Human impacts on generalist and specialist
bird and mammal species
Name Marike Knegtering Student Number 11700696
University University of Amsterdam Department Institute of Biodiversity and
Ecosystem Dynamics Date June 26th 2020
Supervisors dr. J.R. (James) Allan, dr. rer. nat. W.D. (Daniel) Kissling
Abstract
Species can be broadly classified into generalists or specialists based on traits such as habitat preference and diet. Increasing numbers of species are threatened with extinction due to the impact of human threats. These threats, including hunting and land conversion into agriculture, are widespread within species distributions. Species respond differently to human threats, which function as an evolutionary filter, selecting for species with certain traits. In particular, research has suggested that generalists are replacing specialists, because specialists are more dependent on a narrow range of resources and habitats, making them more vulnerable to environmental changes. However, it is unclear if certain specialist traits make a species more vulnerable to such changes, and are therefore indicators of this process of biotic homogenization. In this study, I investigated whether species with particular specialisations (diet, habitat, forest and vegetation stratum use) have a higher proportion of their range impacted by human threats, for 1277 mammal and 2120 bird species (threatened or near-threatened) globally. I found that habitat and forest specialised birds have significantly higher proportions of their range impacted by threats than habitat generalist birds. Relationships between other specialisation variables and proportion of a species range impacted by threats were not significant. Only small percentages of the data could be explained by a linear model, reflecting that there are multiple drivers of the proportion of a species range impacted by threats, beyond just specialist or generalist traits. Therefore, further research that incorporates other potential drives of species-specific differences of human impacts within their range, such as dispersal limitation or categorical specialist traits, is recommended. Nevertheless, my work constitutes an important first step, since it contributes to supporting decisions on conservation management options for species with different traits, and towards developing new theoretical investigations to understand and stem biotic homogenization of species assemblages; a crucial step for maintaining ecosystem resilience and ecosystem services.
Summary
Humans are harming nature and threatening species worldwide by activities including hunting and land conversion into agriculture, which are widespread within species distributions. These human activities function as a filter for species with specific characteristics, since species sensitivity to these human activities differs. Previous research has pointed out that generalist species – for example species that can sustain populations in a broad range of environments – are replacing specialist species. This causes biotic homogenization, meaning that composition of species characteristics are becoming more similar over space and time. Specialists are more vulnerable to environmental changes due to human activities than generalists because specialists are dependent on, for example, one type of habitat, or a narrow range of food sources. However, which kind of specialism mainly determines this vulnerability is still unclear. Therefore, I investigated whether threatened bird and mammal species that are specialised in diet and habitat among others, have a higher proportion of their geographic range impacted by these human activities. Specialised bird species appeared to be more impacted that generalist bird species, but relationships did not follow a straight line, indicating that other drives are present. Therefore, research that incorporates more mechanisms, such as dispersal ability of species, is necessary to identify other factors that might better explain human impacts on species. This is an important step in finding priorities for conservation action to stem the process of biotic homogenization.
Keywords
Functional traits, specialisation, human footprint, human impact, global change, conservation, biotic homogenization.
Content
Abstract ... 2 Summary ... 2 Keywords ... 2 Introduction ... 4 Method ... 6 Data extraction ... 6 Quantification of specialisation ... 6 Statistical analysis ... 9 Results ... 12 Discussion ... 18 Conclusion ... 20 References ... 21 Appendices ... 23Appendix A: IUCN Habitat Classification Scheme (Version 3.0) (IUCN, n.d.)... 23
Introduction
Human pressures on nature such as hunting and land conversion into agriculture threaten tens of thousands of species with extinction (Allan et al., 2019). Despite increased conservation efforts, biodiversity continues to decline globally (Díaz et al., 2019). However, species differ in their sensitivity to threatening processes and respond to threats in different ways (Allan et al., 2019). One reason why species respond differently to human threats, is because they have different functional traits, which determine how they interact with the environment (Newbold et al., 2020). Human‐induced environmental changes therefore act as non‐random filters, favouring species best able to survive within modified ecosystems. This causes ‘biotic homogenization’, the process of species becoming genetically, taxonomically or functionally more similar over space and time (Devictor et al., 2010). One pattern within this biotic homogenization is that specialist species are being replaced by generalist species, which is being increasingly documented (Devictor et al., 2010; Colles, Liow, & Prinzing, 2009; Julliard et al., 2004).
To illustrate, Munday (2004) found that decline in coral abundance causes proportionally greater losses of specialist coral-dwelling fish species than generalist species. In addition, Devictor et al. (2007) showed that populations of specialised bird species become more unstable with increasing urbanization than populations of generalist bird species. Underlying mechanisms have been described by among others Clavel, Julliard & Devictor (2011), who state that disturbances and degradation, caused by humans, may lead to increased competition between specialists and generalists, since changing environments leave little opportunity for specialised species to adapt to new environments, creating an advantage of being a generalist over being a specialist (Clavel, Julliard & Devictor, 2011). However, some contradictory results have also been found. For example, Attum et al. (2006) investigated the effect of habitat degradation on specialist and generalist desert lizards and found the opposite relationship; desert generalist species were not able to persist in degraded sites and were only found in more intact protected sites. They propose that effects of human threats on generalist or specialist lizards depend on the way these threats change the environmental harshness or conditions that favour specialisation. Therefore, in harsh environment species could respond in an opposite way to species in more productive environments.
Specialists are thought to be more vulnerable to environmental changes, because species responses are dependent on life history traits (Eskilden et al., 2015). In turn, aspects of ecological specialisation can be inferred from variability in functional traits (Carboni, Zelený, & Acosta, 2016; Eskilden et al., 2015). Considering different aspects of ecological specialisation (in other words, different ways of specialisation), is therefore important when assessing species vulnerability to environmental changes (Eskilden et al., 2015) and can contribute to the overall understanding of drivers of taxonomic differences in responses (Newbold et al., 2020). The distinction between generalist and specialist species is made based on their niche width, which is defined as all the biotic and abiotic variables that a species requires to ensure its population viability in a given environment (Clavel, Julliard & Devictor, 2011). The difference in niche width is the result of a trade-off between a species capacity to exploit a wide range of environmental conditions and their ability to highly efficiently use a single or narrow range. In other words, specialist species can greatly increase their fitness in a narrow range of a given environmental variable that is part of its niche, whereas a generalist can somewhat increase its fitness across a wider range of a given environmental variable that is part of its niche (Clavel, Julliard & Devictor, 2011). For example, species that are specialised in diet, efficiently use a narrower range of food resources than generalist species. Likewise, species that are specialised in habitat, can increase their fitness in fewer environmental types than generalist species.
