environmental resources. Often we are confronted with multiple reasons for land use decisions of multiple actors that have a stake on one resource (e.g. land, water resources, soil fertility etc.).

Knowledge Elicitation Tools (KnETs) represent a new and reproducible way to formalise this knowledge using computational techniques and to implement scenario techniques within the interviews. KnETs link qualitative and quantitative representations of stakeholder knowledge.

An iterative process that incorporates interviews, a formalisation phase and an empirical data collection phase with respective respondent groups of interest and a resulting decision tree creation and interpretation phase. Finally, another game round is conducted with a non-involved respondents’ group to verify and validate the results of the models and to assess our gained knowledge on decision making.

The behavioural heuristics of decision-making in a given scenario-situation helps to explore the role of local knowledge – knowledge that is voiced and knowledge that is actually used may be different. This may be due to the tacit nature of this knowledge which can also cause problems of communication among the stakeholders themselves.

In the case discussed interventions in flood risk management must take socio-cultural context relevant perceptions into account to understand what drives adaptive and non-adaptive options, changes in behaviour and initiates learning – this can be described as the capacity of stakeholders to adapt. Where gaps in these decision-making structures exist may be exactly where development interventions may be most valuable. In our case, government funding which is specifically targeted toward long-term adaptation planning, such as an early warning system, technological support and improved insurance mechanisms would resonate most strongly with the needs expressed by those whose responsibility it is to prepare communities for flood events.

For application purpose one needs a laptop in the field running the scenario game. The software is free-of-charge available (Java and a Text Editor which enables syntax highlighting). Instead running the game on the computer one could also use cards to replicate the game. According to the local language spoken there will be one person needed who can translate the game.

Eliciting knowledge on soft flood-risk management strategies in the Ukrainian Tisza river basin Svetlana Kuptsova 1, Dagmar Haase2, Sukaina Bharwani3, Michael D. Fischer4, Thomas E. Downing3

1 Zacarpathian Water Board and Management Direction Ushgorod, Ukraine 2 Helmholtz Centre for Environmental Research – UFZ, Department of Computational Landscape Ecology,

Germany 3 Stockholm Environment Institute – Oxford Office, UK. 4 Centre for Social Anthropology & Computing (CSAC), University of Kent, UK. Abstract This paper focuses on a participatory knowledge elicitation process (KnETs) to explore decision-making criteria regarding ‘soft’ techniques for flood risk management in the Ukrainian Tisza river basin. Communities in this region are faced with frequent floods and limited governmental budgets to cope with flood impacts. To identify the potential for soft flood protection measures as opposed to traditional technical solutions, we explored the decision-making heuristics of village council heads and the conditions under which they do or do not prepare for a flood event. Tacit knowledge, which is often unconscious and therefore difficult to describe, is complex to uncover through conventional interview techniques. To address this issue, a participatory process has been designed to reveal this knowledge without losing its connection to the context in which it is applied. That is, the KnETs process has been designed to understand context-relevant adaptive strategies and the reasons they are chosen in a natural resource management context. The process can be adapted to explore the contextual specificities of many situations ranging from flood and drought risk management to livelihood choices and the adaptation options considered in each set of circumstances. This interdisciplinary approach integrates ethnographic methods from the social sciences domain with classical computer science knowledge engineering techniques to address current bottlenecks (related to time and resource requirements) in both areas of research. This provides a participatory process, from knowledge elicitation to knowledge representation, verification and validation, providing a greater clarity of local data and thus possibly a greater understanding of social vulnerability and adaptive behaviour in flood situations. Keywords Knowledge elicitation, tacit knowledge, vulnerability, flood management, Ukraine

1. Introduction In the Ukrainian Zacarpathian Tisza valley, hazardous flood events regularly produce enormous damage in particularly poor regions that also have very limited financial budgets. Local monitoring systems are based administratively on sectoral, rather technical and data collection related objectives. They ignore the complex situation of the “risk of being flooded” for local municipalities. As such water management institutions and local municipalities risk addressing the wrong problems within the complex flood context, as they are not clearly linked to local decision makers’ needs and perceptions of risk. The perception of stakeholders of their own risk is significant at it will ultimately influence their behaviour. At present, communities living in the valley and their surrounding ecosystems are extremely vulnerable since these needs are not being addressed. This situation has not improved as the costs of a better flood protection system would be hard to recover with the limited finances available. Thus, there is a need to implement a new and more participatory flood management analysis to identify ways to reduce flood risk by enhancing local capacity for coping with or adapting to the situation (e.g. using existing knowledge, social networks and resources). Eliciting existing local knowledge and jointly uncovering and learning about existing decision-making procedures is seen as key to introducing stakeholder participation in the Zacarpathian flood management, one branch of the regional water resources management authority in Ushgorod. Specifically, the need to understand the multiple stresses which interact to form complex vulnerabilities, such as those observed in the Tisza basin, has led to the design of participatory Knowledge Elicitation Tools (KnETs). These tools represent a new and reproducible way to formalise local socio-environmental knowledge while exploring future scenarios during interviews. KnETs can be understood as departing from classical empirical tools for qualitative social science research using a more structured, yet more flexible and interactive interview method which results in a ‘game’ that is played iteratively (Bharwani, 2006). However, this does not replace the exploratory, open-ended and often less structured exploratory phase of social science research, which is stage 1 of the KnETs process illustrated in Figure 1.

