Biodiversity Informatics, 19, 2025, pp. 33-45 33 WELCOME TO THE MACHINE: A PAN-CONTINENTAL OVERVIEW OF MACHINE LEARNING APPLICATIONS IN ECOLOGY AND CONSERVATION Juliano A. Bogoni1*, Derick Victor de Souza Campos2*, Claumir C. Muniz2, Manoel dos Santos-Filho1, Jéssica Eloá Poletto3 1 Universidade do Estado de Mato Grosso, Centro de Pesquisa de Limnologia, Biodiversidade e Etnobiologia do Pantanal, Programa de Pós-Graduação em Ciências Ambientais, Laboratório de Mastozoologia, Cáceres, MT, Brazil. 2 Universidade do Estado de Mato Grosso, Centro de Pesquisa de Limnologia, Biodiversidade e Etnobiologia do Pantanal, Programa de Pós-Graduação em Ciências Ambientais, Laboratório de Investigação Ambiental do Pantanal Norte – LIPAN, Cáceres, MT, Brazil. 3 Faculdade de Ciências Médicas da Universidade Estadual de Campinas, Piracicaba, SP, Brasil. Abstract. Machine-learning emerged as an excellent alternative to understanding ecological patterns and pro- cesses at different spatiotemporal scales. The study aimed to offer a global overview of the status quo on the use of machine-learning in ecology and conservation globally. Using keywords in the Scopus engine, we in- dexed all publications in ecology and conservation using machine-learning. We employed descriptive statistics and regressions models to provide an overview and predict geopolitical patterns. The majority of manuscripts were condensed in economically affluent countries, such as the United States (USA) and China (CHN) which together amount to 91 (36.8%) studies. There is a spatial aggregation in the authors’ affiliations, once 182 (73.7%) studies derived from both Nearctic and Palearctic teams, whereas Tropical teams published 65 (26.3%) manuscripts and the most-cited papers also are concentrated in northern regions. In ecology and conservation, machine-learning first appear in the literature in 2003. Since then, the number of publications has increased exponentially, from 09 manuscripts in 2010, to 120 manuscripts 10 years later. Most studies (N = 173; 70.1%) focused on landscape and vertebrate ecology. The primary aims of the publications were widely variable but strongly adherent to providing the best-information on both landscape-scale classifications and species distribu- tion modelling. The manuscripts encompass different methods, from maximum entropy to boosted regression trees and random forest, sometimes using a range of deep-learning architectures. Finally, the predictive varia- bles (i.e., mammal diversity and per capita GDP) do not exert significant influences on the number of studies published. Finally, we recommend a well-structured and collaborative agenda aiming to integrate less-resourced countries into scientific advancements, fostering more equitable and effective responses to global environmen- tal challenges. Keywords: data analysis, global-scale, informatics, numerical ecology, tropical forest. *corresponding authors: bogoni.ja@gmail.com | derick@unemat.br mailto:bogoni.ja@gmail.com mailto:derick@unemat.br Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 34 Introduction Ecology is a relatively young science that fundamen- tally seeks to understand the causes and consequences (i.e., processes) of diversity patterns and species distribu- tions across global environments (Brown, 1995; Haeckel, 1866). Universal features of ecological processes exhibit mathematical properties that are inherently non-linear and complex, historically addressed only through mathemati- cal approximations (Bogoni et al., 2019; Conway, 1977; May, 1976). Since the 1920s, explicit models as in Vol- terra (1926), Lotka (1925), Elton (1924) have been em- ployed in ecology to predict and describe synchronization mechanisms in animal behaviour as Araujo et al. (2013), predator-prey relationships and dynamics in Sherratt et al. (1997), Kar et al. (2010), host-parasitoid interactions in Hassell (2000), and various other ecological dynamics, such as seed predation and dispersal (Bogoni et al., 2019) and species distribution (Bogoni and Tagliari, 2021; Tag- liari et al., 2023). As global biodiversity faces unprecedented and wide- spread declines in modern history (Bogoni et al., 2020; Ceballos et al., 2020), addressing contemporary ecological challenges—such as biodiversity loss, climate change, and the growing demand for ecosystem services—has become a pressing priority for ecologists (Rammer and Seidl, 2019). This issue is particularly critical in the global trop- ics, where habitat loss is most severe. For example, within just half a decade in the early 21st century, from 2000 to 2005, tropical forests in both wet and dry regions expe- rienced a staggering loss of approximately 475,000 km² of forest cover (around 50%) (Hansen et al., 2010). Thus, understanding how to mitigate natural habitat degradation processes, particularly across tropical landscapes, is im- perative for safeguarding biodiversity on a global scale. Machine-learning methods have emerged as an excel- lent alternative for predicting and understanding ecologi- cal patterns and processes across various spatiotemporal scales. These methods consist of computational algorithms designed to analyse the often non-linear structures of com- plex data, thereby generating predictive models based on estimated patterns (Rammer and Seidl, 2019; Olden et al., 2008). In comparison to classical statistical approaches (e.g., regression models), machine learning relies on com- putational power to identify and model complex relation- ships, prioritizing predictive accuracy over the estimation of parameters and confidence intervals (Breiman, 2001). Positioned at the intersection of statistics and computer science—and driven by advances in artificial intelligence and data science—the application of machine learning in ecology and conservation has emerged as a rapidly expanding field (Rammer and Seidl, 2019; Jordan and Mitchell, 2015). However, the conducted search reveals a lack of studies examining how machine-learning methods are being applied in ecology and conservation or providing a pan-continental overview of how these methods can assist ecologists in ensuring biodiversity persistence across glob- al landscapes. Additionally, disparities between local and