The concept of niche can be mainly divided into two dimensions. The first distinction that is made is the fundamental versus the realized niche. The fundamental niche represents all potential biotic and abiotic environmental variables in which a species is able to survive and the realized niche is where a species actually occurs and survives. The realized niche therefore also incorporates processes such as source-sink dynamics, which cause species to occur in unsuitable habitats, and dispersal limitation, which cause species to not occur in
suitable habitats. The second distinction that is made is the Eltonian versus Grinnelian niche. The Grinnelian niche describes species responses in terms of performance to resources (what a species needs, with resource in its largest sense), whereas the Eltonian niche describes a species functional role and position in an environment (what a species does) (Devictor et al., 2010).
In summary, previous research has shown that generalists are replacing specialists (Clavel, Julliard & Devictor, 2011; Devictor et al., 2007), that land use effects differ between functional groups (Newbold et al., 2020) and that functional traits can predict species extinction risk or vulnerability to environmental changes (Gaynor et al., 2018; Hill et al., 2019). However, evidence that links particular facets of species specialisation explicitly to large-scale human driving forces is lacking, especially on a global scale and for different taxa and multiple threats (Devictor, Julliard & Jiguet, 2008). This project contributes to the field of knowledge by analysing human impacts within the ranges of specialist and generalist species and tests the hypothesis that generalists are less impacted than specialists for four different types of specialisation; diet, forest, vegetation stratum use and habitat breadth (Diamond et al., 2011; Buechley, & Şekercioğlu, 2016; Jiquet et al., 2007; Martin & Possingham, 2005). I expect a positive relationship between degree of specialisation and proportion of a species range impacted, since specialised species are dependent on a narrow range of resources (Clavel, Julliard & Devictor, 2011). For diet, vegetation stratum use and habitat breadth I expect that specialists are relatively more impacted in their geographic ranges than generalists, since generalists can survive in various environments (Diamond et al., 2011; Buechley, & Şekercioğlu, 2016; Jiquet et al., 2007). For forest specialisation I expect that forest specialists are more impacted than open habitat specialists, since open habitat is more widespread (Martin & Possingham, 2005).
Method
Data extraction
Data on human impacts within species distributions
Allan et al. (2019) present the first global dataset of species ranges and human induced threats within those ranges that accounts for species-specific vulnerability to threats. It distinguishes areas with impacts, in which threat is present and sensitive species is present and refuges, in which species is present but threats are absent. Allan et al. (2019) obtained maps of threats from the Human Footprint database, which contains global data on eight human pressures: roads, nightlights, croplands, pasturelands, built environments, navigable waterways, railways and human population density at a 1 km2 resolution for the year 2009 (Venter, 2016). Allan et al. (2019)
calculated the proportion of each species geographic range that is impacted by at least one human threat (%) for 1277 mammal and 2120 bird species that are threatened or near-threatened.
Data on species functional traits
I extracted data on species functional traits from the EltonTraits1.0 database (Wilman et al., 2014), which contains standardized and continuous measurements of species ecological traits (for example diet or vegetation stratum use). The EltonTraits1.0 data for mammals and birds is highly complete and contains variables that quantitatively measure the relative importance of different categories of certain traits for each species. This enables me to calculate degree of specialisation in a continuous way. I extracted data on diet and vegetation stratum use from this database for the same mammal and bird species as in the human impact database. Diet is quantified as percent use of 10 diet categories, which include categories such as fruit Fruit) and seeds (Diet-Seed) among others (Wilman et al., 2014). For birds, vegetation stratum use is quantified as percent use of 7 vegetation strata, which include categories such as ground ground) and understorey (ForStrat-understorey) among others. For mammals, vegetation stratum use is a categorical variable and each species is therefore assigned to use one out of five vegetation stratum categories (A – Aerial, Ar – Arboreal, M – Marine, S – Scansorial and G – ground).
I used the IUCN Habitat Classification Scheme (Version 3.0) (IUCN, n.d.) (Appendix A) and database (IUCN, 2020) to obtain data on habitat types in which species occur. The habitat classification scheme is a well-established and a widely used standard division of habitat types, including categories of forest (with subcategories of temporal and boreal forest among others) and savanna (with subcategories of dry and moist savanna) among others. The database consists of columns that correspond to each habitat subcategory in the habitat classification scheme. The data is binary (0 or 1) for each habitat subtype, in which 1 means that species occurs in that habitat subtype and 0 means that species does not occur in that habitat subtype.
Quantification of specialisation
An increasing number of specialisation indices have been proposed in the recent literature. In addition, the quantification of ecological specialisation is highly dependent on the data used. Consequently, the term specialisation is used inconsistently throughout literature. Although definition of specialisation is flexible, one needs to carefully consider the definition and quantification metrics that will be applied in research (Devictor et al., 2010). Devictor et al. (2010) presented a toolbox for applying the right metric in quantifying niche
specialisation along the two dimensions of the niche concept. Data from the EltonTrait1.0 and IUCN database reflect species Grinnelian niche, since this data contain the range of resources that species utilize, which reflects a species needs. Data that I utilized also reflect the realized niche of specialisation, since data is about where species occur and not where they could potentially occur. I used diversity metrics to quantify specialisation, since diversity metrics are suitable to investigate specialisation when data reflects the realized Grinnelian niche of
species as described in the toolbox of Devictor et al. (2010). The next paragraphs explain which diversity metrics are used for each variable and how they are computed. An overview of these metrics is provided in Table 2.
Diet specialisation
Specialisation in diet is quantified with the Simpson’s diversity index (D). This index was originally used to measure species richness in samples, which also takes into account relative abundances of species. However, this metric has also been applied to quantify degree of specialisation, since specialisation can be viewed as lack of evenness in distribution of traits that are related to specialisation (Devictor et al., 2010; Hallman & Robinson, 2020; Poisot et al., 2012). The formula that was applied for calculating D is stated in equation 1. In equation 1 i is the category of the trait (for example fruit (Diet-Fruit) in the case of diet), k is the total number of categories (10 in the case of diet) and p is the proportion of measures that falls into category i. Because p is a proportion (ranging between 0 and 1), D also ranges between 0 and 1, with increasing values of D representing greater degree of specialisation. This metric correctly takes into account species differences in distribution of use of resources. To illustrate, a species that distributes its diet across four categories less equally (for example large use of one category and less use of the other three), is quantified as more specialised than another species that uses the same four categories equally.