Figure 1. The 4-stage KnETs gaming process The KnETs process supports stakeholder-led research by providing a more formal approach to knowledge elicitation, representation, verification and validation with iterative stakeholder engagement and feedback. The results are a set of production rules or decision trees for a given context which can also be used as input data for the rule-based logic of agent-based models (cf. Bharwani et al., 2008). KnETs allows the exploration of changing vulnerability by focussing on the multiple stressors to which differential exposure groups (people that are

faced with a specific environmental stress or danger) are vulnerable, and explores how these stressors influence different decision-making pathways over time by overlaying these with scenario analysis (Downing et al., 2005, Bharwani et al., 2005, Bharwani, 2006; Ziervogel et al., 2006). This process has been significantly influenced by the work of Gladwin (1989), Sinclair (1993) and Dixon (2005) and the importance they attach to emic and tacit knowledge. Emic categories can be described as the socio-cultural or context-specific influences which drive decision-making (Harris, 1979). This refers specifically to units of meaning drawn from the society and culture of interest resulting in local perceptions and meaning which are not necessarily easily observable. As emic categories are also defined in relation to other emic categories, this further complicates such an analysis. This is in contrast to etic categories which are observable (in theory)and selected to identify or describe a phenomena, situation or object but need not have meaning or significance for the community in question (Harris, 1979). That is, there are things people do, which we can observe and categorise. However, to understand the motivations for what people do we have to use relations on observables to 'uncover' the underlying logic and set of relations. Not only is this a difficult task, but often the researcher is not even aware that such relations exist. 1.1 Tacit knowledge Identifying the decision criteria which motivate human behaviour is fundamental to understanding decision-making processes and actions that shape our landscape and its environmental resources (Bharwani, 2006). Often we are confronted with multiple reasons for land use decisions of multiple actors that have a stake on one or more resources (e.g. land, water resources, soil fertility etc.; cf. also Ribarova et al., 2008). Problems of understanding the motivation for behaviour can also arise because people do not recognize that they have knowledge (though this informs their decisions) and therefore they rarely communicate it – tacit knowledge (Spradley, 1979; Werner and Schoepfle, 1987; Fazey, 2006). People find it hard to give descriptions of their knowledge and how it is used because much knowledge has been learnt through observation and experience, and is understood, but is not generally expressed. Although people identified for interview may be ‘experts’, it is unlikely that they have previously been required to describe their knowledge and decision-making procedures. Further, this is compounded when the outside researcher is not aware of the 'emic' knowledge relationships or logic in use. Because people organise their knowledge in meaningful terms and relationships, they use distinctions and make classifications that are not empirically observable until this underlying 'emic' logic is available. Familiarity with the domain can allow the researcher to access local knowledge, but ‘matching methods’ (Kemp-Benedict and Bharwani, 2006), such as KnETs, which incorporate ethnographic techniques combined with computer-aided tools, are powerful in bridging this qualitative local expert knowledge with a more formal representation in order to verify and validate it (Bharwani, 2006) and further to broach the realm of tacit knowledge. That is, etic observations combined with emic classification and logic does indeed make the relation between context and action observable. It is not enough to try to directly relate behaviour to context. Although one can develop correlations, because we are assuming that the behaviour is mediated by decisions (and thus not necessarily linear or continuous) our resolution will improve when we understand the underlying decision logic, and thus have

access to second order 'emic' variables which improve our capacity to observe in a manner compatible to the decision makers. The application of the KnETs methodology is successful due to the complementary nature of the ethnographic methodology and computer science techniques used, as each addresses current bottlenecks in both areas of research (for example ethnographic methods are usually situated at the local scale which provides a low degree of comparability and knowledge transfer and the process is very lengthy); on the other hand, computer-science does not generally capture qualitative perception and meaning in a realistic, non-abstract way. In summary, this innovative methodology for knowledge elicitation allows the construction of production rules which represent:

1 the multiple stresses that create the vulnerability context; 2 some ‘controls and checks’ (Wood and Ford, 1993) to fieldwork including

verification and validation of knowledge; 3 tacit knowledge, which is quite difficult to access otherwise; and 4 a way to formalize qualitative knowledge for use in more quantitative models.

