Elsevier

Biological Conservation

Volume 143, Issue 11, November 2010, Pages 2647-2657
Biological Conservation

Overcoming the rare species modelling paradox: A novel hierarchical framework applied to an Iberian endemic plant

https://doi.org/10.1016/j.biocon.2010.07.007Get rights and content

Abstract

Rare species have restricted geographic ranges, habitat specialization, and/or small population sizes. Datasets on rare species distribution usually have few observations, limited spatial accuracy and lack of valid absences; conversely they provide comprehensive views of species distributions allowing to realistically capture most of their realized environmental niche. Rare species are the most in need of predictive distribution modelling but also the most difficult to model. We refer to this contrast as the “rare species modelling paradox” and propose as a solution developing modelling approaches that deal with a sufficiently large set of predictors, ensuring that statistical models are not over-fitted. Our novel approach fulfils this condition by fitting a large number of bivariate models and averaging them with a weighted ensemble approach. We further propose that this ensemble forecasting is conducted within a hierarchic multi-scale framework. We present two ensemble models for a test species, one at regional and one at local scale, each based on the combination of 630 models. In both cases, we obtained excellent spatial projections, unusual when modelling rare species. Model results highlight, from a statistically sound approach, the effects of multiple drivers in a same modelling framework and at two distinct scales. From this added information, regional models can support accurate forecasts of range dynamics under climate change scenarios, whereas local models allow the assessment of isolated or synergistic impacts of changes in multiple predictors. This novel framework provides a baseline for adaptive conservation, management and monitoring of rare species at distinct spatial and temporal scales.

Introduction

Worldwide loss of biodiversity is of great global concern and has increasingly become a research focus since the Rio Summit in 1992 (Matern et al., 2007). Following the convention on Biological Diversity (CBD; http://www.biodiv.org), European countries agreed on ‘halting biodiversity loss by 2010’. Conversely, many other countries around the world have committed to a significant reduction of the current rate of biodiversity loss at the global, regional and national levels (Balmford et al., 2005, Pereira and Cooper, 2006). Therefore, identifying processes driving the distribution and dynamics of species, communities and ecosystems has become the core of nowadays ecological research. However the task is challenging due to the complex nature of species range dynamics and related factors acting at multiple spatial scales (Elith and Leathwick, 2007, Guisan and Thuiller, 2005, Guisan and Zimmermann, 2000, Murphy and Lovett-Doust, 2007).

The collection and analysis of ecologically relevant information about biodiversity and the development of a wide set of statistical tools able to relate diversity patterns with environmental conditions are thus crucial to understand such patterns (Elith et al., 2006, Guisan and Thuiller, 2005). Species Distribution Models (SDMs; Guisan and Thuiller, 2005, Guisan and Zimmermann, 2000) aim at quantifying species-environment relationships, and thus quantify a species’ realized environmental niche (sensu Hutchinson; see Araújo and Guisan, 2006) or ‘ecological niche’ as we call it hereafter, and have been recently used to address fundamental macro-ecological questions such as the ecological impacts of climate and land-use changes or biological invasions (Broennimann et al., 2006, Guisan and Thuiller, 2005, Guisan and Zimmermann, 2000, Heikkinen et al., 2007, Parmesan and Yohe, 2003). However, extinction and decline in species ranges still remain as core conservation problems and a motive of concern in the scientific community (Balmford et al., 2005, Guisan and Thuiller, 2005, Walther et al., 2007). Several studies suggest that biodiversity is already facing the effects of climate changes, expressed for example in modifications of the phenology and physiology of species, or even by induced displacements of species distributions that may ultimately lead to increased extinction rates (Parmesan and Yohe, 2003, Walther et al., 2007).

Habitat destruction and fragmentation, intensification (or abandonment) of human land use, and biological invasions, constitute severe threats to biodiversity today (Sala et al., 2000, Schröter et al., 2005). In order to preserve biodiversity worldwide, research efforts should be directed to protect the species that are assumed to undergo the highest risk of extinction. Rare species, which are characterized by a low number of occurrences, are particularly threatened. There is thus a pressing need to determine how climate change and human activities may further threaten the distribution of these species, not only for monitoring purposes but also for prioritizing sites or species to preserve and prevent further loss. Therefore gathering knowledge and distribution data is of particular importance for these species, but paradoxically the scientific literature is in deficit of rare species modelling studies (Aitken et al., 2007, Engler et al., 2004, Matern et al., 2007).