(1)
𝐷 = ∑
𝑘
𝑖=1
𝑝
𝑖
2
Habitat specialisation
The IUCN Habitat Classification Scheme contains 18 habitat categories (first level), with for example 1 – Forest and 2 – Savanna among others (Appendix A). These categories are subdivided in habitat subcategories (second level). For example, the category 1 - Forest contains subcategories of 1.1 Boreal and 1.2 Subarctic among others. The numbers of subcategories differ between habitat types, and some habitat categories (for example category 6 – Rocky Areas and 17 – Other) do not have subcategories. Habitat specialisation (habitat breadth) is calculated as the number of habitat subcategories (second level) plus the number of habitat categories that do not have subcategories in the IUCN Habitat Classification Scheme in which a species occurs (Pacifici, Visconti & Rondinini, 2018; Julliard et al., 2006). Species with a high habitat count are therefore habitat generalists and species with a low habitat count are habitat specialists.
Forest specialisation
I made an ordered gradient of habitat types from the IUCN Habitat Classification Scheme that ranges from most forested habitats (order 1 in the gradient) to least forested areas (open habitats, order 11 in the gradient) (Table 1). This gradient excludes marine and aquatic habitats. I converted the IUCN habitat classification database to a matrix with rows corresponding to species and columns corresponding to each habitat class in the gradient. In this matrix ‘NA’ means that a species does not use that habitat. Species that do use that habitat have the value of the order in the gradient in the corresponding cell, which enabled me to calculate a weighted average. For each species I computed a measure of forest specialisation using the barycentre (arithmetic mean), which is the average habitat class along a forest-open gradient in which a species is specialised (Barnagaud et al., 2011).
Table 1 Ordered gradient of forest openness, based on the IUCN Habitat Classification Scheme (Version 3.0), ranging from most forested habitats (order 1 in the gradient) to least forested areas (open habitats, order 11 in the gradient).
Level IUCN Habitat Class Order in gradient
1 Forest 1
14.3 Plantations 2
2 Savanna 3
3 Shrubland 4
14.4 Rural Gardens 5
14.6 Subtropical/Tropical Heavily Degraded Former Forest 6
4 Grassland 7
14.2 Pastureland 8
14.1 Arable Land 9
8 Desert 10
6 Rocky Areas [e.g. inland cliffs, mountain peaks] 11
Vegetation stratum specialisation
Investigating vegetation stratum specialisation is done differently for mammals and birds, since data properties are different. As described in the ‘Data extraction’ section, vegetation stratum use by mammals is a categorical variable and each mammal species is assigned to use one out of five vegetation stratum categories. Therefore, for mammals, vegetation stratum use cannot be quantified in terms of degree of specialisation. Instead, these categorical data give information about in which vegetation stratum a mammal species is mainly specialised.
For birds, I quantified specialisation of vegetation stratum use in two ways. Firstly, I calculated degree of specialisation with the Simpson’s index (D) in the same way as for diet specialisation, in which a Simpson’s index of 0 means that a species is minimally specialised in use of vegetation stratum for foraging, and thus uses multiple vegetation strata, and a Simpson’s index of 1 means that species is maximally specialised in vegetation stratum use for foraging, and thus only uses one type of vegetation stratum. Secondly, I calculated a barycentre as the relative height of vegetation stratum, along a gradient from ground to air, that a species uses for foraging, weighted by scores of use (%) (Barnagaud et al., 2011). To calculate this barycentre, vegetation stratum categories were ranked from lowest to highest in the order: ground (ForStrat-ground), understorey (ForStrat-understory), midhigh (ForStrat-midhigh), canopy (ForStrat-canopy) and aerial (ForStrat-aerial). The formula that was applied to calculate this barycentre (b) is stated in equation 2, in which k is the total number of vegetation categories (5), p the proportion of measures that falls into category i, and vegk is the kth class of vegetation (from 1 (ground) to 5
(aerial)). A barycentre of 1 therefore means that a species is specialised in lowest vegetation stratum for foraging and a barycentre of 5 means that a species is specialised in highest vegetation stratum for foraging.
(2)
𝑏 =
∑
(𝑣𝑒𝑔
𝑘× 𝑝
𝑖)
𝑘 𝑖=1
Table 2 Summary of all specialisation indices and properties that quantify specialisation for the four variables (forest, diet, habitat and vegetation stratum specialisation).
Variable Taxon Specialisation index Range Interpretation Diet
specialisation
Birds and mammals
Simpson’s index (no unit) (D) 0 to 1 0 = least specialised, 1 = most specialised
Forest specialisation
Birds and mammals
Barycentre (no unit) 1 to 11 1 = most forested habitat, 11 = least forested habitat
Habitat specialisation
Birds Number of categories in which species occurs (#)
0 to 16 1 = most habitat specialised,16 = least habitat specialised Mammals Number of categories in which
species occurs (#)
0 to 24 1 = most habitat specialised, 24 = least habitat specialised
Vegetation stratum use specialisation
Birds Simpson’s index (no unit) (D) 0 to 1 0 = least specialised, 1 = most specialised
Birds Barycentre (no unit) 1 to 5 1 = lowest vegetation stratum (ground), 5 = highest vegetation stratum (aerial)
Quantifications of all specialisation variables for each mammal and bird species were combined with data on the proportion of each species range impacted by threats into one data frame. Since the number of species in the human impact database is smaller than in the other databases, only data for species that occur in both the human impact database and the related trait or habitat database were part of the total dataset. Therefore, the final number of species that I could consider in this study is 1850 bird and 1240 mammal species.
Statistical analysis
Wilcoxon Rank Sum test for difference between groups
Data on birds and mammals were investigated separately, because these taxonomic groups could show difference in trends due to fundamental taxonomic differences (Buchmann et al., 2013).
Firstly, I tested if degree of specialisation is significantly different between impacted and non-impacted species. An impacted species has a threat present across > 0 % of its distribution (essentially it is impacted somewhere). A non-impacted species has no threats present within its geographic range. This was tested with the Wilcoxon Rank Sum test (also called Mann-Whitney U test), since this is a nonparametric test that can be used to investigate whether two independent samples (which is in this case the two groups of impacted and non-impacted species) were selected from populations having the same distribution. Applied in this case, this test compares whether the medians of the data on degree of specialisation significantly differ between impacted and non-impacted species.
For vegetation stratum use by mammals, difference of distribution of data between impacted and non-impacted species had to be tested with the Kruskal-Wallis test, since this is a nonparametric test that is suitable for categorical variables.
Linear regression analysis
Relationships between proportion of a species range impacted by threats and numerical variables on species specialisation (diet, forest and habitat specialisation for birds and mammals and vegetation stratum specialisation for birds) were analysed with linear regression models in RStudio (version 1.2.5033) (RStudio Team, 2015), in which the proportion of a species range impacted by at least one threat (%) is the response variable (independent of range size) and the specialisation variable the explanatory variable.