1.2 Objectives In this case study we aim to apply and test the KnETs methodology in the Ukrainian part of Tisza river basin to explore the determinants of decision making on flood protection issues, in the context of climate change and the uncertainty associated with it. The potential of introducing ‘soft’ measures in a more or less ‘technically’ dominated flood protection system will be investigated. After a short introduction of the case study we will expand on the KnETs methodology before presenting the results in form of a set of decision trees. We will discuss the findings, including any tacit knowledge that was uncovered as well as the advantages and disadvantages of this methodological approach. 2. The case study Alongside climate change, there are a number of anthropogenic factors increasing flood risk and flood damage in the Tisza river basin (Jolonkai and Pataki, 2005). Among them the most important are: (1) reduction of water storage capacity in watersheds by (a) river regulation (straightening, shortening, drying small tributaries, wetlands and riparian zones along rivers), (b) deforestation and degradation of vegetation, (c) urbanization and increasing areas with impermeable surfaces in the river valley, in combination with related (2) human activities (settlements, commercial infrastructure and agriculture) in flood-prone areas (Haase and Bohn, 2007). In the Zacarpathian Tisza valley, Ukraine, flood events of the last decade produced enormous material damage in low income regions that already have very limited financial budgets from the government. This makes people and ecosystems extremely vulnerable since most of the costs for an improved flood protection system are hard to recover. Monitoring systems are based administratively, on sectoral, rather technical and data collection related objectives. They quasi ignore the complex situation of the “risk of being flooded” for both humans and ecosystems. As such water management institutions and local municipalities risk addressing the wrong problems within the complex flood context, as they are not clearly linked to local decision makers’ needs and perceptions of risk.

Assuming that existing technical flood protection measures will not be improved, and, furthermore, that each kind of construction leads to more severe flooding elsewhere, the participatory approach described here explores local adaptive capacities and strategies for successful adaptation. The focus of this is on stakeholders’ existing knowledge and coping strategies, in addition to their own knowledge and networking resources (Haase and Bohn, 2007). Flood protection measures can generally be classified into structural, more technical measures, and soft, non-structural measures, usually driven by local knowledge as outlined in Table 1 (Krysanova et al., 2008). The failure of structural measures to prevent extreme floods has long been recognized and it is clear that physical protection measures alone such as dams, storage reservoirs and embankments cannot completely protect against flood impacts (Kundzewicz and Takeuchi, 1999). Table 1. Hard (technical) and soft flood protection measures and risk response strategies (adopted and modified from Krysanova et al., 2008)

Non-structural measures Structural (technical) measures Interventions in river basins Socio-economic “soft” measures

Dams increase of natural water retention and water storage in watersheds (extending floodplains, creation of wetlands and polders)

flood mitigation systems of forecasting, warning, evacuation, and post-flood recovery

Levees improvement of infiltration and retardation of water (reducing impermeable areas, building groundwater cisterns etc.),

land use planning zoning maps (protection zones, evacuation zones)

water storage reservoirs for flood control

agriculture practices reducing runoff (catch crops, no black fallow in set-aside areas),

emergency committees (also transboundary) involvement of public, civil and societal institutions like church

off-stream polders or flood retardation ponds

zoning (delineation of floodplain zone, where only low-value infrastructure is allowed),

household mitigation and municipality preparedness actions

Dykes ensuring appropriate construction methods in flood-prone areas

capacity building (improving flood awareness, understanding and preparedness),

River embankments and floodwalls

risk spreading method: flood insurance

However, the incorporation of both short- and long-term social mitigation and adaptation measures such as land-use planning, flood forecast and warning systems, community emergency planning, community and household mitigation actions require context-specific socio-environmental knowledge of the local municipalities to increase the likelihood of successful implementation (Kundzewicz and Takeuchi, 1999). Thus, the KnETs method has been employed in the Ukrainian case to identify what form this context-specific socio-environmental knowledge may take.

