M8: Place Prioritization

 

Learning Objectives: This module describes the process of prioritizing areas on the basis of what they contribute to the biodiversity representation targets for a region.

 

      The purpose of place prioritization is to rank areas in terms of their quantitative contribution to achieving representation targets for biodiversity surrogates.

      Such a procedure makes sense only if there are quantitative targets of representation for all surrogates.

      Place prioritization has previously been known by the name "reserve selection" sometimes called "conservation area network (CAN) selection".

       What is critical is that all areas/ places/ sites are prioritized with respect to their importance for biodiversity conservation, not necessarily selected for any specific management plan.

       Different terminology have different nuances: "reserve" is taken to imply only one specific kind of management viz., exclusion of human activity - this is why that term is best avoided.

 

      There are two basic representation problems that differ based on whether there is a budget or a maximum area that can be put under protection.

      Area minimization problem: find the minimum set of sites in which all surrogates meet their targets.

       One can generalize the problem by replacing number of sites with total area (this is relevant if cells have different areas).

       There is no maximum budget or number of sites that can be selected.

       Technical name: Expected Surrogate Set Covering Problem (ESSCP) (Sarkar et al. 2004b).

      Representation maximization problem: given a maximum number of sites that can be put under conservation, find the set of sites that maximizes the number of surrogates that meet their targets.

       One can generalize the problem by replacing number of sites with total area of sites (this is relevant if cells have different areas).

       Technical name: Maximal Expected Surrogate Covering Problem (MESCP) (Sarkar et al. 2004b).

      Both problems are routinely encountered in practice during conservation planning.

      The terminology used here for the different problems is that of Sarkar et al. (2004b) as there is no standard terminology yet.

 

      Place Prioritization Algorithms (PPAs):

      Algorithms are step-by-step procedures that can be carried out by machines (computers).

      Use of algorithmic computer-assisted design for decision support is central to systematic conservation planning.

      Three types of algorithms:

       Exact or "optimal" algorithms: guaranteed to achieve optimal (or most "economical"—sometimes called "efficient") solutions to the area minimization and representation maximization problems.

       Heuristic algorithms: use various rules such as complementarity to select sites. This provides solutions that are typically economical but are not guaranteed to be optimal.  Typically these algorithms are fast (computationally "efficient") which is why they are often preferred in practice over optimal algorithms.

       Metaheuristic algorithms: repeatedly use heuristic rules to get solutions that get better and thus closer to the optimal solution. (e.g., simulated annealing, tabu search, genetic algorithms.)

 

      In some planning contexts, heuristic algorithms may be preferred to optimal algorithms. For instance, for place prioritization problems with constraints on the spatial configuration of the selected sites, significantly more time may be required to obtain an optimal solution (for example, using generic branch-and-bound algorithms) than would be required to obtain a heuristic solution (for example, using tabu search). However, the running time of a branch-and-bound algorithm tailored to the particular place prioritization problem may well be competitive with that of the heuristic algorithm (in addition to being as economical as possible). In general, the efficiency and economy of heuristic and optimal algorithms needs to be compared in each planning exercise. Comparisons of this sort are now becoming standard practice in systematic conservation planning.

      Technically, the area minimization and representation maximization problems are solved by mathematical programming algorithms ( for "mixed integer-linear" [MIP] problems):

      Problems are NP-hard: this means that these are among the hardest computational problems to be solved. (Here hardest means most time-consuming.)

      The only advantage is that it can guarantee that the optimal solution is found.

      Nevertheless, optimal algorithms are useful in some circumstances-see Example 8.1.

 

 

 

      Four characteristics of well-designed heuristic algorithms are:

      Transparency: each heuristic rule must have a biological interpretation.  If a site is lost, we should know what the effect is so that the relevant biodiversity features can be recovered by inclusion of other sites.

      Efficiency: heuristic algorithms are typically chosen to be as fast as possible.

      Flexibility: algorithms should not be restricted to particular geographic regions or surrogate types.

      Modularity: different heuristic rules should be incorporated in such a way that each can be used independently as well as in combination with the others.

 

      Disadvantages of heuristic algorithms:

      They may sometimes may produce noticeably suboptimal solutions.

      In all uses of heuristic rules, for economy/ optimality, a redundancy check should be carried out at the end: for each site included in the network, test to see whether it can be dropped without any of the surrogates falling below its targeted representation. Redundant sites can be dropped to increase economy/ optimality of the solution.

 

      Use of heuristic rules:

      A variety of heuristic rules are used, including complementarity, rarity, and adjacency (see below).

      Typically the rules are used hierarchically. First one rule, say complementarity, is used. If there are ties, these are broken using the second rule, say rarity. Ties remaining at the end are typically broken by random selection.

 

      Significance of complementarity:

      Complementarity measures what a new site adds to the representation of biodiversity already present in an existing set of sites.