Before performing the regression analysis, assumptions of normality of residuals, homogeneity of variance, linearity of response variable and influential cases (outliers) were checked with the plot-function in RStudio. Data on proportion of a species range impacted by threats is zero inflated and right skewed for both mammals and birds (Figure 1). In order to remove the zero inflation and achieve normally distributed data, I separated the data for mammals on proportion of a species range impacted by threats into two binary categories of impacted and non-impacted species (Figure 2). Thus for mammals, I performed the linear regression analysis only using species that have > 0% of their range impacted by threats. Data for birds on proportion of a species range impacted by threats was separated into bird species that are impacted in ≤ 5% of their range and > 5% of their range, since distribution of data on proportion of a species range impacted > 0% still showed a strong zero inflation. Thus for birds, I performed the linear regression analysis only using species that have > 5% of their range impacted by threats.
Birds Mammals
Figure 1 Distribution of the data in the database on proportion of a species range impacted by threats is zero inflated and strongly right skewed for both birds (left) and mammals (right) (Allan et al., 2019).
Assumption of homogeneity of variance was met, but assumptions of normality of the response variable and residuals were not met due to strong right skewness of data distribution of both mammal and birds (Figure 2). Log, square, square root and reciprocal data transformations were attempted, but did not or only slightly improved the normality of the distribution. To illustrate, data transformations that showed most improvement (log transformation for birds and square root transformation for mammals) are shown in Figure 2. Even though assumption of normality of response variable is violated, I chose to nevertheless conduct the linear regression analysis with the untransformed data, since conclusions were drawn from regression line and scatter plot combined, in order to judge whether a linear regression line could fit the data. Also, I took R2-values into account
when drawing conclusions, since they provide information on how well a linear model fits the data.
Also some outliers were removed. For mammals, habitat specialisation values of higher values than 20 (21 data points in total) were excluded from the analysis. For birds, habitat specialisation values of higher than 12 (5 data points in total) were excluded from the analysis. These values were removed, since they range more than 1.5 interquartile above the third quartile (Q3), as could be seen by making boxplots. For birds, also vegetation stratum use specialisation (barycentre) values of higher than 4 (8 data points in total) were excluded from the analysis. These decisions were affirmed by improvement of the Residuals vs. Leverage plot (which can visualize
influential cases) after removing these outliers. I also I carried out a sensitivity analysis by re-running the analyses with the full dataset (Appendix B). The significance and direction of slope remained the same for all three models for which outliers were removed, but the R2-value increased. Therefore I chose to analyse the data
without these outliers.
Figure 2 Data of mammals range impacted > 0% (top left) and birds range impacted > 5% (top right) are not normally distributed and strongly right skewed. Data transformations that showed most improvement (log transformation for birds (bottom left) and square root transformation for mammals (bottom right)) did not clearly improve the normality of the distribution.
Vegetation stratum use for foraging by mammals is the only variable that I could not analyse with a linear regression model, since it is a categorical variable. Therefore, the Tukey-Kramer test was used to test for significance of difference in proportion of a species range impacted by threats (%) between mammals that use a different vegetation stratum for foraging. The Tukey-Kramer is a nonparametric test that is suitable for single-step multiple comparison of means from different groups.
Results
Comparison of specialisation between impacted and non-impacted species
Impacted and non-impacted mammals differ significantly in degree of specialisation in diet, forest and habitat, as is the result from the Wilcoxon Rank Sum test (Table 3). In addition, impacted and non-impacted mammals differ significantly in the categories of vegetation stratum that they use for foraging, as is the result of the Kruskal-Wallis test. The differences in specialisation between impacted and non-impacted species are visualized in Figure 3. Some of the specialisation variables are log-transformed to improve the visibility of the difference in specialisation.
The results of the Wilcoxon Rank Sum test also point out that non-impacted birds significantly differ in degree of diet, forest, habitat and vegetation stratum use specialisation (Simpson’s index) from impacted birds (Table 3 and Figure 3). They also differ significantly in the relative height of the vegetation stratum (barycentre) that they use for foraging. Note that difference in vegetation stratum use specialisation (Simpson’s index) of birds is very small, but nevertheless the Wilcoxon Rank Sum test points out a significant difference.
Table 3 Summary of output values of the Wilcoxon Rank Sum test, which compares whether non-impacted species (proportion of range impacted by threats = 0%) significantly differ in degree of diet, forest, habitat and vegetation stratum use specialisation from impacted species (proportion of range impacted by threats > 0%). The results delivered a significant p-value for all variables, implying that non-impacted species significantly differ in specialisation from impacted species. Note that the output values of vegetation stratum use specialisation for mammals are the result of the Kruskal-Wallis test and the corresponding test-statistic is the Chi2-value. This test
was used instead of the Wilcoxon Rank Sum test, because vegetation stratum use for mammals is a categorical variable.
Plot Variable Taxon Test-statistic (W) p-value
A Diet specialisation Mammals 155538 <0.001
B Diet specialisation Birds 199233 <0.001
C Forest specialisation Mammals 153720 <0.001
D Forest specialisation Birds 267933 <0.001
E Habitat specialisation Mammals 210228 <0.001
F Habitat specialisation Birds 334020 <0.001
G Vegetation stratum specialisation (barycentre)
Birds 198265 <0.001
H Vegetation stratum specialisation (Simpson’s index)
Birds 276570 <0.05
Figure 3 Overview that shows the differences in specialisation (y-axis) between non-impacted species (N) and impacted species (Y) (x-axis) for all quantified specialisation variables. Some specialisation variables are log-transformed to improve the visibility of the difference in specialisation (plot C, D, E, F, G and H). Note that difference in vegetation stratum use specialisation (Simpson’s index) of birds is not clearly visible, whereas the Wilcoxon Rank Sum test pointed out a significant difference (plot G). Also note that vegetation stratum use for mammals is not included in this overview, since this is a categorical variable and therefore analysed separately.
Linear model outcomes
As discussed in the Methods section, linear models for mammals were built only using data on the proportion of a species range impacted by threats > 0% and linear models for birds were built only using data on proportion of a species range impacted by threats > 5%. I will now discuss the outcomes of the linear models, which are summarised in Table 4. Also graphs with proportion of a species range impacted by threats (%) plotted against the specialisation variables are summarised in Figure 4.
Firstly, for mammals, diet and forest specialisation are not significantly related to the proportion of a species range impacted by threats (Model B and D). Therefore, mammals that are specialised in diet and forest do not have significantly higher proportions of their range impacted by human threats. However, habitat specialisation is significantly positively related to proportion of a species range impacted by threats (Model H). Therefore, mammals that are specialised in habitat have significantly lower proportions of their range impacted by human threats. This is the opposite direction of the relationship as found for birds (Model G). Namely, bird species that are habitat generalists, meaning that they occur in a broad range of habitat types, have significantly lower proportions of their range impacted by human threats, since the slope of this model is negative.
