Testing the effect of sample prevalence and sampling methods on probability- and favourability-based SDMs

Marchetto, Elisa; Re, Daniele Da; Tordoni, Enrico; Bazzichetto, Manuele; Zannini, Piero; Celebrin, Simone; Chieffallo, Ludovico; Malavasi, Marco; Rocchini, Duccio

doi:10.1016/j.ecolmodel.2022.110248

Cited by 7 publications

(6 citation statements)

References 58 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Notwithstanding the positive results obtained in terms of predictive performance, we argue that a comparison of metrics of model predictive accuracy may not be the best means for evaluating the adequacy of different sampling strategies carried out within the environmental rather than the geographical space. Indeed, previous studies showed that these metrics are affected by several factors, including sample prevalence (Guisan et al, 2017; Leroy et al, 2018; Marchetto et al, 2023), sample bias (Dubos et al, 2022; Rocchini et al, 2023) or the spatial extent of the study area (Lobo et al, 2008). Moreover, AUC and TSS tend to score high even in case of poor models calibrated on data exhibiting strong sample location bias (Fourcade et al, 2018; Jiménez‐Valverde, 2021).…”

Section: Discussionmentioning

confidence: 99%

USE it: Uniformly sampling pseudo‐absences within the environmental space for applications in habitat suitability models

Da Re,

Tordoni,

Lenoir

et al. 2023

Methods Ecol Evol

Self Cite

View full text Add to dashboard Cite

Habitat suitability models infer the geographical distribution of species using occurrence data and environmental variables. While data on species presence are increasingly accessible, the difficulty of confirming real absences in the field often forces researchers to generate them in silico. To this aim, pseudo‐absences are commonly sampled randomly across the study area (i.e. the geographical space). However, this introduces sample location bias (i.e. the sampling is unbalanced towards the most frequent habitats occurring within the geographical space) and favours class overlap (i.e. overlap between environmental conditions associated with species presences and pseudo‐absences) in the training dataset. To mitigate this, we propose an alternative methodology (i.e. the uniform approach) that systematically samples pseudo‐absences within a portion of the environmental space delimited by a kernel‐based filter, which seeks to minimise the number of false absences included in the training dataset. We simulated 50 virtual species and modelled their distribution using training datasets assembled with the presence points of the virtual species and pseudo‐absences collected using the uniform approach and other approaches that randomly sample pseudo‐absences within the geographical space. We compared the predictive performance of habitat suitability models and evaluated the extent of sample location bias and class overlap associated with the different sampling strategies. Results indicated that the uniform approach: (i) effectively reduces sample location bias and class overlap; (ii) provides comparable predictive performance to sampling strategies carried out in the geographical space; and (iii) ensures gathering pseudo‐absences adequately representing the environmental conditions available across the study area. We developed a set of R functions in an accompanying R package called USE to disseminate the uniform approach.

show abstract

Section: Discussionmentioning

confidence: 99%

USE it: Uniformly sampling pseudo‐absences within the environmental space for applications in habitat suitability models

Da Re,

Tordoni,

Lenoir

et al. 2023

Methods Ecol Evol

Self Cite

View full text Add to dashboard Cite

show abstract

“…Despite the generally favourable results in terms of predictive performance for both modelling approaches, we argue that comparing predictive accuracy metrics may not be the most effective way to assess how appropriate different models are. Prior studies demonstrated that these metrics are influenced by a variety of factors, such as sample prevalence (Guisan et al., 2017; Leroy et al., 2018; Marchetto et al., 2023), sample location bias (Dubos et al., 2022; Fourcade et al., 2018; Jiménez‐Valverde, 2021; Rocchini et al., 2023) and the size of the study region (Lobo et al., 2010).…”

Section: Discussionmentioning

confidence: 99%

“…effective way to assess how appropriate different models are. Prior studies demonstrated that these metrics are influenced by a variety of factors, such as sample prevalence (Guisan et al, 2017;Leroy et al, 2018;Marchetto et al, 2023), sample location bias (Dubos et al, 2022;Fourcade et al, 2018;Jiménez-Valverde, 2021;Rocchini et al, 2023) and the size of the study region (Lobo et al, 2010).…”

