DataPDF Available


This PDF file includes: Materials and Methods Supplementary Text Figs. S1 to S6 Tables S1 to S4 References
Supplementary Materials for
Global distribution of earthworm diversity
Helen R. P. Phillips*, Carlos A. Guerra, Marie L. C. Bartz, Maria J. I. Briones, George Brown,
Thomas W. Crowther, Olga Ferlian, Konstantin B. Gongalsky, Johan van den Hoogen, Julia Krebs,
Alberto Orgiazzi, Devin Routh, Benjamin Schwarz, Elizabeth M. Bach, Joanne Bennett, Ulrich Brose,
Thibaud Decaëns, Birgitta König-Ries, Michel Loreau, Jérôme Mathieu, Christian Mulder,
Wim H. van der Putten, Kelly S. Ramirez, Matthias C. Rillig, David Russell, Michiel Rutgers, Madhav P. Thakur,
Franciska T. de Vries, Diana H. Wall, David A. Wardle, Miwa Arai, Fredrick O. Ayuke, Geoff H. Baker,
Robin Beauséjour, José C. Bedano, Klaus Birkhofer, Eric Blanchart, Bernd Blossey, Thomas Bolger,
Robert L. Bradley, Mac A. Callaham, Yvan Capowiez, Mark E. Caulfield, Amy Choi, Felicity V. Crotty,
Andrea Dávalos, Darío J. Diaz Cosin, Anahí Dominguez, Andrés Esteban Duhour, Nick van Eekeren,
Christoph Emmerling, Liliana B. Falco, Rosa Fernández, Steven J. Fonte, Carlos Fragoso, André L. C. Franco,
Martine Fugère, Abegail T. Fusilero, Shaieste Gholami, Michael J. Gundale, Mónica Gutiérrez López,
Davorka K. Hackenberger, Luis M. Hernández, Takuo Hishi, Andrew R. Holdsworth, Martin Holmstrup,
Kristine N. Hopfensperger, Esperanza Huerta Lwanga, Veikko Huhta, Tunsisa T. Hurisso, Basil V. Iannone III,
Madalina Iordache, Monika Joschko, Nobuhiro Kaneko, Radoslava Kanianska, Aidan M. Keith,
Courtland A. Kelly, Maria L. Kernecker, Jonatan Klaminder, Armand W. Koné, Yahya Kooch,
Sanna T. Kukkonen, H. Lalthanzara, Daniel R. Lammel, Iurii M. Lebedev, Yiqing Li, Juan B. Jesus Lidon,
Noa K. Lincoln, Scott R. Loss, Raphael Marichal, Radim Matula, Jan Hendrik Moos, Gerardo Moreno,
Alejandro Morón-Ríos, Bart Muys, Johan Neirynck, Lindsey Norgrove, Marta Novo, Visa Nuutinen,
Victoria Nuzzo, Mujeeb Rahman P, Johan Pansu, Shishir Paudel, Guénola Pérès, Lorenzo Pérez-Camacho,
Raúl Piñeiro, Jean-François Ponge, Muhammad Imtiaz Rashid, Salvador Rebollo, Javier Rodeiro-Iglesias,
Miguel Á. Rodríguez, Alexander M. Roth, Guillaume X. Rousseau, Anna Rozen, Ehsan Sayad, Loes van Schaik,
Bryant C. Scharenbroch, Michael Schirrmann, Olaf Schmidt, Boris Schröder, Julia Seeber, Maxim P. Shashkov,
Jaswinder Singh, Sandy M. Smith, Michael Steinwandter, José A. Talavera, Dolores Trigo, Jiro Tsukamoto,
Anne W. de Valença, Steven J. Vanek, Iñigo Virto, Adrian A. Wackett, Matthew W. Warren, Nathaniel H. Wehr,
Joann K. Whalen, Michael B. Wironen, Volkmar Wolters, Irina V. Zenkova, Weixin Zhang, Erin K. Cameron,
Nico Eisenhauer
*Corresponding author. Email:
Published 25 October 2019, Science 366, 480 (2019)
DOI: 10.1126/science.aax4851
This PDF file includes:
Materials and Methods
Supplementary Text
Figs. S1 to S6
Tables S1 to S4
Materials and Methods
Literature Search
Web of Science was searched on 18th December 2016, using the following search term:
((Earthworm* OR Oligochaeta OR Megadril* OR Haplotaxida OR Annelid* OR Lumbric* OR
Clitellat* OR Acanthodrili* OR Ailoscoleci* OR Almid* OR Benhamiin* OR riodrilid* OR
Diplocard* OR Enchytraeid* OR Eudrilid* OR Exxid* OR Glossoscolecid* OR Haplotaxid* OR
Hormogastrid* OR Kynotid* OR Lutodrilid* OR Megascolecid* OR Microchaetid* OR
Moniligastrid* OR Ocnerodrilid* OR Octochaet* OR Sparganophilid* OR Tumakid* ) AND
(Diversity OR “Species richness” OR “OTU” OR Abundance OR individual* OR Density OR
“tax* richness” OR “Number” OR Richness OR Biomass))
This search returned 7783 papers. All titles and abstracts of papers post-2000 were screened
(6140 papers), and were excluded if they did not reference data suitable for the analysis
(suitability discussed below). Since it was anticipated that raw data would need to be requested,
papers published before 2000 were not screened, as it was unlikely that available author contact
details were up-to-date. We note however that earlier publications may be useful for future
research, e.g., focusing on long-term monitoring and temporal analyses. After this initial
screening, PDFs of all remaining papers (n = 986) were manually screened to determine whether
data were suitable.
