**2. New challenges for monitoring systems and information management in Adaptive Management (AM)**

Incorporating uncertainties about future pressures on river basins into water resources management sets new challenges for environmental resources management. One learning process being developed to address this challenge is Adaptive Management (AM) (Holling 1978). Learning more about the resources or system to be managed and its responses to management actions, in order to develop a shift in understanding, is an inherent objective of AM (Walters, 1997; Fazey et al., 2005). Learning in AM leads to a

Monitoring Information Systems to Support Adaptive Water Management 429

other scales, i.e. the level below, to understand the important processes that lead to the emerging characteristics of the level considered, and the level above it. Two sets of variables have to be considered for every system-subsystem pair. One set is required to describe the properties of the subsystem, whereas the second set is needed to describe the contribution of the subsystem to the performance of the whole system. This duality should be repeated at

Therefore, during the participatory process aimed at developing the cognitive model, participants should be required to think about their understanding of the total system, its essential component systems and the relationships that exist between them. The variables forming the cognitive model have to be able to describe the performance of the individual system and its contribution to the performance of the other systems. Using this inter-scale cognitive model as a basis for the design phase allows us to define a monitoring system capable of dealing with complex relationships between different scales, thus overcoming

However, adopting this inter-scale approach usually results in a demand to monitor a broader set of monitoring variables than traditional monitoring approaches. Some of these variables are fairly cheap to measure, but others, such as trends in very rare and important species, can be very expensive to monitor (Walkers, 1997). Thus, the development of an affordable monitoring program to support Adaptive Management involves substantial, scientific innovation in both method and approach, aimed at simplifying the set of

The key components of the system, or key variables, are those that influence the system dynamics and bring about the most important changes (Walker et al., 2006; Campbell et al., 2001). Since these variables influence the overall dynamics of the system, they are of direct interest to managers, who are frequently focused on fast variables. These variables operate at different scales and with different speeds of change. The slowly changing variables determine the dynamics of the ecological system, whereas the social systems can be influenced by slow and/or fast variables (Walker et al., 2006). The conceptual models developed integrating the stakeholders' understanding of the system can be used as a basis for identifying the key variables (Campbell et al., 2001). To this aim, the analysis of CM can provided information about the relative importance of the different variables, by analysing the complexity of the causal chain. Those nodes whose immediate domain is most complex

The identification of the key variables can also be supported by a strict integration between system monitoring and system modelling. This, in turn, is essential to any analysis of the implications of water policies. It allows the difficulties in understanding the dynamic feedback of the systems to be overcome, a particularly difficult task in an environmental context because of the number of factors involved. Moreover, humans have a limited capacity to understand the complexity of feedback in ecological systems (Fazey et al., 2005). This leads to erroneous connections between cause and effect and, thus, to erroneous conclusions about the impact of management actions. Conversely, models suggest which variables may be critical to monitor the impact of management actions, by posing elaborate hypotheses of which variables and relationships are critical to understanding the problem in question. The models then consider the dynamic implications of these hypotheses through the simulation of different scenarios. This allows monitoring networks to be designed (and re-designed) according to the model results. The potential of models to simulate future scenarios can be exploited to support the categorisation of the variables according to speed

every level of the system hierarchy (Bossel, 2001).

one of the main drawbacks of traditional monitoring practices.

monitoring variables by identifying the key components of the system.

are taken to be those most central and, thus, the most important.

focus on the role of feedback from the implemented actions. Such feedback-base learning models stress the need for monitoring the discrepancies between intentions and actual outcomes (Fazey et al., 2005). Monitoring becomes the primary tool for learning about the system and its performance under different management alternatives (Campbell et al., 2001).

