High uncertainty in the effects of data characteristics on the performance of species distribution models

Tessarolo, Geiziane; Lobo, Jorge M.; Rangel, Thiago F.; Hortal, Joaquín

doi:10.1016/j.ecolind.2020.107147

Cited by 41 publications

(42 citation statements)

References 82 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This is an important constraint which typically precludes many relevant predictors from being available for distribution modeling. Also, a predictor layer may itself be the output of a modeling process, e.g., climate data interpolated from observations at weather stations or composite layers (Simensen et al, 2020;Tessarolo et al, 2021). Depending on their origin, explanatory (predictor) data used for distribution modeling are subject to many sources of inaccuracies and errors, such as imprecise or erroneous variable values and inappropriate resolution (in space or time).…”

Section: Explanatory (Predictor) Datamentioning

confidence: 99%

Reliability in Distribution Modeling—A Synthesis and Step-by-Step Guidelines for Improved Practice

et al. 2021

View full text Add to dashboard Cite

Information about the distribution of a study object (e.g., species or habitat) is essential in face of increasing pressure from land or sea use, and climate change. Distribution models are instrumental for acquiring such information, but also encumbered by uncertainties caused by different sources of error, bias and inaccuracy that need to be dealt with. In this paper we identify the most common sources of uncertainties and link them to different phases in the modeling process. Our aim is to outline the implications of these uncertainties for the reliability of distribution models and to summarize the precautions needed to be taken. We performed a step-by-step assessment of errors, biases and inaccuracies related to the five main steps in a standard distribution modeling process: (1) ecological understanding, assumptions and problem formulation; (2) data collection and preparation; (3) choice of modeling method, model tuning and parameterization; (4) evaluation of models; and, finally, (5) implementation and use. Our synthesis highlights the need to consider the entire distribution modeling process when the reliability and applicability of the models are assessed. A key recommendation is to evaluate the model properly by use of a dataset that is collected independently of the training data. We support initiatives to establish international protocols and open geodatabases for distribution models.

show abstract

Section: Explanatory (Predictor) Datamentioning

confidence: 99%

Reliability in Distribution Modeling—A Synthesis and Step-by-Step Guidelines for Improved Practice

et al. 2021

View full text Add to dashboard Cite

show abstract

“…2) Statistical weightings have been used to adjust the representativeness of the data sample e.g., by up-weighting under-represented regions or taxa (e.g., as employed by the Living Planet Index 82 and often with citizen science data 31,73 ) but this approach can over emphasize the effect of very small portions of the overall data 83 and potentially inflate errors associated with those data 36,60,83,84 . 3) Bias can be explicitly modelled using fixed effects for continuous variables of driver intensity and random effects to represent geographic, temporal and taxonomic structure (e.g., as in 85 ), but care must be taken to ensure all uncertainties are propagated through to the global mean estimate [86][87][88][89] . 4) Baselines, time since disturbance and changing intensity of impact of global change drivers can be explicitly incorporated into analyses of time series data 13,43 .…”

Section: Recommendation 2: Account For Data Representation Across Multiple Axes In Existing Syntheses Of Observational Datamentioning

confidence: 99%

Representation of global change drivers across biodiversity datasets

Daskalova¹,

Bowler²,

Myers‐Smith³

et al. 2021

Preprint

View full text Add to dashboard Cite

Global change has altered biodiversity and impacted ecosystem functions and services around the planet. Understanding the effects of anthropogenic drivers like human use and climate change on biodiversity change has become a key challenge for science and policy. However, our knowledge of biodiversity change is limited by the available data and their biases. Over land and sea, we test the representation of three worldwide and multi-taxa biodiversity databases (Living Planet, BioTIME and PREDICTS) across spatial and temporal variation in global change and across the tree of life. We find that variation in global change drivers is better captured over space than over time around the world and across the previous 150 years. Spatial representation of global change was as high as 78% in the marine realm and 31% on land. Our findings suggest ways to improve the use of existing biodiversity data and better target future ecological monitoring.