In order to be suitable for the analysis, papers had to present (or make reference to) the
following information and data:
1. Sampled earthworm communities using standard earthworm extraction methodologies,
which would adequately capture quantitative information of the earthworm community,
such as hand-sorting of a sufficient soil volume [e.g., (39)] or chemical expulsion from a
quadrat [e.g., (40)] at two or more sites. At a minimum, total fresh biomass and/or total
abundance of the earthworms at each site had to be measured. Ideally, there was data on
identification of all individuals to species level, with the abundance/biomass data of
each species;
2. Available geographic coordinates for all sampled sites, or maps that could be
3. Measurements of at least one soil property at each site (see below);
4. Information on the habitat cover and/or land use;
5. Differences in land use/habitat cover or soil properties (see below for information on the
land use/habitat cover and soil properties) across the sites.
Where possible, all suitable data were taken from the 477 papers that were identified as
containing suitable data. Data were extracted from figures where necessary (using IMAGEJ
(41)). If data were not provided in the text or the supplementary materials, authors were
contacted to obtain the raw data from each site. As some datasets remain unpublished, or are yet
to be published, individual earthworm researchers were also contacted to enquire as to whether
they had suitable data. Including unpublished data helps to reduce publication bias (42).
Data collation
The data taken or requested from one publication or an unpublished field campaign was
considered a ‘dataset’. If a dataset contained data sampled using different methodologies, we
split it into different ‘studies’ based on the methodology, as measured diversity of earthworms is
highly dependent on the methods used (43). For datasets where sites were repeatedly sampled
over time, both within years and across years, we used only the first and the last sampling
campaign and these were split into two studies. The modelling approach used (linear mixed-
effects models, with random effects accounting for different studies) dealt with non-
independence of such datasets (44).
Site level information
Sites were described as a location of one or more samples, which, when taken together,
adequately captured the earthworm community. Sampling methodology, and therefore the
number of samples per site, were determined by the original data collectors. But sampling effort
was constant within a study. For each dataset, we collated the following information into a
standardised data template: geographic coordinates for each of the sampled sites, start and end
dates of sampling (month and year), and the sampling method used. For each dataset, we
requested at least one soil property (pH, cation exchange capacity (CEC) or base saturation,
organic carbon, soil organic matter, C/N ratio, soil texture, soil type, soil moisture) for each site,
but only pH, CEC, organic carbon and soil texture (silt and clay) variables were used for this
analysis. Most sites contained pH values (63.7%), 14% of sites contained organic carbon, 40% of
sites contained silt and clay, but only 7.3% contained CEC. Any missing soil properties were
filled with SoilGrids data, described below. If soil properties were given for different soil depths,
then we calculated a weighted average (maximum soil depth = 1 m, but typically collected down
to 30 cm). Using information within the published articles, and additional information provided
by the data collectors, the habitat cover at each site was classified into categories based on the
ESA CCI-LC 300m map (; Table S1).
Recorded community metrics
For each dataset, the following site-level community metrics were calculated where
possible: total (adults and juveniles) abundance of earthworms at the site, total (adults and
juveniles) fresh biomass of earthworms at the site, and number of species at the site. Using the
area sampled at the site, both abundance and biomass were transformed to individuals per m2 and
grams per m2, respectively, if they were not already given in that unit, to standardize the data into
commonly used units. Species richness of each site was calculated from available species lists, if
not already provided. Two issues arose when calculating species richness of earthworms. Firstly,
many specimens were not identified to species level. Where data collectors identified a specimen
as a unique morphospecies (species delineation based solely on morphological characteristics,
typically identified to genus level with a unique ID differentiating from other species of the same
genus, as determined by the original data collector), they were included in the species richness
estimate as an additional species. Records that were not identified to species level, or identified
as a morphospecies, were excluded. Secondly, typically only adult specimens of many
earthworm species can be identified to species level (43), so juveniles were excluded from the
calculation. Therefore, a more appropriate term would be ‘number of identified adult (morpho-)
species’, but for brevity this will be referred to as ‘species richness’. Species richness was not
calculated per unit area (i.e., density), as within each study the sampled area was consistent.
