Marxan User Manual: Introduction

1.1 What is Marxan?
Marxan is software that delivers decision support for reserve system design.WARNING: Plugin disabled TAG!1 The basic idea behind a reserve design problem is that a conservation planner has a large number of potential sites (or planning units) from which to select new conservation areas. They may wish to devise a reserve system which is made up of a selection of these planning units which will solve a problem that includes several ecological, social and economic criteria and principles. Marxan is primarily intended to solve a particular class of reserve design problem known as the ‘minimum set problem’, where the goal is to achieve some minimum representation of biodiversity features for the smallest possible cost (McDonnell et al. 2002). The rationale is that cheaper or less socially disruptive reserve networks are more likely to be implemented. Furthermore, meeting a set of targets for all conservation features provides a solid platform for expanding a reserve system in the future; reserve systems biased to habitats of little commercial value are often hard to expand. In minimum set problems the elements of biodiversity that you wish to conserve are entered as constraints to solutions of the problem (Possingham et al 2000). Given reasonably comprehensive data on species, habitats and/or other relevant biodiversity features, Marxan aims to identify the reserve system (a combination of planning units) that will meet user-defined biodiversity targetsWARNING: Plugin disabled TAG!2 for the minimum cost (Ball and Possingham 2000; Possingham et al 2000).

As an example, a possible biodiversity target could be to ensure that at least 30% of the abundance of every vegetation type is represented in a protected area network. If this protection must be achieved through the purchase of land, then a conservation planner (and politicians) will probably desire a system of reserves that minimises the total monetary cost required for purchasing the necessary land and meeting those targets (Carwardine et al. 2006). Where information on the actual cost of land is not available, reserve area might be used as a surrogate for cost, based on the assumption that the larger the entire reserve system the more costly it will be to implement and manage (although this is not always the case). The cost used in Marxan can also be any relative social, economic or ecological measure of costs, or combination thereof.

The number of possible solutions to even a small reserve selection problem is vast (for only a modest number of 200 planning units there are over 1.6 x 1060 possible ways a reserve system could be configured). Because finding the best solution to this problem is complex and time consuming, computer algorithms have been developed. An algorithm is a mathematical process or set of rules used for problem solving. Two general types of reserve design tools have been devised to efficiently solve “reserve design” problems: exact algorithms and heuristic (non-exact) algorithms. Although exact algorithms can identify the single optimal solution, for large reserve design problems it is difficult (and often impossible) to find this solution in a reasonable amount of time (Possingham et al. 2000; Cabeza 2003). Heuristics, on the other hand, provide a number of good, near-optimal solutions, which not only provide a set of options for planners and stakeholders to consider but can also be generated very quickly (Possingham et al. 2000; McDonnell? et al. 2002; Cabeza 2003). As a result, heuristics are generally preferred over exact algorithms. Marxan is able to find a range of near-optimal solutions quickly (even for very large planning problems), using a powerful heuristic known as ‘simulated annealing’ (Appendix B-2.1). Simulated annealing will generally get much closer to the optimal solution than other heuristics such as the Greedy Heuristic, (Appendix B-2.3.1). If desired, Marxan is also able to find solutions using a variety of less sophisticated, but often faster, heuristic algorithms (see Section 3.2.1.2.1). Marxan is part of a lineage of reserve design tools including its direct predecessor, SPEXAN.

WARNING: Plugin disabled TAG!1Though it can be used for other purposes as well, as noted below in section 1.3.
WARNING: Plugin disabled TAG!2 We use the term ‘target’ to refer not to species or features present in the planning region, but to the desired representation of these inside a reserve system.

1.1.1 Other versions of Marxan
In this manual we only describe Marxan version 1.8.10., the classic Marxan software. There are, however, a number of Marxan variations with modified functionalities that are either available or in development from The Ecology Centre at The University of Queensland (http://www.ecology.uq.edu.au). These include: an optimised version of Marxan for handling very large problems involving greater than 20,000 planning units (people have successfully found solutions to problems with half a million planning units); a version that allows probabilistic information on threats or the presence of conservation features at sites to be included in the reserve design problem; and Marxan with Zones, which is being developed to handle multiple objective zoning. Although the basic operation of these versions is the same as described in this manual, they each have idiosyncrasies which will be described in an appendix provided with each of the versions. While Marxan and its variations are freely downloadable, their development has always been partially dependent on external funding bodies who have stepped forward to support this work.

1.2 Systematic Conservation Planning
Systematic conservation planning is widely considered 'best practice' in conservation because it facilitates a transparent, inclusive and defensible decision making process. Transparency refers to how well people understand the decision-making procedures and output products. A highly transparent planning process will tend to increase the accountability and credibility of conservation planning and decision-making. Inclusive planning processes aim to incorporate information and values from stakeholders to reduce conflicts amongst interests. This, in turn, results in stronger, more widely accepted decisions. Defensibility is derived from the ability of systematic plans to explicitly consider how well a particular selection of reserves meets its objectives, and the validity of the reasoning used to get there. The MGPH discusses each of these principals in detail.

