Assessing the reliability of species distribution projections in climate change research

Diversity and Distributions, Journal Year: 2021, Volume and Issue: 27(6), P. 1035 - 1050

Published: Feb. 19, 2021

Abstract Aim Forecasting changes in species distribution under future scenarios is one of the most prolific areas application for models (SDMs). However, no consensus yet exists on reliability such drawing conclusions species’ response to changing climate. In this study, we provide an overview common modelling practices field and assess model predictions using a virtual approach. Location Global. Methods We first review papers published between 2015 2019. Then, use approach three commonly applied SDM algorithms (GLM, MaxEnt random forest) estimated actual predictive performance parameterized with different settings violations assumptions. Results Most relied single (65%) small samples ( N < 50, 62%), used presence‐only data (85%), binarized models' output (74%) split‐sample validation (94%). Our simulation reveals that tends be over‐optimistic compared real performance, whereas spatial block provides more honest estimate, except when datasets are environmentally biased. The binarization predicted probabilities presence reduces models’ ability considerably. Sample size main predictors accuracy, but has little influence accuracy. Finally, inclusion ecologically irrelevant violation assumptions increases accuracy decreases projections, leading biased estimates range contraction expansion. Main predict low average, particularly binarized. A robust by spatially independent required, does not rule out inflation assumption violation. findings call caution interpretation projections climates.

Language: Английский

---

Top ten hazards to avoid when modeling species distributions: a didactic guide of assumptions, problems, and recommendations

Ecography, Journal Year: 2024, Volume and Issue: 2024(4)

Published: Jan. 31, 2024

Species distribution models, also known as ecological niche models or habitat suitability have become commonplace for addressing fundamental and applied biodiversity questions. Although the field has progressed rapidly regarding theory implementation, key assumptions are still frequently violated recommendations inadvertently overlooked. This leads to poor being published used in real‐world applications. In a structured, didactic treatment, we summarize what our view constitute ten most problematic issues, hazards, negatively affecting implementation of correlative approaches species modeling (specifically those that model by comparing environments species' occurrence records with background pseudoabsence sample). For each hazard, state relevant assumptions, detail problems arise when violating them, convey straightforward existing recommendations. We discuss five major outstanding questions active current research. hope this contribution will promote more rigorous these valuable stimulate further advancements.

Language: Английский

---

Integrating food webs in species distribution models can improve ecological niche estimation and predictions

Ecography, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 14, 2025

Biotic interactions play a fundamental role in shaping multitrophic species communities, yet incorporating these into distribution models (SDMs) remains challenging. With the growing availability of interaction networks, it is now feasible to integrate SDMs for more comprehensive predictions. Here, we propose novel framework that combines trophic networks with Bayesian structural equation models, enabling each be modeled based on its predators or prey alongside environmental factors. This addresses issues multicollinearity and error propagation, making possible predict distributions unobserved locations under future conditions, even when predator are unknown. We tested validated our realistic simulated communities spanning different theoretical ecological setups. scenarios. Our approach significantly improved estimation both potential realized niches compared single SDMs, mean performance gains 8% 6%, respectively. These improvements were especially notable strongly regulated by biotic factors, thereby enhancing model predictive accuracy. supports integration various SDM extensions, such as occupancy integrated offering flexibility adaptability developments. While not universal solution consistently outperforms provides valuable new tool modeling community known assumed.

Language: Английский

---

Trade‐offs in covariate selection for species distribution models: a methodological comparison

