Landscape delineation strategy and size of mapping units impact the performance of habitat suitability models
(2018) Ecological Informatics : an international journal on ecoinformatics and computational ecology — Vol. 47, p. 55-60 (2018)
Files
Landscape-delineation-strategy-and-size-of-mapping-units-im_2018_Ecological-.pdf
Open Access - Adobe PDF
- 938.9 KB
Details
- Authors
- Abstract
- In view of the current biodiversity crisis, the implementation of efficient conservation actions is crucial to the preservation of biodiversity and ecosystem services. However, the design of an ecologically sound conservation strategy requires an extensive knowledge of the species distributions and habitat requirements. Although the amount of biological data available has substantially increased in the last few years, field surveys are highly time-consuming (Niamir et al., 2011). The collection of occurrence data remains thus mainly opportunistic and our knowledge of species distributions is still far from exhaustive. This issue is known as the “Wallacean shortfall” (Brown and Lomolino, 1998). The recent development of correlative species distribution modelling techniques offers an interesting alternative to deal with data incompleteness and optimize future field surveys (Aizpurua et al., 2015; Feria and Peterson, 2002). In these techniques, an algorithm is used to detect statistical relationships between known species occurrences and environmental predictors. This allows the identification of additional sites in which environmental conditions are similar to occurrence localities, thereby providing a habitat suitability map (Guisan and Zimmermann, 2000). Species distribution models can be used to address various conservation issues such as reserve selection, invasive species risk assessment and site selection for species reintroductions (e. g. Broennimann et al., 2007; Roura-Pascual et al., 2009; Guisan et al., 2013 and references therein). These models can also provide predictions of the future impacts of climate and land use changes on species distributions, in order to establish efficient mitigation strategies (e.g. Pearson et al., 2004; Polansky et al., 2000; Thuiller et al., 2005; Williams et al., 2005). Our perception of ecological processes and species distribution patterns is influenced by the spatial resolution of observations (Levin, 1992; Stoms, 1994). In species distribution models, resolution is defined as the area of the mapping unit for which presence probability is predicted (McGeoch and Gaston, 2002). Those mapping units are usually delineated using regular square grids (termed “raster” in GIS) or, more rarely, hexagonal grids (Birch et al., 2007). However, the limits of these grid cells are arbitrary and do not necessarily reflect ecological boundaries. The aggregation of environmental variables into mapping units of different sizes (scaling) and shapes (zoning) can modify their mean and/or variance: this issue is known as the Modifiable Areal Unit Problem (Jelinski and Wu, 1996). Several studies have assessed the scaling effect on the predictive ability of species distribution models. For example, by comparing different resolutions ranging from 1 to 6400 ha, Li et al. (2006) showed that a grain size of 2 ha was optimal to model nest site selection by the crested ibis. Graf et al. (2005) found an optimal grain size of 253 ha for the capercaillie, which corresponds to a small home range size for this species. In multi-species studies, Ferrier and Watson (1997), Guisan et al. (2007) and Gottschalk et al. (2011) have revealed a general negative correlation between grain size and predictive ability, although the results varied between species. Those studies provided interesting comparisons of different resolutions, but did not address the zoning component of the MAUP. Hanberry (2013) was the only one to compare the performance of species distribution models based on mapping units of different shapes. However, in this study mean polygon area (209 ha) was larger than grid cell area (0.81 ha), thus making impossible to fully dissociate the effects of scaling and zoning on model performance. In the present study, we assess the impact of landscape segmentation type on the performance of species distribution models. Therefore, the traditional grid approach was compared to a landscape delineation based on land cover and topography homogeneity. Contrary to a raster grid, the shape and size of the delineated polygons varied according to landscape configuration. Using generalized additive models, this comparison was performed at four different scales in order to also quantify the impact of spatial resolution on predictive performance.
- Affiliations
Citations
Delangre, J., Radoux, J., & Dufrêne, M. (2018). Landscape delineation strategy and size of mapping units impact the performance of habitat suitability models. Ecological Informatics : an international journal on ecoinformatics and computational ecology, 47, 55-60. https://doi.org/10.1016/j.ecoinf.2017.08.005 (Original work published 2018)