An Approach to Support Collaborative Spatio-Temporal Analysis of Complex Systems

Andrew Fall, Dave Daust, Don G. Morgan and Marvin Eng

  1. Andrew Fall. School of Resource and Environmental Management, Simon Fraser University, Burnaby, B.C.
    V5A-1S6. Email: fall@cs.sfu.ca
  2. Dave Daust. Research Consultant, Francois Lake, B.C
  3. Don G. Morgan. Research Branch, B.C. Ministry of Forests
  4. Marvin Eng. Research Branch, B.C. Ministry of Forests

Keywords:

spatially-explicit simulation, complex adaptive systems, landscape ecology, adaptive resource management, collaborative modelling, participatory design

Abstract Landscape management requires analysis of complex interactions among ecosystems and management regimes. Spatio-temporal simulation models are increasingly being used to assess potential long-term consequences of decisions on ecological, social and economic values.

To be applied successfully in management situations, models must address appropriate questions, include relevant processes and interactions, be perceived as credible and involve people affected by decisions. There are two requirements for success: a tool that supports rapid model prototyping and modification, that makes a clear link between a conceptual and implemented model, and that has the ability to implement a wide range of model types; and a framework for collaborative model building.

Our approach has roots in adaptive management, computer-supported collaborative work, complexity theory, logic programming, computer simulation and landscape ecology. SELES (Spatially Explicit Landscape Event Simulator) is a tool for building and running spatio-temporal models that supports this process by providing a high-level means of specifying complex model behaviours. The workshop process and resulting models have efficiently provided insight into the dynamics of large landscapes over long time frames in several forest modelling projects in British Columbia, Canada.

1. Introduction

Forest managers face complex problems, due to the intricate nature of forest ecosystems, often elaborate socio-economic and policy environments, and increasing local and global public scrutiny. Given this complexity, forest management problems may require a decision-support process to foster communication and shared understanding among decision-makers and stakeholders. Landscape modelling can provide a focus for gaining insight into the complexities of landscape dynamics and for showing logical outcomes to sets of assumptions. It is particularly useful when problems include aspects of natural disturbance, succession and management over long time frames and large areas (Hunsaker et al. 1993, Sklar and Costanza 1991). These situations often involve moderate to high levels of uncertainty and risk, and include complex spatial and temporal interactions, such as connectivity, feedback, emergent properties and cumulative effects.
Generating model results and presenting them to appropriate people, however, is not sufficient to increase understanding. The target audience must have confidence in model results, rather than blind trust or unwarranted scepticism, and the knowledge of how to interpret them. This implies that people must be involved in the process of landscape model development and analysis. Due to the wide range of interests, time availability and technical skills of participants, model development should be at the conceptual, rather than implementation, level. The structured thinking imposed by the process of developing conceptual models forces participants to focus on key elements and interactions in a problem. Improved understanding of model assumptions, behaviour and limitations can improve communication and interpretation of results from simulation experiments.
The notion that a group with appropriate expertise can combine talents to produce better models and solutions is not new (Clark et al. 1979, Holling 1978, Kyng 1991, Maxwell and Constanza 1997). Participatory model design combines diverse sources of expertise and may generate novel solutions by exploring a wide range of scenarios, including alternatives that might otherwise not be considered (Kyng 1991, Schuler and Namioka 1992). The benefits of a participatory design process extend well beyond the model produced (Kyng 1991, McLain and Lee 1996), bringing benefits to participants such as improved understanding, communication and co-operation among stakeholders and designers; conflict resolution and consensus building; and an opportunity to influence the decision-making process. However, collaborative model development has not been widely practised at the landscape scale due to a lack of technological support, data, system knowledge, and the difficulty of formulating, implementing and validating complex models (Maxwell and Costanza 1997). The growing availability of landscape models will not solve these problems. Applying an existing model, developed for a different purpose, forces current questions to be adapted to fit the model rather than vice versa. Collaborative tools and processes should not constrain participants within a specific viewpoint, but instead support free thought, flexibility and reorganisation (Maxwell and Villa 1999). As well as constraining questions, an existing model can bring more insidious problems by masking their implicit assumptions, increasing the risk that interpreted model behaviour reflects artefacts of these assumptions rather than the system under study. In addition, existing models do not easily allow for input from local people. Technology to enable this process must provide a transparent basis to implement and analyse the resulting models and management scenarios.
We have developed a structured framework to guide collaborative development of landscape models (for more details see Fall et al. 2001). The focus on application of a collaborative process to address complex spatio-temporal issues at the landscape scale, through participation in conceptual model development, makes our approach novel. We describe modelling software that facilitates this framework by capturing conceptual landscape models fairly easily and by allowing rapid model implementation. We illustrate our landscape analysis process with example applications, and conclude with observations on our approach.