Ecography, Journal Year: 2019, Volume and Issue: 43(1), P. 11 - 24

Published: Oct. 2, 2019

Species distribution models (SDMs) are a common approach to describing species’ space‐use and spatially‐explicit abundance. With myriad of model types, methods parameterization options available, it is challenging make informed decisions about how build robust SDMs appropriate for given purpose. One key component SDM development the covariates, such as inclusion covariates that reflect underlying processes (e.g. abiotic biotic covariates) act proxies unobserved space time covariates). It unclear different apportion variance among suite influence accuracy performance. To examine trade‐offs in covariation SDMs, we explore attribution spatiotemporal environmental variation across SDMs. We first used simulated species distributions with known preferences compare three types SDM: machine learning (boosted regression tree), semi‐parametric (generalized additive model) mixed‐effects (vector autoregressive model, VAST). then applied same comparative framework case study fish (arrowtooth flounder, pacific cod walleye pollock) eastern Bering Sea, USA. Model type covariate both had significant effects on found including either or typically reproduced patterns abundance tested, but performance was maximized when framework. Our results reveal current generation tools between accurately estimating abundance, spatial patterns, quantifying species–environment relationships. These comparisons can help users better understand sources bias estimate error.

Language: Английский

---

A quantitative review of abundance‐based species distribution models

Ecography, Journal Year: 2021, Volume and Issue: 2022(1)

Published: Dec. 15, 2021

The contributions of species to ecosystem functions or services depend not only on their presence but also local abundance. Progress in predictive spatial modelling has largely focused occurrence rather than As such, limited guidance exists the most reliable methods explain and predict variation We analysed performance 68 abundance‐based distribution models fitted 800 000 standardised abundance records for more terrestrial bird reef fish species. found a large amount models. While many performed poorly, subset consistently reconstructed range‐wide patterns. best predictions were obtained using random forests frequently encountered abundant within same environmental domain as model calibration. Extending outside conditions used training generated poor predictions. Thus, interpolation abundances between observations can help improve understanding patterns, our results indicate extrapolated under changing climate have much greater uncertainty. Our synthesis provides road map key property distributions that underpins theoretical applied questions ecology conservation.

Language: Английский

---

Evaluation metrics and validation of presence-only species distribution models based on distributional maps with varying coverage

Scientific Reports, Journal Year: 2021, Volume and Issue: 11(1)

Published: Jan. 15, 2021

Abstract We examine how different datasets, including georeferenced hardcopy maps of extents and herbarium specimens (spanning the range from 100 to 85,000 km 2 ) influence ecological niche modeling. check 13 available environmental modeling algorithms, using 30 metrics score their validity evaluate which are useful for selection best model. The validation is made an independent dataset comprised presences absences collected in a range-wide field survey Carpathian endemic plant Leucanthemum rotundifolium (Compositae). Our analysis models’ predictive performances indicates that almost all datasets may be used construction species distributional range. Both very local general can produce predictions, more detailed than original ranges. Results also highlight possibility data manually archival sources reconstructions aimed at establishing species’ niches. discuss possible applications those associated problems. For evaluation models, we suggest employing AUC, MAE, Bias. show example AUC MAE combined select model with performance.

Language: Английский

---

Target‐group backgrounds prove effective at correcting sampling bias in Maxent models

Diversity and Distributions, Journal Year: 2021, Volume and Issue: 28(1), P. 128 - 141

Published: Nov. 19, 2021

Abstract Aim Accounting for sampling bias is the greatest challenge facing presence‐only and presence‐background species distribution models; no matter what type of model chosen, using biased data will mask true relationship between occurrences environmental predictors. To address this issue, we review four established correction techniques, empirical with known effort, virtual distributions. Innovation Occurrence come from a national recording scheme hoverflies ( Syrphidae ) in Great Britain, spanning 1983 – 2002. Target‐group backgrounds, distance‐restricted travel time to cities human population density were used account 58 hoverfly. Distributions generated by techniques compared geographical space produced accounting Schoener's distance, centroid shifts range size changes. validate our results, performed same comparisons 50 randomly species. We effort hoverfly structure regime, emulating complex real‐life bias. Main conclusions Models made without any typically distributions that mapped rather than underlying habitat suitability. backgrounds best at unbiased occurrences, but also showed signs overcompensation places. Other methods better no‐correction, often differences difficult visually detect. In line previous studies, when unknown, target‐group provide useful tool reducing effect should be inspected biological realism identify areas potential overcompensation. Given disparity corrected un‐corrected models, constitutes major source error modelling, more research needed confidently issue.