Thus, due to the modelling framework, issues of diversity increasing with sampled area were
accounted for.
Species identity
For datasets where the earthworms had been identified to species level, all species names
were checked for spelling errors and synonyms. Scientific names were standardised using expert
opinion (MJIB, GB, MLCB) and DriloBASE ( Following
standardisation, earthworm species were categorised into the three main ecological groups:
epigeics, endogeics, and anecics (45), plus a fourth minor group, epi-endogeic (species which
exhibit traits of both epigeics and endogeics). Earthworms provide a variety of ecosystem
functions, for example, increasing crop yield by enhancing decomposition and nutrient
minerialization rates (12), but each ecological group contributes in different ways, often on the
basis of their feeding or habitat preferences (45). Epigeic species are typically found in the upper
layers of the soil and litter, and, amongst other roles, are important in the first stages of
decomposition through burial of the litter layer (11, 46, 47). Endogeic species live in the mineral
soil layers, creating horizontal burrows (45). One function they have been shown to provide is a
decrease in the density of root-pathogenic nematodes (48, 49), reducing nematode populations
and disease incidence, which can contribute to increased crop yields (50, 51). Anecic species mix
the litter and mineral soil via surface cast production (45, 46). In addition, the vertical burrows
created by anecic species increase water infiltration into deeper soil layers, increasing water
holding capacity (52, 53), and regulating water availability.
Data extraction and harmonisation across global layers
In order to predict earthworm communities across the globe, we required harmonised sets of
spatially distributed variables. We collected 15 globally distributed layers that are described as
predictors of earthworm distribution (Table S2). For the SoilGrids data (54;;
modelled global layers of soil properties based on soil profiles and remotely-sensed products),
which provides soil properties for different layers within the soil profile, we calculated the
weighted average for the values of the top four layers (corresponding to the top 30 cm of the soil
profile, which matches the soil depth of the earthworm sampling techniques). For sites missing
one or more sampled soil properties, the soil properties associated with the 1km pixel
corresponding to the site’s geographical coordinates were used in the analyses. For CEC, for all
sites, values were taken from SoilGrids.
Where possible, the land cover global layer (ESA CCI-LC 300 m; https://www.esa- was re-categorised to amalgamate similar habitat cover categories matching
the ones collected within the dataset (see Table S1). Where not possible, the categories were
ignored (i.e., classified as NA) during later steps, as estimates could not be produced for
unknown habitat cover categories.
No climate variables were taken from the papers or raw data provided, as there was little
consistency in climate variables across the papers. Instead, five global climate layers
(climatologies) obtained from the CHELSA climate dataset (55) were used (annual mean
temperature, temperature seasonality, temperature annual range, annual precipitation, and
precipitation seasonality) and, from other sources, the number of months of snow cover (56), and
the aridity index and potential evapotranspiration (PET; 57, 58). The within-year standard
deviation of PET (PETSD) was calculated as well. Finally, a globally distributed layer of
elevation (59) was also included in the analysis. For all of these layers, the value within the 1 km
pixel that matched the site’s coordinates was used in the analyses.
For an initial harmonisation across all global layers, it was necessary to aggregate or
disaggregate - when appropriate - the spatial resolution of the different layers to match a one-
kilometre square grid. A nearest neighbour disaggregation algorithm was applied without
changing the pixel values, but changing the pixel resolution using the one-kilometre square
resolution from SoilGrids as a reference.
Following the spatial harmonisation, the global layers were matched with the collated
dataset based on the geographic coordinates of the sampled sites. In the case of the climate
layers, all variables were appended to the dataset. Soil variables were only appended if the sites
were missing sampled measures, with all studies lacking at least one soil property.
To help prevent extrapolation, all global layers were truncated to values represented by each
subset of data, i.e., the minimum and maximum values used in each of the three community
metric models. The exception was the number of months of snow cover, which was truncated at
four months, thus any sites or areas of the globe with a greater number of months than four were
modelled and predicted (respectively) as four months. This ensured an even spread across the
range of values (many sites were within 0-4 months, only 9% of sites were greater than four).