2. A Collaborative Framework for Landscape Modelling

The framework we propose rests on the foundations of adaptive environmental planning, assessment and management (Holling 1978, Selin and Chavez 1995) and computer-supported collaborative work (Grudin 1991), but differs somewhat by emphasizing opportunities for stakeholders to have a more extensive role in conceptual model development, thereby increasing opportunities to learn, and by emphasising the use of a high-level model support tool, SELES (Fall and Fall, 2001). Maxwell and Costanza (1997) propose a framework for collaborative model construction that helps multiple teams to implement various portions of a large model. While useful for some circumstances, in many resource management problems, stakeholders do not have the time, mandate and/or technical background to actually implement models. We focus on conceptual model construction, rather than model implementation, as the hub for collaboration. As with the collaborative planning approach advanced by Selin and Chavez (1995), our framework can be flexibly adapted to meet the unique needs of each new situation.

Our framework divides participants into three embedded groups, each of which is involved in different aspects of development (Figure 1). The basic steps in our nested, iterative process are:

(1) A set of workshops allows interested parties to define relevant questions and objectives, to develop conceptual models, including key model outputs, and to provide expert knowledge and empirical data. Usually one workshop for the entire set of participants is followed by smaller workshops with domain experts that focus on specific model components to refine and formalize the conceptual models.

(2) A core modelling team assembles and analyzes the available data (e.g. GIS data, fire history information) and implements the formal conceptual models. Simulation experiments are used to verify equivalence between the implemented and the formal conceptual models. The team documents the conceptual models and experimental results to communicate the model status and limitations to interested participants.

(3) A second round of workshops is held with the goal of completing model verification. Each of these workshops usually focuses on a specific component of the project, and includes only a subset of participants interested in that area (mainly domain experts). Participants refine the formal conceptual models, and provide new information to help parameterise models.

(4) The core team updates model implementation and documentation. With successive iterations and increasing confidence in conceptual and implemented models, experimentation shifts from verification towards validation, sensitivity analysis, hypothesis testing and scenario evaluation. Documentation of the assumptions, input parameters and results of scenario experiments is critical at this stage.

(5) A final workshop is held with all participants to present and discuss results and to discuss the status and future of the project. Participants may identify modifications to questions addressed, processes and interactions included, indicators tracked, management scenarios, and conceptual models. This completes one iteration of the modelling cycle (Figure 1). These steps can be repeated as needed given the time constraints and complexity of the problem, where the final workshop initiates the subsequent iteration.

Figure 1.

Our nested, iterative model development process. Groups participate in all circles that surround them. All participants set objectives, select scenarios, develop conceptual models, and discuss model results. Domain experts and the core team develop and verify the formal models. The core modelling team is responsible for organizing workshops, communication, gathering information, implementing models, running simulations, analyzing outputs and documentation.

We separate model development into three parts: (possibly informal) conceptual models, formal conceptual models, and implemented models. Conceptual model development focuses on identifying the key ecological and management processes and interactions, relevant spatial and temporal scales, and appropriate levels of detail included. Conceptual models can form a longer-term vision that is relatively unrestricted by current knowledge, data and technological limitations.

The goal of model formalization is to create unambiguous model specifications in conjunction with domain experts. It is critical that the core team can gather the required data, do the required background analyses and implement the formal models within the project time frame. Identifying model aspects that are currently not feasible to implement due to limitations in understanding, data, time, or technological capability may lead to recommendations for required research, data collection, or technology advance.