global scales have historically biased scientific advance- ments (Bogoni et al., 2021; Hughes et al., 2021). Modern ecological methods, such as machine learning, tend to be more widely utilized in regions with more affluent econ- omies, where access to advanced computational resources is readily available (e.g., Bogoni et al., 2021). In contrast, countries where scientific development is still in its early stages—and which often host the highest levels of biodi- versity—have likely made less use of these computational advancements. This limitation is particularly concerning when it comes to understanding ecological patterns and processes and addressing the severe biodiversity crises in these regions. Our primary aim was to provide a global pictorial overview of the use of machine learning methods in ecol- ogy and conservation by assembling a comprehensive database through an extensive literature review. Specif- ically, we sought to: (1) quantitatively and qualitatively evaluate how machine learning methods have been ap- plied in peer-reviewed ecology and conservation publica- tions across a pancontinental scale; (2) assess the research questions, focal areas, and taxonomic groups addressed in studies involving machine learning within the disciplines of ecology and conservation; and (3) discuss how coun- tries with lower investment in science or less frequent use of modern analytical methods can leverage these theoreti- cal and analytical advancements to address their environ- mental and ecological challenges. We tested the following hypotheses: (1) the vast majority of publications originate from authors affiliated with economically affluent coun- tries; (2) the application of machine learning in ecology and conservation is predominantly focused on species distribution models and landscape ecology; and (3) at a predictive level, a country’s per capita income positive- ly correlates with the number of its publications, while mammal diversity (used as a proxy for country-scale bio- diversity) is negatively correlated with publication output, as poorer countries often harbor greater biodiversity than wealthier nations. Our goal is to provide a global synthesis of how machine learning has been integrated into ecologi- Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 35 cal and conservation research. By identifying geographic, taxonomic, and methodological patterns in the literature, it discusses existing disparities in scientific output and highlights opportunities for greater inclusion of biodiver- sity-rich yet underrepresented countries. Material and Methods Our study is characterized as data gathering research, with a posteriori application of plans with descriptive sta- tistical analysis method and mapping. Thus, in early May (04th-May-2023), we used the Scopus search engine1 to in- dex and aggregate all publications (published or in press) in ecology and conservation that used machine-learning methods. To do so, we used a conjunct of keywords al- lied to AND/OR operators, as follows: “machine learn- ing” AND/OR “deep learning” AND “ecology” AND/ OR “conservation (similar to Magioli et al., 2015). This allowed us to compile the number of scientific publica- tions that included these terms in their manuscript titles, abstracts, and keywords. After capturing the studies, we extracted the follow- ing information: (1) the total number of studies; (2) the publication year of each manuscript; (3) the taxonomic group(s) studied or the primary approach (e.g., landscape ecology); (4) the country of primary affiliation of the first author; (5) the city-based geolocation of the study in latitu- dinal and longitudinal decimal degrees; (6) the set of study keywords; (7) the study title; and (8) the total number of citations for each manuscript. Documents with incomplete information, missing location, author, or keywords, were not consider in our analysis. Once the database was consolidated, we used descrip- tive statistics and mapping to provide a pancontinental overview of the state of machine-learning applications in ecology and conservation. We analysed the number of published articles, the annual publication trends, and the distribution of articles by country. Additionally, we generated a word-cloud based on the top-100 keywords provided by the authors. We also explored the proportion of taxonomic groups or approaches studied in a pictori- al synthesis of studies. Then, we identified the top-cited manuscripts (i.e., all those with more than 100 citations) and provided detailed insights into their primary aims, methods, and the specific sub-disciplines of ecology and conservation they addressed. The data analyses were performed in R v.4.0.5 (R Core Team, 2023) using stats (R Core Team, 2023), sp (Bivand et al., 2013; Pebesma and Bivand, 2005), map- tools (Bivand and Lewin-Koh, 2021) and the dependent 1 https://www.scopus.com/home.uri. R-package. Additionally, we utilized data from the Inter- national Union for Conservation of Nature (IUCN) Red List of Threatened Species2 to obtain country-level mam- mal richness as a proxy for national biodiversity. Mam- mals were chosen for this analysis as they represent one of the most extensively studied and well-documented taxo- nomic groups worldwide (Bogoni et al., 2021). We also sourced data from the World Bank3 to obtain country-level per capita Gross Domestic Product, hereaf- ter, GDP. Both variables mentioned above were subse- quently used to predict the number of studies published per country. To do so, we used a generalized linear regres- sion model (glm), using Poisson distribution under the data variance given that the response variable is a count (non-negative integers) (Dobson, 1990), correcting the predictive data asymmetry by log x + 1. Thus, the glm was performed as follows: Studies(ith country) ~ log(MammalRichness(ith country )) + log(GDP(ith country)) where Studies(ith country) is the number of studies published in the country, log(MammalRichness(ith country )) is the natu- ral log of mammalian species richness in the country, and log(GDP(ith country)) is the natural log of the Gross Domestic Product for country. Then, deriving a bivariate effect plot for both predic- tive variables vs. number