Machine learning and deep learning—A review for ecologists

Methods in Ecology and Evolution, Journal Year: 2023, Volume and Issue: 14(4), P. 994 - 1016

Published: Feb. 13, 2023

Abstract The popularity of machine learning (ML), deep (DL) and artificial intelligence (AI) has risen sharply in recent years. Despite this spike popularity, the inner workings ML DL algorithms are often perceived as opaque, their relationship to classical data analysis tools remains debated. Although it is assumed that excel primarily at making predictions, can also be used for analytical tasks traditionally addressed with statistical models. Moreover, most discussions reviews on focus mainly DL, failing synthesise wealth different advantages general principles. Here, we provide a comprehensive overview field starting by summarizing its historical developments, existing algorithm families, differences traditional tools, universal We then discuss why when models prediction where they could offer alternatives methods inference, highlighting current emerging applications ecological problems. Finally, summarize trends such scientific causal ML, explainable AI, responsible AI may significantly impact future. conclude powerful new predictive modelling analysis. superior performance compared explained higher flexibility automatic data‐dependent complexity optimization. However, use inference still disputed predictions creates challenges interpretation these Nevertheless, expect become an indispensable tool ecology evolution, comparable other tools.

Language: Английский

---

A comparison of machine learning and statistical species distribution models: Quantifying overfitting supports model interpretation

Ecological Modelling, Journal Year: 2023, Volume and Issue: 481, P. 110353 - 110353

Published: April 3, 2023

Species distribution models are commonly applied to predict species responses environmental conditions. A wide variety of with different properties exist that vary in complexity, which affects their predictive performance and interpretability. Machine learning algorithms increasingly used because they capable capture complex relationships often better prediction. However, inform management, it is important a model predicts well for the right reasons. It remains challenge select reasonable level complexity captures true relationship between response explanatory variables as good possible rather than fitting noise data. In this study we ask: 1) how much can gain by using models, 2) does affect degree overfitting, 3) do inferred differ among what learn from them? To address these questions, eight probability occurrence freshwater macroinvertebrate taxa based on 2729 Swiss monitoring samples. We compared terms during cross-validation generalization out calibration domain ("extrapolation" or transferability). agnostic tools shed light Contrary our expectation, all predicted similarly cross-validation, while no null out-of-domain average over taxa. Performance was best intermediate prevalence. More slightly standard statistical but were prone overfitting. Overfitting indicates describes not only signal data also part noise. This impedes interpretation shapes learned model, one cannot distinguish Furthermore, strongly overfitting irregular strong interactions ecologically plausible. Thus, study, minor more outweighed Ecological field input typically sources variability, sampling, measurement process stochasticity. therefore call caution when data-driven about management. such cases, recommend compare range regarding performance, understand robustness responses.

Language: Английский

---

Impacts of climate change on vegetation pattern: Mathematical modeling and data analysis

Physics of Life Reviews, Journal Year: 2022, Volume and Issue: 43, P. 239 - 270

Published: Oct. 25, 2022

Language: Английский

---

Comparing the performance of global, geographically weighted and ecologically weighted species distribution models for Scottish wildcats using GLM and Random Forest predictive modeling

Ecological Modelling, Journal Year: 2024, Volume and Issue: 492, P. 110691 - 110691

Published: April 8, 2024

Species distribution modeling has emerged as a foundational method to predict occurrence and suitability of species in relation environmental variables advance ecological understanding guide conservation planning. Recent research, however, shown that species-environmental relationships habitat model predictions are often nonstationary space, time context. This calls into question approaches assume global, stationary realized niche use predictive describe it. paper explores this issue by comparing the performance models for wildcat hybrid based on (1) global pooled data across individuals, (2) geographically weighted aggregation individual models, (3) ecologically (4) combinations geographical weighting. Our study system included GPS telemetry from 14 hybrids Scotland. We developed both using Generalized Linear Models (GLM) Random Forest machine learning compare these differing algorithms how they analyses. validated predicted four different ways. First, we used independent hold-out collared hybrids. Second, 8 additional previous were not training sample. Third, sightings sent public researchers expert opinion. Fourth, collected camera trap surveys between 2012 – 2021 various sources produce combined dataset showing where wildcats had been detected. results show validation individuals train provides highly biased assessment true other locations, with particular appearing perform exceptionally (and inaccurately) well when same models. Very obtained three sources. Each sets gave result terms best overall model. The average datasets suggested produced potential was an ensemble Model GLM suggests debate over whether which vs is superior or aggregated may be false choice. presented here prediction applies combination all framework.

Language: Английский

---

Spatial heterogeneity of ecosystem service bundles and the driving factors in the Beijing-Tianjin-Hebei region

Journal of Cleaner Production, Journal Year: 2024, Volume and Issue: unknown, P. 144006 - 144006

Published: Oct. 1, 2024

Language: Английский

---

Modeling the effects of climate change scenarios on the potential distribution of Vespa crabro Linnaeus, 1758 (Hymenoptera: Vespidae) in a Mediterranean biodiversity hotspot

Ecological Informatics, Journal Year: 2025, Volume and Issue: unknown, P. 103006 - 103006

Published: Jan. 1, 2025

Language: Английский

---

Not seeing the forest for the trees: Generalised linear model out-performs random forest in species distribution modelling for Southeast Asian felids

Ecological Informatics, Journal Year: 2023, Volume and Issue: 75, P. 102026 - 102026

