Improving the use of environmental diversity as a surrogate for species representation

Abstract

The continuous p‐median approach to environmental diversity (ED) is a reliable way to identify sites that efficiently represent species. A recently developed maximum dispersion (maxdisp) approach to ED is computationally simpler, does not require the user to reduce environmental space to two dimensions, and performed better than continuous p‐median for datasets of South African animals. We tested whether maxdisp performs as well as continuous p‐median for 12 datasets that included plants and other continents, and whether particular types of environmental variables produced consistently better models of ED. We selected 12 species inventories and atlases to span a broad range of taxa (plants, birds, mammals, reptiles, and amphibians), spatial extents, and resolutions. For each dataset, we used continuous p‐median ED and maxdisp ED in combination with five sets of environmental variables (five combinations of temperature, precipitation, insolation, NDVI, and topographic variables) to select environmentally diverse sites. We used the species accumulation index (SAI) to evaluate the efficiency of ED in representing species for each approach and set of environmental variables. Maxdisp ED represented species better than continuous p‐median ED in five of 12 biodiversity datasets, and about the same for the other seven biodiversity datasets. Efficiency of ED also varied with type of variables used to define environmental space, but no particular combination of variables consistently performed best. We conclude that maxdisp ED performs at least as well as continuous p‐median ED, and has the advantage of faster and simpler computation. Surprisingly, using all 38 environmental variables was not consistently better than using subsets of variables, nor did any subset emerge as consistently best or worst; further work is needed to identify the best variables to define environmental space. Results can help ecologists and conservationists select sites for species representation and assist in conservation planning.

Stocktaking the environmental coverage of a continental ecosystem observation network

“…Without prior knowledge of the species that would be sampled at potential monitoring sites, we relied on environmental differences as a surrogate for expected ecological differences (Albuquerque and Beier 2018, Ware et al 2018). However, different vegetation may occur in similar macro‐environments (Bruelheide et al 2018).…”

“…We therefore limited comparisons of the Ausplots stratification to relatively simple strategies that have been commonly implemented. More sophisticated algorithms could weight potential locations by distance or accessibility, or seek to iteratively complement the environmental coverage of an accumulating sample (Kumar et al 2016, Albuquerque and Beier 2018). Indeed, a recent aim of Ausplots sampling has been to gap‐fill the network of monitoring plots with surveys targeting regions predicted to be the most ecologically different in a generalized dissimilarity modeling framework (Guerin et al 2020 b ).…”

“…We describe the stratification method used to determine the locations of Ausplots. We then consider how well the environment heterogeneity of Ausplots locations (Christianson and Kaufman 2016) acts as a surrogate for their observed ecological diversity (Albuquerque and Beier 2018). Our assessment is relevant internationally to the implementation of new monitoring networks and to the application of data from existing networks built under various sampling strategies.…”

Using generalised dissimilarity modelling and targeted field surveys to gap‐fill an ecosystem surveillance network

“…Conservation planners have long grappled with the problem of how to maximise ecological representation in the absence of perfect information on biodiversity (Albuquerque & Beier, 2018). For example, finite budgets mean that the acquisition of incompletely surveyed land parcels to complement reserve systems must be prioritised.…”

“…(2011) used Maxent modelling of the environmental coverage of monitoring site locations to select the most dissimilar putative sampling sites. Another approach is to select new sites that maximise the pairwise distances among all sample sites, known as ‘Environmental Diversity’ (ED), which results in even ecological sampling across scaled environmental space (Albuquerque & Beier, 2018; Faith et al., 2004; Faith & Walker, 1996).…”