FLORES is intended to be a model to help explore the consequences at the landscape scale of policies and other initiatives intended to influence land use in tropical developing. It seeks to provide an accessible platform to foster interdisciplinary collaboration between researchers and resource managers, and to facilitate empirical tests of hypotheses and other propositions.

1. Introduction

Policies and incentives to promote sustainable forestry and better land use do not always achieve the desired effect. Proponents rarely foresee all the consequences, and that those best able to offer alternative views may be unable to contribute to the decision-making process. This leads to inefficient and sometimes counter-effective initiatives. How can we better equip policy makers and their advisors to envisage fully the efficacy and consequences of initiatives? not covered by plot data or other models.

Here's an analogy: What makes air transport so safe and pilot error so rare? Good design, careful planning, diligent maintenance and competent supervision are factors, but training is crucial. Before crew members take the controls of a commercial airliner, they will have studied the theory of flight, trained in light aircraft, spent hours in a flight simulator, and flown with more experienced colleagues. They know how each meter, indicator and control device should be used. They know how to respond when something goes wrong. And they rarely need to use their training, because an autopilot takes care of most problems.

Now contrast this with our management of forests:

• Do we know what to do when things go wrong?
• Can we tell when things are beginning to go wrong?
• Do we know which controls we can use to change things?
• Do we know what the controls are, where to find them, and how to activate them?
• Can we recognize and interpret the indicators?

Why don't we have an "autopilot" to give advice?

Why is it that so many amongst those who make important decisions about the world's forests have never raised a tree, tended a garden, gathered food from the forest, or even used a simulator to explore the implications of a decision? Would a forest landscape simulator make a difference?

2. A Forest Simulator

The popular computer game SimCity provides a useful analogy of the forest simulator I envisage. The game offers the player an "aerial view" of a city, a menu of policies and incentives (e.g., expenditure on education, transport, sanitation, etc.), and indicators of performance (e.g., unemployment, GNP, pollution, etc.).

A forest simulator would replace the cityscape with a landscape of forest and non-forest land. Its menu would include a range of options to manipulate the forest and land use patterns, such as tenure, transport, and subsidies. Performance indicators could include biodiversity and rural poverty. Such a forest simulator should have a strong factual basis, and could be customized to suit different situations. It would:

• Synthesize existing knowledge and identify gaps and other deficiencies
• Express present knowledge concisely, completely, explicitly and unambiguously as a model
• Create a framework to promote collaborative interdisciplinary research
• Provide a basis for strong empirical tests of hypotheses relating to land use policy
• Create a planning tool to allow planners and policy makers to explore future scenarios
• Provide an educational game to improve general knowledge of tropical forest environments.

3. Flores as a Model

FLORES aims to improve our understanding of land use patterns in time and space, especially in forested landscapes, and to facilitate quantitative analyses of policy options intended to manipulate these patterns. FLORES is spatially explicit, and operates at the landscape scale, spanning both forest and agricultural lands. Agricultural lands and villages form a critical component of the landscape, and must be modeled to fully understand the processes at work in and near the forest.

The basic concepts of this work are not new; what is new is the way concepts are integrated and applied. FLORES seems most closely related to work by Bousquet et al. (1993, 1994), who constructed a multi-agent simulation (MAS) model of an inland fishery in the Central Niger Delta as basis for focusing discussion, evaluating options and formulating recommendations. There is an interesting contrast between FLORES and MAS: both are concerned with agents that can modify and respond to their environment, but the emphasis differs. Generally, MAS attempts to find the simplest set of rules that can reproduce a particular pattern from a defined scenario. In essence, the usual question for MAS is "What are the rules that might explain this pattern that we have observed?" FLORES considers the converse: Given what we know about human behavior, can we predict future outcomes for a range of scenarios?

Generally we do not know what future outcomes should look like, except in a few specific cases that may be used to test the model. FLORES also recognizes that people may have complex reasons for their behavior, and attempts to represent our present understanding of those reasons, rather than seeking the simplest rules that may reproduce a given pattern. It is inevitable, and quite deliberate, that the initial version of FLORES will be simplistic. We are building a platform that will be the basis for on-going work over a long period.

Assumptions, Actors and Activities

FLORES relies on four basic assumptions (Vanclay 1995):

1. Actors, individuals or groups of individuals who collaborate as families, clans, associations and corporations create land use patterns.
2. These actors make rational decisions based on available information, obligations and expectations, social as well as economic. Note that an actor's perception may be more important than reality. For example, doubt about security of land tenure may lead an owner to adopt a shorter time frame than would otherwise be the case.
3. When choosing an activity, actors explore all options available to them, within the constraints imposed by resources (land, time, capital, etc.), knowledge, and their comfort zone (cultural attachments, willingness to attempt novel activities, etc.).
4. Actors tend to undertake activities that maximize expected benefits or minimize anticipated risks to themselves and their beneficiaries (families, clans, shareholders, etc.). It may be possible to model both benefit-seeking and risk-avoiding behavior by considering risk-adjusted benefits.