3. Methods It has been important to maintain the involvement of the stakeholder in entire knowledge elicitation process, from acquiring the data to refining the resulting decision trees, to fill gaps in existing knowledge and to potentially reveal areas of tacit knowledge. Following ethnographic guidelines, the questions in a game (such as KnETs) are derived from an understanding of the informants’ own world-view (emic viewpoint), rather than imposed on them from a scientific/outsiders’ conceptual (mis)framing (etic viewpoint) which may not fully appreciate or represent the relationships between emic categories (Spradley, 1979; Ellen, 1984). Data is then collected during a phase of ‘gaming’ with a ‘training’ set of respondents to produce a set of decision trees which are generalised into production rules using a pattern extraction algorithm. These heuristics are presented to the same stakeholder groups in an iterative process to verify the decision rules, to access tacit knowledge where there are any inaccuracies or inconsistencies and to fill any remaining gaps in the knowledge base. Finally, another phase is conducted with a different ‘testing’ group of respondents with the same profile as the original stakeholders to validate the quality of the rules produced and to assess the representativeness of these decision rules for this profile of stakeholders, with the recognition that this is simply our understanding the relationships between decision-making criteria (both emic and etic), though this may not be the actual case (Bharwani 2006; cf. again Figure 1). 3.1 Designing the game for the Tisza case Knowledge elicitation can be a big bottleneck in the research process due to its qualitative, perceptual and fuzzy nature but KnETs includes tools that automate parts of this process, to reduce the time involved. The 4-stages (cf. Figure 1) include interviews to identify salient aspects of decision-making, different exposure units, their context-specific goals and the multiple stressors which drive or constrain decisions. In the Tisza case, the stressors include climate (low - high precipitation), personal awareness of risk (awareness of flood risk areas or of the flood action plan.) and economic factors (existence of state funds, personal capital or compensation potential). As we are interested in community and household preparedness strategies which can provide protection in the absence of additional assistance from the government, adaptation options are also included (e.g. relocating away from vulnerable places, paying for insurance, education on floods, involving the Church, reliance on social networks, simply following the flood action plan and restocking First Aid resources) which represent the adaptive capacity of the community and ultimately lead to a different decision pathway. These combinations of variables are used to produce a scenario which is explored using the computer-aided interactive ‘game’ to isolate the specific variables required for the decision-making process to proceed under differing conditions (Bharwani, 2006; Figure 2). Therefore, the aim of this KnETs game is to assess whether the local village council heads (VCH: the target stakeholder group with whom the game is played) of various upland and lowland farming communities would decide to help their communities ‘prepare’ for a potential flood situation or ‘not to prepare’ (their decision/goal). VCH are responsible for decision-making in the case of flooding which is why they are a significant target group for the knowledge sharing and learning process.

Figure 2. Variables which affect decisions on flood risk and adaptation management in the Zacarpathian Tisza valley, Ukraine. Exploring their decision pathways and the adaptive capacity of the group to deal with differing scenarios is important, as comparing what the VCH say they would do, with what they have actually done in the past and what they do at present provides information on where capacity is lacking and where interventions would be most valuable (cf. results; Table 2). That is, short and long-term adaptation strategies undertaken to make a scenario easier to cope with and to potentially reduce future vulnerability. Such strategies including switching to soft risk adaptation paths, such as providing insurance mechanisms, improving social networks, construction of floodplain management plan (e.g. dike construction, bank reinforcement, strengthening river channel), improving the early warning system and technical support, education on soft FPM, flood education at school, involving the Church, improving informational networks, reforestation and evacuation of vulnerable populations are all important to explore in hypothetical scenarios. People may indicate that they would opt for alternative adaptive management techniques if they had the capacity to do so, which may lead to a different management solution altogether. However, as mentioned all of these management techniques in the game would originate from stakeholder suggestions in the first instance, as it is their perception of their vulnerability and views on how to alleviate this which are critical for achieving successful transitions to new water and resource management regimes. For example, certain actors may feel their decisions are constrained by a lack of alternatives, by traditional or cultural beliefs, by a lack of knowledge or by a lack of institutional support to make the necessary changes (Table 2; Figure 3).

Table 2. Component attributes of the game designed for the Tisza case study Game component Attributes

Goals Prepared (meaning: whether village council heads will take any FPM andwhat FPM exactly);

Not prepared (village council heads won’t take any FPM)

Drivers Climatic: low precipitation, medium precipitation, increasing flood levels,high precipitation

Economic: lack of capital, additional state funds, state compensation exists,no state compensation.

Perception: Awareness of flood risk areas, Awareness of flood action plan.

Strategies Improve informational network and strengthen social network, Improvewarning system, Technical support (computer, Internet access), Involvechurch, Simple following action plan, Insurance system, Asking for floodrisk study, Relocate away from vulnerable places, Education on soft FPMand flood education at school, Construction FPM (to build/reconstruct dike,strengthen river channel, bank reinforcement), Organizational preparedness,Fill up first aid resources, Trainings of rescue team, Reforestation, Improveof infrastructure conditions, Responsibility of separate household.

Reasons No choices or alternatives, No experience yet, Do not have institutionalsupport, No state funds, It is tradition.

