Global mammalian zooregions reveal a signal of past human impacts

TLDR: This work quantifies the relative importance of human land use from ∼5000 years ago to predict the current assemblage of terrestrial mammals in biogeographical regions across the Earth and highlights the far-reaching effect that past anthropogenic actions have had on the organization of biodiversity globally.

Abstract: Understanding how the world’s biodiversity is organized and how it changes across geographic regions is critical to predicting the effects of global change1. Ecologists have long documented that the world’s terrestrial fauna is organized hierarchically in large regions - or realms - and continental scale subregions2–6, with boundaries shaped by geographic and climatic factors2,7. However, little is known about how global biodiversity is assembled below the continental level and the factors, including the potential role of human impacts, triggering faunistic differences as the biogeographical scale becomes smaller. Here we show that the hierarchical organization of global zoogeographical regions extends coherently below the region level to reach a local scale, and that multiple determinants act across varying spatial and temporal scales. Among these determinants, anthropogenic land use during the Late Holocene stands out showing a footprint across biogeographical scales and explaining 22% of the faunistic differences among the larger bioregions. The Late Holocene coincided with the development of large cities and substantial transformation of ecosystems into agricultural land8,9. Our results show that past human activity has played a role in the global organization of present-day animal assemblages, leaving a detectable signal that warns us about significant time-lag effects of human-mediated impacts on biodiversity.

Figures

Figure 2 | Venn diagrams describing the predictive power of the three most important determinants of the two largest bioregion levels. Values indicate percentages of samples correctly classified (1-OBB error) and are obtained from running RF classification models for the individual determinants, by pairs and for the three most important determinants together. Venn diagrams for the smaller bioregion levels (141 and 1128) are not shown because the predictive value of RF models using only the three most important determinants was low (32.4% for the 141 bioregions model and 7.3% for the 1128 bioregions). But see Extended Data Table 1 for more complete information, where the determinants needed to reach 50% of samples correctly classified for each biogeographical level is shown.

Figure 1 | Hypothesized scenarios (A, B) and results describing variable importance from random forest classification models (C). A) This scenario proposes that remote past determinants are largely irrelevant to explain the assembly of species in smaller bioregions. The signal of the past has been diluted as the biogeographic resolution increases, such that large bioregions can be described by remote past determinants, medium-sized bioregions by recent past determinants, and smaller bioregions by contemporary determinants. B) This scenario proposes nested importance of determinants, with remote past determinants being important at all scales, recent past determinants being important at medium and small scales, and contemporary determinants being only relevant at local scale. C) Observed global variable importance

Figure 3 | Local importance values of the human impact 2000 years ago for the four biogeographical scales. The legends show the impact on correct classification of single samples (grid cells): negative, 0 (the variable is neutral) and positive. See Extended Data Fig. 2 to compare these results with the map showing human land use 2000 years ago.

Full text

1
Global mammalian zooregions reveal a signal of past human impacts
Marta Rueda
1*
, Manuela González-Suárez
2
, Eloy Revilla
1
1. Department of Conservation Biology, Estación Biológica de Doñana (EBD-CSIC),
Seville, Spain
2. Ecology and Evolutionary Biology, School of Biological Sciences, University of
Reading, Reading, UK.
*Corresponding author
certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was notthis version posted March 24, 2019. ; https://doi.org/10.1101/586313doi: bioRxiv preprint