Mixed effects modelling
Earthworm species richness, abundance, and biomass models
Three (generalised) linear mixed effects models were constructed, using lme4 (60), one for
each of the site-level community metrics: species richness, total abundance (individuals per m2),
and total biomass (grams per m2). Prior to modelling, the full dataset was split into three subsets,
based on the response variables (i.e., a dataset containing all sites with a species richness value).
Within each dataset, we tested for multicollinearity between the elevation, climate, and soil
variables using Variance Inflation Factors (VIFs) and removing the variable with highest VIF in
turn until all remaining variables were below the predetermined threshold of 3 (61).
Abundance and biomass were log transformed (log(x + 1)) prior to modelling and were then
modelled using a Gaussian error structure. Species richness was not log transformed, but instead
modelled with a Poisson error structure. All models had random effects that accounted for
variation between each of the different studies, using an intercept only structure. Fixed effects
included habitat cover, elevation, soil properties, and climate variables. All continuous variables
(i.e., elevation, all soil variables, and most climate variables) were centred and scaled (variables
were centred on the mean value and divided by the standard deviation) to aid model fitting and
interpretability. Number of months of snow cover was modelled as a categorical variable (and
therefore not centred and scaled) to allow for a non-linear relationship. As it is expected that
earthworm diversity will peak with some snow cover, due to increased precipitation, and soil
protection during freezing months (62), but be restricted by prolonged snow cover (63). This also
improved the modelling process, as sites were skewed towards the lower number of months, with
not enough data at the higher latitudes to fit a non-linear regression.
For each of the three models, the structure of the fixed effects in the maximal model was the
same. Habitat cover and elevation were included as additive effects with no interactions. The
other variables were grouped into four themes: ‘soil’, ‘precipitation’, ‘temperature’ and ‘water
retention’ (Table S3). For example, all precipitation variables that remained (i.e., were not
removed due to their VIF score) were grouped together. Within the soil and two climate groups,
all two-way interactions were considered. The water retention group contained specific two-way
interactions between soil structure variables (clay and silt percentage) and climate variables
relating to water availability that were present in the two climate themes (annual precipitation,
precipitation seasonality, PET, PETSD, and aridity). These specific interactions were to account
for soil moisture and how quickly moisture might leave the soil.
Each maximal model was then simplified using Akaike information criterion (AIC) values.
All interactions were tested first, and removed if AIC values were reduced compared to the more
complicated model. Any main effects that were not involved with interactions were tested, and
removed if AIC values were reduced (44, 64) (Table S3).
Ecological group responses
The same modelling approach was used to investigate changes in earthworm ecological
groups across the different habitat types. For each site, the diversity, abundance and biomass of
the three main ecological groups (epigeic, endogeic, and anecic) and one minor ecological group
(epi-endogeic) were calculated, based on the category assigned following species name
standardisation. Three (generalised) linear mixed effects models with diversity, abundance and
biomass as response variables were constructed as detailed above, with the exception that habitat
cover interacted with the ecological group (i.e., the biomass of epigeics, endogeics, and anecics
at each site). The model was simplified following details given above.
The community metrics of each ecological group in each habitat cover was then predicted,
using the ‘predict’ function in ‘lme4’ (when all other variables were at zero, i.e., the mean). The
predicted values for the three main ecological groups (epigeic, endogeic and anecic, which had
sufficient underlying data. Epi-endogeics were modelled but did not have enough underlying
data for robust predictions) were plotted using the ‘triangle.plot’ function in ‘ade4’ (65). The
predicted total biomass, i.e., the total of the predicted biomass of the three main ecological
groups, was used to determine size of the points within the triangle plot.
Creating maps of earthworm communities
The global patterns of earthworm communities (species richness, abundance, and biomass)
were predicted using each of the three models. The values from the relevant global layers (i.e.,
those corresponding to the variables that remained in each model following simplification) were
used in the ‘predict’ function in the ‘lme4’ package, being predicted based on the coefficients of
the final models.
A global layer of predicted values was then presented as maps of local communities of
earthworms. Although all global layers had been capped at values represented in the underlying
dataset, extrapolation still occurred during the prediction (there were instances where grid cells
in multiple layers were at the extreme values, and such combinations were not represented in the
underlying data, most evident in the predictions of earthworm biomass, see ‘Interpreting the
model validation’). To prevent outliers skewing the visualization of results, the colour of maps
were curtailed at the extreme low and high values. Curtailing was based on where the majority of
values laid. Thus, values lower or higher than the number marked on the scale are coloured the
same but may represent a large range of values.