Figure 3. The game designed for the Tisza valley village council heads (VCH)

3.2 Applying KnETs in the field Stage 1 Before beginning the fieldwork we undertook some theoretical research on the local situation relating to flood protection measures (FPM) in the area was undertaken in order to achieve a better understanding of the domain and background to FPM in the Ukrainian Tisza valley generally (Table 3). Part of this involved analysis of the results of a questionnaire (stage 1) that was conducted during an earlier phase of research in the villages of the Tisza valley as this was related to different interrelations between people and institutions involved in the case of flooding. It was also important to meet the VCH that would be interviewed to understand their concerns relating to FPM, the drivers of their decisions and what do they actually do at present in case of flooding. Table 3. Villages where interviews and games had been conducted and number of damage cases

Name of the municipality

Number of inhabitants Number of damage cases during last

hazardous flood in 2001 or 2005

Number of scenario game runs conducted

Training, testing and verification group

Zarichchya 3700 168 57

Siltce 3113 48 49

Chorniy Potik 3372 56 85

Drotincy 2000 36 50

Fanchikovo 2059 50 74

Validation group

Korolevo 8064 702 55

Sasovo 2400 499 55

Chepa 1943 495 50

Trosnyk 2222 19 58

Steblivka 2231 21 50

Stage 2 Between March and June 2008, the local Ukrainian team designed the KnETs game by conducting interviews with VCH in combination with other information collection procedures such as a secondary literature review, local press and planning document studies. A local stakeholder analysis (ODA, 1995) was conducted and first contacts with the stakeholders were made. From July to September 2008 the refinement of the field study game continued based on input from the interviews. The game was translated into the Ukrainian language and

on to cards since most of the VCH have only limited computer access. From October to November 2007 10 VCH were interviewed (in Ukrainian). After a general introduction to the theme, various scenarios were proposed using cards to reflect those generated by the computer, allowing relevant responses regarding strategies and motivations for decision-making to be electronically recorded and then filtered through a rule induction algorithm. In order to get adequate and comparable results, the profile of the respondents in the chosen sample – the experienced local VCH in our case – had been the same during all three phases of the KnETs gaming process – the game itself, the verification and the validation phases. Stage 3 The interview-based data gathered in stages 1 and 2 was then filtered through a rule induction algorithm. KnETs uses Weka (Witten and Frank 2005) (open source software tools to support data mining), to produce decision trees based on stakeholder judgements with respect to a single goal. We currently use the J48 algorithm which is an implementation of the C4.5 algorithm (Quinlan 1993; this replaced our earlier use (Bharwani, 2006) of the ID3 algorithm (Quinlan 1986)). KnETs is being extended to make use of a wider range of pattern extraction algorithms available in Weka (such as Bayesian algorithms), as well as multiple algorithms for comparison. J48 (C4.5) identifies relationships within large datasets by building decision trees. The J48 algorithm implements exception based filtering: it learns by forming entailment rules and actively seeking instances that do not follow the rule. It then uses these exceptions to enhance its representation of the data set. Similar use of the older ID3 algorithm has been used by Reynolds (2000) to generate rules for predicting the creation of complex societies in the Valley of Oaxaca, Mexico (Reynolds, 2000). While J48 can only evaluate a single goal variable, it may have many exclusive states. The module was modified to output results in an XML format which can be exported for further processing or be translated in KnETs to production rules suitable for JESS (Friedman-Hill, 2000) or other expert system/production system software. Currently nominal data is being used for pattern extraction by Weka, although it is possible to use quantitative values for the criteria in J48 (though not the goal, which must be nominal in value). The resulting rules from stage 3 are then further refined by both the researcher and the stakeholder group in an iterative verification process. The following section provides details on how this is carried out and whether any new knowledge is gained through this process. 4. Results 4.1 Creation of decision-making heuristics After the decision trees were created by the rule induction algorithm (Figure 4) they were reorganised by the researcher on the basis of different criteria, such as ‘enabling’ and ‘maximising’ conditions (Gladwin 1989) based on the researcher’s own knowledge of the domain (Figure 5). The first number beside each rule represents the number of responses from the stakeholder which supports the rule (success), while the second number represents the number of times the response does not correspond with the rule (failure). According to rules produced we found that flood preparedness necessity at the local municipality level was closely tied to the climate situation. That is, only in case of low rainfall

did municipality heads not decide to mobilise any local resources such as manpower, knowledge or information distribution for flood preparation as risk was perceived to be ‘low’. In the case of extremely high rainfall they chose either to “simply follow the flood action plan” or to “relocate to less vulnerable places”. However, measures that represented the mobilisation of local potential such as organisational preparedness were also considered. In terms of longer-term measures, education on FPM or land-use zoning were options that were only considered when state funds were available. Figure 6 shows the decision tree created from these rules after re-ordering according to ‘enabling’ and ‘maximising’ criteria based on the researchers’ own knowledge of the domain.

Figure 4. Rules created by the rule induction algorithm at the first and tenth game.

Figure 5. Resulting rules after re-ordering according to ‘enabling’ and ‘maximising’ criteria by researchers.