A variety of techniques may be used for model formalization, such as graphs, equations, transition tables, Markov chains, and stock and flow diagrams. It is important to express model structure according to the viewpoints of the experts involved to maximize model clarity. The goal of model implementation is to accurately and quickly encode the formal conceptual models as a computer simulation. This highlights the need for modelling tools that support transparent and timely implementation.

Our framework is based on interactions between four components:

(1) Decision-makers and stakeholders set project objectives, defining the issues at stake and the range of potential management actions to consider. They should be the primary beneficiaries of improved understanding gained by developing and applying the model. These include people who make decisions, are affected by decisions or who advocate certain values (e.g. non-governmental organizations).

(2) Domain experts and information sources provide the basis for formalizing and parameterizing conceptual models. The people involved include scientists (both local experts and domain specialists) familiar with the ecological processes involved, and planners familiar with the management regime. Sources of information include published literature, internal reports and studies, and databases.

(3) Enabling technology supports the landscape analysis process as transparently as possible. Geographic information systems (GIS) are used to organize spatial information, and for static spatial analysis of input and output maps. Statistical packages are useful to analyze background data (e.g. fire history) and simulation results. Databases, spreadsheets, graphing packages and scripting tools support the processing of potentially vast amounts of simulation data. An array of tools is presently available for these tasks, with a gap in tools to support the construction of spatio-temporal landscape simulation models.

(4) A small core team (3-5 people) is responsible to co-ordinate the process, and collectively must possess expertise in several disciplines: (i) Problem domain: they should minimally have a basic knowledge of the important ecological and management processes in order to understand the issues at stake. (ii) People skills: communication and facilitation is required to organize workshops and to present model assumptions, behaviour and results. (ii) Technical skills: a solid understanding of GIS, ecology, forestry, formal analysis, statistics and spatio-temporal modelling is critical.

3. Enabling Technology for Constructing Spatio-Temporal Simulation Models

Options for model implementation range from direct programming to parameterizing a pre-existing model. Intermediate approaches include using software libraries and general, flexible models. Pitfalls of directly programming a complex model or using a pre-built model are many (See Derry 1998; Fall and Fall 2001). In (Fall and Fall 2001) we describe a range of tools to assist with model development, and argue that the use of domain-specific modelling languages which lie between programming languages and generalized models provide an appropriate balance between ease of implementation and model flexibility and transparency.

They support construction of classes of models rather than individual model types. They make fewer assumptions about the underlying structure of the implemented model than do pre-programmed models. For example, Petri nets (Groenwold and Sonnenschein 1998) provide a graphical technique for describing the dynamics of cellular automata (synchronous cell-based models). PCRaster integrates a declarative language for iterative environmental modelling that is integrated with a GIS (Wesseling, 1996). WESP-TOOL (Lorek and Sonnenschein, 1999) aims to support conceptual individual-based modelling.

4. Applications of Methodology

We have applied our framework in a number of landscape scale problems. All share a broad spatial scale (of at least one million hectares) at a relatively fine resolution (1 to 6.25 ha/cell). These problems also represent complex systems, with several interacting agents of change and with a number of competing values of interest. As a result they present difficult management problems for which landscape modelling provides one avenue to gain insights into the effects of potential management options. Due both to the uncertainty and complexity of these systems as well as to the competing values at stake, the process of model application must be transparent. Due to space limitations, we describe only a few key features of a few of these problems.

Natural disturbance is a recurring yet unpredictable process in many forest landscapes. To address how uncertainty in natural disturbances, primarily fire and insect outbreaks, interact with timber supply over several centuries, we developed a landscape model for the Robson Valley Forest District in east-central British Columbia. We held several general workshops to develop scenarios and more specific workshops to derive the formal conceptual models of forest management, natural disturbance and forest succession. The model allowed us to demonstrate a number of emergent, yet logical consequences of some assumptions regarding natural disturbance regimes that are commonly made in timber supply analyses.