show abstract

“…Uncertainty is unavoidable in any statistical process in which a prediction is made from data. In the case of SDMs, uncertainties are introduced in all stages of the modelling process, including the spatial precision and selection of distributional data and the ancillary variables used as predictors, the algorithms used for modelling, model evaluation, model projection into geographical space, or the application of thresholds to distinguish areas of predicted presence (Heikkinen et al 2006, Hortal et al 2008, Diniz‐Filho et al 2009, Buisson et al 2010, Grenouillet et al 2011, Watling et al 2015, Tessarolo et al 2021). Such issues are not solely of academic interest; SDM outputs are frequently taken at face value (Wilson 2010), potentially leading to the ineffective use of limited conservation resources.…”

Section: Introductionmentioning

confidence: 99%

“…The quantification, visualization and effective communication of the uncertainties of the data underpinning SDMs is thus a critical step to evaluate the outputs of these ubiquitous conservation tools (Rocchini et al 2011, Beale and Lennon 2012, Gould et al 2014, McInerny et al 2014, Araújo et al 2019). Indeed, evaluating uncertainty in model predictions has attracted much attention (Barry and Elith 2006, Graham et al 2007, Wisz et al 2008, Syphard and Franklin 2009, Lobo and Tognelli 2011, Naimi et al 2011, 2014, Kramer‐Schadt et al 2013, Swanson et al 2013, Tessarolo et al 2014, 2021). These studies cover different stages of the SDM process, including model selection, parametrization and selection of predictors.…”

Section: Introductionmentioning

confidence: 99%

Using maps of biogeographical ignorance to reveal the uncertainty in distributional data hidden in species distribution models

et al. 2021

Self Cite

View full text Add to dashboard Cite

Species distribution models (SDMs) are subject to many sources of uncertainty, limiting their application in research and practice. One of their main limitations is the quality of the distributional data used to calibrate them, which directly influences the accuracy of model predictions. We propose a standardized methodology to create maps, describing the limitations of occurrence data for covering the distribution of a species. We develop a set of tools based on the general framework of Maps of Biogeographical Ignorance to describe the main sources of data‐driven uncertainty: taxonomic stability, environmental similarity, geographical proximity and temporal decay of the underlying biodiversity data. The so‐derived indicators of data‐driven uncertainty account for inventory completeness, taxonomic quality, time since the surveys and geographical (and environmental) distance to localities with information. These indicators form the basis of ignorance maps, which can be used to visualize the reliability of SDM projections in geographical space, to estimate the uncertainty of these predictions and to identify target survey areas. To demonstrate the application of our approach, we use data on fourteen Iberian species of Scarabaeidae dung beetles. Data‐driven uncertainty is widespread even for this well‐surveyed group; more than 60% of the region has distributional uncertainty values higher than 0.6, and 30% higher than 0.7. Ignorance maps can be jointly evaluated with SDM predictions to generate spatially explicit maps of uncertainty, identifying where predictions are reliable/unreliable. Neglecting such uncertainty can severely affect SDM effectiveness, as it can introduce biases and inaccuracies into the measured species–environment relationships. These errors could result in incorrect theoretical or practical applications, including ill‐advised conservation actions. We therefore advocate for the routine use of ignorance maps or similar techniques as supporting information in SDM applications.

show abstract

High uncertainty in the effects of data characteristics on the performance of species distribution models

Cited by 41 publications

References 82 publications

Reliability in Distribution Modeling—A Synthesis and Step-by-Step Guidelines for Improved Practice

Reliability in Distribution Modeling—A Synthesis and Step-by-Step Guidelines for Improved Practice

Representation of global change drivers across biodiversity datasets

Using maps of biogeographical ignorance to reveal the uncertainty in distributional data hidden in species distribution models

Contact Info

Product

Resources

About