Variable Importance
In order to determine which themes (soil, elevation, habitat cover, precipitation,
temperature, water retention) were the most influential in driving earthworm communities,
Variable Importance was performed using random forest models (66, 67).
For each of the three community metrics, random forest models were constructed (67) with
all the variables that were present in the final (i.e. simplified) model. Random forest models use
multiple regression trees to classify data (67). This method was chosen as these models can
handle non-linear data, whilst interactions are not specified but can be learnt from the data (68).
Random forest models are an ensemble of individual regression (or classification) trees (66, 67).
Each tree is created using around two-thirds of the available data, i.e., “out-of-bag” regression,
and the process is repeated until the ‘forest’ is complete (ntree default = 500 trees). At each node
in the tree, the subset of response variables is split using the best predictor variable. Unlike
regression trees, where at each node the best predictor is used from all available predictor
variables, random forest models use only a random sample of the predictor variables (“Mtry”) to
determine the best predictor to split the response variable at each node (66, 67). The default Mtry
value was used (number of predictors divided by 3), so in our case of 10 to 12 predictor variables
Mtry = 3 (biomass model) and 4 (species richness and abundance models). The “out-of-bag” data
is then predicted using the average prediction of all trees (67).
In addition, random forest models can be used to assess the importance of individual
variables (66). One such measure is the mean decrease in node impurity calculated from the
decrease in the residual sum of squares for the variable that was used at the node. The average
decrease for each variable is averaged across all the trees to create the node impurity (67). An
alternative importance variable is the mean decrease in accuracy. For each tree, when the “out-
of-bag” data (~one-third of the data) is being predicted, a single predictor variable is permuted,
and the increase in prediction error calculated (67). This mean decrease in accuracy is often
considered the best of those available (69), but results between the decrease in node impurity and
mean decrease in accuracy often correlate well (70).
For each of the three random forest models, the mean decrease in node impurity and mean
decrease in accuracy was calculated (using the ‘importance’ function in ‘randomForest’) for each
predictor variable in the random forest. In order to determine which theme of variables (habitat
cover, elevation, soil, temperature, precipitation, and water retention) was most important in
driving patterns in earthworm communities, the mean decrease in node impurity was averaged
for all variables within each theme and weighted by the number of times each variable was used
in the random forest compared to the other variables in the same theme.
Model validation and sensitivity analysis
A number of additional analyses were performed to determine the robustness of the models
and the ability to predict new values. Firstly, the influence of combining both measured soil
properties and soil properties from SoilGrids was tested. For the three response variables (species
richness, abundance, and biomass) models were created that only included data from SoilGrids.
The same modelling process was used as described above (using VIFs to determine starting
variables, then simplification). Once the final model had been identified for each of the three
community metrics, 10-fold cross-validation was performed (71).
Cross-validation was performed in two ways. Firstly by randomly splitting the dataset
underlying each of the three models into 10 nearly-equal size groups. Using the model structures
produced following simplification, the model was built using 9 of the groups of data. The 10th
group of data was predicted from the re-built model. The predicted data was plotted against the
observed data. This process was repeated until all 10 groups of data had been predicted. This
process was done for the models that contained only SoilGrids data, and the main models (that
used a mixture of soil property data, Fig. S3). Secondly, by splitting the dataset into 10 near-
equal sized groups based on study. Thus, the site-level community metrics for a 10th of the
studies were predicted by the remaining 9/10 of the data. This process was only performed on the
main models (Fig. S4).
For the site-level cross-validated models, the mean squared error (MSE) was calculated
from the results of the cross-validation. MSE measures the ability of the model to predict new
data, and the result are easily interpretable as they are on the same scale as the original data.
MSE was calculated for the total of all models, but as the models may be better at predicting
certain ranges of values, MSE was also calculated for the tertiles of the observed data (i.e., the
ability of the model to predict the low, medium, and high values of earthworm communities).
Secondly, the R2 values of all models (the main models, and the models with only SoilGrids
data) were calculated using the MuMIn package (72). The R2 values describe the fit of the model
to the data. The R2marginal is the variance explained by the fixed effects, whilst the R2conditional is the
variance explained by the fixed and random effects.
To determine the confidence of the globally predicted values, we followed methods of (9).
For each of the biodiversity models, we investigated how well the underlying data represented
the full multivariate environmental covariate space of the global layers. We performed a
Principal Components based approach on each of the datasets. The centering values, scaling
values, and eigenvectors were then used to transform all global layers into the same PCA spaces.