of studies. The statistics of glm model were presented for z- and p-values, estimates and standard errors. The regression analysis and bivariate plots were performed in R v.4.0.5 (R Core Team, 2023) based on the stats R-package (R Core Team, 2023). Results Our literature search identified 251 documents relat- ed to machine learning and/or deep learning with a focus on ecology and conservation. After applying an intuitive filter to exclude documents with incomplete information, our study was able to analyse pancontinental trends based on 247 manuscripts. This overview revealed that the vast majority of published manuscripts were concentrated in economically affluent countries, such as the United States (USA) and China (CHN), which together accounted for 91 studies (36.8%) (Fig. 1A). Furthermore, there was a notable spatial concentration in the primary affiliations of corresponding authors: 182 studies (73.7%) originated from teams in the Nearctic and Palearctic regions, particu- larly Europe, while Neotropical and Paleotropical teams 2 https://www.iucnredlist.org/. 3 https://www.worldbank.org/en/home. https://www.scopus.com/home.uri https://www.iucnredlist.org/ https://www.worldbank.org/en/home Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 36 contributed 65 studies (26.3%). Of these, only 11 studies (4.5%) were published in the biodiversity-rich Neotropical realm (Fig. 1B). Similarly, the most-cited papers also pre- dominantly originated from northern countries, mirroring the geographic bias in publication origins (Fig. 1B). In terms of publication accumulation, the ma- chine-learning publications in ecology and conservation first appear on literature in 2003. Yet, this topic increased exponentially since the 2010s (Fig 1B). In 2010 this over- view indicated nine published manuscripts, whereas 10- yrs later this number reached 120 publications (Fig. 1C), an increment of 111 publications. Moreover, in 2021 and 2022, these techniques seem to have become a hot-topic and consolidating more than 100 (>40.5%) publications in ecology and conservation (≥50 manuscripts per year; Fig 1C). Across the countries, most studies (N = 173; 70.1%) are fundamentally focused on landscape ecology (which here includes species distribution), bird ecology and con- servation, wildlife ecology with more than one taxonomic group, mammal ecology and conservation, plant ecology and conservation, fish ecology and conservation, and in- vertebrate ecology and conservation (Fig 2A). Proportion- ally, these areas respond to 28, 10, 10, 7, 6, 5, and 5%, respectively, of all approaches (Fig. 2B). log(N of studies) C ou nt ry c od e 0 50 100 150 200 250 20 03 20 07 20 09 20 11 20 13 20 15 20 17 20 19 20 21 20 23 Year C um m ul at iv e pu bl ic at io ns A B C USA CHN AUS GBR IND CAN FRA ESP JPN DEU ITA TUR ZAF BRA MEX NLD FIN PRT LKA RUS COL IRN IDN SWE SVN MAR VNM HKG POL CHE KOR LVA URY CZE SVK BEL BEN SGP NZL ETH NPL EGY SRB ROU MYS 0 1 2 3 4 Figure 1. Number of studies (log-scaled) involving machine- learning in ecology and conservation among countries (A), its global geo-distribution and citation number (log-scaled; B), and the cumulative publications over the years (C). Source: Original search results. Landscape.ecology Bird.ecology Wildlife.ecology Mammal.ecology Plant.ecology Fish.ecology Invertebrate.ecolgoy Climate.change Data.obtaining Freshwater.ecology Marine.ecology Urban.ecology Acoustic.ecology Soil.conservation Citizen.science Disease.ecology Green.energy Reptile.ecology Amphibian.ecology Behaviour.ecology Conservation.ecology Genetic Multidisciplinary Transdisciplinary Agroecology Education Polution USA CHN AUS GBR IND CAN FRA ESP JPN BRA DEU ITA TUR ZAF FIN LKA MEX NLD PRT COL IDN IRN RUS SVN SWE MAR BEL BEN CHE CZE EGY ETH HKG KOR LVA MYS NPL NZL POL ROU SGP SRB SVK URY VNM A B Landscape ecology: 28% Bird ecology: 10% Wildlife ecology: 10% Mammal ecology: 7% Plant ecology: 6% Fish ecology: 5% Invertebrate ecology: 5% Climate change: 3% Data obtaining: 3% Freshwater ecology: 3% Marine ecology: 3% Urban ecology: 3% Acoustic ecology: 2% Other approachers: <2% Figure 2. (A) Network between publication-based country vs. major approach of machine-learning studies in ecology and conserva- tion; and (B) Percentage (and main authors keywords) of approaches in ecology and conservation using machine-learning methods. Source: Original search results Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 37 Analysis of the top-cited manuscripts (N = 14; 6%), totaling 9887 citations, reveals that the primary objectives of these publications varied significantly, yet consistently focused on providing high-quality information on both landscape-scale classifications and species distribution modeling (Table 1). Whereas, some top-cited manuscripts provide comparisons between machine-learning algo- rithms in solving an ecological or conservation issue (Ta- ble 1). These 100% northern-based researches, published in high-impact journals, encompasses different methods, from maximum entropy models to boosted regression trees and random forest, and sometimes uses a vast gamma of deep learning architectures (e.g., AlexNet, NiN, VGG, Goog- LeNet, and ResNet; see Norouzzadeh et al. (2018); Table 1). Finally, the mammal diversity (species richness) ex- erts significant influences on the number of country-scale studies, whereas per capita GDP does not. Mammal di- versity had a positive and significant tendency [z-score = 2.362; estimates = 0.41; se = 0.17; p = 0.02] while per capita GDP had a positive but non-significant tendency [z = 1.357; estimate = 0.14; se = 0.10; p = 0.175] in influenc- ing the patterns of global distribution of studies (Fig. 3). log(Mammal diversity) N um be r o f s tu di es log(per capita GDP) Nearctic Neotropic Palearctic Paleotropic 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Biogeographic realm lo g (N um be r o f s tu di es ) Nearctic Neotropic Palearctic Paleotropic 3.5 4.0 4.5 5.0 5.5 6.0 6.5 Biogeographic realm lo g (M am m al d iv er si ty ) Nearctic Neotropic Palearctic Paleotropic 1 2 3 4 5 6 7 Biogeographic realm lo g (p er c ap ita G PD ) C D E 0 5 10 15 20 4 5 6 5 10 15 2 4 6 BA Figure 3. Bivariate plot between the number of studies per country