2
Understanding how the world’s biodiversity is organized and how it changes across
geographic regions is critical to predicting the effects of global change
1
. Ecologists
have long documented that the world's terrestrial fauna is organized hierarchically
in large regions - or realms - and continental scale subregions
2-6
, with boundaries
shaped by geographic and climatic factors
2,7
. However, little is known about how
global biodiversity is assembled below the continental level and the factors,
including the potential role of human impacts, triggering faunistic differences as the
biogeographical scale becomes smaller. Here we show that the hierarchical
organization of global zoogeographical regions extends coherently below the region
level to reach a local scale, and that multiple determinants act across varying spatial
and temporal scales. Among these determinants, anthropogenic land use during the
Late Holocene stands out showing a footprint across biogeographical scales and
explaining 22% of the faunistic differences among the larger bioregions. The Late
Holocene coincided with the development of large cities and substantial
transformation of ecosystems into agricultural land
8,9
. Our results show that past
human activity has played a role in the global organization of present-day animal
assemblages, leaving a detectable signal that warns us about significant time-lag
effects of human-mediated impacts on biodiversity.
The questions of how the world’s biodiversity is organized, and why large-scale patterns
of taxonomic diversity change through natural geographic regions have attracted the
attention of naturalists since the early 19
th
century
2,10-15
. The answers to these questions
are important to satisfy our curiosity about the natural world, but have also become
certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was notthis version posted March 24, 2019. ; https://doi.org/10.1101/586313doi: bioRxiv preprint

3
critical to forecast the future of biodiversity in the face of global change
1
. A key step in
understanding the organization of biodiversity is the assemblage of regions based on their
shared elements
14
. Alfred R. Wallace was among the first to propose that the world’s
fauna is organized hierarchically in broad regions shaped by geographic and climatic
factors
2
. About 150 years later, the development of multivariate analytical techniques has
led to the revaluation of Wallace's proposal
3-6
and refining of the extrinsic factors
explaining the major dissimilarities among zooregions
7
. However, biogeographic
boundaries and the signal of evolutionary processes associated to species isolation are not
so evident at smaller scales, and importantly, still remain globally unexplored. Smaller
regions, which are generally the units of conservation actions, contain more similar biota
and thus, the factors determining faunistic dissimilarities among them are likely to be
more diverse and include spatial and taxonomic idiosyncracies
7,16,17
.
We hypothesize that global biodiversity patterns can be characterized by a hierarchical
system of biogeographic regions extending from global to local scales, with regions at
different levels explained by determinants that represent varying temporal scales. To test
this, we considered determinants already identified as important: plate tectonics, climate -
including Quaternary climate changes, orography and changes in habitat type
7,16,17
. But
critically, we also explore the role of largely overlooked predictors associated with
present and past anthropogenic global impacts. The effects of recent human actions on
current species distributions are undeniable
18
, already affecting current biogeographic
patterns
19,20
, but past human actions are generally portrayed as localized and insignificant
in comparison
21
. The increasing evidence that Quaternary human impacts induced shifts
certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was notthis version posted March 24, 2019. ; https://doi.org/10.1101/586313doi: bioRxiv preprint

4
in the plant and animal communities we see today
22,23
challenges this view and poses the
question of whether past anthropogenic land impacts may have been large enough to
induce changes detectable today at biogeographical scales.
To address these questions we first applied an affinity propagation clustering algorithm
24
to a co-occurring species matrix of global terrestrial mammals generating a hierarchical
bioregionalization upscaling from the smallest detectable bioregions to the largest realms.
We then used random forest classification models to identify the determinants that best
predict taxonomic differences among bioregions within the framework of two
hypothesized scenarios (Fig. 1). These scenarios always consider remote past, recent past
and contemporary determinants but assume their influence will differ across the
hierarchical levels. Differentiation between large realms should require longer
evolutionary times, and therefore, both scenarios assume that factors related to historical
and macroevolutionary processes of speciation and extinction will be most important to
explain taxonomic dissimilarities of the largest realms. As bioregions decrease in size we
predict processes related to tolerances to given habitats or climates (which are also forged
over evolutionary time) and human impact would gain importance. The scenarios differ
in how we suggest this process may occur: linearly (Fig. 1A) or with a nested structure
(Fig. 1B).
The clustering algorithm generated a hierarchical system of biogeographic regions with
four levels showing that global biodiversity patterns can be cohesively shaped from local
(area of the smallest bioregions detected is ~93 km
2
) to regional and to realm scales
certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was notthis version posted March 24, 2019. ; https://doi.org/10.1101/586313doi: bioRxiv preprint