Then, we created convex hulls for each of the bivariate combinations from the first 6 (total
biomass data) and 7 (richness and total abundance data) i.e., half of the number of variables
within the model, principal components to cover more than 90% of the sample space variation.
Using the coordinates of these convex hulls, we classified whether each pixel of each global
layer falls within or outside each of these convex hulls. Therefore, if all global layers within a
pixel were within the convex hull, the interpolation percentage would be 100%, while if only
half of the layers were within the convex hull, the interpolation percentage would be 50%. This
analysis was performed in Google Earth Engine (73).
Regional latitudinal diversity gradients
In order to ascertain whether there is a species diversity gradient with latitude, the site-level
diversity data underlying the species richness model (i.e., contained sites with species level or
morphospecies identification) was used. The sites were split into latitude zones that contained
roughly equal numbers of sites. Sites were assigned to a zone based on their latitude, with the
intention that each zone would contain close to 250 sites. However, all sites with the same
coordinates were kept within the same band, so the number of sites within a zone did vary (min =
209, max = 341, mean = 267.6). The number of unique species, based on species binomials,
across all sites within each zone was calculated. Within each zone, it was also assumed that each
uniquely named morphospecies was different from any of the named species (number of
morphospecies across zones, min = 0, max = 21, mean = 3.05). Some of the sites also contained
genus-level only identification. When this was the case, a genus was included as one additional
species if the genus was unique within the zone (i.e., no named species belonged to that genus).
As the number of morphospecies was biased with latitude (i.e., greater taxonomic expertise in
the temperate regions, Table S4), the analysis was repeated excluding morphospecies (Fig. S2).
The two methods resulted in similar patterns, but reduced richness in some of the zones in the
All statistics, data manipulation and processing of global data layers was implemented in R
[version 3.3.1; (74)].
Supplementary Text
Interpreting the model validation
The results of the biomass model highlighted an issue with the modelling technique used.
All of the global data layers were cut at values represented by the underlying datasets. However,
during the prediction, it was often the case that multiple data layers were at the extreme ends of
the possible range of values. This led to, especially in the case of the biomass model, and to
some extent the abundance model, unrealistically high values being predicted. This issue could
only be fixed with additional data, but does not affect the visual maps produced in this study. For
the global predictions of biomass, values greater than 2 kg per m2 were deemed to be unrealistic.
This threshold is over 4 times the maximum recorded biomass of earthworm communities (75),
and thus is highly unlikely to be realistic. 98.9% of pixels were less than 400 g per m2 [the
maximum recorded earthworm community biomass recorded in the temperate region (75)].
Overall, the models had reasonably good fit to the data, assessed using the R2 values (Table
1C). However, the predictive power of the models was variable. With all models, the total MSE
(Table 1A) increased mainly due to the ill-fitting of the sites with higher values. It is unclear why
high values cannot be fitted well with the models; however, it is highly likely that increasing the
number of sites would help either identify the issue or improve model fit.
For the majority of the datasets (182 out of 228 studies), the models contained the measured
soil properties for some of the variables. Where this was missing, we used the SoilGrids data.
Models which contained only SoilGrids data had a better fit to the data (Table 1C) and were
typically better at predicting during cross-validation (lower MSE values; Table 1B). However, in
most instances, the change in MSE was negligible between the different types of models (Table 1
A versus B). Despite the models that contained only SoilGrids data performing slightly better in
terms of R2 and MSE, there are other reasons why using a mixture of the measured variables and
the SoilGrids variables is the best option in the modelling process. Firstly, modelled global
estimates of the soil properties may not accurately depict site-level conditions (76), which could
result in the variables appearing less important than they would be if they matched the measured
communities. Secondly, some of the coordinates within a study were identical which would
result in identical SoilGrids data (for these datasets, often small-scale field experiments, the
measured soil properties variables were not identical). Using only SoilGrids data would reduce
the gradient of soil properties within each study, reducing the number of available gradient
comparisons across all datasets. And given that a number of studies (106 out of 228 studies) had
identical climate variables across all sites, having variety in all other variables prevented this
being an issue within the modelling framework. We call on soil ecologists to collect data on soil
properties when they measure diversity of soil taxa, as this permits more robust modelling at
both the small scale, and across larger scales.
Regardless of whether the model contained measured soil properties or only SoilGrids data,
the models were consistently worse at predicting when observed values were high (Table 1).
This is likely due to the small number of studies where sampled values were high. Only 5 studies
had more than 10 species of earthworms in at least one site, and only 6 studies had more than
300 grams per m2 of earthworm biomass in at least one site. There were a greater number of
studies that contained high abundance of earthworms, with 34 studies having at least one site that
contained more than 600 individuals per m2. Increasing the number of studies and sites would
help identify whether this, or another cause, is responsible. Ideally, this would improve the
predictive power of the models. It is hoped that efforts will continue to collate earthworm
diversity data from across the globe.
