Site-Selection Algorithms
By Reed Noss
So, how do we put all this information on special elements (e.g., imperiled species and the sites where they occur), ecosystem representation, and focal species together into a conservation plan? Planners today usually employ computerized site-selection algorithms to integrate multiple datasets and meet conservation planning goals. These algorithms are
- linked to GIS (and hence are spatially explicit);
- highly efficient in achieving stated goals for each conservation target (feature);
- transparent and can be applied interactively in a workshop format, where one alters inputs and goal levels and sees the results in the form of a mapped reserve network;
- address two kinds of problems: meeting a variety of biodiversity goals while minimizing net expected costs – the minimum set problem, or maximizing biodiversity benefit (e.g., representation of features) within a fixed budget – the maximal coverage problem.
Simulated annealing algorithms, which are commonly used, attempt to minimize the “cost” of a reserve network or “portfolio.” In plain language, the Total Portfolio Cost = (cost of selected sites) + (penalty cost for not meeting the stated conservation goals for each element) + (cost of spatial dispersion of the selected sites as measured by the total boundary length of the portfolio).
An example of a simulated annealing algorithm is Marxan, which is probably the most frequently used spatial prioritization algorithm in the world. Another excellent algorithm, Zonation, works by iteratively removing the least valuable cell (accounting for complementary) from the landscape until no cells remain. In this way, landscapes can be zoned according to their value for conservation.
This level of ambition in conservation planning goals greatly influences the size of the conservation network that is assembled by site-selection algorithms. The figure below, where the Marxan algorithm was applied to the Great Sand Hills region of Saskatchewan, illustrates how increasing the target levels for representation of various features (in this case, vegetation types and modeled habitat for rare and declining grassland birds and plants) dramatically affects the area selected by the algorithm. At the left, an ambitious average goal level of 65% of the theoretical biodiversity optimum was set, going down to 20% at the right. Site-selection was limited to the less-developed portion of the region, but goals were set based on habitat availability across the planning region, forcing MARXAN to meet those goals in the less-developed region where conservation options are less restricted. The large area in the northern part of the study region, which is included in all solutions, is a previously-established reserve, which the government did not want to alter (hence, this reserve was “locked in” to all solutions). After much discussion, the planners settled on the average 30% goal level, where core biodiversity areas are most clearly delineated. One lesson from this example is that site-selection tools are objective, but require calibration by goals set by subjective expert opinion.
An important caution: Application of site-selection algorithms does not usually guarantee that the landscape will be connected enough to meet the needs of wide-ranging or fragmentation-sensitive species. Connectivity is important because a connected system of reserves can be a whole greater than the sum of its parts. For example, although no single reserve might maintain viable populations of focal species, a well-connected network of reserves might contain a viable metapopulation. Of course, this is not necessarily true – planners have to make it true through good, science-based design.
Reed Noss, the Davis-Shine Professor of Conservation Biology at the University of Central Florida, has been one of the lead voices in conservation biology for nearly three decades. He has served as President of the Society for Conservation Biology, Editor-in-Chief of its journal, Conservation Biology, and was a co-founder of the Wildlands Project.