vs. mammal diversity (A) and per capita GDP (B), and distribu- tion of per country studies (C), mammal diversity (D), and per capita GDP (E) across the major biogeographic realms. Source: Original search results Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 38 This model presented deviance and goodness-of-fit (G-sta- tistic) with p < 0.001, low overdispersion (OD = 0.346) and diagnostic residuals plots suggest that the distribution is satisfactory in terms of residuals. Discussion The complexities of ecosystems and the challenges of managing and protecting biodiversity require constant computational advances (Chaves, 2013), in which, ma- chine-learning approaches, poses as strong candidate to help ecologists perform sophisticated data analysis. Ma- chine-learning algorithms have a high capacity of process- ing large datasets, identify patterns, and make predictions that can support in decision-making and resource alloca- tion (Rammer and Seidl, 2019). Our main results showed that machine-learning has been widely used in habitat mapping particularly within landscape ecology, predictive modelling, and other ecological data analysis devoted to understanding patterns and drivers of diversity and species distribution, especially in vertebrate and plant ecology. The integration of satellite imagery and machine-learn- ing provides a powerful tool for mapping and monitoring habitats, tracking deforestation, and assessing land-use changes (McLaren et al., 2018; Kampichler et al., 2010). Moreover, machine-learning can predict species distribu- tion, movement patterns, and migration routes, supporting in conservation planning (Pittman and Brown, 2011). In the context of ecological data analysis, machine-learning approaches were used to process large datasets derived from field observations and sensors (e.g., Christin et al., 2019; Stowell and Plumbley, 2014) as to identify complex relationships and patterns in ecological systems. These main approaches address important ecological and conservation challenges. Both landscape ecology and predictive modelling are essential tools for understanding and responding to the accelerating changes in habitats and climate. Ecologists are increasingly required to improve their analytical skills to better predict biodiversity and spe- cies distribution responses to highly modified landscapes and pervasive climatic changes. Understanding the pro- cess that influences the species distributions is a critical is- sue in ecology and conservation (Franklin, 2009; Hutchin- son, 1957), especially for species under threats (Phillips et al., 2004). In this context, machine-learning algorithms have become widely used in biogeography, ecology, and conservation biology to estimate the relationship between species occurrences and environmental variables (Elith et al., 2011; Elith and Leathwick, 2009; Franklin, 2009). For instance, species distribution models (SDMs) en- able researchers to explore key questions in conservation, ecology, and evolution, such as: (1) determining priority areas for conservation and contributing to the protection of species (Faleiro et al., 2013; García, 2006; Chen and Pe- terson, 2002); (2) understanding invasive process of spe- cies (Giovanelli et al., 2008; Ficetola et al., 2007); (3) gen- erating ecocultural niche modelling that reflects ecological influences on past human culture distributions (Banks et al., 2008); (4) proposing past distributions of species (Car- naval and Moritz, 2008; Hugall et al. 2002); and (5) pre- dicting future distributions of species under changing of climatic and environmental characteristics (Tagliari et al., 2023; Bogoni and Tagliari, 2021; Siqueira and Peterson, 2003). Our results on the use of machine learning methods in ecology and conservation have shown an exponential in- crease since 2017. In 2010, our search identified nine pub- lished manuscripts, whereas ten years later, this number had increased by 111 publications. Furthermore, our find- ings reveal a strong geographic concentration of ecologi- cal studies using machine learning approaches in northern countries. Historically, the vast majority of scientific pub- lications are located in these regions. For instance, only in 2020, the US authors signed about ~755,000 scientific publications, while Brazilian authors reached ~100,000 publications on Scopus (see search logs4), representing 86.6% less. These values are reflected in the context of machine learning approaches in ecology, as our results in- dicate that Brazilian authors published 92.0% fewer man- uscripts than their counterparts in the United States. Brazil is undoubtedly Earth’s most biodiverse coun- try, harbouring the largest set of known and unknown spe- cies (Moura and Jetz, 2021), but still presenting glaring knowledge gaps such as the multispecies Wallacean short- falls (Bogoni et al., 2021). Addressing these discrepancies requires substantial investment in scientific research. In contrast, with the exception of 2022 – when investments in science were temporarily resumed –, the Brazilian govern- ment has systematically deepened budget cuts for science since 2017 (see Angelo, 2017; Escobar, 2017). In terms of machine-learning, we advocated that the investments should be addressed to form and solidify hu- man resources and increase the computational power of institutions, especially all those located across the coun- tryside, given that both these factors are critical to per- forming high-complex analysis. Moreover, our findings based on proxies of biodiversity and financial resources in 4 https://www.scopus.com/results/results.uri?st1=United+States&st2= &s=AFFILCOUNTRY%28United+States%29&limit=10&origin=resul tslist&sort=plf-f&src=s&sot=b&sdt=cl&sessionSearchId=1f06790275- 90dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 | https:// www.scopus.com/results/results.uri?st1=United+States&st2=&s=AF FILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort =plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c44- 7bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28United+States%29&limit=10&origin=resultslist&sort=plf-f&src=s&sot=b&sdt=cl&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28United+States%29&limit=10&origin=resultslist&sort=plf-f&src=s&sot=b&sdt=cl&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28United+States%29&limit=10&origin=resultslist&sort=plf-f&src=s&sot=b&sdt=cl&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28United+States%29&limit=10&origin=resultslist&sort=plf-f&src=s&sot=b&sdt=cl&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort=plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort=plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort=plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort=plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 https://www.scopus.com/results/results.uri?st1=United+States&st2=&s=AFFILCOUNTRY%28Brazil%29&limit=10&origin=searchbasic&sort=plf-f&src=s&sot=b&sdt=b&sessionSearchId=1f0679027590dc7c447bf72ed3d85980&yearFrom=2020&yearTo=2020 Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 39 predicting the pancontinental patterns in the application of machine-learning in ecology showed no significant influ- ence on the number of studies published globally. Howev- er, mammal diversity showed a positive tendency whereas per capita GDP had a negative tendency in potentially influencing the distribution patterns of published studies. Economically affluent regions — despite having low bio- diversity when compared to tropical countries — typically account for the vast majority of scientific projects research fundings, while biodiversity-rich countries tend to have lower per-capita incomes. Furthermore, despite the high- ly-recognised quality of researches led by global south scientists—often conducted under highly adverse field and financial conditions and whose legacy chips away at much of our knowledge gaps (Bogoni et al., 2021)—rarely is a tropical country a protagonist in terms of publications number. In this context, for example, Canada and China to- gether harbour a total of 753 mammalian species (195 and 558, respectively; Smith and Xie, 2013; Banfield, 1974), whereas Brazil has 775 mammal species described (Abreu et al., 2022). Therefore, a kind of per capita publication in relation to mammalian species reaches 0.07 and 0.006, re- spectively, representing 13-fold more publications lead by these northern countries weighted by its biodiversity when compared to Brazil. Yet, our results indicate that this is not an exclusivity for Neotropical countries, such as Brazil. Excepting for Australia, Paleotropical countries amount to 25 studies, representing only ~10% of the total. Moreover, the top-10 richest countries in terms of GDP per capita (70% of them located in the Nearctic of Palearctic) amount by 38.1% of publications (N = 94), while the top-10 poorer countries (60% located in the Paleotropical region) signed only 10.1% of scientific publications (N = 27) focused in machine-learning in ecology. Our initial hypotheses were partially corroborated. Our results indicated that the vast majority of publica- tions derive from authors affiliated in economically afflu- ent countries, and the major focus of machine-learning in ecology and conservation was widely applied to species distribution models and landscape ecology. In predictive terms, although there is a clear tendency for richer coun- tries to publish more than poorer ones, and a negative re- lationship between biodiversity and publication output, mammal diversity showed a positive and significant trend, while per capita GDP showed a positive but non-signifi- cant trend. Based on our global overview of the use of machine learning methods in ecology and conservation, we con- clude that a wide range of ecological and conservation issues has already been addressed using these techniques. Machine learning thus represents a strong promise for the coming years, contingent only on the availability of hu- man and financial resources. This promise can therefore be solidified by a fine-tuned agenda aiming to initiate a discussion concerning countries with lower investments in scientific research and those that employ fewer mod- ern analytical methods. Such an agenda would also allow for an exploration of how these nations can benefit from theoretical and analytical advancements to address their persistent environmental and ecological challenges. Final- ly, collaboration between scientific communities across different economic levels could enable more equitable and effective responses to shared global challenges. Declarations Funding - Not Applicable Conflicts of interest/Competing interests The authors have no conflicts of interest to declare. Availability of data and material Data used here are widely available at Scopus search engine (https://www.scopus.com/home.uri). Code availability See Supporting Information Code 1 Author Contributions Statement JAB and DVSC: conceptualization, data acquisition, data analysis and figures, writing & revising the original draft; DVSC, CCM, MSF & JEP: conceptualization, review and editing. References Abreu, E.F., Casali, D., Costa-Araújo, R., Garbino, G.S.T., Libardi, G.S., Loretto, D., Loss, A.C., Marmontel, M., Moras, L.M., Nascimento, M.C., Oliveira, M.L., Pavan, S.E., Tirelli, F.P., 2022. Lista de Mamíferos do Brasil (2022-1) [Dataset]. Zenodo. https://doi. org/10.5281/zenodo.7469767 Angelo, C., 2017. Scientists plead with Brazilian govern- ment to restore funding. Nature 550: 166–167. https:// doi.org/10.1038/nature.2017.22757 Araujo, S.B.L., Rorato, A.C., Perez, D.M., Pie, M.R., 2013. A spatially explicit model of synchronization in fiddler crab waving displays. PLoS One 8(3): e57362. https://doi.org/10.1371/journal.pone.0057362 Banfield, A.W.F., 1974. The Mammals of Canada. Univer- sity of Toronto Press. Banks, W.E., d’Errico, F., Peterson, A.T., Vanhaeren, M., Kageyama, M., Sepulchre, P., Ramstein, G., Jost, A., https://www.scopus.com/home.uri https://doi.org/10.5281/zenodo.7469767 https://doi.org/10.5281/zenodo.7469767 https://doi.org/10.1038/nature.2017.22757 https://doi.org/10.1038/nature.2017.22757 https://doi.org/10.1371/journal.pone.0057362 Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 40 Lunt, D., 2008. Human ecological niches and ranges during the LGM in Europe derived from an applica- tion eco-cultural niche modeling. J. Archaeol. Sci. 35: 481–491. https://doi.org/10.1016/j.jas.2007.05.011 Bogoni, J.A., Navarro, A.B., Graipel, M.E., Peroni, N., 2019. Modeling the frugivory of a plant with incon- stant productivity and solid interaction with relictual vertebrate biota. Ecol. Model. 