Mountain pine beetle (Dendroctonus ponderosae Hopk.) (MPB) can cause wide scale mortality in mature pine forests in western North America, yet the complexity of beetle dynamics has limited development of landscape scale models to assess management implications. We used SELES to scale a stand level MPB model developed at the Canadian Forest Service (Safranyik et al. 1999) up to the landscape. The stand model drives within-cell dynamics, while beetle dispersal and pheromone diffusion are landscape scale phenomena. This complex population model then interacts with a forest management model that can apply various beetle management options. We held workshops in the Kamloops Forest District of south-central British Columbia and with provincial experts to develop model objectives. Detailed workshops were held with appropriate experts to develop the formal beetle and management models. Directly integrating the knowledge and models of MPB experts gave credibility and confidence in our landscape model, which allowed us to address how current beetle management efforts have influenced outbreak dynamics over a time horizon of one decade.

Mountain caribou are an endangered sub-species of woodland caribou (Rangifer tarandus caribou) that inhabit mountainous areas of south-eastern British Columbia. Their need for mid-elevation old growth forest in early winter presents a conflict with forestry values. Apps et al. (2001) developed a model of habitat based on several years of telemetry data using logistic regression. To explore long-term consequences of management options, we integrated this static statistical model with a forest management model for the Columbia Mountains Forest District. We held several workshops to develop scenarios and formal sub-models. The results of this project have been useful to highlight the likelihood of management impacts on caribou habitat, and also to explore the challenges of applying statistical models in projected landscapes.

5. Discussion and Conclusions

For spatio-temporal modelling to adequately support decision-making, the model development process must adapt to fit the decision-making process, not vice versa. Because each management problem has unique characteristics and questions, frameworks to build models serve better than application of existing models. The framework we propose is people and issue centred, not technology centred, with the focus on the conceptual models, scenarios and outputs.

A major challenge is the need for technology to support our framework and to allow model development to be focused at the conceptual level. A variety of available geographic information systems, statistical packages, etc. fit the needs of some portions of our framework, but there are few software tools for constructing spatio-temporal models. Such tools must be flexible to handle a diverse range of model types and to accelerate implementation so that it does not become the focus of the project. SELES provides a flexible environment for implementing spatio-temporal models that facilitates participatory model design, allows rapid model implementation, and supports dynamic spatial analysis of landscape pattern and process. In the context of our framework, SELES allows the modelling technology to stay in the “back seat” of the process, allowing participants to focus on the conceptual model. The core modelling team ensures that the gap between the conceptual and the implemented models is as small as possible and well documented.

We have applied our framework in a number of projects addressing complex spatio-temporal problems. The model development process and results have provided a forum to develop shared understanding and to discuss impacts of management decisions on landscape structure. In the past, modelling has failed to reach its potential as a decision-support tool in the forest management arena. Yet the complexity and large spatial and temporal scales that characterize forest management problems suggest modelling should help. While our framework and software aim to improve the quality of decisions, similar frameworks, passable software and modelling expertise have existed for many years. Critics have identified flaws in existing frameworks (McLain and Lee 1996) and software has had limitations, but we believe decision-support projects have failed primarily for two reasons. First, modellers and decision-makers have not understood each other’s objectives and limitations. Decision-makers must often decide quickly, using available information and subjective judgement, and they may have unreasonable expectations about what can be modelled. Modellers, on the other hand, often aim for efficient elegant models, backed by thorough research, and are not unduly constrained by time. Second, influential stakeholders have not been adequately included in modelling projects (McLain and Lee 1996). In one sense, the framework and software we propose simply aims to align the objectives of the modellers and the decision-makers (including stakeholders) and to ensure that technology is used to enhance, and not to replace, human thought and decision-making.

Problems at the landscape scale have a number of challenging characteristics, including large spatial and temporal extents, complex interacting processes that occur at various scales, and uncertainty in the underlying data and in the understanding of key processes. Large areas hold many values, thus bringing diverse interests to the decision-making process. Increasing public involvement in landscape management requires collaborative decision-support processes. Ultimately, decisions are made in a socio-political setting and depend on perspectives as well as information, so decision-support frameworks must be human centred to provide opportunities for a group of stakeholders to discuss perspectives, to learn about social, economic and ecological values in the forest, and to assess possible consequences of management alternatives, with the final goal of a shared vision.