Although Marxan can be used for a variety of purposes at a variety of stages in the systematic conservation planning process, it was designed primarily to help inform the selection of new conservation areas for minimal “cost” and facilitate the exploration of trade-offs between conservation and socio-economic objectives. Marxan can help set priorities for conservation action by highlighting those places that are likely to be important inclusions in an efficient reserve network. Marxan can also be employed as a tool for evaluating the representation and comprehensiveness of existing reserve networks (Stewart et al. 2003).

It is important to understand that the appropriate role for Marxan, as with other decision support software, is to support decision making. Marxan solutions can form the basis of discussions towards a final plan that incorporates additional political, socio-economic and pragmatic factors. Some of the limitations to the use of Marxan are described in Section 1.4 below.

1.3 Questions Marxan can help answer
The backbone of Marxan is to facilitate the efficient selection of subsets from a large set of mapped, spatially constant features. While Marxan was originally designed to ensure species and ecosystem representation in biodiversity conservation planning, and has primarily been applied to that field, it has proven applicable to a broad range of planning challenges. Marxan can generally assist all problems related to the spatially-explicit selection of ‘minimum sets’. For example, it has been used to identify a spatially efficient suite of “fishing areas” (Ban, Personal Communication); and in the field of coastal and marine natural resource management, Marxan has been employed to support multiple-use zoning plans that balance the varied interests of fisheries, transportation and conservation, amongst others (e.g. Fernandes et al. 2005). Chan et al. (2006) have explored the use of Marxan in achieving ecosystem service, as well as biodiversity targets. Some of these applications will require a more creative use of Marxan than we have space to provide guidance on here. We suggest that once you are familiar with Marxan’s basic operation you seek out some of the many published examples of Marxan in operation, as well as consulting the MGPH.

1.4 Limitations of Marxan
Technical limitations of Marxan will become apparent as you read this manual and in many cases can be overcome through data or scenario manipulation (for examples see the MGPH). More subtle and yet more important, however, are the philosophical limitations of reserve design software. These should be well understood. Marxan operates as part of a planning process and is not designed to act as a stand-alone reserve design solution. Its effectiveness is dependent upon the involvement of people, the adoption of sound ecological principles, the establishment of scientifically defensible conservation goals and targets and the development and inclusion of quality spatial datasets. Marxan should be used as part of a systematic conservation planning process and in collaboration with other forms of knowledge. These other forms of knowledge are essential to the refinement of Marxan inputs, the interpretation of Marxan outcomes and the refinement of final conservation area boundaries.

1.5 The Objective Function
In order for Marxan to find good solutions to a problem it must have some basis by which to compare alternate solutions (i.e. collections of planning units) and hence identify good ones. This is achieved through the use of a mathematical objective function that gives a value for a collection of planning units based on the various costs of the selected set and the penalties for not meeting conservation (or other) targets. Thus, a solution containing zero planning units, though cheap to implement, would probably not meet any biodiversity goals and so the objective function value should be very poor. Having an objective function which gives any possible reserve system a value, allows us to automate the selection of good reserve networks (at least according to the objective function). Marxan works simply by continually testing alternate selections of planning units, aiming to improve the whole reserve system value.


The objective function’s value must of course reflect the desirability of that particular reserve system. In its simplest form, the Marxan objective function is a combination of the total cost of the reserve system and a penalty for any of the ecological targets that are not met. This objective function is designed so that the lower the value the better. Marxan also allows a measure of reserve system fragmentation to be taken into account, so that it will generally be desirable for a reserve system not to be too fragmented. Not only will a fragmented reserve system often lead to undesirable fragmentation of ecological communities, it is also likely to make management and compliance more challenging and costly. A more fragmented reserve network will have a greater overall boundary length. It is this boundary length, plus a weighting on its importance relative to the other components of the objective (cost and meeting targets), that can be included in the objective function. The final possible addition to the objective function is a penalty for exceeding some pre-set cost.WARNING: Plugin disabled TAG!3 Although Marxan always tries to find the cheapest satisfactory reserve network, there may occasionally (or frequently) be immoveable fiscal constraints on conservation actions. In these cases we want to ensure the best solutions, given the available budget, will be found.

Thus, the objective function in Marxan takes the form:



1. The total cost of the reserve network (required)
2. The penalty for not adequately representing conservation features (required)
3. The total reserve boundary length, multiplied by a modifier (optional)
4. The penalty for exceeding a preset cost threshold (optional – see footnote 3)

Terms one and three can be thought of as ‘costs’, whereas terms two and four are penalties for breaching various criteria. In general we do not advise using the cost threshold penalty. More detail on the objective function and how each of the different terms is calculated can be found in Appendix B-1. Section 3 of this manual contains details of how to control which features contribute to the objective function and what the size of the penalties will be.