The constraints implied by an actor's comfort zone and previous experience mean that many actors consider a rather small number of activities, often only those done in the past, plus a few new activities pursued profitably by neighbors. However, there are usually a few innovators who consider an extended list of activities and may attempt a diverse range of enterprises. Disposition is only one determinant of willingness to accept risk: age, assets and income also feature prominently in many explanations.

The fourth assumption deals with benefits and utility functions. These benefits may be expressed in dollars, or in other quantitative ways. Maximizing benefits may be realistic for some communities, but is only one way to represent behavioral tendencies. The role of FLORES is to provide a way to calibrate and test alternatives, and to establish which alternative is most consistent with the available evidence. Note that decisions may depend on many things, including:

• anticipated yields of an activity (e.g., cropping, hunting, handicraft, share-farming, wages, etc.)
• anticipated prices, net of costs incurred in initiating (e.g., seed, fertiliser, raw materials, etc.) and realising a return (e.g., harvesting, packing, transport, marketing, commissions, etc.), discounted as necessary for any delays;
• reductions for real or imagined risks including pests, disease, fire, theft, loss of tenure, spoiling during transport, viability of an employer, etc.allowances for shares that others may have in the activity, including for example, clan
• obligations as well as landlords who may share revenues but not costs satisfaction experienced by an actor in producing an item.

The decision made for any particular resource is not independent of decisions made for other resources, since price and risk may depend on total production across all resources, and many options may have off-site impacts such as erosion and pollution. Lagged adjustments may also be needed to account for time taken to learn and implement new technologies and to meet transition costs in adopting the technology. However, in the initial prototype of the model, we may avoid these complexities by making the prevailing market prices exogenous to the model, and assuming that they remain constant. We can then assume that decisions on any site are independent of other sites, and the utility function can be solved without taking topology into account.

Decision-making by actors is just one component of FLORES, and several other sub-models are needed to predict the growth of trees and crops, changes in the soil and water balance, interactions between key plant and animal species, and other ecosystem processes. Fortunately, many such models already exist (e.g., Vanclay 1994, Anon 1997), and some are amenable to calibration and integration within the FLORES framework.

4. Implementation

FLORES deals with land, people who interact with that land, and the land-use and related decisions that they make. The landscape is made amenable to modeling by tessellating it into land units that are relatively homogeneous with regard to key parameters such as tenure, vegetation, accessibility, and soil fertility. Utility functions deal with actors, resources such as land and capital, and activities such as clearing land, planting crops, hunting, making things, working for wages, and so on. We assume that actors compile a "menu" of possible activities from which they select the item that appeals most under the prevailing circumstances. For many individuals in a forested landscape, such a menu could be relatively small, comprising those activities that have been entertained in the past, or which have proved successful for neighbors. The innovators may have a very extensive menu, and a stochastic selection may be appropriate, at least for their novel enterprises. Here's how FLORES could be implemented in an algorithmic language such as Fortran, Pascal or C.

Initialize land attributes, actor profiles, potential activities For each time_step until end_date For each land unit Update land attributes by predicting natural changes (growth, erosion, etc.) Next land unit. For each actor, taken in "pecking order" Formulate the feasible subset of activities by examining thresholds Rank feasible activities in order of decreasing utility Attempt each activity in order of utility until all resources are used Update land units impacted by this activity Update actor profile (assets, etc.)Next activity Next actor. Report intermediate results Next time_step.

However, this algorithm may make decision-making by actors too structured, because unless it is programmed carefully, decisions will be taken in turn, with each actor having sufficient time to reach a decision. Most algorithmic languages imply a serial representation, whereas in reality, decisions are taken in parallel. FLORES should allow all actors to make decisions simultaneously, sometimes independently, sometimes in concert, with some options foreclosing for those who are too slow to decide.

5. Outputs and Clients

Too many models languish under-utilized, because they do not satisfy the needs of potential users and because system developers did not explicitly contact clients, ascertain their needs, and stimulate their interest. To encourage uptake, potential users must be involved in the development of the model.

FLORES will provide a range of outputs to suit different user requirements. One output will be the forested landscape of a SimForest implementation. The single greatest contribution that information science could make for conservation and wise use of forests would be to construct a virtual reality interface for a forest management system. This could allow a minister and his advisors to don a virtual reality headset and take a "magic carpet" ride over a forest estate. They could observe the spatial pattern of their forest and watch how it changes over time, and under different scenarios. They could "zoom in" to examine particular issues, and stand back to get overall perspective. The technology to do this exists, and it is possible to link forest inventory systems, growth models, geographic information systems and virtual reality systems in this way. However, it has not been done at this time, and awaits further software and hardware development to make it more affordable. However, the SimCity-style interface of FLORES is adequate for many applications, and would be particularly useful for educational applications and general information dissemination.