Language: Английский

---

Evaluating presence‐only species distribution models with discrimination accuracy is uninformative for many applications

Journal of Biogeography, Journal Year: 2019, Volume and Issue: 47(1), P. 167 - 180

Published: Sept. 20, 2019

Abstract Aim Species distribution models are used across evolution, ecology, conservation and epidemiology to make critical decisions study biological phenomena, often in cases where experimental approaches intractable. Choices regarding optimal models, methods data typically made based on discrimination accuracy: a model's ability predict subsets of species occurrence that were withheld during model construction. However, empirical applications these involve making inferences continuous estimates relative habitat suitability as function environmental predictor variables. We term the reliability ‘functional accuracy.’ explore link between accuracy functional accuracy. Methods Using simulation approach we investigate whether good predictions distributions correctly infer underlying relationship predictors habitat. Results demonstrate is only informative when simple similar structure true niche, or partitioning geographically structured. utility for selecting with high was low all cases. Main conclusions These results suggest many studies criteria unrelated models’ usefulness their intended purpose. argue modelling need place significantly more emphasis insight into plausibility current maximizing at expense other considerations detrimental both methodological literature this active field. Finally, future development field must include an increased simulation; may be largely uninformative about best practices interpretation relies estimating ecological processes, will unduly penalize biologically approaches.

Language: Английский

---

Testing the ability of species distribution models to infer variable importance

Ecography, Journal Year: 2020, Volume and Issue: 43(12), P. 1801 - 1813

Published: Sept. 2, 2020

Models of species’ distributions and niches are frequently used to infer the importance range‐ niche‐defining variables. However, degree which these models can reliably identify important variables quantify their influence remains unknown. Here we use a series simulations explore how well 1) discriminate between with different 2) calibrate magnitude relative an ‘omniscient’ model. To variable importance, trained generalized additive (GAMs), Maxent boosted regression trees (BRTs) on simulated data tested sensitivity permutations in each predictor. Importance was inferred by calculating correlation permuted unpermuted predictions, comparing predictive accuracy predictions using AUC continuous Boyce index. In scenarios one influential uninfluential variable, failed when training occurrences were < 8–64, prevalence > 0.5, spatial extent small, environmental had coarse resolution autocorrelation low, or pairwise |r| 0.7. When two influenced distribution equally, underestimated species narrow intermediate niche breadth. Interactions they shaped did not affect inferences about importance. acted unequally, effect stronger overestimated. GAMs discriminated more than BRTs, but no algorithm consistently well‐calibrated vis‐à‐vis omniscient Algorithm‐specific measures like Maxent's change‐in‐gain metric less robust permutation test. Overall, high connote inferential capacity. As result, requirements for measuring likely stringent creating accuracy.

Language: Английский

---

Comparing sample bias correction methods for species distribution modeling using virtual species

Ecosphere, Journal Year: 2021, Volume and Issue: 12(3)

Published: March 1, 2021

Abstract A key assumption in species distribution modeling (SDM) with presence‐background (PB) methods is that sampling of occurrence localities unbiased and any bias proportional to the background environmental covariates. This rarely met when SDM practitioners rely on federated museum records from natural history collections for geo‐located occurrences due inherent found these collections. We use a simulation approach explore effectiveness three developed account PB frameworks. Two careful filtering observation data—geographic thinning (G‐Filter) (E‐Filter)—while third, FactorBiasOut, creates selection weights data locations toward areas where dataset was sampled. While have been assessed previously, evaluation has emphasized spatial predictions habitat potential. Here, we dig deeper into by exploring how not only affects potential, but also our understanding niche characteristics such as which explanatory variables response curves best represent species–environment relationships. simulate 100 virtual ranging generalist specialist their preferences introduce geographic at intensity levels measure each correction method (1) predict true probability across study area, (2) recover relationships, (3) identify variables. find FactorBiasOut most often showed greatest improvement recreating known distributions did no better correctly identifying covariates or relationships than G‐Filter E‐Filter methods. Narrow are problematic biased calibration datasets, can, some cases, make worse.

Language: Английский

---