5
(Extended Data Fig. 1 and Supplementary Fig. 1). The broadest delineation of nine large
bioregions was strikingly similar to the six zoogeographical regions and boundaries
proposed by Wallace
2
, showing our method is a suitable approach to define bioregions.
Particular differences include the delineation of Madagascar and Chilean subregions
(sensu Wallace) as regions
2,6
, differences in the limits of the Palearctic also detected in
previous analytical regionalizations
3,5,6
,
and an extension towards the arid steppes of
Mongolia of the 'Saharo-Arabian' realm
3,6
.
We found a nested effect of temporal determinants of bioregion assemblages (Fig. 1C),
similar to our proposed nested scenario (Fig. 1B). The signal of events occurring millions
of years ago, such as tectonic movements or orographic barriers, remained apparent from
the largest to the smallest bioregions, while recent past and contemporaneous
determinants acquired importance at smaller scales. Overall, results for the two broadest
scales (nine and 27 bioregions respectively) supported findings from prior work
7
. Plate
tectonics drove the main taxonomic dissimilarities between large landmasses in interplay
with variability in climate and orographic barriers, the latter with less weight in the global
model but important for determining differences between specific regions (Figs. 1C, 2,
Extended Data Fig. 3 and Supplementary Table 1 and 2). For the smaller scales detected
(141 and 1128 bioregions), we found that a combination of multiple determinants, that
varied spatially in importance, was critical to predict assemblages (Fig. 1C, Extended
Data Fig. 4, and Supplementary Table 3 and 4). Among them, the association of
geological factors, past climate change and current variability in temperature resulted
decisive (Fig. 1C, Extended Data Table 1).
certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.
The copyright holder for this preprint (which was notthis version posted March 24, 2019. ; https://doi.org/10.1101/586313doi: bioRxiv preprint

Frequently asked questions

Q1. What are the main barriers to dispersal for mammals?

Mountain ranges represent major barriers to dispersal for most mammals, whereas plate tectonics are responsible of the long-term isolation of the biotas on some plates42. 

Q2. How do random forests disentangle interacting effects?

Random forests are able to disentangle interacting effects and identify nonlinear and scale-dependent relationships that often occur at the scale of the analysis performed here among multiple correlated predictors 

Q3. What predictors were used to determine the current climate?

Contemporary predictors included current climate (annual total precipitation, mean annual temperature, seasonality in temperature and seasonality in precipitation), habitatrelated predictors (primary productivity, main land cover type and tree cover), and human impact related predictors (human appropriation of net primary production, and the human influence index). 

Q4. What are the predictors of the global human influence index?

The global human influence index (HII) is a composite predictor covering population density, human land use and infrastructure, and human access37. 

Q5. What is the way to detect exemplars?

The algorithm detects special data points called exemplars, and by a message-passing procedure it iteratively connects every data point to the exemplar that best represent it until an optimal set of exemplars and clusters emerges. 

Q6. What are the predictors of the current climate?

Mean annual temperature and annual total precipitation for the MH and the LGM were calculated using the Model for Interdisciplinary Research on Climate (MIROC-ESM)39, whereas forthe LIG the authors used the model of Otto-Bliesner et al. (2006)40. 

Q7. What is the local importance score for a tree?

The local importance score is derived from all trees for which the sample was not used to train the tree (i.e. its value is OBB). 

Q8. What is the significance of the local importance score?

Although random forests are generally assumed to not be affected by highly correlated predictor variables, the authors eliminate some predictors showing a high correlation (r > 0.70, Supplementary Table 5) as some evidence from genomic studies suggests that variable importance measures may show a bias towards correlated predictor variables 

Q9. How is the random forest algorithm trained?

Each tree is trained by selecting a random set of variables and a random sample from the training dataset (i.e., the calibration data set).