408: 108728. https:// doi.org/10.1016/j.ecolmodel.2019.108728 Bogoni, J.A., Tagliari, M.M., 2021. Potential distribu- tion of piscivores across the Atlantic Forest: From bats and marsupials to large-bodied mammals under a trophic-guild viewpoint. Ecol. Inform. 64: 101357. https://doi.org/10.1016/j.ecoinf.2021.101357 Bogoni, J.A., Peres, C.A., Ferraz, K.M.P.M.B., 2021. Me- dium- to large-bodied mammal surveys across the Ne- otropics are heavily biased against the most faunally intact assemblages. Mammal Rev. 52(2): 221–235. https://doi.org/10.1111/mam.12274 Bivand, R.S., Pebesma, E., Gomez-Rubio, V., 2013. Ap- plied Spatial Data Analysis with R. 2nd ed. Springer, NY, USA. Bivand, R.S., Lewin-Koh, N., 2021. Maptools: Tools for Handling Spatial Objects. R package version 1.1-1. https://CRAN.R-project.org/package=maptools (ac- cessed 2 Mar 2023). Breiman, L., 2001. Statistical modeling: the two cultures. Stat. Sci. 16: 199–231. Brown, J.H., 1995. Macroecology. 1st ed. University of Chicago Press, Chicago, IL, USA. Carnaval, A.C., Moritz, C., 2008. Historical climate mod- elling predicts patterns of current biodiversity in the Brazilian Atlantic forest. J. Biogeogr. 35: 1187–1201. https://doi.org/10.1111/j.1365-2699.2007.01870.x Ceballos, G., Ehrlich, P.R., Raven, P.H., 2020. Vertebrates on the brink as indicators of biological annihilation and the sixth mass extinction. Proc. Natl. Acad. Sci. USA 117: 13596–13602. https://doi.org/10.1111/ j.1365-2699.2007.01870.x Chaves, J., 2013. The problem of pattern and scale in ecol- ogy: what have we learned in 20 years? Ecol. Lett. 16(1): 4–16. https://doi.org/10.1111/ele.12048 Chen, G., Peterson, A.T., 2002. Prioritization of areas in China for the conservation of endangered birds using modeled geographical distributions. Bird Conserv. Int. 12: 197–209. https://doi.org/10.1017/ S0959270902002125 Christin, S., Hervet, É., Lecomte, N., 2019. Applications for deep learning in ecology. Methods Ecol. Evol. 10(10): 1632–1644. https://doi.org/10.1111/2041- 210X.13256 Conway, G.R., 1977. Mathematical models in applied ecology. Nature 269: 291–297. Dobson, A.J., 1990. An Introduction to Generalized Linear Models. Chapman and Hall. Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation and prediction across space and time. Annu. Rev. Ecol. Evol. Syst. 40: 677–697. https://doi.org/10.1146/annurev.ecol- sys.110308.120159 Elith, J., Phillips, S.J., Hastie, T., Dudík, M., Chee, Y.E., Yates, C.J., 2011. A statistical explanation of MaxEnt for ecologists. Divers. Distrib. 17: 43–57. Elton, C.S., 1924. Periodic fluctuations in the num- bers of animals: their causes and effects. Br. J. Exp. Biol. 2: 119–163. https://doi.org/10.1111/j.1472- 4642.2010.00725.x Escobar, H. 2017. In Brazil, researchers struggle to fend off deepening budget cuts. Scienceinsider. Available at: https://www.science.org/content/article/brazil-re- searchers-struggle-fend-deepening-budget-cuts Faleiro, F.V., Machado, R.B., Loyola, R.D., 2013. Defin- ing spatial conservation priorities in the face of land- use and climate change. Biol. Conserv. 158: 248–257. https://doi.org/10.1016/j.biocon.2012.09.020 Ficetola, G.F., Thuiller, W., Miaud, C., 2007. Prediction and validation of the potential global distribution of a problematic alien invasive species – the American bullfrog. Divers. Distrib. 13: 476–485. https://doi. org/10.1111/j.1472-4642.2007.00377.x Franklin, J., 2009. Mapping Species Distributions: Spa- tial Inference and Prediction. Cambridge University Press. García, A., 2006. Using ecological niche modeling to iden- tify diversity hotspots for the herpetofauna of Pacific lowlands and adjacent interior valleys of Mexico. Biol. Conserv. 130: 25–46. https://doi.org/10.1016/j. biocon.2005.11.030 Giovanelli, J.G.R., Haddad, C.F.B., Alexandrino, J., 2008. Predicting the potential distribution of the alien in- vasive American bullfrog (Lithobates catesbeianus) in Brazil. Biol. Invasions 10: 585–590. https://doi. org/10.1007/s10530-007-9154-5 Haeckel, E., 1866. Generelle Morphologie der Organis- men. G. Reimer, Berlin, Germany. Hansen, M.C., Stehman, S.V., Potapov, P.V., 2010. Quan- tification of global gross forest cover loss. Proc. Natl. Acad. Sci. USA 107(19): 8650–8655. https://doi. org/10.1073/pnas.0912668107 Hassell, M.P., 2000. Host-parasitoid population dynamics. J. Anim. Ecol. 69: 543–566. https://doi.org/10.1046/ j.1365-2656.2000.00445.x https://doi.org/10.1016/j.jas.2007.05.011 https://doi.org/10.1016/j.ecolmodel.2019.108728 https://doi.org/10.1016/j.ecolmodel.2019.108728 https://doi.org/10.1016/j.ecoinf.2021.101357 https://doi.org/10.1111/mam.12274 https://CRAN.R-project.org/package=maptools https://doi.org/10.1111/j.1365-2699.2007.01870.x https://doi.org/10.1111/j.1365-2699.2007.01870.x https://doi.org/10.1111/j.1365-2699.2007.01870.x https://doi.org/10.1111/ele.12048 https://doi.org/10.1017/S0959270902002125 https://doi.org/10.1017/S0959270902002125 https://doi.org/10.1111/2041-210X.13256 https://doi.org/10.1111/2041-210X.13256 https://doi.org/10.1146/annurev.ecolsys.110308.120159 https://doi.org/10.1146/annurev.ecolsys.110308.120159 https://doi.org/10.1111/j.1472-4642.2010.00725.x https://doi.org/10.1111/j.1472-4642.2010.00725.x https://www.science.org/content/article/brazil-researchers-struggle-fend-deepening-budget-cuts https://www.science.org/content/article/brazil-researchers-struggle-fend-deepening-budget-cuts https://doi.org/10.1016/j.biocon.2012.09.020 https://doi.org/10.1111/j.1472-4642.2007.00377.x https://doi.org/10.1111/j.1472-4642.2007.00377.x https://doi.org/10.1016/j.biocon.2005.11.030 https://doi.org/10.1016/j.biocon.2005.11.030 https://doi.org/10.1007/s10530-007-9154-5 https://doi.org/10.1007/s10530-007-9154-5 https://doi.org/10.1073/pnas.0912668107 https://doi.org/10.1073/pnas.0912668107 https://doi.org/10.1046/j.1365-2656.2000.00445.x https://doi.org/10.1046/j.1365-2656.2000.00445.x Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 41 Hugall, A., Moritz, C., Moussalli, A., Stanisic, J., 2002. Reconciling paleo distribution models and compar- ative phylogeography in the Wet Tropics rainforest land snail Gnarosophia bellendenkerensis (Brazier 1875). Proc. Natl. Acad. Sci. USA 99: 6112–6117. https://doi.org/10.1073/pnas.092538699 Hughes, A.C., Orr, M.C., Ma, K., Cotello, M.J., Waller, J., Provoost, P., Yang, Q., Zhu, C., Qiao, H., 2021. Sampling biases shape our view of the natural world. Ecography 44(9): 1259–1269. https://doi.org/10.1111/ ecog.05926 Hutchinson, G.E., 1957. Concluding remarks. Cold Spring Harb. Symp. Quant. Biol. 22: 415–427. Jordan, M.I., Mitchell, T.M., 2015. Machine learn- ing: trends, perspectives, and prospects. Science 349(6245): 255–260. https://www.science.org/ doi/10.1126/science.aaa8415 Kampichler, C., Wieland, R., Calmé, S., Weissenberger, H., et al., 2010. Classification in conservation biology: A comparison of five machine-learning methods. Ecol. Inform. 