Secondly, for birds, species that are specialised in forest habitats have significantly higher proportions of their range impacted by threats than bird species that are specialised in open habitats. Namely, the barycentre, as calculated from the forest-open gradient, is negatively related to proportion of a species range impacted by threats and in turn, high values of the barycentre correspond to more open habitats. In addition, bird species that are specialised in a narrow range of vegetation strata (Simpson’s index close to 1) have significantly lower proportions of their range impacted by human threats than bird species that are specialised in a broader range of vegetation strata.
Thirdly, for birds, relationships between diet specialisation and vegetation stratum barycentre and proportion of a species range impacted by human threats were not significant. Therefore, bird species that are specialised in diet (Simpson’s index close to 1) do not have significantly higher or lower proportions of their range impacted by human threats and bird species that are specialised in a higher vegetation stratum (barycentre close to 5) do not have significantly higher or lower proportions of their range impacted by human threats.
However, for all models counts that the R2-value is very low, implying that a linear relationship does
not explain a large amount of the variation in the data. Data on all specialisation variables is always widely scattered around the regression line and therefore does not clearly follow a linear relationship (Figure 4). Table 4 Summary of output values of the linear regression models that were tested for birds and mammals. Response variable is the same in all models (proportion of a species range impacted by at least one human threat (%)).
Model Explanatory variable Taxon Slope p-value R2
A Diet specialisation Birds 0.27 >0.05 -0.001008
B Diet specialisation Mammals -1.65 >0.05 -0.001095
C Forest specialisation Birds -1.90 <0.001 0.0228
D Forest specialisation Mammals -0.028 >0.05 -0.00118 E Vegetation stratum specialisation (barycentre) Birds 0.93 >0.05 -0.00019 F Vegetation stratum specialisation (Simpson’s index) Birds -8.87 <0.05 0.005599
G Habitat specialisation Birds -2.46 <0.001 0.02519 H Habitat specialisation Mammals 1.68 <0.001 0.0113
Figure 4 Proportion of a species ranges impacted by threats (%) (y-axis) plotted against all specialisation variables that were used in the linear models (x-axis). Regression line is added to plots that have a significant p-value in the linear model (Model C, E, G and H). For all models count that the R2-value is low, implying that a
linear relationship does not explain a large amount of the variation in the data. Data on all specialisation variables is always widely scattered around the regression line and therefore does not clearly follow a linear relationship. Note that the analysis for vegetation stratum specialisation of mammals is not included in this overview, because it is an categorical variable and therefore separately analysed.
Vegetation stratum use for mammals
The results of the Tukey-Kramer test, which was used to test difference in the proportion of a species range impacted by threats between mammals that use different vegetation stratum for foraging, show that mammals that are classified to use Aerial (A) and Marine (M) vegetation stratum for foraging have significantly lower proportions of their range impacted by human threats than mammals that are classified to use Ground (G), Arboreal (Ar) or Scansorial (S) vegetation strata for foraging (Table 5 and Figure 5). To obtain a test-statistic (t), I also performed the two-sample t-test for the each comparisons of the categories in Table 5. Significance levels of the t-test were the same as for the Tukey-Kramer test.
Table 5 Output of Tukey-Kramer test for specialisation in vegetation stratum use for foraging by mammals shows that mammals that use Aerial (A) and Marine (M) vegetation stratum for foraging significantly differ from mammals that are classified to use other vegetation stratum. Mammals that use aerial and marine vegetation stratum for foraging thus have significantly lower proportions of their range impacted by human threats. To obtain a test-statistic (t), I also performed the two-sample t-test for the same comparisons of categories. Significance levels of the t-test were the same as for the Tukey-Kramer test.
Comparison of categories
Difference Test-statistic (t) p-value
Ar-A 39.52 9.375 <0.001 G-A 34.62 8.4317 <0.001 M-A -8.33 -1.1104 >0.05 S-A 36.61 5.9352 <0.001 G-Ar -4.90 -1.8836 >0.05 M-Ar -47.85 -7.0449 <0.001 S-Ar -2.91 -0.55108 >0.05 M-G -42.95 -6.3867 <0.001 S-G 1.99 0.38237 >0.05 S-M 44.94 5.5142 <0.001
Figure 5 Proportion of a species range impacted by threats (%) (y-axis) plotted against vegetation stratum categories used by mammals (x-axis). Output of Tukey-Kramer test for specialisation in vegetation stratum use for foraging by mammals shows that mammals that use Aerial (A) and Marine (M) vegetation stratum for foraging significantly differ from mammals that are classified to use other vegetation stratum (significance level is indicated in top part of the figure). Mammals that use aerial and marine vegetation stratum for foraging thus have significantly lower proportions of their range impacted by human threats.
Legend: A = Aerial Ar = Arboreal G = Ground M = Marine S = Scansorial *** ***
Discussion
This research investigated whether species with particular specialisations (diet and habitat among others) have higher proportions of their range impacted by human threats. I did this by constructing and testing linear models with specialisation metrics as explanatory variables and proportion of a species range impacted by threats as response variable. The results advanced knowledge about the determinants of human impacts within species geographic ranges, by using quantified measurements of multiple types of specialisations separately for 1277 mammal and 2120 bird species on a global scale.
Overall, I found that birds that are specialised in habitat and forest have a significantly higher proportions of their range impacted by threats than generalist birds. This trend confirms our expectations, since we expected forest birds to be more impacted on average, because forest habitats are less widespread than open habitats, and habitat specialised birds to be more impacted on average, because they are limited in the habitat types in which they can occur. This result also confirms previous research that investigated population trends of 71 bird species on a national scale and found that habitat specialised species (narrow habitat breadth) are declining at a much higher rate than habitat generalist species (Jiguet et al., 2007). This means that habitat specialist birds could be increasingly replaced by habitat generalist birds in ecological assemblages. Consequently, ecosystem functioning could be impacted, since alteration of community trait composition alters the ecosystem processes that species contribute to (Díaz et al., 2013). To illustrate, scavenger birds are highly threatened and scavenging is a critical component of ecosystem ecology. For example, if scavengers become scarce, carcasses that would otherwise have been cleaned by scavengers could increase the risk of diseases impacting other wildlife (Buechley & Şekercioğlu, 2016).
In contrast, the relationship between habitat specialisation and proportion of a species range impacted by threats was in the opposite direction for mammals. This does not confirm my expectation, since I expected habitat specialists to have a higher proportion of their range impacted. However, I suspect that this relationship is sensitive to outliers, since removal of more outliers quickly decreased the significance level. Therefore, future research has to further investigate the relationship between habitat specialisation and human impacts within species geographic ranges to find a solid result.