Even though great methodological progresses have been recently achieved, such as the improvement of predictive algorithms (Elith et al., 2006, Elith et al., 2008, Guisan and Thuiller, 2005) and the use of more causal environmental predictors at a better spatial resolution (Lassueur et al., 2006, Randin et al., 2009, Wu and Smeins, 2000), modelling the distribution of rare species remains a challenge (Aitken et al., 2007, Elith et al., 2006, Ferrier et al., 2002, Guisan et al., 2006, Guisan and Thuiller, 2005, Wisz and Guisan, 2009). Rare species, either naturally rare or as a result of human impact, are characterized by restricted geographic ranges, habitat specialization (species occur only in one or few specific habitat types) and small population sizes (Aitken et al., 2007, Irfan-Ullah et al., 2007, Lavergne et al., 2005, Medrano and Herrera, 2008, Rabinowitz, 1981). Datasets on rare species distribution are usually characterized by small numbers of observations, often gathered over long periods of time (e.g. the whole 20th century) and of limited spatial accuracy, and a lack of valid absences (Engler et al., 2004, Pearson et al., 2004, Rushton et al., 2004, Vaughan and Ormerod, 2005, Wisz and Guisan, 2009). On the other hand, the small number of available observations can in many cases provide a fairly comprehensive view of the species’ distribution. When rare species are additionally endemic to the studied area, it allows modellers to realistically capture a large part of their ecological niche, thus avoiding issues of truncated response curves biasing future projections (Thuiller et al., 2004). Having good data on the whole distribution of these taxa also allows addressing questions related to the causes of rarity (Guisan and Thuiller, 2005, Sala et al., 2000).

Rare species additionally constitute an interesting application for spatially-hierarchic modelling approaches. As for other species, their distribution and abundance are known to be determined by distinct sets of conditions and resources, acting or being reflected at different spatial scales (Guisan and Thuiller, 2005). For instance, to predict the distribution of plants in England, climate requirements need to be captured at the European scale, while it is sufficient to consider land cover at the national scale, a situation that is best modelled in a hierarchical framework (Pearson et al., 2004). Species considered as habitat specialists, restricted and uniquely adapted to highly fragmented habitat types, are the ones most in need of predictive habitat suitability modelling. In addition to that, approaches that include hierarchical levels of refinement at distinct scales can also contribute to strategic conservation planning at distinct scales (Elith and Leathwick, 2007). By using different levels of information at different spatial scales – including climate, land cover, history, dispersal ability, and biotic interactions – hierarchical approaches may help refine distribution models and to make them more informative (Luoto et al., 2007, Thuiller et al., 2003). However, although such approaches could be of high relevance for rare species conservation at local or regional scales, most studies so far focussed on common species, and were often conducted at broad spatial scales. One example of a multi-scale approach to rare species modelling was presented by Wu and Smeins (2000), who modelled eight rare plant species in Texas at regional, landscape and site scales based on the generation of habitat mapping rules from field assessments and expert judgment.

Summarizing, rare species are the most in need of predictive distribution modelling, for both monitoring and conservation management purposes, but at the same time they are also the most difficult to model due to the limited number of occurrences available for them. We refer to this contrasting situation as the “rare species modelling paradox”.

In this paper we address this paradox and propose a novel approach to overcome it. We illustrate this approach with a test species, Narcissus cyclamineus DC., a small rare daffodil species, endemic to the northwest corner of the Iberian Peninsula. We propose a modelling framework that can deal with a sufficiently large set of predictors, this way allowing to capture as much of the species’ ecological niche as possible, while ensuring that the statistical models are not over-fitted by including too many predictors for the limited number of available observations. Our novel approach fulfils these two conditions, by fitting a large number of bivariate models (i.e. two predictors only at a time in a model) and averaging all models with a weighted ensemble approach (Araújo and New, 2007) to obtain a final model that can capture as much as possible of the species’ ecological requirements. We further propose that this novel ensemble forecasting of rare species be conducted within a hierarchic multi-scale framework allowing the entire climatic niche to be fitted at the whole species’ range scale and then refining the regional predictions with more local predictors at a finer scale of management. Finally, we discuss whether this improved hierarchical framework could provide reliable forecasts of impacts of future multi-scale environmental changes on rare species and thus provide support to conservation planning and decision making.

