Biodiversity Informatics, 3, 2006, pp. 59-72 USES AND REQUIREMENTS OF ECOLOGICAL NICHE MODELS AND RELATED DISTRIBUTIONAL MODELS A. TOWNSEND PETERSON Natural History Museum, The University of Kansas, Lawrence, Kansas 66045 USA E-mail [email protected] Abstract.—Modeling approaches that relate known occurrences of species to landscape features to discover ecological properties and predict geographic occurrences have seen extensive recent application in ecology, systematics, and conservation. A key component in this process is estimation or characterization of species’ distributions in ecological space, which can then be useful in understanding their potential distributions in geographic space. Hence, this process is often termed ecological niche modeling or (less boldly) species distribution modeling. Applications of this approach vary widely in their aims, products, and requirements; this variety is reviewed herein, examples are provided, and differences in data needs and possible interpretations are discussed. Key words.− ecological niche, ecological niche modeling, distribution modeling, geographic distributions. Recent years have seen impressive growth in A basic dichotomy that pervades both the list use of modeling approaches based on relationships of uses to which these methods are put and even between known occurrences of species and the terminology used to refer to them is that of features of the ecological and environmental ecological niche modeling (ENM) versus landscape (Guisan and Zimmermann 2000; distributional modeling (DM). A recent paper Pearson and Dawson 2003; Peterson 2003a; (Soberón and Peterson 2005) formalized the idea Soberón and Peterson 2004). These models are of niche modeling, and clarified the differences often termed ‘distribution models,’ ‘climatic between these two views. Niches and distributions envelope models,’ or (most generally) ‘ecological of species were visualized as a set of 3 intersecting niche models.’ The aim of these studies is circles, representing diagrammatically 3 classes of generally to reconstruct species’ ecological determinants: physical conditions necessary for a requirements and/or predict geographic species’ survival and reproduction (“abiotic distributions of species. niche”; e.g., correct combinations of humidity, The earliest applications in this realm were temperature, other biophysical variables, substrate undoubtedly those of Joseph Grinnell, in the 1910s types, disturbance regimes), biotic conditions and 1920s (Grinnell 1917; Grinnell 1924), who necessary for a species’ survival and reproduction used the spatial distribution of occurrences of (“biotic niche”; e.g., presences of mutualists, species to infer factors limiting their distributions, absences of diseases and predators), and and laid a firm foundation for subsequent work in accessibility (i.e., within the dispersal capabilities this field. The diversity of such applications, of the species, either historically or at present) however, has now grown considerably. (Figure 1, top). This latter set of factors is not a Distributional models and ecological niche models niche dimension, but rather is a set of non- are being used not just to understand species’ ecological factors that constrain the species to ecological requirements, but also to understand inhabit less than its full distributional potential, and aspects of biogeography, predict existence of may indeed not be permanent—as shown in the unknown populations and species, identify sites for case of invasive species, dispersal limitations often translocations and reintroductions, plan area with time are overcome. selection for conservation, forecast effects of In this framework, where abiotic conditions are environmental change, etc. (Table 1). appropriate can be compared with the fundamental 59 PETERSON – USES AND REQUIREMENTS OF ECOLOGICAL NICHE MODELS Table 1. Summary of uses to which ecological niche models have been put, and the requirements that these uses have in terms of output and information content. Understand Understand Conservation Predict Predict ecological biogeography Identify sites for planning and effects of Predict climate Quality of requirements and dispersal Find unknown translocations and reserve system habitat species’ change interest of species barriers populations Find new species reintroductions design loss invasions effects Form of Any Any Any Any Absolute Absolute Any Any Any prediction (e.g., binary, ranked, absolute) Grain required Any Any Population Any Individual- Individual- Any Any Any, but may population population be limited by resolution of climate change data sets Causal variables Causal Causal Surrogates Surrogates OK if Surrogates OK if Surrogates Causal Causal Causal needed, or within range; within range; surrogates OK? causal necessary causal necessary if extrapolating if extrapolating Need model Yes Yes No No Useful, to avoid Useful, to No No No 60 response curve error avoid error or parameter retrieval Error needs Overall low Overall low Low omission Low omission Low commission Low Overall Overall Overall error needed error needed (don’t mind (don’t mind (very high cost of commission searching extra searching extra errors) (very high cost localities, but localities, but of errors) don’t want to don’t want to leave anything leave anything out) out) Potential Potential Potential Realized Potential Realized Realized Realized Potential Both distributional model versus realized distribution? Uncertainty No No No No Yes Yes No No No estimates needed? 60 PETERSON – USES AND REQUIREMENTS OF ECOLOGICAL NICHE MODELS ecological niche of the species, and where abiotic DM proponents, on the other hand, include and biotic conditions are fulfilled can be compared effects of abiotic, biotic, and accessibility with the realized ecological niche of the species considerations in their models from the outset. (Hutchinson 1957), although Hutchinson focused They argue that because distributional information mostly on competition among the broader suite of is an expression of a realized ecological niche, as potential biotic intereractions; these interactions such the realized niche (only) is the target of could potentially be integrated more intimately into modeling. As such, DM proponents would often the niche modeling framework (Araújo and Guisan include in the modeling approach independent 2006). The geographic projection of these variables that summarize biotic considerations conditions (i.e., where both abiotic and biotic (e.g., distributions of other species in the region) requirements are fulfilled) represents the potential and that bring in spatial considerations that may be distribution of the species (Figure 1, top, blue relevant to dispersal ability and accessibility area)—areas where the species could survive if (Leathwick 1998; Latimer et al. 2006). Whereas introduced. Finally, those areas where the potential DM is simpler in producing estimates of species’ distribution is accessible to the species is likely to actual geographic distributions directly, approximate the actual distribution of the species predictivity across scenarios of change is largely (Figure 1, top, black area). Other authors (Araújo lost, and assumptions regarding accessibility of and Guisan 2006) have further distinguished areas may still required (Soberón and Peterson between my ‘actual’ distribution and the area 2005). A further discussion of the differences actually occupied at any point in time, taking into between ENM and DM is provided below. account stochasticity, metapopulation dynamics, This diversity of ideas can place different etc. demands on features of algorithms and approaches ENM proponents are interested in using used to develop models—and clearly has the distributional information (i.e., known occurrences potential to lead to debate and perhaps sampled from the actual distribution) to estimate misunderstanding between workers with distinct ecological niches and potential distributions of needs and interests. Nonetheless, contrasts between species, which then provides a means of different conceptualizations of the process (e.g., understanding and anticipating ecological and ENM versus DM) have seen little direct discussion geographic features of species’ distibutional in the literature (Araújo and Guisan 2006; Soberón biology (Soberón and Peterson 2005). This and Peterson 2005). Such is the purpose of this approach has the advantage of allowing the effects contribution: to survey the diverse uses to which of the three components listed above to be these approaches have been put, and discuss distinguished, which offers greater interpretability differences in data needs and interpretation that as to causation of phenomena, and permits this diversity demands. predictability of phenomena that depend on the differences between components—e.g., the ECOLOGICAL NICHES AND EVOLUTIONARY invasive potential of a species depends on the CONSERVATISM OF NICHES difference between potential and actual In general, readers are referred to recent distributional areas. This approach, nonetheless, conceptual reviews (Chase and Leibold 2003; requires additional complexities of interpretation to Pulliam 2000; Soberón and Peterson 2005) of produce estimates of actual geographic relationships among autecology, synecology (i.e., distributions, given that the data on which the species interactions), and history and accessibility. niche models are based are not drawn from the Once again, this general approach was pioneered entire abiotic niche or even from the potential