Using Gradient Forests to summarize patterns in species turnover across large spatial scales and inform conservation planning

Abstract

AimProducing quantitative descriptions of large‐scale biodiversity patterns is challenging, particularly where biological sampling is sparse or inadequate. This issue is particularly problematic in marine environments, where sampling is both difficult and expensive, often resulting in patchy and/or uneven coverage. Here, we evaluate the ability of Gradient Forest (GF) modelling to describe broad‐scale marine biodiversity patterns, using a large dataset that also provided opportunity to investigate the effects of sample size on model stability.LocationNew Zealand's Extended Continental Shelf to depths of 2,000 m.MethodsGF models were used to analyse and predict spatial patterns of demersal fish species turnover (beta diversity) using an extensive demersal fish dataset (>27,000 research trawls) and high‐resolution environmental data layers (1 km2 grid resolution). GF models were fitted using various sized, mutually exclusive subsets of the demersal fish data to explore the effect of variation in numbers of training observations on model performance and stability. A final GF model using 13,917 samples was used to transform the environmental layers, which were then classified to produce 30 spatial groups; the ability of these groups to identify fish samples with similar composition was evaluated using independent sample data.ResultsModel fitting using varying sized subsets of the data indicated only minimal changes in model outcomes when using >7,000 observations. A multiscale spatial classification of marine environments created using results from a final GF model fitted using ~14,000 samples was highly effective at summarizing spatial variation in both fish assemblage composition and species turnover.Main conclusionsThe hierarchical nature of the classification supports its use at varying levels of classification detail, which is advantageous for conservation planning at differing spatial scales. This approach also facilitates the incorporation of information on intergroup similarities into conservation planning, allowing greater protection of distinctive groups likely to support unusual assemblages of species.

Modelling the spatial distribution of cetaceans in New Zealand waters

“…Some species with widely distributed recorded locations had poorer model fits than species with restricted ranges, perhaps reflecting the cosmopolitan distribution of the former (e.g., moderate explained deviance and AUC scores for killer whale and bottlenose dolphin) and the more aggregated nature of others for the latter (e.g., high explained deviance and AUC scores for Māui dolphin, Hector's dolphin). Evidence from previous studies have indicated that species with limited geographic ranges and/or environmental tolerances are generally better modelled than those with greater ranges (Morán‐Ordóñez, Lahoz‐Monfort, Elith, & Wintle, ; Stephenson et al, ; Thomson et al, ) because widespread species are less likely to have sharp easily identifiable environmental thresholds that clearly delineate their environmental niche (Morán‐Ordóñez et al, ). For species with limited ranges, the best model fits were commonly located closer to shore where sampling effort was highest.…”

“…Additionally, reduced model fit could be influenced by historical events, human activities, population and species dynamics (e.g., migration, competition, predation, and for many cetacean species, social interactions; Elith & Leathwick, ) and temporal environmental patterns (e.g., diurnal, tidal, seasonal and annual patterns; fluctuating weather patterns; and prey distributions) which were not accounted for here. Despite these factors not being considered in a quantitative manner, model outputs are still valid for management purposes, but it should be noted that the representation of species' probability of occurrence are a smoothed representation of the raw data (spatially and temporally; Stephenson et al, ). Both RES and BRT analyses are correlative models and, in many cases, rely on biotic processes such as predation to be represented by environmental variables as proxies.…”

A New Zealand demersal fish classification using Gradient Forest models

“…In order to address some of these concerns, more recent approaches have combined continuous environmental data with biological samples (Dunstan et al 2012;Anderson et al 2016) with various statistical techniques used to quantitatively assess the role that different environmental factors play in influencing biodiversity patterns (Ferrier et al 2004;Pitcher et al 2012). Results can then be used to control the selection, weighting and transformation of environmental predictors, increasing the ability of the classification groups to represent spatial variation in biodiversity character (Leathwick et al 2011;Stephenson et al 2018).…”

“…One recently developed approach, Gradient Forests (GF, Pitcher et al (2011)) uses an aggregation of Random Forests (RF, (Breiman 2001)), each of which describes the environmental relationships of an individual species, to inform the selection, weighting and transformation of environmental predictors to maximise their correlation with species turnover (Ellis et al 2012). A GF-trained classification was recently used to describe spatial patterns of demersal fish species turnover (beta diversity) in New Zealand using an extensive demersal fish dataset (>27,000 research trawls) and high-resolution environmental data layers (1 km 2 grid resolution) (Stephenson et al 2018). Using a large set of independent data for evaluation, this classification was found to be highly effective at summarising spatial variation in both the composition of demersal fish assemblages and species turnover (Stephenson et al 2018).…”

“…A GF-trained classification was recently used to describe spatial patterns of demersal fish species turnover (beta diversity) in New Zealand using an extensive demersal fish dataset (>27,000 research trawls) and high-resolution environmental data layers (1 km 2 grid resolution) (Stephenson et al 2018). Using a large set of independent data for evaluation, this classification was found to be highly effective at summarising spatial variation in both the composition of demersal fish assemblages and species turnover (Stephenson et al 2018).…”