O.R. & natural resources management

Applying forestry models in Argentina demonstrates huge potential for operations research throughout resource-rich Latin America.

By Gustavo Braier and Javier Marenco

A typical forestry model aims at planning long-term planting and harvesting operations.

Proper management of natural resources is a key feature in Latin American countries, given that an important slice of their gross domestic product comes from the exploitation of land and sea. Although still on an incipient scale in Latin America, operations research can prove a useful tool for successfully accomplishing these tasks. Our experience with forestry models demonstrates a huge potential in Latin America for introducing operations research techniques in this area, at both private companies and public agencies.

A typical forestry model aims at planning long-term planting and harvesting operations within one or more parcels of land. The modeled decisions include when, where and what species to plant, and when and where to harvest in order to meet the expected demand. A usual setting considers a planning horizon of at least 40 years, with annual granularity. Going into a finer granularity does not usually pay off, since the models become quite large and potentially intractable, and the obtained detail does not further enhance the planning process.

Forestry models may also include conversion processes for the harvested wood, thus integrating into the decision process industrial activities that generate manufactured products. The wood demand may come directly from internal or external markets or be pulled from the demand for the manufactured products – when integrated into the model. A typical model may seek to optimize one of the following objective functions:

  • cost of planting and harvesting (given minimum production targets or demand satisfaction constraints),
  • total benefit for the forestry sector or for some selected special companies within the model, or
  • total benefit for the forestry complex, including the manufacturing companies (and in this case reasonable benefits for the foresters must be incorporated as model constraints).
As this area near the Chilean border shows, Argentina is rich in forest land.

Since the decisions are naturally divisible, it is customary to apply linear programming techniques for these applications. Furthermore, imponderable factors – such as weather, quality of roads and availability of resources – ask for approximate planning, so going into more sophisticated modeling techniques is not usually worth the effort. The resulting models may have millions of variables and constraints, which do not impose too high a burden for reasonably priced commercial solvers.

From a computational point of view, the main issue in these applications is the management of huge amounts of data and a relatively high number of different constraints. A major challenge for putting these tools into the hands of unassisted users is providing good interfaces for handling all the data sets and identifying sources of infeasibilities, which are unfortunately quite frequent given the large number of constraints and the interactions among them.

Customers for these tools vary from forestry and manufacturing companies (e.g., paper mills and sawmills) seeking to maximize their benefits while ensuring a steady provision of wood, to local governments looking for good stimulus policies for the forestry sector or planning infrastructure related to future changes, to regulatory agencies trying to assess the impact of their policies.

The Commercial Side

Commercial applications for forestry-based companies usually aim at planning the company’s planting and harvesting activities, taking into account the estimated demand for wood. The demand can come either from market considerations for wood or for manufactured products, and typical constraints include demand satisfaction, internal policy requirements and growing yields. Forestry managers are the primary users of these tools, and the main benefit for them is the assurance that the obtained plans are sustainable in the long term. Indeed, a main risk for manual planners is to make myopic decisions that compromise the future supply of wood in the long term. In contrast, the model naturally avoids such issues since it includes constraints for ensuring that the demand is fully covered within the planning interval.

Usual customers of these tools are big paper mills and sawmills that often have their own forests. Besides the management of the company forests, the models may include the purchase of raw wood in case the supply from the company forests is not enough to cover the planned demand. Models may also include other major players in the local forestry market in order to take into account their expected policies and market share.

Sometimes unexpected results provided by a model may activate latent tensions within the company, and handling this situation may be a difficult task. We actually had to face such an issue in a project for a private company: The model’s outcome fully supported for the proposal of one of the planners, which was opposed by other planning team members. In this situation, reassuring the users that the model was impartial and “free to decide” the best option was an important task. Eventually the rationale leading the model to the initially unexpected outcome was understood by the whole team and is now accepted as a common-sense practice within the company, but in general the initial outcome may come as a blow, and the practitioner’s conciliatory skills may be put to the test.

An example of the model failing to convince the decision-makers surfaced in another project for another private company. In this case, the model was employed to analyze whether buying additional land was convenient or not. This evaluation was not trivial because buying land would imply a lower supply from other foresters, hence a model-based approach was ideal for the analysis. The model-based study showed that buying additional land was very convenient. The results were shared with the company board, which rejected the proposal. After several years, real data showed that buying additional land would have been very convenient for the company but, unfortunately, the land prices were much higher by then and the decision could not be reverted.

Assisting Government Policymakers

As mentioned earlier, a second source of interesting forestry applications comes from government agencies. Interesting challenges appear when an operations research practitioner interacts with policymakers. The usual practitioner’s objective is to model reality carefully but without over-modeling (so the resulting model is computationally tractable). In our experience, the main contribution of such a modeling process within a policymaking environment usually lies in providing a framework for analysis rather than providing actual figures or precise production plans. The interaction between the practitioner and the policymakers is crucial and must develop into a process of mutual learning and trust-building. In this case, the labor of the practitioner is not limited to installing an out-of-the-box computational tool, but it includes providing the policymakers a framework for conceptual thinking and thorough analysis.