Published: Feb. 18, 2023

Species Distribution Models (SDMs) are a powerful tool to derive habitat suitability predictions relating species occurrence data with features. Two of the most frequently applied algorithms model species-habitat relationships Generalised Linear (GLM) and Random Forest (RF). The former is parametric regression providing functional models direct interpretability. latter machine learning non-parametric algorithm, more tolerant than other approaches in its assumptions, which has often been shown outperform algorithms. Other have developed produce robust SDMs, like training bootstrapping spatial scale optimisation. Using felid presence-absence from three study regions Southeast Asia (mainland, Borneo Sumatra), we tested performances SDMs by implementing four modelling frameworks: GLM RF bootstrapped non-bootstrapped data. With Mantel ANOVA tests explored how combinations influenced their predictive performances. Additionally, scale-optimisation responded species' size, taxonomic associations (species genus), area algorithm. We found that choice algorithm had strong effect determining differences between SDMs' predictions, while no effect. followed species, were main factors driving scales identified. trained showed higher performance, however, revealed significant only explaining variance observed sensitivity specificity and, when interacting bootstrapping, Percent Correctly Classified (PCC). Bootstrapping significantly explained specificity, PCC True Skills Statistics (TSS). Our results suggest there systematic identified produced vs. RF, but neither approach was consistently better other. divergent inconsistent abilities analysts should not assume inherently superior test multiple methods. implications for SDM development, revealing inconsistencies introduced on optimisation, selecting broader RF.

Language: Английский

---

Exploring nonstationary limiting factors in species habitat relationships

Ecological Modelling, Journal Year: 2024, Volume and Issue: 490, P. 110663 - 110663

Published: Feb. 29, 2024

Species distribution modeling is widely used to quantify and predict species-environment relationships. Most past applications methods in species assume context independent stationary relationships between patterns of occurrence environmental variables. There has been relatively little research investigating dependence nonstationarity modeling. In this paper we explore spatially varying limiting factors using high resolution telemetry data from 14 individual wildcat hybrids distributed across geographical gradients Scotland. (1) We proposed that nonstationary would be indicated by significant association statistical measures variability predictors the predictive importance those (2) further most factor observed related spatial variation a lesser amount mean value variables within study sites. (3) Additionally, anticipated when there was relationship an its as predictor positive, such higher associated with variable (following theory factors). (4) Conversely, roughly evenly split positive negative relationships, given could become either they are highly abundant or value, rare low particular landscape, depending on nature for ecological variable. (5) Finally, hypothesized frequency supported differ among groups, were directly key resources more likely than have indirect impacts hybrid habitat selection foraging. Our results show assumptions global, associations not met many models, requiring explicit consideration scale paradigm. found both standard deviation strong whether will differentially important occurrence. confirmed it sampled data, abundant. The differed essential ecology

Language: Английский

---

Interactions between vegetation and river morphodynamics. Part I: Research clarifications and challenges

Earth-Science Reviews, Journal Year: 2024, Volume and Issue: 253, P. 104769 - 104769

Published: May 17, 2024

Rivers have an intricate relationship with the vegetation that colonizes them. Riparian plants, capable of thriving within river corridors, both respond to and influence geomorphology. Yet interactions between morphodynamics tend be context specific, making it challenging generalize findings locations. The current comprehension interaction physical processes, especially its effects on morphodynamics, still lacks clarity. This article examines numerous sources variation in plant responses to, on, morphodynamics. Vegetation influences geomorphological parameters vary terms intensity spatial extent along gradient energy according fluvial style. Whilst feedbacks are readily discernible at a local scale, larger scales, can remain difficult precisely determine cause-and-effect relationships link hydrogeomorphic drivers outcomes their feedbacks. is problematic for those give rise emergent system landscape behaviour meandering island braided rivers. By contrast, certain configurations, such as anabranching rivers, imprint riverscape clearly evident. also supported by evidence from ancient alluvial record. Through this review, we highlight key perspectives wide range modern rivers varied configuration order inform future studies

Language: Английский

---

Modeling Dynamic Processes in the Black Sea Pelagic Habitat—Causal Connections between Abiotic and Biotic Factors in Two Climate Change Scenarios

Sustainability, Journal Year: 2024, Volume and Issue: 16(5), P. 1849 - 1849

Published: Feb. 23, 2024

The paper contributes to the Sustainable Development Goals (SDGs) targeting Life Below Water by introducing user-friendly modeling approaches. It delves into impact of abiotic factors on first two trophic levels within marine ecosystem, both naturally and due human influence. Specifically, study examines connections between environmental parameters (e.g., temperature, salinity, nutrients) plankton along Romanian Black Sea coast during warm season over a decade. research develops models forecast zooplankton proliferation using machine learning (ML) algorithms gathered data. temperature significantly affects copepods “other groups” densities season. Conversely, no discernible is observed dinoflagellate Noctiluca scintillans blooms. Salinity fluctuations notably influence typical phytoplankton proliferation, with phosphate concentrations primarily driving widespread explores scenarios for forecasting growth: Business as Usual, predicting modest increases in constant nutrient levels, Mild scenario, projecting substantial salinity alongside significant decrease 2042. findings underscore high under scenarios, particularly pronounced second surpassing around 70%. These findings, indicative eutrophic potential implications altered ecosystem health, aligning SDGs focused Water.

Language: Английский

---