6. Challenges

There are several specific problems that need to be addressed before this model can be realized as anything more than a simple prototype.

Most utility functions appear innocent enough, but they require a lot of data: anticipated yields and prices of all possible crops under a range of situations, detailed tenure and demographic data, and a good understanding of the socioeconomic culture of the community. This is a major undertaking, and may be one limitation of the model. We envisage that initial prototypes will be restricted to a limited geographic area. Crop yields may be inferred from models, but price and elasticity must be gleaned from field survey work.

Superficially, the model appears tractable, but it involves many challenges. Is it really possible to quantify the social profile of all actors in a community in sufficient detail to provide meaningful predictions from a simple utility function? There is no clear answer.

Two further issues for methodological research are evident at this stage: whether to model individual actors or classes of actors, and how to quantify risk and willingness of actors to accept risk.

It is presently assumed that an actor's willingness to accept risk can be quantified, in part through the historic variation in benefits accruing from a particular activity, and from the actor's age, tangible assets and income. However, this assumption warrants closer scrutiny since attitudes to risk have a major influence on land use decisions. Our ability to quantify risks and attitudes to risk will have a major influence on the accuracy of FLORES predictions.

Satisfactory ways to value the intangibles involved with land use decisions pose a major challenge. One particular aspect that needs to be addressed is how to value prestige. Prestige may take many forms, and may explain land purchases at prices inconsistent with production (e.g., prestige of owning a bigger estate), herd sizes (e.g., prestige of large flocks leads to overstocking, even though smaller flocks may offer equivalent returns and lower risks), and possession or production of certain items.

A FLORES-type model is easy to conceive for a small village, where we can simulate every individual actor. However, when we scale up our efforts to model larger landscapes, it may become impractical to examine decision-making by all actors, and it may be necessary to extrapolate from a sample of actors. The choice of sample may be critical to the outcome, and suitable sampling strategies must be investigated before the approach can be scaled-up to the provincial or national level. A crucial part of this investigation will be to identify the minimum essential set of prime determinants.

In theory, it is possible to conduct experiments to gather rigorous data to test FLORES, but there are ethical questions that would need to be considered carefully. For example, is it feasible to go to a village and buy locally produced goods at prices higher than the prevailing market rate, and watch how the community responds? Fortunately, this experiment is not necessary, because in many developing countries, governments conduct such "experiments" all the time. For instance, new bridges and roads can markedly change transport costs. Thus the data required for model testing may be obtained by strategically choosing and monitoring selected communities over an extended period.

7. Prognosis

Perhaps the best test of a model is how well can the modeler answer the questions 'What do you know now that you did not know before?' and 'How can you find out if it is true?' FLORES has many limitations, but it provides a fertile test bed for ideas, and offers ample scope for furthering our knowledge of policies, incentives and land use patterns in forested landscapes. We need the product, and we need the process. We need to bring together scientists from diverse disciplines to work towards a common goal. We also need to add more rigor to forest policy research. FLORES can help realize it.

8. References

--. Crop Yield Forecasting Methods: Proceedings of the seminar, Villefranche-sur-Mer, 24-27 Oct 1994. Eurostat, JRC-ISPRA, DG VI and FAO, ECSC-EC-EAEC, Luxembourg, 1997.

Bousquet F, Cambier C, Mullon C, Morand P, Quensiere J, and Pavè, A, 1993. Simulating the interaction between a society and a renewable resource. Journal of Biological Systems 1:199-214.

Bousquet, F, Cambier C, Morand P, 1994. Distributed artificial intelligence and object-oriented modelling of a fishery. Mathematical and Computer Modeling 20:97-107.

Muetzelfeldt, R.I. and J. Taylor, 1997. The suitability of AME for agroforestry modeling. Agroforestry Forum 8(2): 7-9.

Vanclay, J.K., 1993. Inventory and yield prediction for natural forest management. In: Thwaites RN and Schaumberg B.J. (eds) Australasian Forestry and the Global Environment. Institute of Foresters of Australia, Brisbane, pp. 163-169.

Vanclay, J.K., 1994. Modelling Forest Growth and Yield: Applications to Mixed Tropical Forests. CAB International, Wallingford, England. xvii+312 pp.

Vanclay, J.K., 1995. Modelling to explore land use patterns at the forest edge: objectives and model design. In: P. Binning, H. Bridgman and B. Williams (eds.) Proceedings, International Congress on Modelling and Simulation (MODSIM'95), 27-30 November, University of Newcastle, Australia, pp. 1:113-116.

Vanclay, J.K. and J.P. Skovsgaard, 1997. Evaluating forest growth models. Ecological Modeling 98: 1-12.

9. About the Author

Jerome K Vanclay, a CIFOR associate, is a professor at the School of Resource Science and Management at Southern Cross University - Lismore in New South Wales, Australia.

Contact Jerry Vanclay at JVanclay@scu.edu.au for more details.