4.2 Verification of the decision heuristics and elicitation of tacit knowledge The next stage in the KnETs process is to try and improve the re-ordered decision trees with further input from the informants. This aggregates our decision trees with monotonic decision paths to a generalised production system with potentially dynamic outcomes at each node. The verification phase was carried out with the same VCH (the “training” group) to correct any inaccuracies such as missing conditions, incorrect branches or decision nodes and to access new and potentially tacit knowledge that was driving the decision to ‘prepare’ or ‘not to prepare’ for floods. Stage 4 During the verification phase the same 5 VCH (from Zarichchya, Siltce, Chorniy Potik, Drotincy and Fanchikovo villages) interviewed in stage 1 were interviewed again. Since they had been already involved in the design of the game, conversations about future scenarios and flood management options were not difficult, and they were very interested in the results of the game. Each scenario or branch of the tree was explored in further discussions to investigate ‘levels of preparedness’. Their agreement or lack thereof with the outcome ‘prepared’ or ‘not prepared’ allowed them to mention further information which we did not have originally and to verify the outcomes we had recorded (Figure 6).

Figure 6. Updated rules with new information from stakeholders during the verification phase.

For example, in Zarichchya, we found the need to add new knowledge to the production rules as the VCH indicated that in case of increasing flood levels they are prepared organizationally (meaning that they do observe the flood levels), but if this is not the case, they are also able to involve the Church to warn people about coming flood. In Drotincy, the council head remarked that even under conditions of low rainfall, some long-term measures can be taken to prevent the consequences of future floods such as implementing financial insurance mechanisms. Previously, game responses had indicated that no measures were taken in the case of low rainfall, perhaps due to the emphasis of the research on flooding impacts. However, a further iteration allowed us to establish that even under these circumstances; some flood impact prevention measures are taken due to the historical experience of these events. This is an example of knowledge that is tacit – common-sense or subconscious information – which may not be mentioned to the researcher as it is perceived as ‘obvious’ to the stakeholder. 4.3 Validation phase Once the rules were verified with the original stakeholder group, our aim was to identify another sample of stakeholders with a similar profile (a “testing group”) who were not involved in the initial phase to validate the rules and to see how well they predict the decision-making processes of the group – village council heads – as a whole. Therefore, five new VCH (in Vary, Galabor, Svoboda, Chetfalva and Orosiyevo) were interviewed. In each new village a short introduction about the research was given. Some scenarios were then proposed based

on the decision trees from the previous game and their responses were incorporated into the game as new nodes where appropriate. In these villages, most technical flood prevention measures involved strengthening the dikes with sand bags. However, the role of the Church to prepare people was mentioned again under both conditions of high precipitation as well as increasing flood levels. This would seem to serve as a safety net in all high risk situations where risk mitigation and preparation for impacts has not worked. An additional precaution taken under the condition of increasing flood levels is training the flood rescue team. In these discussions it also became apparent that additional state funding was important under conditions of low as well as medium precipitation to adequately plan for and reduce future risk. In this case, they are able to take some more long-term strategies such as asking for a flood risk study and to improve infrastructure. If funds are not available, other measures are taken such as household insurance. In this regard, the general responsibility of individual households to “be prepared” for a flood while having such an insurance mechanism and communicating risky situations to family members or neighbours was highly valued in these circumstances.

Figure 7. Further verification and validation of the production rules. A further level of validation was carried out in Korolevo, Sasovo, Chepa, Trostnyk and Stevlivka to establish how well these decisions resonated with the original group. The VCH observe flood levels on the river (organisational preparedness) not only under obvious conditions of increasing flood levels but also when rainfall is high. In the case of long-term

preparedness to reduce future risk when conditions of medium and low precipitation allow, further strategies using additional finance appear. One important new strategy – improvement of flood warning system, appears since the VCH stated that it is very important to ‘be prepared for flood before the water comes’. This indicates that there is awareness for the need to implement some long-term preparedness measures in years of low rainfall and flood risk if additional funding is available. In addition to this the VCH also voiced the need to receive information about an approaching flood as soon possible. This need requires a long-term strategy in the form of technological support including investments in hardware, software and internet access. Moreover, VCH need support from governmental bodies in order to implement long-term actions. This was seen as an important use of any additional funding available in low risk/rainfall periods. This new information is represented in the production rules in Figure 7.

5. Discussion Our analysis of flood risk management and flood preparedness measures in the Ukrainian Tisza valley indicate that a long term adaptation can only be planned when risk is low (medium or low precipitation periods) and when funding is available. However, when funding is not available in both low and high risk periods, adaptation planning is not undertaken and responses are simply short-term coping strategies or highly dependent on individual household and social responsibility or on the Church. This implies that above all else, governmental support is critical for long-term adaptation. That is, long-term planning is not neglected due to a lack of knowledge of adaptation strategies but rather due to a lack of financial capacity to undertake these options. New knowledge in terms of potentials and barriers for the implementation of soft flood risk prevention became accessible using KnETs as opposed to straightforward interviews because:

1. The research process can explore the role of local knowledge – knowledge that is voiced and knowledge that is actually used may be different. This may be due to the tacit nature of this knowledge which can also cause problems of communication among the community themselves (e.g. new and incoming VCH).