To this aim, we assume that learning can be defined as a change in a person-system relationship, that is, the understanding of a person's place in the system and how they perceive it (Fazey et al., 2005). This definition implies that, because understanding is the goal which is achieved by the learner, each person may understand the environmental system differently and, therefore, act differently (Fazey et al., 2005). From the information production and management point of view, this implies that mental models influence an actor's perception of a problematic situation by influencing not only what data the actor perceives in the real world and what knowledge the actor derives from it (Timmerman and Langaas, 2004; Pahl-Wostl, 2007; Kolkman et al., 2005), but also what is noticed and what is taken to be significant (Checkland, 2001). It is important in information production and management that there should be a clear understanding and sharing of information users' mental models.

Therefore, contrarily to the traditional approach, in which information needs elicitation was intended in a top-down perspective, the design of a monitoring system for AM should begin by bringing together the interested parties to discuss their understanding of the system, the management problem, the information needed and how this information should be used. This implies involving a wide variety of stakeholders (i.e. scientists, managers, policy makers and members of the public at large) in a debate in which assumptions about the world are teased out, challenged, tested and discussed (Checkland, 2001), leading to the establishment of a common understanding about the system to be managed (Pahl-Wostl, 2007). This shared understanding can be structured in a system cognitive model, which allows the emergent properties of the system (i.e. variables to be monitored, thresholds, etc.) to be identified.

Among the different methods for Cognitive Modelling, an integration between Cognitive Maps (CM) and Causal Loop Diagrams (CLD) would seem particularly interesting to support monitoring system design. Given the peculiarities of the two modelling devices, CM can be used to disclose individual understanding of the system and to support the debate among participants, whereas CLD has great potentialities to simulate system dynamics.

When defining the cognitive model to be used as basis for a monitoring system, it is essential to address certain issues related to complex system dynamics. Firstly, the issue of scale must be tackled, since complex systems have structures and functions that cover a wide range of spatial and temporal scales. The impact of a given management action may vary at different scales (Campbell et al., 2001). Moreover, structures and processes are also linked across scales. Thus, the dynamics of a system at one particular scale cannot be analysed without taking into account the dynamics and cross-scale influences from the scales above and below it (Walker et al., 2006).

To deal with interaction between scales, we assume that the complex web of interacting systems can be broken down recursively into a network of individual systems, each of which determines its own fate and affects that of one or more other systems. The hierarchical structure of relationships between systems and subsystems (Campbell et al., 2001) implies that working on a particular scale often requires insights from at least two

focus on the role of feedback from the implemented actions. Such feedback-base learning models stress the need for monitoring the discrepancies between intentions and actual outcomes (Fazey et al., 2005). Monitoring becomes the primary tool for learning about the system and its performance under different management alternatives (Campbell et al.,

To this aim, we assume that learning can be defined as a change in a person-system relationship, that is, the understanding of a person's place in the system and how they perceive it (Fazey et al., 2005). This definition implies that, because understanding is the goal which is achieved by the learner, each person may understand the environmental system differently and, therefore, act differently (Fazey et al., 2005). From the information production and management point of view, this implies that mental models influence an actor's perception of a problematic situation by influencing not only what data the actor perceives in the real world and what knowledge the actor derives from it (Timmerman and Langaas, 2004; Pahl-Wostl, 2007; Kolkman et al., 2005), but also what is noticed and what is taken to be significant (Checkland, 2001). It is important in information production and management that there should be a clear understanding and sharing of information users'

Therefore, contrarily to the traditional approach, in which information needs elicitation was intended in a top-down perspective, the design of a monitoring system for AM should begin by bringing together the interested parties to discuss their understanding of the system, the management problem, the information needed and how this information should be used. This implies involving a wide variety of stakeholders (i.e. scientists, managers, policy makers and members of the public at large) in a debate in which assumptions about the world are teased out, challenged, tested and discussed (Checkland, 2001), leading to the establishment of a common understanding about the system to be managed (Pahl-Wostl, 2007). This shared understanding can be structured in a system cognitive model, which allows the emergent properties of the system (i.e. variables to be monitored, thresholds, etc.)