Section snippets

The modelling framework

Species Distribution Models (SDMs; Guisan and Thuiller, 2005) rely on pattern recognition approaches, i.e. assessing the determinants of species’ ecological niche and geographic distributions by relating occurrences of a species with values of predictor variables across a series of observation sites (Elith et al., 2006, Guisan and Thuiller, 2005, Guisan and Zimmermann, 2000). Our novel ensemble and hierarchical approach for modelling the distribution (and dynamics) of rare species is based on

Results

The climatically based ensemble model predicted the occurrence of N. cyclamineus along a narrow stripe ranging from Western Galicia (Spain) to Central Portugal, which is rather coherent with the described geographic range of the species. The major part (80%) of the known occurrence records at the Iberian scale can be recognized in the highest probability class for potential distribution (Fig. 3). Nevertheless, this is not the case for occurrences located in the North and South extremes of

Ensemble models and the “rare species modelling paradox”

Rare species have been considered a modelling challenge due to the generalized idiosyncrasies of their distribution datasets (Farnsworth and Ogurcak, 2006, Guisan et al., 2006, Walther et al., 2007, Wu and Smeins, 2000). Scarce datasets, based on presence-only data and low geographic accuracy, raise severe limitations to the application of the main core of modelling algorithms, especially concerning the number of predictors that can be included jointly in a model. This in turn hampers the

Acknowledgements

This research was supported by the Portuguese Science and Technology Foundation (FCT) through PhD Grant SFRH/BD/31576/2006 to A. Lomba. AG benefitted support from the European Commission (ECOCHANGE Project, Grant No. FP6-036866) and from the Swiss National Science Foundation (BIOASSEMBLE Project, Grant No. 31003A-125145). The authors would like to thank two anonymous referees for their comments that contributed to the improvement of the manuscript.

References (55)

  • Anthos, 2008. Sistema de Información de las plantas de España. Real Jardín Botánico, CSIC - Fundación Biodiversidad....
  • M.B. Araújo et al.

    Five (or so) challenges for species distribution modelling

    Journal of Biogeography

    (2006)
  • A. Balmford et al.

    The convention on biological diversity’s 2010 target

    Science

    (2005)
  • O. Broennimann et al.

    Do geographic distribution, niche property and life form explain plants’ vulnerability to global change?

    Global Change Biology

    (2006)
  • O. Broennimann et al.

    Evidence of climatic niche shift during biological invasion

    Ecology Letters

    (2007)
  • J. Elith et al.

    Novel methods improve prediction of species’ distributions from occurrence data

    Ecography

    (2006)
  • J. Elith et al.

    Predicting species distributions from museum and herbarium records using multiresponse models fitted with multivariate adaptive regression splines

    Diversity and Distributions

    (2007)
  • J. Elith et al.

    A working guide to boosted regression trees

    Journal of Animal Ecology

    (2008)
  • R. Engler et al.

    An improved approach for predicting the distribution of rare and endangered species from occurrence and pseudo-absence data

    Journal of Applied Ecology

    (2004)
  • ESRI, 2009. ArcMap 9.3. Environmental Systems Research Institute...
  • E.J. Farnsworth et al.

    Biogeography and decline of rare plants in New England: historical evidence and contemporary monitoring

    Ecological Applications

    (2006)
  • S. Ferrier et al.

    Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling

    Biodiversity and Conservation

    (2002)
  • E.A. Freeman et al.

    PresenceAbsence: an R package for presence absence analysis

    Journal of Statistical Software

    (2008)
  • C.H. Graham et al.

    The influence of spatial errors in species occurrence data used in distribution models

    Journal of Applied Ecology

    (2008)
  • A. Guisan et al.

    Using niche-based models to improve the sampling of rare species

    Conservation Biology

    (2006)
  • A. Guisan et al.

    Predicting species distribution: offering more than simple habitat models

    Ecology Letters

    (2005)
  • F.E. Harrell et al.

    Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors

    Statistics in Medicine

    (1996)
  • Cited by (184)

    View all citing articles on Scopus
    View full text