      Complementarity is thus a measure of beta-diversity, focusing on differences between sites.

      In general, complementarity must be assessed relative to explicit targets for each biodiversity surrogate. Thus, what is at stake is how much a new site adds to the representation of only those biodiversity surrogates which do not have their targets already met in the existing set of sites. 

      When no site is yet selected, complementarity is measured by richness.

      Complementarity versus richness: once sites are being selected, complementarity is not the same as richness. Two sites may both have very high richness but very similar composition (which surrogates occur in each of them). Then, both may not have a high value for complementarity.

      Note that the complementarity value of a site must be updated as each site is selected.

      Throughout, complementarity must be assessed quantitatively for use in systematic conservation planning.

       At least 12 measures of complementarity have been proposed.

       The simplest (and most commonly used) measure: for presence-absence data, count the number of occurrences at a site of surrogates that have not yet met their targets. This will be called standard complementarity. Box 8.1 shows how this measure of complementarity is calculated.

       Complementarity-based algorithms can also be run backwards.Start will all cells in solution and then iteratively drop least valuable cell from the point of view of complementarity.

 

 

 

 

 

 

      Significance of rarity:

      After complementarity, rarity is the most commonly used heuristic rule for place prioritization.

      Rarity is typically interpreted as geographical rarity (the inverse of a taxon's range).

       This concept of rarity can be used to capture endemism-which is known to be important in identifying taxa of conservation concern.

      Three different criteria have been used to define rarity.

       Geographical rarity (see above).

       Abundance rarity, measured by the total number or abundance of a taxon. Even geographically widespread taxa can be rare in this sense, for instance, in the case of many large birds.

       Niche specificity: some taxa can occupy very little of a habitat because of ecological requirements.

 

      Using complementarity and rarity together:

      Joint use is known be economical, computationally efficient, and transparent (Csuti et al. 1997).

       For binary data rarity-complementarity seems to work best with respect to economy/ optimality of solutions.

       For probabilistic data complementarity alone has been found to work better than rarity-complementarity (Sarkar et al. 2004a, b).

 

      Other heuristic rules are also used and often to break ties after the use of complementarity and rarity.

      Adjacency: This is giving preference to a site that is adjacent to one that has already been selected over a site that is not so adjacent.

       The main reason for its use is that it results in larger, connected conservation areas—this is the chief way to select for size in heuristic algorithms.

       If the landscape is divided into a (roughly) rectangular grid, lateral adjacency can be distinguished from diagonal adjacency. The former is typically given preference over the latter.

      Various measures of alpha-diversity have sometimes been used.

       Both Shannon and Simpson indices incorporate some evenness considerations.

      Many other heuristic rules, pertaining to size, shape, connectivity, spatial configuration, dispersion, etc., of conservation areas have been proposed.

 

 

Example 8.3 Place Prioritization for Québec

In the early 1990s, Canada's provinces, including Québec, announced their intention to create representative networks of conservation areas comprising at least 12 % of the area of each province. Québec was supposed to achieve this target by 2000. By 1998 Québec had only 63 960 km.2 or 4.2 % of its land area under conservation management. Sarakinos et al. (2001) developed a nominal conservation area network for Québec. The analysis was done using an algorithm that selected sites primarily on the basis of complementarity, with ties being broken using adjacency. Their analysis, besides prioritizing sites for conservation action, also showed how place prioritization can be used to inform social policy - the last aspect will be emphasized here. Québec (approximately 1 522 842 km.2) was divided into 21 403 cells at a 12´ × 12´ longitude × latitude resolution. As (true) surrogates for biodiversity they used 400 faunal and floral species at risk (346 plant species and 54 animal species), 22 native small mammals, six game mammals, and 92 fish species. Data on the distribution of species at risk were obtained from the Québec Natural Heritage Data Centre of this Ministry of Forests and the Environment. Small and game mammal and fish data came from other governmental agencies. The ready availability of so much georeferenced data on species at risk is an important aspect of this analysis. The data were a mixture of presence-absence and presence-only records. No data treatment was attempted. The rationale was that, for species at risk, false presences, as would at least occasionally be generated by modeling treatments, constituted an unacceptable risk. All data were treated as presence-absence. The rationale was again precautionary: if areas selected also included presences of unrecorded species at risk, these would only be even better represented in the conservation area network. Figure 8.3 shows the areas selected using the algorithm discussed there with a target of 50 representations for each of the species at risk, and 250 representations for the other species. In Figure 8.3 the arrow shows a band of selected cells in the north-west. This band only begins to be selected when targets of representation are set at all records for species at risk and 200 representations for all other species. They become continuous at a target of 250 representations. This band runs through boreal forest which is the most northerly and abundant of Québec's three forest zones and straddles the Canadian Shield and upper lowlands region of the province. Because of the milder climate, the diversity of organisms is also higher, with approximately 850 plant species and 281 vertebrates. The boreal forest covers 27 % of Québec and only 0.1 % is under any legal protection though 87 % of it is owned by the provincial government. The main threat to it is due to logging; most logging in Québec occurs in this ecozone and about 80 % of logging in Québec still uses clear-cutting methods. Logging contributes only 1 % to Québec's GDP even though about 0.5 % of the forests are logged annually producing about 80 000 jobs. In the 1990s logging of the boreal forests had emerged as an issue of public concern. This analysis showed that, while these concerns were partly justified, since these areas are not selected when species at risk are used as surrogates, logging could be phased out gradually over a period of years, allowing alternative employment to be found for the affected people, without endangering species.