Furthermore, for mammals, I did not find a linear relationship between species specialisation in diet, forest and vegetation stratum use and proportion of a species range impacted by human threats. And also the relationship between diet specialisation and proportion of a species range impacted by human threats for birds was not significant. Therefore, our results do not confirm previous research that found a positive relationship between specialisation sensitivity to human-induced environmental changes (Munday, 2004; Heikkinen et al., 2010). In addition, the relationship between height of vegetation stratum use for foraging and the proportion of a species range impacted by human threats for birds was not significant, whereas previous research has shown that use of vegetation stratum can predict the impact environmental changes on bird species (Martin & Possingham, 2005).
In summary, a linear relationship could only explain a low amount of the variation of the data. This reflects that there are multiple drivers of proportion of a species range impacted by human threats, that I did not include in my models. Possible drivers to include in future studies are for example, competitive exclusion and dispersal limitation (Devictor, Julliard & Jiguet, 2008; Pulliam, 2000). Due to competitive exclusion, species are dependent on presence or absence of other species whether they can survive in certain areas or not. Namely, if a species is dominant, it can competitively exclude other species and survive in the area. Conversely, species that are competitively excluded can not occur in that area. Therefore, competitive exclusion shapes distribution of species and is therefore also likely to shape proportion of species range impacted by human threats, since human threats are also spatially distributed. Likewise, dispersal limitation causes species to not occur in areas where it could potentially survive. Consequently, this will be related to proportion of a species range that is impacted by
threats. For example, species that are limited in dispersal are less able to escape ranges with human impacts to which they are sensitive. In addition, research has shown that species with lower dispersal are more vulnerable to climate-related environmental changes (Hellman, Alkemade & Knol, 2016).
We know that effects of threats are governed by species traits. However, multiple species traits and mechanisms like dispersal limitation can show additive effect and synergy, determining species vulnerability to threats. Likewise, also threats interact and consequently magnify their effect on species (Ewers & Didham, 2006). These interactive effects are complex and could have contributed to the low R2-values in the models, as data was
widely scattered around the regression line.
In addition, I used specialisation metrics that could not take differences of resource value or commonness into account. To illustrate, shrub dominated wetlands (habitat category 5.3 in the IUCN Habitat Classification Scheme) could possibly be less widespread than boreal forest habitats (habitat category 1.1 in the IUCN Habitat Classification Scheme), whereas in our analysis they have same weight in the habitat specialisation calculation. As a result, species that use shrub dominated wetlands instead of boreal forests may have a higher proportion of their range impacted by human threats. For future research I recommend investigating if there is a relationship between proportion of a species range impacted by threats and a species habitat specialisation weighted by habitat total area across the world. Other possible explanations for finding no linear relationship, could be that proportion of a species range impacted by threats is not mainly determined by degree of specialisation (quantity), but what resources a species is specialised to use (quality).This idea is supported by the results of specialisation of vegetation stratum for mammals. Namely, I found that mammal species that use aerial and marine vegetation stratum for foraging have significant lower proportions of their range impacted by human threats than mammal species that use arboreal, ground or scansorial vegetation stratum for foraging. This could be due to that aerial and marine vegetation strata are more widespread. It could also be due to that arboreal, ground or scansorial mammals on average live closer to humans and therefore have on average a higher proportion of their range impacted by threats, which is intuitive, since extinction risks of mammals have been shown to correlate with human population density (Parks & Harcourt, 2002). One could explore the effect of use of resources on proportion of range impacted by separating resource categories (for example habitat, diet and vegetation stratum) into subcategories (for example seeds and fruits) and test whether species that use a resource from one subcategory have lower or higher proportions of their range impacted by threats than species that use a resource from another subcategory.
Nevertheless, the results are innovative, since this research was the first effort that investigated the relationship between human threats within species geographic ranges and multiple quantified specialisation variables, on a global scale, for different taxa and multiple human threats. I applied the habitat classification scheme in quantifying species habitat specialisation, which is an authorized and detailed habitat classification structure. And even though I could not take relative differences of resource value or commonness into account, I could take into account relative difference in use of resource. To illustrate, a species that distributes its diet across four categories less equally (for example large use of one category and less use of the other three), is quantified as more specialised than another species that uses the same four categories equally. In addition, previous
neoecological, paleoecological and phylogenetical approaches that inferred decline of specialist species, have delivered divergent results due to, among others, coding specialism as binary or not accounting for species differences in range size (Colles, Liow & Prinzing, 2009). I was able to account for species difference in range size, since I used species proportion of a species range impacted as the response variable, which is independent of total range size (Allan et al., 2019). In addition, I used a continuous measurement of specialisation by quantifying specialisation with the use of functional trait data from the EltonTrait1.0 database (Wilman et al., 2014).
To round up, species differences in responses to human impacts is a pattern that causes a shift in the functional trait composition of communities (Newbold et al., 2020). In particular, if generalist species replace specialist species, spatial distribution of functional traits becomes more equal, which is called functional
homogenization (Clavel, Julliard & Devictor, 2011). As climate change and habitat modification are expected to continue due to human activities, I measured the opposing responses of more or less specialised species to these human activities, which contributes in narrowing down conservation options and developing new theoretical investigations in order to stem this homogenization process (Devictor, Julliard & Jiguet, 2008). Namely, functional homogenization, one symptom of biodiversity loss, requires more attention because species identity, abundance and geographic range have been shown to matter for resilient ecosystems and ongoing delivery of ecosystem services (Díaz et al., 2006; Eskilden et al., 2015).
Conclusion
This research investigated whether species with particular specialisations (diet, habitat, forest and vegetation stratum use) have higher proportions of their range impacted by human threats. I constructed and tested linear models with specialisation metrics as explanatory variables and proportion of range impacted by threats as response variable. Birds that are specialised in habitat and forest have significantly higher proportions of their range impacted by threats than generalist birds. However, no other significant relationships were found, which does not confirm previous research that showed a positive relationship between specialisation and species sensitivity to environmental changes. Clearly more research is needed to understand the complex interactions of drivers to human impacts within species ranges. It is recommended to investigate the effects of mechanisms like dispersal limitation on the proportion of a species range impacted by human threats and to account for habitat differences in commonness when calculating a measure of habitat specialisation. Differences of human impacts between species with different traits cause a shift in the functional trait composition of communities, resulting in functional homogenization. This work contributes to supporting decisions on conservation management options for species with different traits and towards developing new theoretical investigations to understand and stem biotic homogenization of species assemblages; a crucial step for maintaining ecosystem resilience and ecosystem services.
References
Allan, J. R., Watson, J. E., Di Marco, M., O’Bryan, C. J., Possingham, H. P., Atkinson, S. C., & Venter, O. (2019). Hotspots of human impact on threatened terrestrial vertebrates. PLoS Biol, 17(3), e3000158.
Attum, O., Eason, P., Cobbs, G., & El Din, S. M. B. (2006). Response of a desert lizard community to habitat degradation: do ideas about habitat specialists/generalists hold?. Biological conservation, 133(1), 52-62.