5(6): 441–450. https://doi.org/10.1016/j. ecoinf.2010.06.003 Kar, T.K., Chakraborty, K., Pahari, U.K., 2010. A prey– predator model with alternative prey: Mathemati- cal model and analysis. Can. Appl. Math. Q. 18(2): 137–168. Lotka, A.J., 1925. Elements of Physical Biology. Williams and Wilkins Co., Baltimore, MD, USA. McLaren, J.D., Buler, J.J., Schreckengost, T., Smolinsky, J.A., et al., 2018. Artificial light at night confounds broad-scale habitat use by migrating birds. Ecol. Lett. 21(3): 356–364. https://doi.org/10.1111/ele.12902 Magioli, M., Ribeiro, M.C., Ferraz, K.M.P.M.B., Ro- drigues, M.G., 2015. Thresholds in the relationship between functional diversity and patch size for mam- mals in the Brazilian Atlantic Forest. Anim. Conserv. 18(6): 499–511. https://doi.org/10.1111/acv.12201 Magurran, A.E., 2004. Measuring Biological Diversity. Blackwell Publishing, Malden, MA, USA. May, R.M., 1976. Thresholds and breakpoints in ecosys- tems with a multiplicity of stable states. Nature 269: 471–477. https://doi.org/10.1038/269471a0 Moura, M.R., Jetz, W., 2021. Shortfalls and opportunities in terrestrial vertebrate species discovery. Nat. Ecol. Evol. 5: 631–639. https://doi.org/10.1038/s41559- 021-01411-5 Olden, J.D., Lawler, J.J., Poff, N.L., 2008. Machine learn- ing methods without tears: a primer for ecologists. Q. Rev. Biol. 83: 171–193. https://www.journals.uchica- go.edu/doi/10.1086/587826 Pebesma, E.J., Bivand, R.S., 2005. Classes and methods for spatial data in R. R News 5(2). https://cran.r-pro- ject.org/doc/Rnews/ (accessed 2 Mar 2023). Pittman, S.J., Brown, K.A., 2011. Multi-scale approach for predicting fish species distributions across coral reef seascapes. PLoS One 6(5): e20583. https://doi. org/10.1371/journal.pone.0020583 R Core Team, 2023. R: A language and environment for statistical computing. R Foundation for Statistical Computing. http://www.R-project.org/ (accessed 2 Mar 2023). Rammer, W., Seidl, R., 2019. Harnessing deep learn- ing in ecology: An example predicting bark beetle outbreaks. Front. Plant Sci. 10: 1327. https://doi. org/10.3389/fpls.2019.01327 Sherratt, J.A., Eagan, B.T., Lewis, M.A., 1997. Oscilla- tions and chaos behind predator-prey invasion: math- ematical artifact or ecological reality? Philos. Trans. R. Soc. Lond. B Biol. Sci. 352: 21–38. https://doi. org/10.1098/rstb.1997.0003 Siqueira, M.F., Peterson, A.T., 2003. Consequences of global climate change for geographic distributions of Cerrado tree species. Biota Neotrop. 3(2): 1–14. https://doi.org/10.1590/S1676-06032003000200005 Smith, A.T., Xie, Y., Hoffmann, R.S., Lunde, D., MacKin- non, J., Wilson, D.E., Wozencraft, W.C., Gemma, F. (Illustrator), Wang, S. (Honorary Editor), 2013. Mam- mals of China. Princeton University Press. (Princeton Pocket Guides). Stowell, D., Plumbley, M.D., 2014. Automatic large-scale classification of bird sounds is strongly improved by unsupervised feature learning. PeerJ 2: e488. https:// doi.org/10.7717/peerj.488 Tagliari, M.M., Bogoni, J.A., Blanco, G.D., Cruz, A.P., Peroni, N., 2021. Disrupting a socio-ecological sys- tem: could traditional ecological knowledge be the key to preserving the Araucaria Forest in Brazil under climate change? Clim. Change 176: 1–20. https://doi. org/10.1007/s10584-022-03477-x Volterra, V., 1926. Variazioni e fluttuazioni del numero d’individui in specie animali conviventi. Mem. Acad. Lincei Roma 2: 31–113. https://doi.org/10.1073/pnas.092538699 https://doi.org/10.1111/ecog.05926 https://doi.org/10.1111/ecog.05926 https://www.science.org/doi/10.1126/science.aaa8415 https://www.science.org/doi/10.1126/science.aaa8415 https://doi.org/10.1016/j.ecoinf.2010.06.003 https://doi.org/10.1016/j.ecoinf.2010.06.003 https://doi.org/10.1111/ele.12902 https://doi.org/10.1111/acv.12201 https://doi.org/10.1038/269471a0 https://doi.org/10.1038/s41559-021-01411-5 https://doi.org/10.1038/s41559-021-01411-5 https://www.journals.uchicago.edu/doi/10.1086/587826 https://www.journals.uchicago.edu/doi/10.1086/587826 https://cran.r-project.org/doc/Rnews/ https://cran.r-project.org/doc/Rnews/ https://doi.org/10.1371/journal.pone.0020583 https://doi.org/10.1371/journal.pone.0020583 http://www.R-project.org/ https://doi.org/10.3389/fpls.2019.01327 https://doi.org/10.3389/fpls.2019.01327 https://doi.org/10.1098/rstb.1997.0003 https://doi.org/10.1098/rstb.1997.0003 https://doi.org/10.1590/S1676-06032003000200005 https://doi.org/10.7717/peerj.488 https://doi.org/10.7717/peerj.488 https://doi.org/10.1007/s10584-022-03477-x https://doi.org/10.1007/s10584-022-03477-x Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 42 Journal Im pact factor Year A uthors M ain country N citation Prim ary aim M achine learning m ethod(s) or fram ew ork M ain approach Ecography 6.802 2006 Elith et al. A U S 6282 To provide effective guidance on how best to use this inform ation in the context of num erous approaches for m odelling distributions M axim um entropy m odels (M A X EN T and M A X EN T-T) and boosted regression trees (B RT, also called stochastic gradient boosting) Landscape ecology & species distribution Ecological M odelling 3.512 2009 K ocev et al. SV N 139 To applied various m achine learning m ethods to the problem of predicting the condition or quality of the rem nant indigenous vegetation across an extensive area of south-eastern A ustralia R egression trees, M