The impact of data realities on conservation planning
Highlights
► We explore how data uncertainty and bias can impact conservation planning decisions. ► We conducted a sensitivity analysis on the Protection Index using a real world dataset. ► Data depletion assessed effects of data paucity and bias on site priorities. ► Certain high value sites were more sensitive to data depletion scenarios. ► Demonstrates how uncertainty and data quality can affect planning outcomes.
Introduction
The pressures on biodiversity today are well documented (MEA, 2005) and the importance of making the best possible decisions about what to do, when and where is clear (Wilson et al., 2007). To guide the process of allocating scarce resources to maximise the persistence of biodiversity, the field of conservation planning has developed rapidly over the last 30 years, moving from ad hoc reserve design, to applying a decision theory approach, where goals and objectives are explicit (Margules and Pressey, 2000, Possingham et al., 2001, Sarkar et al., 2006). However, one key component of decision theory has been neglected in conservation planning: the quantification of uncertainty and its explicit consideration in the decision making process (Regan et al., 2009).
Typically conservation planning relies on species distribution data, such as occurrences, range maps or predictive models. The ideal data for conservation planning, assuming that the priority is the protection of all biodiversity, would have complete coverage spatially and across taxonomic groups, but of course such data do not exist. Often planners’ priorities are explicitly stated, such as Birdlife Internationals’ Important Bird Areas programme e.g. (Chan et al., 2004). Priorities are also implicit, and certain groups may be targeted for personal preferences, such as charisma, or ease of detection, accessibility or proximity to roads (Freitag et al., 1998, Reddy and Dávalos, 2003, Grand et al., 2007). Regardless of their priorities, planners often pragmatically use the existing data from datasets and inventories collected with a different set of objectives and constrained by budgets (Nichols and Williams, 2006, Field et al., 2007, Grantham et al., 2009). Often data collection is reactive, for example, a site may be surveyed if threatened by development or for an environmental impact assessment, or a species may be targeted for monitoring if there is a perceived threat to its viability, or simply because it is popular (Kautz et al., 2006, Chaineau et al., 2010).
The gap between the ideals of conservation planning data and the realities of data collection render conservation planning outputs vulnerable to uncertainty (Pressey, 2004, Wolman, 2006). Uncertainty can take the form of sampling bias on taxonomic or spatial grounds, errors of omission due to problems in surveying and detecting all occurrences, or errors of commission as old records are used even after populations are extirpated (Gaston and Rodrigues, 2003, Rondinini et al., 2006). Despite the many sources of uncertainty, many researchers have noted that biological data are often accepted at face value in conservation planning (Moilanen et al., 2006). This should give cause for concern about the knock-on effects of inadequate data on the outcomes of conservation planning decisions.
Sensitivity testing is the exploration of the effect of variation in inputs on process outputs, and has been used widely across a range of disciplines, including GIS, IUCN Red listing, population viability analysis and ecological modelling (Crosetto et al., 1999, Akcakaya et al., 2000, Cariboni et al., 2007, Rae et al., 2007, Naujokaitis-Lewis et al., 2009). Its application to assess the impacts of data uncertainty, bias and quality in conservation planning is less well represented in the literature. A number of studies have directly tested the sensitivity of conservation planning tools to different habitat suitability models (e.g., Gaston and Rodrigues, 2003, Rondinini et al., 2006) while a plethora of studies have examined the impact of different surrogate groups on conservation planning outcomes (Grand et al., 2007, Rodrigues, 2007, Grantham et al., 2010). The broad body of work in the literature dealing with issue of the impacts of surrogates contrasts with the few studies that explore the link between the constraints of data realities and the resultant patterns in data and their impacts on planning outputs. Two studies have conducted sensitivity analyses on reserve selection algorithms using occurrence (Grand et al., 2007) and distribution data (Freitag and Van Jaarsveld, 1998). Both these studies used existing biodiversity inventory data and investigated the sensitivity of a site selection algorithm to perturbations of the dataset designed to replicate biases commonly associated with occurrence and distribution data; survey extent, survey intensity and taxonomic representation in the case of Freitag and van Jaarsveld (1998) and proximity to roads and abundance as a surrogate for detectability by Grand et al. (2007). Our study takes a novel approach; investigating patterns of data representation within an existing dataset currently used to guide the application of conservation decision making and using this as the basis for a subsequent sensitivity analysis involving a ‘real-world’ network of sites as opposed to a hypothetical reserve network.