Barnagaud, J. Y., Devictor, V., Jiguet, F., & Archaux, F. (2011). When species become generalists: on‐going large‐ scale changes in bird habitat specialisation. Global Ecology and Biogeography, 20(4), 630-640.
Buchmann, C. M., Schurr, F. M., Nathan, R., & Jeltsch, F. (2013). Habitat loss and fragmentation affecting mammal and bird communities—The role of interspecific competition and individual space use. Ecological
Informatics, 14, 90-98.
Buechley, E. R., & Şekercioğlu, Ç. H. (2016). The avian scavenger crisis: looming extinctions, trophic cascades, and loss of critical ecosystem functions. Biological Conservation, 198, 220-228.
Carboni, M., Zelený, D., & Acosta, A. T. (2016). Measuring ecological specialisation along a natural stress gradient using a set of complementary niche breadth indices. Journal of Vegetation Science, 27(5), 892-903.
Clavel, J., Julliard, R., & Devictor, V. (2011). Worldwide decline of specialist species: toward a global functional homogenization?. Frontiers in Ecology and the Environment, 9(4), 222-228.
Colles, A., Liow, L. H., & Prinzing, A. (2009). Are specialists at risk under environmental change? Neoecological, paleoecological and phylogenetic approaches. Ecology letters, 12(8), 849-863.
Devictor, V., Julliard, R., Couvet, D., Lee, A., & Jiguet, F. (2007). Functional homogenization effect of urbanization on bird communities. Conservation Biology, 21(3), 741-751.
Devictor, V., Julliard, R., & Jiguet, F. (2008). Distribution of specialist and generalist species along spatial gradients of habitat disturbance and fragmentation. Oikos, 117(4), 507-514.
Devictor, V., Clavel, J., Julliard, R., Lavergne, S., Mouillot, D., Thuiller, W., Venail, P., Villeger, S. & Mouquet, N. (2010) Defining and measuring ecological specialisation. Journal of Applied Ecology, 47, 15–25.
Diamond, S. E., Frame, A. M., Martin, R. A., & Buckley, L. B. (2011). Species traits predict phenological responses to climate change in butterflies. Ecology, 92(5), 1005-1012.
Díaz, S., Purvis, A., Cornelissen, J. H., Mace, G. M., Donoghue, M. J., Ewers, R. M., ... & Pearse, W. D. (2013). Functional traits, the phylogeny of function, and ecosystem service vulnerability. Ecology and evolution, 3(9), 2958-2975.
Díaz, S., Settele, J., Brondízio, E. S., Ngo, H. T., Agard, J., Arneth, A., ... & Garibaldi, L. A. (2019). Pervasive human-driven decline of life on Earth points to the need for transformative change. Science, 366(6471).
Eskildsen, A., Carvalheiro, L. G., Kissling, W. D., Biesmeijer, J. C., Schweiger, O., & Høye, T. T. (2015). Ecological specialisation matters: long‐term trends in butterfly species richness and assemblage composition depend on multiple functional traits. Diversity and distributions, 21(7), 792-802.
Ewers, R. M., & Didham, R. K. (2006). Confounding factors in the detection of species responses to habitat fragmentation. Biological reviews, 81(1), 117-142.
Hallman, T. A., & Robinson, W. D. (2020). Deciphering ecology from statistical artefacts: Competing influence of sample size, prevalence and habitat specialisation on species distribution models and how small evaluation datasets can inflate metrics of performance. Diversity and Distributions, 26(3), 315-328.
Hellmann, F., Alkemade, R., & Knol, O. M. (2016). Dispersal based climate change sensitivity scores for European species. Ecological Indicators, 71, 41-46.
IUCN. (n.d.). IUCN Habitat classification scheme. Retrieved from https://www.iucnredlist.org/resources/habitat-classification-scheme.
IUCN. (2020). The IUCN Red List of Threatened Species. Version 2020-1. Retrieved from https://www.iucnredlist.org.
Jiguet, F., GADOT, A. S., Julliard, R., Newson, S. E., & Couvet, D. (2007). Climate envelope, life history traits and the resilience of birds facing global change. Global Change Biology, 13(8), 1672-1684.
Julliard R, Jiguet F, and Couvet D. 2004. Evidence for the impact of global warming on the long-term population dynamics of common birds. P Roy Soc Lond 271: S490– 92.
Julliard, R., Clavel, J., Devictor, V., Jiguet, F., & Couvet, D. (2006). Spatial segregation of specialists and generalists in bird communities. Ecology letters, 9(11), 1237-1244.
Martin, T. G., & Possingham, H. P. (2005). Predicting the impact of livestock grazing on birds using foraging height data. Journal of Applied Ecology, 42(2), 400-408.
Munday, P. L. (2004). Habitat loss, resource specialisation, and extinction on coral reefs. Global change
biology, 10(10), 1642-1647.
Newbold, T., Bentley, L. F., Hill, S. L., Edgar, M. J., Horton, M., Su, G., ... & Purvis, A. (2020). Global effects of land use on biodiversity differ among functional groups. Functional Ecology, 34(3), 684-693.
Pacifici, M., Visconti, P., & Rondinini, C. (2018). A framework for the identification of hotspots of climate change risk for mammals. Global change biology, 24(4), 1626-1636.
Parks, S. A., & Harcourt, A. H. (2002). Reserve size, local human density, and mammalian extinctions in US protected areas. Conservation Biology, 16(3), 800-808.
Poisot, T., Canard, E., Mouquet, N., & Hochberg, M. E. (2012). A comparative study of ecological specialisation estimators. Methods in Ecology and Evolution, 3(3), 537-544.
Pulliam, H. R. (2000). On the relationship between niche and distribution. Ecology letters, 3(4), 349-361.
RStudio Team (2015). RStudio: Integrated Development for R. RStudio, Inc., Boston, MA URL http://www.rstudio.com/.
Venter, O., Sanderson, E. W., Magrach, A., Allan, J. R., Beher, J., Jones, K. R., ... & Levy, M. A. (2016). Global terrestrial Human Footprint maps for 1993 and 2009. Scientific data, 3(1), 1-10.
Wilman, H., Belmaker, J., Simpson, J., De La Rosa, C., Rivadeneira, M. M., & Jetz, W. (2014). EltonTraits 1.0: Species‐level foraging attributes of the world's birds and mammals: Ecological Archives E095‐178. Ecology, 95(7), 2027-2027.
Appendices
Appendix A: IUCN Habitat Classification Scheme (Version 3.0) (IUCN, n.d.)