When cross-validation was performed at the study level (Fig. S4) the predictions were not
scattered around the 1:1 line. However, this is to be expected, as when sites are randomly
selected and predicted, the study level random-effect is most likely still present in the model.
This ensures that the community metrics of each sites can be predicted using the variance from
the study it is within. When an entire study is removed, and so no random-effect level exists for
it in the model, all study-level random effects are averaged in order to produce the prediction.
Thus, the prediction error is increased, and more concentrated around the overall mean.
For the species richness (Fig. S1A) and total abundance data (Fig. S1B), the interpolation
percentage across the globe was relatively high (i.e., the underlying datasets adequately captured
the majority of the multivariate environmental conditions). Regions surrounding the Eurasian
Steppe, and the Himalayas were some of the most extrapolated regions, with arid regions in
Africa and boreal regions also having lower interpolation percentages. For the total biomass data,
more regions of the globe had low interpolation percentages (Fig. S1C). These low-value regions
were spread across the tropics, particularly Brazil and Indonesia, and large parts of Africa, the
sub-tropics, such as India, and temperate regions, including northern China and Russia. Overall,
we would expect the globally predicted values of the biomass model to be more extrapolated,
than the diversity and total abundance models.
Fig. S1.
Assessment of global extrapolation and interpolation for the (A) species richness data, (B) total
abundance data, and (C) total biomass data. Scale shows the percentage of pixels (from each of
the global layers) falling within the convex hull spaces of the first 6 (biomass) and 7 (richness,
abundance) Principal Components collectively explaining >90% of the variation. Low
interpolation percentage values (in blue) indicate that few global layers were represented by data,
thus extrapolation would have occurred during prediction, whilst high interpolation percentages
(in yellow) indicated that many or all global layers were represented by data, thus interpolation
would have occurred during prediction.
Fig. S2.
The number of unique species within each latitudinal zone, when the number of sites within each
zone was kept relatively equal. The height of the bar indicates the number of unique species
across all sites. The width of the bar shows the latitude range the sites cover. Within each zone
only the species with binomials, or genera with no other identified species, were included in the
calculations (morphospecies were excluded).
Fig. S3.
10-fold cross validation of the three main community metric models, (A) species richness, (B)
ln-abundance, and (C) ln-biomass. X-axis shows the observed value, and Y-axis the predicted
value, black line is the 1:1 line. The underlying dataset of each model was randomly split into 10
nearly-equal size groups. Using the model structures produced following simplification, the
model was built using 9 of the groups of data. The 10th group of data was predicted from the re-
built model. This process was repeated until all 10 groups of data had been predicted. The
predicted data was plotted against the observed data.
Fig. S4.
10-fold cross validation of the three main community metric models, (A) species richness, (B)
ln-abundance, and (C) ln-biomass. X-axis shows the observed value, and Y-axis the predicted
value, black line is the 1:1 line. The underlying dataset of each model was randomly split into 10
nearly-equal size groups, so that each group contained all the data of a tenth of the studies. Using
the model structures produced following simplification, the model was built using 9 of the
groups of studies. The 10th group of studies was predicted from the re-built model. This process
was repeated until all 10 groups of studies had been predicted. The predicted data was plotted
against the observed data.
Fig. S5.
Changes in (A) species richness and (B) ln-abundance across the different habitat cover
categories (+/- SD). Values of species richness and abundance are predicted from the main
models when all other variables are at zero, i.e., the mean. Not all habitat cover categories had
sampled estimates (i.e., species richness could not be estimated for ‘Cropland/Other vegetation
Fig. S6.
The (A) total abundance and (B) total biomass of the three ecological groups (epigeic, endogeic
and anecic earthworms) within each habitat cover category based on modelled estimates. Circle
size is relative to the total biomass predicted for the habitat cover, and circle colour indicates the
habitat cover. Position within the three axes indicates the proportion of each of the three
ecological groups within the community, based on the interaction term between habitat cover and
ecological group. During simplification, the interaction term between habitat cover and
ecological group was removed in the species richness model, thus those results are not shown.
Not all habitat cover categories had sampled estimates (i.e., biomass could not be estimated for
‘Broadleaf evergreen forest’ or ‘Cropland/Other vegetation mosaic’). This figure shows, for
example, that “Broadleaf deciduous forests” have a rather even predicted biomass distribution
across the three ecological groups (but low total biomass), while “Production sites” (“Plantation”
and “Herbaceous”) have high total earthworm abundance, but are dominated by endogeic
Table S1.