We performed a sensitivity analysis on conservation priorities based on the Protection Index and using a large dataset from the Florida Natural Areas Inventory (FNAI). The Protection Index (PI) is a metric developed to measure conservation progress and identify priorities (Turner et al., 2006) and has been used in Florida to measure conservation progress in the state’s managed area networks and as a planning tool to assess the benefit of site acquisition portfolios (Turner et al., 2006, Nicholson et al., 2007). The FNAI database is currently used to inform conservation planning in Florida, and therefore the results of this study are highly relevant to on-the-ground decisions based on a real-world dataset. We explored how real datasets diverge from conservation planning ideals due to limited resources and differing goals in data collection, and how this impacts on the decision making process: How sensitive are our metrics to the mismatches between data assumptions and realities? How do these mismatches affect our measure of conservation progress? And most importantly, how do gaps in data and uncertainty affect conservation priorities?
Section snippets
The study area and data
Florida is an area of high biodiversity value that is highly threatened by human development, and consequently has been a focus of much conservation activity (Oetting et al., 2006, Endries et al., 2007). The managed area (MA) network currently covers over 27% of the state, spread over 1850 sites. One major conservation initiative is The Nature Conservancy’s (TNC) acquisition portfolio for the central Florida Peninsula Ecoregion (FPE), an area of 76,200 km2, in which 186 sites are proposed for
Exploration of the dataset
Plants made up >65% of the 181 species in the study, although a substantial number had only a handful of EOs (Table 1), reflecting in many cases genuine rareness of endemic plant species. Mammals, often cited as an over-represented ‘charismatic’ group, were surprisingly unrepresented, in terms both of the number of species and EOs. This is partly explained by species with large home ranges such as the Florida panther Puma concolor having fewer, geographically large EOs. A small number of
Discussion
Conservation planning tools are widely used, but rely on good input data to produce useful outputs (Pressey, 2004). Although the biased nature of species-based data and the potential significance of data quality for conservation planning outputs are well documented (Ponder, 1999, Pressey, 2004, Wolman, 2006, Moilanen et al., 2006), relatively little work has been done to explore their impacts on decision making (for rare examples see Freitag and Van Jaarsveld, 1998, Grand et al., 2007). The
Acknowledgements
We thank the Florida Natural Areas Inventory (FNAI) for their co-operation and for providing data, in particular Amy Knight and Jon Oetting for advice and comments on earlier drafts. We thank David Wilcove, Hilary Swain, Doria Gordon and Gene Kelly for discussions and comments. EN was supported by a Marie Curie Fellowship (IIF-221050) and a Leverhulme Trust grant at Imperial College, and a grant from the Rodney Johnson/Katharine Ordway Stewardship Endowment (RJ/KOSE) from The Nature Conservancy
References (53)
- et al.
The role of sensitivity analysis in ecological modelling
Ecological Modelling
(2007) - et al.
A new predictor of the irreplaceability of areas for achieving a conservation goal, its application to real-world planning, and a research agenda for further refinement
Biological Conservation
(2000) - et al.
How much is enough? Landscape-scale conservation for the Florida panther
Biological Conservation
(2006) - et al.
Uncertainty analysis favours selection of spatially aggregated reserve networks
Biological Conservation
(2006) - et al.
Planning for robust reserve networks using uncertainty analysis
Ecological Modelling
(2006) - et al.
Integrating economic costs into conservation planning
Trends in Ecology & Evolution
(2006) - et al.
Monitoring for conservation
Trends in Ecology & Evolution
(2006) - et al.
Systematic reserve design as a dynamic process: F-TRAC and the Florida Forever program
Biological Conservation
(2006) - et al.
Implications of error and uncertainty for an environmental planning scenario: a sensitivity analysis of GIS-based variables in a reserve design exercise
Landscape and Urban Planning
(2007) - et al.
Making consistent IUCN classifications under uncertainty
Conservation Biology
(2000)