2. Furthermore, in the Tisza case, the knowledge that emerged in the list of strategies (stage 2 – game design) after initial conversations with VCH (stage 1), were in form of new measures that were previously unknown to the researcher.

3. The method also allowed the investigation of the specific driving conditions that allow particular strategies to be chosen by VCH. The research was interesting and important for the VHC as their knowledge of what should be done for flood preparedness of their villages became more structured and explicit.

4. A concrete use of the output of this research is that it may allow newly elected village council heads who are not experienced in FPM to become quickly accustomed with the necessary information in the domain. This is very valuable particularly where experienced VCH may not realise that certain knowledge does need to be articulated or where they may find it difficult to describe their knowledge.

5. Finally, in contrast to cognitive mapping techniques, which take more time and a lot of explanation, KnETs provide a way to rapidly access, represent, verify and validate local knowledge (both tacit and explicit). Furthermore, as both techniques have been piloted in the Tisza basin, the results complement each other very well.

6. Conclusions The application of the KnETs game methodology revealed the salient criteria and thresholds of decision-making by municipal representatives concerning ‘soft’ mitigation decision pathways in flood risk management in the Ukrainian Tisza valley. The resulting production rules shed light on what knowledge is used for decision-making and how different criteria are prioritised in these choices. Interventions, be they related to water management or vulnerability reduction generally, must take socio-cultural context relevant perceptions into account to understand what drives adaptive and non-adaptive options, changes in behaviour and initiates learning – this can be described as the capacity of stakeholders to adapt. Where gaps in these decision-making structures exist may be exactly where development interventions may be most valuable. In this case, government funding which is specifically targeted toward long-term adaptation planning, such as an early warning system, technological support and improved insurance mechanisms would resonate most strongly with the needs expressed by those whose responsibility it is to prepare communities for flood events. At present, these areas are still highly dependent on individual households, social networks and the Church for support and therefore reinforcing these institutions would also provide greater stability and security for these communities in times of ‘high’ risk. This would appear to be a high priority for local government. What is most striking is that adaptation planning is not neglected due to a lack of knowledge of adaptation strategies but rather due to a lack of institutional and financial capacity to undertake these options to their maximum benefit. Ideally these needs could be addressed together and draw on the current strengths of the community. For example, the use of social networks, the Church and innovative information communication technologies (ICTs) could be drawn together to design a community-based flood early warning system. As these conclusions are derived from participatory, grounded, bottom-up research methods, where both scenarios and responses were stakeholder-defined, one anticipates that interventions at this level have the potential to be most effective. Acknowledgements First and foremost we would like to thank the municipal heads of all 10 villages in the Tisza valley who participated in the game. Further the authors are thankful to Katherine Daniell and Sabine Möllenkamp for very useful advice to improve the manuscript at an early stage as well as Yorck von Korff for launching the Special Issue in the Journal of Ecology and Society. Last not least the European Commission in “person” of the Newater EU project (EC contract No. 511179 GOCE) supported this work financially. References Becker, A. & Grunewald, U. 2003. “Disaster management - Flood risk in central Europe.”

Science 300 (5622): 1099-1099 Benioff R., Guill S. & Lee J. 1996. Vulnerability and Adaptation Assessments. An

International Handbook. Dordrecht, Kluwer Academic Publishers. Version 1.1. Berz, G. 2001. Climatic change: Effects on and possible responses by the insurance industry,

In: J. L. Lozán, H.Graßl, P. Hupfer (eds) Climate of the 21st Century: Changes and Risks, Office: and Wissenschaftliche Auswertungen, H., pp. 392-399.

Bharwani, S., Downing, T., Haase, D., Pahl-Wostl, C., Taylor, A., Shale, M., Taylor, R., Matin, N., Krywkow, J., van der Veen A., Hinkel, J., Bisaro, S., Sendzimir, J., Magnuszewski, P., Sullivan, C., Varela-Ortega, C., Esteve, P. 2008. Dynamic vulnerability. Learning from NeWater Case Studies. Internal report at www.newater.info.

Bharwani, S., 2006. Understanding complex behaviour and decision making using ethnographic knowledge elicitation games (KnETs). Social Science Computer Review 24(1), 78-105.

Downing, T. E., Aerts, J., Soussan, J., Barthelemy, O., Bharwani, S., Ionescu, C., Hinkel, J., Klein, R.J.T., Mata, L., Moss, S., Purkey, D. and Ziervogel, G. 2006. Integrating social vulnerability into water management. Oxford: Oxford: Stockholm Environment Institute.

EEA 2004. Impacts of Europe’s changing climate – An indicator-based assessment. EEA Report No. 2/2004.