The habitat types listed below are standard terms used to describe the major habitats in which taxa occur. The three levels of the hierarchy are self-explanatory, as they use familiar habitat terms that take into account biogeography, latitudinal zonation and depth in marine systems.• 1 Forest o 1.1 Boreal o 1.2 Subarctic o 1.3 Subantarctic o 1.4 Temperate o 1.5 Subtropical/Tropical Dry
o 1.6 Subtropical/Tropical Moist Lowland
o 1.7 Subtropical/Tropical Mangrove Vegetation Above High Tide Level o 1.8 Subtropical/Tropical Swamp
o 1.9 Subtropical/Tropical Moist Montane
• 2 Savanna o 2.1 Dry Savanna o 2.2 Moist Savana • 3 Shrubland o 3.1 Subarctic o 3.2 Subantarctic o 3.3 Boreal o 3.4 Temperate o 3.5 Subtropical/Tropical Dry o 3.6 Subtropical/Tropical Moist o 3.7 Subtropical/Tropical High Altitude o 3.8 Mediterranean-type Shrubby Vegetation
• 4 Grassland
o 4.1 Tundra o 4.2 Subarctic o 4.3 Subantarctic o 4.4 Temperate
o 4.5 Subtropical/Tropical Dry Lowland
o 4.6 Subtropical/Tropical Seasonally Wet/Flooded Lowland o 4.7 Subtropical/Tropical High Altitude
• 5 Wetlands (inland)
o 5.1 Permanent Rivers/Streams/Creeks [includes waterfalls] o 5.2 Seasonal/Intermittent/Irregular Rivers/Streams/Creeks o 5.3 Shrub Dominated Wetlands
o 5.4 Bogs, Marshes, Swamps, Fens, Peatlands o 5.5 Permanent Freshwater Lakes [over 8 ha]
o 5.6 Seasonal/Intermittent Freshwater Lakes [over 8 ha] o 5.7 Permanent Freshwater Marshes/Pools [under 8 ha]
o 5.8 Seasonal/Intermittent Freshwater Marshes/Pools [under 8 ha] o 5.9 Freshwater Springs and Oases
o 5.10 Tundra Wetlands [includes pools and temporary waters from snowmelt] o 5.11 Alpine Wetlands [includes temporary waters from snowmelt]
o 5.12 Geothermal Wetlands o 5.13 Permanent Inland Deltas
o 5.14 Permanent Saline, Brackish or Alkaline Lakes
o 5.15 Seasonal/Intermittent Saline, Brackish or Alkaline Lakes and Flats o 5.16 Permanent Saline, Brackish or Alkaline Marshes/Pools
o 5.17 Seasonal/Intermittent Saline, Brackish or Alkaline Marshes/Pools o 5.18 Karst and Other Subterranean Hydrological Systems [inland]
• 6 Rocky Areas [e.g. inland cliffs, mountain peaks]
• 7 Caves and Subterranean Habitats (non-aquatic)
o 7.1 Caves
o 7.2 Other Subterranean Habitats
• 8 Desert
o 8.1 Hot o 8.2 Temperate o 8.3 Cold
• 9 Marine Neritic (Submergent Nearshore Continental Shelf or Oceanic Island)
o 9.1 Pelagic
o 9.2 Subtidal Rock and Rocky Reefs o 9.3 Subtidal Loose Rock/Pebble/Gravel o 9.4 Subtidal Sandy
o 9.5 Subtidal Sandy-Mud o 9.6 Subtidal Muddy o 9.7 Macroalgal/Kelp o 9.8 Coral Reef
▪ 9.8.1 Outer Reef Channel ▪ 9.8.2 Back Slope
▪ 9.8.3 Foreslope (Outer Reef Slope) ▪ 9.8.4 Lagoon
▪ 9.8.5 Inter-Reef Soft Substrate ▪ 9.8.6 Inter-Reef Rubble Substrate o 9.9 Seagrass (Submerged) o 9.10 Estuaries • 10 Marine Oceanic o 10.1 Epipelagic (0-200 m) o 10.2 Mesopelagic (200-1,000 m) o 10.3 Bathypelagic (1,000-4,000 m) o 10.4 Abyssopelagic (4,000-6,000 m)
• 11 Marine Deep Benthic
o 11.1 Continental Slope/Bathyl Zone (200-4,000 m) ▪ 11.1.1 Hard Substrate
▪ 11.1.2 Soft Substrate o 11.2 Abyssal Plain (4,000-6,000 m)
o 11.3 Abyssal Mountain/Hills (4,000-6,000 m) o 11.4 Hadal/Deep Sea Trench (>6,000 m) o 11.5 Seamount
o 11.6 Deep Sea Vents (Rifts/Seeps)
• 12 Marine Intertidal
o 12.1 Rocky Shoreline
o 12.2 Sandy Shoreline and/or Beaches, Sand Bars, Spits, Etc. o 12.3 Shingle and/or Pebble Shoreline and/or Beaches o 12.4 Mud Flats and Salt Flats
o 12.6 Tidepools
o 12.7 Mangrove Submerged Roots
• 13 Marine Coastal/Supratidal
o 13.1 Sea Cliffs and Rocky Offshore Islands o 13.2 Coastal Caves/Karst
o 13.3 Coastal Sand Dunes
o 13.4 Coastal Brackish/Saline Lagoons/Marine Lakes o 13.5 Coastal Freshwater Lakes
• 14 Artificial - Terrestrial o 14.1 Arable Land o 14.2 Pastureland o 14.3 Plantations o 14.4 Rural Gardens o 14.5 Urban Areas
o 14.6 Subtropical/Tropical Heavily Degraded Former Forest
• 15 Artificial - Aquatic
o 15.1 Water Storage Areas (over 8 ha) o 15.2 Ponds (below 8 ha)
o 15.3 Aquaculture Ponds o 15.4 Salt Exploitation Sites o 15.5 Excavations (open)
o 15.6 Wastewater Treatment Areas
o 15.7 Irrigated Land [includes irrigation channels] o 15.8 Seasonally Flooded Agricultural Land o 15.9 Canals and Drainage Channels, Ditches
o 15.10 Karst and Other Subterranean Hydrological Systems [human-made] o 15.11 Marine Anthropogenic Structures
o 15.12 Mariculture Cages
o 15.13 Mari/Brackish-culture Ponds
• 16 Introduced Vegetation
• 17 Other
Appendix B: Removal of outliers
Birds Habitat specialisation
Full dataset Without outliers
Slope p-value R2-squared Slope p-value R2-squared
-2.13 <0.001 0.022 -2.46 <0.001 0.025
Mammals Habitat specialisation
Full dataset Without outliers
Slope p-value R2-squared Slope p-value R2-squared
1.13 <0.001 0.012 1.68 <0.001 0.011
Birds Vegetation stratum specialisation (barycentre)
Full dataset Without outliers
Slope p-value R2-squared Slope p-value R2-squared