Among the different methods for Cognitive Modelling, an integration between Cognitive Maps (CM) and Causal Loop Diagrams (CLD) would seem particularly interesting to support monitoring system design. Given the peculiarities of the two modelling devices, CM can be used to disclose individual understanding of the system and to support the debate among participants, whereas CLD has great potentialities to simulate system

When defining the cognitive model to be used as basis for a monitoring system, it is essential to address certain issues related to complex system dynamics. Firstly, the issue of scale must be tackled, since complex systems have structures and functions that cover a wide range of spatial and temporal scales. The impact of a given management action may vary at different scales (Campbell et al., 2001). Moreover, structures and processes are also linked across scales. Thus, the dynamics of a system at one particular scale cannot be analysed without taking into account the dynamics and cross-scale influences from the

To deal with interaction between scales, we assume that the complex web of interacting systems can be broken down recursively into a network of individual systems, each of which determines its own fate and affects that of one or more other systems. The hierarchical structure of relationships between systems and subsystems (Campbell et al., 2001) implies that working on a particular scale often requires insights from at least two

2001).

mental models.

to be identified.

dynamics.

scales above and below it (Walker et al., 2006).

other scales, i.e. the level below, to understand the important processes that lead to the emerging characteristics of the level considered, and the level above it. Two sets of variables have to be considered for every system-subsystem pair. One set is required to describe the properties of the subsystem, whereas the second set is needed to describe the contribution of the subsystem to the performance of the whole system. This duality should be repeated at every level of the system hierarchy (Bossel, 2001).

Therefore, during the participatory process aimed at developing the cognitive model, participants should be required to think about their understanding of the total system, its essential component systems and the relationships that exist between them. The variables forming the cognitive model have to be able to describe the performance of the individual system and its contribution to the performance of the other systems. Using this inter-scale cognitive model as a basis for the design phase allows us to define a monitoring system capable of dealing with complex relationships between different scales, thus overcoming one of the main drawbacks of traditional monitoring practices.

However, adopting this inter-scale approach usually results in a demand to monitor a broader set of monitoring variables than traditional monitoring approaches. Some of these variables are fairly cheap to measure, but others, such as trends in very rare and important species, can be very expensive to monitor (Walkers, 1997). Thus, the development of an affordable monitoring program to support Adaptive Management involves substantial, scientific innovation in both method and approach, aimed at simplifying the set of monitoring variables by identifying the key components of the system.

The key components of the system, or key variables, are those that influence the system dynamics and bring about the most important changes (Walker et al., 2006; Campbell et al., 2001). Since these variables influence the overall dynamics of the system, they are of direct interest to managers, who are frequently focused on fast variables. These variables operate at different scales and with different speeds of change. The slowly changing variables determine the dynamics of the ecological system, whereas the social systems can be influenced by slow and/or fast variables (Walker et al., 2006). The conceptual models developed integrating the stakeholders' understanding of the system can be used as a basis for identifying the key variables (Campbell et al., 2001). To this aim, the analysis of CM can provided information about the relative importance of the different variables, by analysing the complexity of the causal chain. Those nodes whose immediate domain is most complex are taken to be those most central and, thus, the most important.

The identification of the key variables can also be supported by a strict integration between system monitoring and system modelling. This, in turn, is essential to any analysis of the implications of water policies. It allows the difficulties in understanding the dynamic feedback of the systems to be overcome, a particularly difficult task in an environmental context because of the number of factors involved. Moreover, humans have a limited capacity to understand the complexity of feedback in ecological systems (Fazey et al., 2005). This leads to erroneous connections between cause and effect and, thus, to erroneous conclusions about the impact of management actions. Conversely, models suggest which variables may be critical to monitor the impact of management actions, by posing elaborate hypotheses of which variables and relationships are critical to understanding the problem in question. The models then consider the dynamic implications of these hypotheses through the simulation of different scenarios. This allows monitoring networks to be designed (and re-designed) according to the model results. The potential of models to simulate future scenarios can be exploited to support the categorisation of the variables according to speed

Monitoring Information Systems to Support Adaptive Water Management 431

Considering the issues described in the previous section, the conceptual architecture of a monitoring system for AM was defined (figure 1). From now onward, we refer to this