Fazey, I., K. Proust, B. Newell, B. Johnson, and J. A. Fazey 2006. Eliciting the implicit knowledge and perceptions of on-ground conservation managers of the Macquarie Marshes. Ecology and Society 11(1): 25. [online] URL: http://www.ecologyandsociety.org/vol11/iss1/art25/

Fischer, M.D. 1994. Modelling Complexity: Social Knowledge and Social Process' in When History Accelerates: essays on the study of rapid social change (ed.) C. Hann. London:

Athlone Friedman-Hill, E.J. 2000. Jess, the Java Expert System Shell, Version 5.0. Sandia National

Laboratories, Livermore, CA,. http://herzberg.ca.sandia.gov/jess/docs accessed July 1, 2008.

Flood risk reduction in Germany. Lessons learned from the 2002 disaster in the Elbe region. 2004. German Committee for Disaster Reduction (DKKV) Publ. 29e, Bonn, 36pp.

Gladwin, C. H. 1989. Ethnographic Decision Tree Modelling. Sage Publications, Beverly Hills, CA.

Haase, D., Bohn, C. 2007. Flood vulnerability and preparedness: model approach to mitigate the risk for local communities. In: Schumann, A., Pahlow, M., Bogardi, J.J., van der Zaag, P. (eds) Reducing the Vulnerability of Societies Against Water Related Risks at the Basin Scale. IAHS Red Book Series 317, 1-7.

Jolonkai, G, Pataki, B. 2005. Summary of the Tisza River Project and its main results Real-life scale integrated catchment models for supporting water and environmental management decisions Contract No: EVK1-CT-2001-00099, Project duration: 01 January 2002 – 31 December 2004, Report for Newater.

Kemp-Benedict, E., Bharwani, S. 2007. Methods for Linking Social and Physical Analysis for Sustainability Planning. SEI Internal Report, Poverty & Vulnerability Programme Stockholm Environment Institute, Boston; Oxford.

Krysanova, V., H. Buiteveld, D. Haase, F. Hattermann, K. van Niekerk K. Roest, P.M. Santos, M. Schlüter, 2008. Practices and lessons learned in coping with climatic hazards (floods and droughts) at river basin scale. Ecology and Society, in press.

Kundzewicz, Z. W. & Kaczmarek, Z. 2000. Coping with hydrological extremes, Water International 25 (1): 66-75.

Kundzewicz, Z. W., S. Budhakooncharoen, A. Bronstert, H. Hoff, D. Lettenmaier, L. & K. Takeuchi. 1999. Flood protection and management: quo vadimus? Hydrological Sciences Journal, 44 (3): 417-432.

Menzel, D. & R. Schulze. 2001. Floods and Droughts: Coping with Variability and Climate Change, Thematic background paper at the Int. Conf. on Freshwater, Bonn 2001.

Overseas Development Administration, ODA. 1995. Guidance Note on How to Do Stakeholder Analysis of Aid Projects and Programmes. ODA Social Development Department (London: ODA [now known as DfiD]) Available at:

http://www.euforic.org/gb/stake1.htm (last retrieved May 2007).

http://www.newater.info/

Pahl-Wostl, C. 2007. Transition towards adaptive management of water facing climate and global change. Water Resources Management. 21(1), 49-62.

Quinlan, J.R. 1986. Induction of Decision Trees, in Machine Learning, 1, pp.81-106, 1986. Quinlan, J.R. 1993. C4.5: Programs for Machine Learning, San Francisco: Morgan

Kaufmann. Reynolds, R.G. 2001. Chapter 11, 'The Impact of Raiding on Settlement Patterns in the

Northern Valley of Oaxaca: An Approach Using Decision Trees' in Dynamics in Human and Primate Societies: Agent-based Modelling of Social and Spatial Processes. Timothy A. Kohler and George J. Gumerman (eds.) New York, NY: Oxford University Press.

Ribarova, I.S., Ninov, P.I., Daniell, K.A., Ferrand, N., Hare, M. 2008. Integration of technical and non-technical approaches for flood identification. Water Down under, pp. 2598-2609.

Spradley, J. P. (1979). The Ethnographic Interview. New York, London Rinehart & Winston, 1979.

Werner, O. & Schoepfle, G. M. 1987. Systematic Fieldwork: Ethnographic Analysis and Data Management, Vol. 1, Sage Publications,

Witten, I. H. & F. Eibe 2005. Data Mining: Practical machine learning tools and techniques, 2nd Edition, San Francisco: Morgan Kaufmann.

Wood, L. E. & Ford, J. M. 1993. Structuring interviews with experts during knowledge elicitation. International Journal of Intelligent Systems, 8(1):71-90.

knowledge elicitation training and dissemination report€¦ · knowledge elicitation tools (knets)...

Documents