WARNING: Plugin disabled TAG!3Due to sometimes inconsistent results, the Cost Threshold Penalty feature of Marxan is currently being re-programmed. Users of Marxan 1.8.10 should be aware of this. It is recommended that this function be used carefully.

1.6 Primary assumptions
The use of an automated reserve selection tool such as Marxan rests upon some key assumptions. Although it can be very powerful in solving difficult site selection problems, some subtleties, such as knowledge about data quality, cannot always be incorporated, so the use of Marxan must necessarily rest on certain assumptions. Perhaps the assumption most difficult to attain, and thus most frequently violated, is that the spatial distribution of data used in a Marxan analysis is assumed to be consistent. This is not to say that the same features are found everywhere, but that the data was collected in a way that the same features would be found everywhere if they existed there, i.e. the data is not spatially biased. For instance, if using species occurrence data to select reserves, it is highly likely that the detection of species has not been uniform across the planning region. Collections or observations may have occurred more intensively around research field stations, populated areas, or easily accessible places, such as near roads. This will be interpreted by Marxan as a true reflection of the species’ full distribution and will subsequently direct the reserve solutions to those well-studied areas. This may have a substantial bearing on the shape of the entire reserve system, particularly if any emphasis is placed on system compactness. One way of partially overcoming this bias is to model the likely distribution of species or habitats based on biophysical data. An additional and perhaps simpler way to overcome biases due to sampling intensity is to use surrogate measures such as habitat type or even physical variables to represent the distribution of biodiversity we wish to conserve. In some cases, however, it would be irresponsible to neglect known occurrences of valuable conservation features such as highly threatened species. A method of dealing with such situations is provided in Section 3.2.3.3. When it is believed that data may be spatially biased, this bias should be documented.

Marxan does not consider uncertainty in the data.WARNING: Plugin disabled TAG!4 It assumes that all feature representations are true, and that all occurrences of that feature are of equal value. In reality, a conservation planner may be very confident about the presence of a feature in some areas of its distribution and less so in others. Subtleties such as this require careful evaluation of the Marxan outputs to ensure that they actually capture the desired conservation features. Always consider the cliché that the quality of the results you get out of Marxan can be no greater than the quality of the raw or modelled data you put in. The MGPH suggests some methods for conducting robust analysis using weaker datasets.

WARNING: Plugin disabled TAG!4Versions of Marxan under development have some ability to deal with levels of uncertainty.

1.7 Pre-processing of data
Actually running Marxan to generate reserve solutions will generally be the quick part of a conservation planning exercise! Before that, a number of often time-consuming steps must be completed.

1.7.1 Choosing planning units
An essential pre-processing step is to divide your planning region into a set of planning units. A tutorial on some methods to create your planning units is provided in Appendix C-2. In their simplest form, planning units may be defined by overlaying your planning region with a grid of squares or lattice of hexagons. They must capture all the areas that can possibly be selected as part of the reserve system and their size should be at a scale appropriate for both the ecological features you wish to capture and the size of the protected areas likely to be implemented. In general, they should be no finer in resolution than the data on conservation features and no coarser than is realistic for management decisions. There is, however, no necessity to have uniformly shaped planning units. Nor is it always true that smaller planning units are better. In some cases it will make more sense to have planning units that are informed by natural ecological divisions such as hydrological units, or even by political/governmental divisions such as cadastral parcels. For other uses, a uniform planning unit will provide more useful results.


Three possible types of planning units that could be used in Marxan.

There is a limit on the number of planning units that Marxan can handle. This is not, however, a fixed number as it depends also on the number of conservation features you wish to plan for and even to some extent on the power of your computer. Unfortunately we know of no good rule of thumb for assessing this number but we have quite comfortably run Marxan analyses with 10,000 planning units and 100 conservation features. Really big analyses (i.e. >20,000 planning units) should be run using the optimised version of Marxan (2.0.2.) also available from The Ecology Centre website (see Section 1.1.1). The number of planning units and features this version can handle are essentially limited only by available memory.

A great deal of care should be taken when deciding upon appropriate planning units, as it will influence the results of your Marxan analyses. References on this topic are provided in the Key References section. While seldom done, there is no reason why two analyses using different planning units could not be run.

1.7.2 Determining the distribution of conservation features
A second essential step prior to using Marxan is to determine the distribution of conservation features across your planning units. This means assembling all the requisite data on your conservation features and then calculating how much of each feature is located within each planning unit. To do this will generally require some knowledge of a geographical information system (GIS); a tutorial on one way to do this is provided in Appendix C. In most cases, compiling the necessary data and working out the representation of conservation features across your planning units is likely to involve greater effort than running Marxan. Project managers must be careful to allow sufficient time for this step.

---