Section: Discussionmentioning

confidence: 99%

Towards causal relationships for modelling species distribution

Da Re,

Tordoni,

Lenoir

et al. 2023

Journal of Biogeography

Self Cite

View full text Add to dashboard Cite

AimUnderstanding the processes underlying the distribution of species through space and time is fundamental in several research fields spanning from ecology to spatial epidemiology. Correlative species distribution models rely on the niche concept to infer or explain the distribution of species, though often focusing only on the abiotic component of the niche (e.g. temperature, precipitation), without clear causal links to the biology of the species under investigation. This might result in an oversimplification of the complex niche hypervolume, resulting in a single model formula whose estimates and predictions lack ecological realism.LocationNot applicable.Time PeriodNot applicable.Major Taxa StudiedVirtual species.Materials and MethodsWe believe that a causal perspective associated with a finer definition of the modelling target is necessary to develop more ecologically realistic outputs. Here, we propose to infer the geographical distribution of a species by applying the modelling relation approach, a causal conceptual framework developed by the theoretical biologist Robert Rosen, which can be formalized through structural equation modelling.ResultsOur findings suggest that building a model relying on a strong conceptual basis improves the stability of the estimated model's coefficients, without necessarily increasing the predictive accuracy metrics of the model.Main ConclusionsIncluding causal processes underlying the spatial distribution of species into an inferential formal system highlights the methodological steps where uncertainty can arise and results in model outputs which are tightly linked to the ecology of the target species.

show abstract

“…Despite the generally favourable results in terms of predictive performance for both modelling approaches, we argue that comparing predictive accuracy metrics may not be the most effective way to assess how appropriate different models are. In fact, prior studies demonstrated that these metrics are influenced by a variety of factors, such as sample prevalence (Guisan et al, 2017;Leroy et al, 2018;Marchetto et al, 2023), sample location bias (Fourcade et al, 2018, Jiménez-Valverde, 2021Dubos et al, 2022Rocchini et al, 2023) and the size of the study region (Lobo et al, 2008).…”

Section: Discussionmentioning

confidence: 99%

Towards causal relationships for modelling species distribution

Da Re,

Tordoni,

Lenoir

et al. 2023

Preprint

View full text Add to dashboard Cite

1. Understanding the processes underlying the distribution of species through space and time is fundamental in several research fields spanning from ecology to spatial epidemiology. Correlative species distribution models (SDMs) involve popular statistical tools to infer species geographical distribution thanks to spatiotemporally explicit observations of species occurrences coupled with a set of environmental predictors. 2. So-called SDMs rely on the niche concept to infer or explain the distribution of species, though often focusing only on the abiotic component of the niche (e.g., temperature, precipitation), without clear causal links to the biology of species under investigation. This might result in an over-simplification of the complex niche hypervolume, resulting in a single model formula whose estimates and predictions lack ecological realism. 3. We believe that a causal perspective associated with a finer definition of the modelling target is necessary to develop ecologically more realistic outputs. Here, we propose to infer the geographical distribution of a species by applying the modelling relation approach, a causal conceptual framework developed by the theoretical biologist Robert Rosen, which can be formalized through structural equation modelling (SEM). 4. Implementing the modelling relation into SDMs would improve the inclusion of the causal processes underlying the spatial distribution of species into an inferential formal system, potentially highlighting the methodological steps where uncertainty arises and eventually resulting in model outputs which are tightly linked to the ecology of the target species.

show abstract

Testing the effect of sample prevalence and sampling methods on probability- and favourability-based SDMs

Cited by 7 publications

References 58 publications

USE it: Uniformly sampling pseudo‐absences within the environmental space for applications in habitat suitability models

USE it: Uniformly sampling pseudo‐absences within the environmental space for applications in habitat suitability models

Towards causal relationships for modelling species distribution

Towards causal relationships for modelling species distribution

Contact Info

Product

Resources

About