A Systematic Review of Marine-Based Species Distribution Models (SDMs) with Recommendations for Best Practice

Abstract

In the marine environment Species Distribution Models (SDMs) have been used in hundreds of papers for predicting the present and future geographic range and environmental niche of species. We have analyzed ways in which SDMs are being applied to marine species in order to recommend best practice in future studies. This systematic review was registered as a protocol on the Open Science Framework: https:// osf.io/tngs6/. The literature reviewed (236 papers) was published between 1992 and July 2016. The number of papers significantly increased through time (R 2 = 0.92, p < 0.05). The studies were predominantly carried out in the Temperate Northern Atlantic (45%) followed by studies of global scale (11%) and studies in Temperate Australasia (10%). The majority of studies reviewed focused on theoretical ecology (37%) including investigations of biological invasions by non-native organisms, conservation planning (19%), and climate change predictions (17%). Most of the studies were published in ecological, multidisciplinary, or biodiversity conservation journals. Most of the studies (94%) failed to report the amount of uncertainty derived from data deficiencies and model parameters. Best practice recommendations are proposed here to ensure that novice and advanced SDM users can (a) understand the main elements of SDMs, (b) reproduce standard methods and analysis, and (c) identify potential limitations with their data. We suggest that in the future, studies of marine SDMs should report on key features of the approaches employed, data deficiencies, the selection of the best explanatory model, and the approach taken to validate the SDM results. In addition, based on the literature reviewed, we suggest that future marine SDMs should account for uncertainty levels as part of the modeling process.

Outstanding Challenges in the Transferability of Ecological Models

“…Colors indicate whether transfers were considered successful (green, unbroken lines; 1-9) or not (dark red, dashed lines; [16][17][18][19][20], as reported by the authors and irrespective of the statistical methods chosen to build the models or the metrics used to evaluate them. Dual colors indicate scenarios in which the quality of transfers varied as a function of modeling algorithms (10-11), space (12), or species (13)(14)(15) contemporaneous sites at different positions along an environmental gradient can approximate temporal variability [72]. Such would be the case, for example, for areas subject to temperature regimes similar to those anticipated in the future, noting it will not be appropriate for species occupying small ranges or those not well-represented in the fossil record.…”

“…Assessments of transferability demand appropriate diagnostics of prediction accuracy and precision [73], yet there is still little consensus on which metrics are most appropriate [6,74]. All else being equal, true validation is possible only with independent data, which are often Figure 7 in Robinson et al [12]). Challenges were identified by a consortium of 50 experts.…”

“…Acknowledging significant overlap and linkages between these challenges (Figure 2), we explore each separately below. Attempts to understand and enhance transferability face many of the same hurdles as ecological modeling generally (e.g., data quality, stochasticity), and adhering to best practice recommendations (e.g., [12,13]) is thus imperative. We do not focus on these well-established standards, but concentrate on the additional challenges posed by transferring models.…”

Dynamic ensemble models to predict distributions and anthropogenic risk exposure for highly mobile species

“…Species distribution models (SDMs) are key tools for describing species habitats and distributions across marine and terrestrial systems (Elith & Leathwick, ; Robinson, Nelson, Costello, Sutherland, & Lundquist, ). Species distribution modelling involves using statistical tools to relate species occurrence or abundance to spatiotemporal patterns of environmental variation (Elith & Leathwick, ).…”

“…First, while marine SDMs have typically used a single‐model type (Robinson et al, ), determining an appropriate modelling approach can be challenging given inherent trade‐offs in the statistical methods available (Elith et al, ; Qiao, Soberón, & Peterson, ). Multi‐model ensembles can reduce uncertainty by overcoming biases inherent in any one model type and providing a “consensus” approach to predictions (Araújo & New, ; Gritti, Duputié, Massol, & Chuine, ).…”