Figure 8.3 Boreal Forests of Québec For explanation, see Example 7.3.

 

      Subsidiary concepts:

      Irreplaceability: this is the extent to which the removal of a site from a set of potential conservation areas constrains possible networks.

       This becomes important when not all selected conservation areas can simultaneously put under a conservation management plan- as is typically the case due to budget considerations.

       It is almost impossible to compute exactly in practice because of computational complexity.

       Various statistical and other estimators for irreplaceability have been proposed.

       The most common one is the frequency with which an area is selected using complementarity.

      Conservation value:

       Sometimes complementarity is combined with a socio-economic or other measure, using trade-off analysis.

       This uses a mixture of biological and socio-economic criteria.

       An alternative is to use multi-criteria analysis - see M11: Multi-Criteria Analysis.

      Software Packages:

      Except for Target and WorldMap, all of these packages can be freely downloaded. WorldMap can be downloaded freely with some restrictions.

      C-Plan.

       C-Plan uses complementarity.

       It is well-integrated into an ArcView-based GIS protocol.

      Marxan, Spexan, and SITES.

       This is a family of software packages that use simulate annealing to solve the representation problem.

       Critical innovation: it allows the incorporation of some shape considerations in the design of CANs.

       However, (i) it cannot handle large data sets and (ii) is typically quite slow (computationally inefficient).

       SITES runs Spexan/ Marxan from within ArcView 3.x.

      ResNet, ResNet-GUI, and Surrogacy.

       Heuristic packages using rarity and complementarity (in that order).

       Computationally the fastest of the packages that exist at present.

       The software can handle probabilistic data.

       Surrogacy uses ResNet for surrogacy analysis - see M5: Surrogacy Identification and Analysis.

       ResNet-GUI runs ResNet from within ArcView 3.x.

      Target.

       Target uses trade-off analysis to accommodate one other criterion besides biodiversity.

       It is the only package that combines complementarity and economic cost for place prioritization.

       It comes with its own graphics for visualization.

      WorldMap.

       WorldMap uses complementarity.

       It comes with its own sophisticated graphics for visualization, including maps of all continents.

      Running ResNet over the ConsNet Portal: ResNet runs on a server at the University of Texas Biodiversity and Biocultural Conservation Laboratory.

      This requires a user account and password.

       To obtain these, follow the instructions at www.consnet.org.

      Each session requires a name.

      It has the following requirements:

       Two uploaded data files are necessary: one the "ResNet Input file" and the other the "Target" file.

       Target files can be created on the portal, provided that the targets can be set uniformly as a percentage of the occurrences of the surrogates.

       The user must know the number of sites/cells and the number of surrogates - these are used for internal error-checking.

      To upload the input and target files:

       Select "Browse" and search for a file that has not been previously uploaded.

      To create a target file from the input file:

       Select "Create Target Files" from the list of options in the left column of the screen.

       Upload a ResNet input.

       Enter the number of cells and the number of surrogates in the file.

       Provide a name to assign to the target file.

      To run ResNet:

       Select a name for the ResNet session.

       Select both a ResNet input file and a target file.

       Specify the number of cells and surrogates in the input file.

      There are three output files:

       The log file provides a record of the ResNet run.

       The GIS file provides a list of cells selected in the run, indexed by their longitude and latitude.

       The parameters file provides a list of the parameters selected in the ResNet run.

       Finally, there is a file with graphics output. That output is shown in Figure 8.5

 

Example 8.4 Place Prioritization for Namibia Using Termite Data and the ResNet on the ConsNet Portal

Namibia was divided into 1 250m cells at a 1 ° × 1 ° longitude × latitude resolution. 35 termite genera were used as biodiversity surrogates. (This analysis is intended entirely as an academic exercise, with no implication for actual policy.) For this analysis targets of 10 % and 20 % of the distribution were used; the graphical output is shown in Figures 8.4a, b respectively.

Figure 8.4a Graphic Output from the ConsNet Portal

Figure 8.4b Graphic Output from the ConsNet Portal