Acknowledgements

We would like to thank Karen Price and Marie-Ange Fall. We would also like to thank participants of the case study projects, in particular those who played a key role in model development: Bill Riel, Les Safranyik and Terry Shore of the Canadian Forest Service, Craig Delong, Fred Hovey, Bruce McLellan, Albert Nussbaum, and Dave Piggin of the B.C. Forest Service, and Glenn Sutherland and Don Sachs.

References

Apps C D, McLellan B N, Kinley T A, and Flaa J P 2001 Scale-dependent habitat selection by mountain caribou in the Columbia Mountains, British Columbia. Journal of Wildlife Management 65:65-77
Baltzer H, Braun P, and Koehler W 1998 Cellular automata models for vegetation dynamics. Ecological Modelling 107: 113-125
Carpendale M S T, Cowperthwaite D, Tigges M, Fall A, and Fracchia F D 1999 The Tardis: a visual exploration environment for landscape dynamics. In Proceedings of the Visual Data Exploration and Analysis VI, SPIE Conference, San Jose, CA. Volume 3643: 110-119
Clark W C, Jones D D, and Holling C S 1979 Lessons for ecological policy design: a case study of ecosystem management. Ecological Modelling 7: 1-53
Derry J F 1998 Modelling ecological interaction despite object-oriented modularity. Ecological Modelling 107: 145-158
Fall A, Daust D and Morgan D 2001 A framework and software tool to support collaborative landscape analysis: fitting square pegs into square holes. Transactions in GIS 5(1): 67-86
Fall A and Fall J 2001 A domain-specific language for models of landscape dynamics. Ecological Modelling 141(1-3): 1-18
Groenwold A and Sonnenschein M 1998 Event-based modelling of ecological systems with asynchronous cellular automata. Ecological Modelling 108: 37-52
Grudin J 1991 Computer-supported cooperative work. Special Issue: Collaborative Computing. Communications of the ACM, 34(12): 30-34
Holling C S 1978 Adaptive Environmental Assessment and Management. Chichester, John Wiley and Sons
Hunsaker C T, Nisbet R A, Lam D C, Browder J A, Baker W L, Turner M G, and Botkin D B 1993 Spatial models of ecological systems and processes: the role of GIS. In Goodchild M F, Parks B O, and Steyaert L T (eds) Environmental Modelling with GIS. New York, Oxford University Press
Itami R 1994 Simulating spatial dynamics: cellular automata theory. Landscape and Urban Planning 30: 27-47
Kyng M 1991 Designing for cooperation: cooperating in design. Special Issue: Collaborative Computing, Communications of the ACM, 34(12): 65-73
Lorek H and Sonnenschein M, 1999 Modelling and simulation software to support individual-based ecological modelling Ecological Modelling 225: 199-216
McLain R J and Lee R G 1996 Adaptive management: promises and pitfalls. Environmental Management 20(4): 437-448
Maxwell T and Costanza R 1997 A language for modular spatio-temporal simulation. Ecological Modelling 103: 105-113
Maxwell T and Villa F 1999 Collaborative modular modeling of environmental systems. WWW Document, http://kabir.cbl.umces.edu/Tom/Papers.html
Safranyik L, Barclay H, Thomson A, and Riel W G 1999 A population dynamics model for the mountain pine beetle, Dendroctonus ponerosae, Hopk. (Coleoptera: Scolytidae). Victoria, BC, Canada, Pacific Forestry Centre Information Report BC-X-386
Schuler D and Namioka A (eds) 1992 Participatory Design. Hillsdale, N J, Lawrence Erlbaum Associates
Selin S and Chavez D 1995 Developing a collaborative model for environmental planning and modelling. Environmental Management 19(2): 189-195
Sklar F H and Costanza R 1991 The development of dynamic spatial models for landscape ecology: A review and prognosis. In Turner M G and Gardner R H (eds) Quantitative Methods in Landscape Ecology. New York, Springer-Verlag: 239-288
de Vasconcelos M and Zeigler B 1993 Discrete-event simulation of forest landscape response to fire disturbance. Ecological Modelling 65: 177-198
Wesseling C G, Karssenberg D, Van Deursen W P A, Burrough P A 1996 Integrating dynamic environmental models in GIS: The development of a dynamic modelling language Transactions in GIS 1:40-48