Such an example appeared in a project carried out with a provincial government in Argentina, which was interested in promoting forestry investments within the province. Oddly enough, the model suggested not to promote planting anymore, but to focus the efforts in promoting the installation of manufacturing companies and providing them with suitable infrastructure. This was a major blow to the planning team. Similarly, the provincial government was pushing hard for constructing a bridge across a main river with a neighboring province, but the model, based on reasonable demand forecasts, proposed to build a harbor instead. After some further analysis these conclusions eventually became a natural outcome of the planning process (since huge amounts of wood with no demand do not make much sense, and a harbor is much preferred over a bridge if most customers are overseas), but they certainly were not clear before the modeling process started.

In this sense, the modeling process provides a framework to absorb the impact and to analyze such unexpected model outputs, giving the policymaker quality information about the dynamics of the market being promoted or regulated, and giving the practitioner a deeper insight on the reality being modeled. At a later stage, the (refined) model can provide precise figures, but in our experience the greater added value for the decision process is gained within the “learning” stage.

A similar project was carried out for another province with no tradition in forestry. The province was interested in stimulating a forestry complex along a major river. The introduction of a model – albeit rudimentary and over-simplified in a first stage – allowed for an enriched discussion to take place that was focused on real data and measurable impacts (as provided by the model results) instead of politicized or pre-formatted arguments. Arguing “within a model” is much more productive than throwing unsupported arguments. A model provides an even set of rules for arguments and allows for an impartial assessment of each proposed policy, which is usually translated into model constraints.

Challenging Interfaces

Providing a suitable interface for properly visualizing the model solution is not a straightforward task. Since most data sets are given by single values indexed by many sets (e.g., the expected amount of wood in tonnes per hectare depends on the species, the location, the planting method and the tree age, among others), a pivot table is in many cases the ideal visualization tool. However, for many analyses just a simple table or a graph plotting summarized values is enough. A reasonable compromise may be attained by linking a pivot table to a dynamical graph, so the skilled user can generate his or her own visualizations starting from raw data.

However, this is just scratching the surface. Good visual aids for enabling fruitful discussions must go beyond plain pivot tables and old-styled graphs, especially when non-technical people are part of the team.

Forestry data are naturally displayed in maps, since the available fields are usually stored in a geographical information system, and it is not difficult to export this information to other computational tools. A natural approach consists in showing the areas planted and harvested, the location of the manufacturing facilities and their production load, and the outgoing points for manufactured products at each period. However, in our experience this kind of output only allows for highly summarized information to be provided, and may not be useful in planning meetings (although it may be a high-impact interface when first presented to the users).

When non-technical users are part of the discussion, more creative interfaces may be useful. For example, an animation showing that harvested logs are being uselessly stocked since there are not enough manufacturing facilities or convenient markets for raw logs is much more telling (and a more powerful message) than a growing sequence of numbers in a table. Similar resources can be employed for showing other model results as, e.g., the construction of new manufacturing facilities or the social impact of the proposed policies. Such interfaces act as facilitators for discussions, in the same manner as commented in the previous section.

Searching for new and creative ways of presenting the model solution to non-technical users is an underexploited line of research in the forestry industry. Existing development tools allow for easy deployment of applications with pivot tables, dynamical graphs and interactive maps, but they lack development environments for providing easy-to-build animations integrated into user interfaces.

Beyond Forestry

In Argentina and elsewhere in Latin America, the application of O.R. can be useful for the management of natural resources.

Although our primary experience lies within forestry models, there is a huge potential for operations research in the management of many other natural resources:

  • Sugar cane crops are subject to renewal decisions, which are usually annual. Each year the sugar cane is harvested and grows again for next year, but the yields decrease as the plants age. At some point, the decision to remove the plants and soil new seeds must be taken, and the main inputs for this decision are the historical yields per age of each crop. In our experience, linear programming models are appropriate for tackling this problem and are quite integer-friendly in case all-or-nothing decisions must be taken.
  • The management of other crops such as tea, grapes and olives is also subject to renewal decisions, although on a larger scale and is thus similar to forestry management. Again, in this case the objective is to plan the planting and renewal in such a way that a sustained demand is guaranteed and the costs are minimized. We are not aware of applications on these particular topics in Latin America, and there seems to be interesting potential here.
  • Seasonal crops such as soybean, corn and wheat have not received much attention from operations research but are nevertheless one of the main engines of the economies of Latin American countries. In this case, the main decisions do not concern what to plant or when to harvest, but involve logistic considerations such as truck movements, warehouse management and sales strategies. Some incipient approaches have been performed in Argentina with these kinds of crops, but the introduction of model-based techniques for decision-making certainly needs more effort.

To summarize these observations, the application of operations research can be very useful for the management of natural resources, both from a commercial and a policymaking point of view. The modeling techniques need not be very sophisticated, as in our experience the decisions are quite well modeled with linear programming. These issues are particularly interesting for Latin American countries, and we encourage practitioners to unleash the huge potential awaiting within this field.

Gustavo Braier is an economist and holds a master’s degree in forestry from the University of Toronto, Canada. He is an operations research practitioner and partner at (, a software development company based in Argentina and devoted to providing operations research solutions for natural renewable resources and the industrial sector throughout Latin America. 

Javier Marenco is a computer scientist and holds a Ph.D. in computer science from the University of Buenos Aires, Argentina. He is an operations research practitioner, project manager and software developer at and an assistant professor at the University of Buenos Aires and at the National University of General Sarmiento (Argentina).