Fig. 1. AMIS conceptual architecture. The figure has been adapted from the Information cycle elaborated by Timmerman and others (2000), to emphasise the two learning

As described previously, the basis for AMIS design is the conceptual model of the system, which simplifies the system and makes the key components and interactions explicit. The definition of this model is based on the integration between a participatory process, allowing experienced stakeholders to provide their understanding of the system, and models able to simulate future scenarios. The conceptual model is structured using the

Two different conceptual models, i.e. the "water management conceptual model" and the "information management conceptual model" are defined as the basis of AMIS. The former concerns the interpretation of the problem considered, while the latter concerns the information needed to solve the problem considered, and the "frames" used to interpret the

The AMIS architecture consists of four main boxes, i.e. Conceptual model elicitation, Design, Data collection and Interpretation. The links between them represent the iterative process of monitoring design, which is at the basis of AMIS. The figure was elaborated starting from the information cycle developed by Timmerman et al. (2000). This cycle depicts a framework where information users and producers communicate information needs that link the

integration between Cognitive Maps and Causal Loop Diagrams.

information (Pahl-Wostl, 2007; Kolkman et al., 2005).

**3. Adaptive monitoring and information system** 

system as Adaptive Monitoring Information System (AMIS).

processes.

of change, i.e. slow changing variables and fast changing variables. Scenario simulation can draw attention to the role of the slow-changing variables in influencing system dynamics (Walker et al., 2006). The categorisation of variables according to speed of change can be used to program the frequency of data collection, making it easier to identify each variable's trend.

The integration between monitoring and modelling has to be considered as an iterative process. In fact, while models can simulate system dynamics, allowing the identification of key variables, the availability of new data allows the revision and updating of models. Moreover, the speed of change of the variables can also be considered iterative. Indeed, variables classified as slow changing in the model may be identified as fast changing by the monitoring system. In this case, the monitoring sample interval has to be changed. Thus, clearly a re-assessment process is needed both in models and in monitoring.

Simulation of system dynamics facilitates the identification of thresholds, which can be broadly defined as a breakpoint between two states of a system. When a threshold is exceeded, a change in system function and structure results. Such changes regard the nature and extent of feedback, resulting in changes of directions of the system itself. The changes can be reversible, irreversible or effectively irreversible (Walker et al., 2006). Two different types of thresholds can be defined, i.e. positive and negative. A positive threshold represents a desirable change in the state of the system. Such a change can be due to implemented management actions. A negative threshold can be considered as the starting point of a non-acceptable system trajectory. The recognition of these thresholds is particularly important in the case of irreversible changes. In this situation, actions are needed in order to avoid exceeding the threshold. The integration between monitoring and modelling provides information about the current state and the future trajectory of the system.

The position of the threshold is strictly linked to past experience. There are no examples where a new kind of threshold has been predicted before it has been experienced. Typically, the identification of thresholds is based on an analysis of systems similar to the one under investigation (Walker and Meyers, 2004). To this aim, a database is going to be implemented to collect empirical data on possible regime shifts in socio-ecological systems (Walker and Meyers, 2004). Some authors suggest using variances in variable trends to detect an impending system change (Brock and Carpenter, 2006). Integrating these two different approaches can be very useful. In other words, the existing experience regarding regime shifts, coming both from other systems and from the tacit knowledge of experienced and highly skilled people, can be structured and included in the system model. The variance can be calculated using monitoring data and the position of the threshold can be changed.

Integrating system modelling and monitoring iteratively highlights the importance of collecting information on trends. In fact, the availability of time series of data on the different variables allows the behaviour of the system variables and the trajectory of the system to be defined. The detection of trends can support the revision of the hypothesis concerning system dynamics, which is at the basis of the models. For these reasons it is fundamental to develop a monitoring system which is sustainable over time. To this aim, two important issues needs to be addressed, i.e. the need firstly to increase the adaptability of the monitoring system to policy and learning processes, and secondly to reduce monitoring costs through the adoption of scientific and technical innovation in information collection.