ulti-target regression trees, Ensem bles, B agging, R andom forests Landscape ecology & species distribution Ecological Inform atics 4.498 2010 K am pichler et al. M EX 120 To com pare five m achine-learning based classification techniques (classification trees, random forests, artificial neural netw orks, support vector m achines, and autom atically induced rulebased fuzzy m odels) in a biological conservation context Techniques of classification trees, random forests, artificial neural netw orks, support vector m achines, and autom atically generated fuzzy classifiers B ird ecology & conservation PLoS O N E 3.752 2011 Pittm an & B row n U SA 181 to: (1) D eterm ine w hether the influence of environm ental predictors on species’ distribution w as scale-dependent; (2) evaluate the utility of environm ental data from a single rem ote sensing device com bined w ith m etrics for surface m orphology to predict and m ap fish species distributions across a com plex coral reef ecosystem ; (3) determ ine w hich om ponents of rem otely sensed seafloor structure contribute m ost to the species distribution m odels; (4) identify threshold effects w here changes in environm ental variables abruptly influence species occurrence; and (5) evaluate the perform ance of tw o different m achine-learning m odelling algorithm s for spatial predictions of m arine fish distributions B oosted regression trees (B RT) and m axim um entropy m odelling (M axEnt) Fish ecology & distribution Table 1. Journal, the im pact factor (IF), year of publication (yr), authors, m ain country, num ber of citations, prim ary aim , m achine-learning m ethods or fram ew ork, and m ain approach of the top-cited m anuscripts (> 100 citations; N = 14) using m achine-learning approach in ecology and conservation across the globe. Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 43 R em ote Sensing of Environm ent 13.85 2012 D ronova et al. U SA 139 C lassify Poyang Lakew etland PFTs using O B IA and to determ ine w hich im age segm entation scales and m achine- learning algorithm s optim ize class discrim ination Six m achine-learning algorith representing m achine-learning principles: a probabilistic B ayes m ethod (N aïveB ayesSim ple in W eka notation), a logistic regression m ethod (Sim pleLogistic in W eka), an artificial neural netw ork algorithm (M ultilayerPerceptron (M LP)), a support vector m achine tool w ith polynom ial kernel and com plexity param eter value of 10 (SM O inW eka), a K -N earest N eighbors (IB k inW eka) and a tree-based classifier (R andom Forest). Landscape ecology & species distribution Journal of B iogeography 4.81 2012 W erneck et al. U SA 174 To investigate the historical distribution of the C errado across Q uaternary clim atic fluctuations and to generate historical stability m aps to test: (1) w hether the ‘historical clim ate’ stability hypothesis explains squam ate reptile richness in the C errado; and (2) the hypothesis of Pleistocene connections betw een savannas located north and south of A m azonia M axim um -entropy C lim ate change M ethods in Ecology and Evolution 8.335 2012 B arbet- M assin et al. FR A 1401 To conduct a com prehensive com parative analysis based on sim ple sim ulated species distributions to propose guidelines on how, w here and how m any pseudo-absences should be generated to build reliable species distribution m odels. B oosted regression trees (B RT) and random forest (R F) Landscape ecology & species distribution Integrative Zoology 2.083 2013 Li & W ang C H N 120 To applying various algorithm s for species distribution m odelling M ultivariate adaptive regression splines, M ixture discrim inant analysis, A rtificial neural netw orks, G eneralized boosting m odels, C lassification and regression tree, R andom forest, H ierarchical m odelling, G enetic algorithm for rule set production, and M axim um entropy Landscape ecology & species distribution Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 44 PeerJ 3.061 2014 Stow ell & Plum bley G B T 185 To introduce a technique for feature learning from large volum es of bird sound recordings, inspired by techniques that have proven useful in other dom ains Spectral features and feature learning B ird ecology & conservation PLoS O N E 3.752 2016 de Souza et al. C A N 152 To assess the accuracy of satellite-based A IS (A utom atic Identification System ) and V M S (Vessel M onitoring System ) in correctly identifying individual fishing events or ‘sets’ by com paring against expert-labeled data H idden M arkov M odel (H M M ) and D ata M ining (D M ) Fish ecology & distribution Energy and B uildings 7.201 2018 Touzani et al. U SA 212 To propose an energy consum ption baseline m odeling m ethod based on a gradient boosting m achine w asproposed. D ecision trees and G radient boosting m achine G reen energy & energy econom y Ecology Letters 11.274 2018 M cLaren et al. U SA 102 To disentangle anthropogenic and landscape-related factors affecting stopover density, and thereby assess w hether artificial light at night m ight be affecting selection of stopover habitat, w e estim ated responses in seasonal- m ean reflectivity to geographic, land cover and anthropogenic predictors using additive regression m odels fit by gradient boosting, a m achine-learning technique A dditive regression m odels fit by gradient boosting B ird ecology & conservation Juliano A. Bogoni et al. – Machine-Learning Applications in Ecology and Conservation 45 Proceedings of the N ational A cadem y of Sciences 12.799 2018 N orouzzadeh et al. U SA 493 Test how w ell deep learning can autom ate inform ation extraction from cam era-trap im ages D eep learning architectures: A lexN et, N iN , V G G , G oogLeN et, R esN et W ildlife ecology M ethods in Ecology and Evolution 8.335 2019 C hristin et al. C A N 187 To review existing im plem entations and show that deep learning has been used successfully to identify species, classify anim al behaviour and estim ate biodiversity in large datasets like cam era‐trap im ages, audio recordings and videos Tensorflow, PyTorch, K eras, M icrosoft C ognitive Toolkit (C N TK ), D eeplearning4J, M ATLA B + D eep Learning Toolbox, A pache M X N ET, PlaidM L W ildlife ecology