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

Abstract

Effective ecosystem management requires spatially distributed measurements that are representative of ecological diversity. When considering which sites complement existing conservation or monitoring networks, there are many strategies for optimising ecological coverage in the absence of ground observations. However, such optimisation is often implemented theoretically in conservation prioritisation frameworks and real‐world implementation is rarely assessed, particularly for monitoring networks. We assessed the performance of adding new survey sites informed by predictive modelling in gap‐filling the ecological coverage of the Terrestrial Ecosystem Research Network's (TERN) continental network of ecosystem surveillance plots, Ausplots. We constructed a generalised dissimilarity model (GDM) in which plant species composition in 531 sites was fitted to environmental parameters. We combined predicted nearest‐neighbour ecological distances for locations across Australia with practical considerations to select regions for gap‐filling surveys of 181 new plots across 18 expeditions. We iteratively tracked the reduction in mean nearest‐neighbour distances in GDM space, and increases in the actual sampling of ecological space, using cumulative multivariate dispersion. The generalised dissimilarity model explained 34% of deviance in species compositional turnover as a function of geographic distance, soil P, aridity, actual evapotranspiration and rainfall seasonality, among 17 significant predictors. We targeted identified gap regions in the Australian jurisdictions of Queensland, New South Wales and Western Australia for surveys in addition to opportunistic or project‐based gap‐filling over 2 years. Approximately 20% of the land area of Australia received increased servicing of biological representation, corresponding to a reduction in mean nearest‐neighbour ecological distances from 0.38 to 0.33 in units of compositional dissimilarity. The corresponding increase in sampled ecological space was 172% more than that achieved from the previous 181 plots. Synthesis and applications. Increases in the representation of ecosystems included in surveillance networks can be achieved efficiently using objective methods for site selection and appraisal. Scaling of environmental variables through ecological models supports practical sampling decisions, while optimising putative survey locations via their ecological distance to a nearest neighbour is useful when the aim is to increase inclusion of habitats. Iterations between modelled gaps and field campaigns provide a pragmatic compromise between theoretical optima and real‐world decision‐making.

Global soil microbiomes: A new frontline of biome‐ecology research

“…Abiotic drivers of eukaryotic and prokaryotic soil communities were identified using distance‐based redundancy analysis (dbRDA; “vegan” package in R; Oksanen et al., 2020) and generalized dissimilarity modelling (gdm; Guerin et al., 2021). dbRDA can incorporate categorical predictor variables, whereas gdm allows nonlinear effects to be identified and a spatial distance matrix to be incorporated directly.…”

A working guide to harnessing generalized dissimilarity modelling for biodiversity analysis and conservation assessment

“…This method relies only on the predicted ecological distance that is derived from a GDM, without undertaking a prediction of pairwise compositional dissimilarity (Ferrier et al., 2007). Applying GDM transformed environmental layers within the environmental diversity (ED) framework (Faith & Walker, 1996), this approach has been applied in a number of settings (Funk et al., 2005; Guerin et al., 2021).…”

Spatial distribution patterns and predictors of fish beta‐diversity in a large dam‐free tributary from a Neotropical floodplain

“…Backward elimination procedures for variables selection were recommended by Ferrier et al (2007) and are commonly used in GDM analyses (e.g. Coccia et al, 2021; Guerin et al, 2021; Saiter et al, 2016; Williamson et al, 2019). The model was originally built with 35 geo‐climatic variables.…”