Original habitat cover variable
Reclassified habitat cover
No Data
Cropland, rainfed
Production - Herbaceous
Cropland - herbaceous cover
Production - Herbaceous
Cropland - Tree or shrub cover
Production - Plantation
Cropland, irrigated or post-flooding
Mosaic cropland (>50%) / natural vegetation (tree, shrub, herbaceous cover)
Cropland/Other vegetation
Mosaic natural vegetation (tree, shrub, herbaceous cover) (>50%) / cropland
Cropland/Other vegetation
Tree cover, broadleaved, evergreen, closed to open (>15%)
Broadleaf evergreen forest
Tree cover, broadleaved, deciduous, closed (>40%)
Broadleaf deciduous forest
Tree cover, broadleaved, deciduous, open (15-40%)
Broadleaf deciduous forest
Tree cover, needleleaved, evergreen, closed (>40%)
Needleleaf evergreen forest
Tree cover, needleleaved, evergreen, open (15-40%)
Needleleaf evergreen forest
Tree cover, needleleaved, deciduous, closed (>40%)
Needleleaf deciduous forest
Tree cover, needleleaved, deciduous, open (15-40%)
Needleleaf deciduous forest
Tree cover, mixed leaf type (broadleaved and needleleaved)
Mixed forest
Mosaic tree and shrub (>50%) / herbaceous cover (<50%)
Tree open
Mosaic herbaceous cover (>50%) / tree and shrub (<50%)
Herbaceous with spare tree/shrub
Lichens and mosses
Sparse vegetation (tree, shrub, herbaceous cover) (<15%)
Sparse vegetation
Tree cover, flooded, fresh or brackish water
Tree cover, flooded, saline water
Shrub or herbaceous cover, flooded, fresh/saline/brackish water
Urban areas
Bare areas - consolidated
Bare area (consolidated)
Bare areas - unconsolidated
Bare area (unconsolidated)
Water bodies
Water bodies
Permanent snow and ice
The re-categorisation of the ESA habitat cover variable. Habitat cover at a sampled site was
classified based on the ‘Reclassified habitat cover’ column. As not all categories of habitat were
available in the data (i.e., due to too detailed categories, or in habitats typically devoid of
sampling), some of the categories of the original habitat cover variable (left-hand column) were
reclassified (right-hand column). Usually, this meant that categories were grouped together (i.e.,
to reduce the categories based on ‘openness’).
Table S2.
Original Spatial Resolution
Habitat Cover
300 m
1 km
Soil Parameters
pH (H20)
SoilGrids (54)
1 km
Organic carbon
1 km
Soil clay content
1 km
Soil silt content
1 km
1 km
Annual Mean Temperature
1 km
Temp. seasonality
1 km
Temp. annual range
1 km
(57, 58)
1 km
1 km
Annual precipitation
1 km
Precip. seasonality
1 km
Number of Months with
1 km
Aridity Index
(57, 58)
1 km
Information for each of the 15 global layers detailed in the methods. Abbreviations: CEC =
Cation exchange capacity, Temp. = Temperature, Precip. = Precipitation, PET = Potential
evapotranspiration, PETSD = within year standard deviation of PET
Table S3.
Results following model simplification of the three community metric models. ‘Main Effect
Only’ column shows the slope for the main effect of each variable in the final species richness
(turquoise), total abundance (green) and total biomass (yellow) models. ‘+’ indicates the slope
was positive, ‘-’ indicates a negative slope, and ‘*’ indicates that the variable was categorical
(with intercepts and slopes depending on the category). The remaining columns show the
interactions between the variables. An upwards arrow indicates that the slope of one variable
would become more positive as the other variable is increased. A downwards arrows indicates
that the slope of the one variable would become more negative as the other variable is increased.
However, it may not necessarily indicate that the slope changes direction. Black symbols
indicate that the coefficient was significant (p < 0.05) within the model, and grey/hatched
symbols indicate they were not significant [NB. P-values are for illustrative purposes only, as
models were simplified based on AIC values]. Habitat cover and elevation were only in the
models as main effects. Also noted is the variable theme in which the variable was grouped.
Variables that interacted within the ‘water retention’ theme are not shown explicitly, but can be
deduced based on interactions between a climate variable and soil property variable.
Abbreviations: CEC = Cation exchange capacity, Temp. = Temperature, Precip. = Precipitation,
PET = Potential evapotranspiration, PETSD = within year standard deviation of PET.
Table S4.
Number of
Number of
named species
Mean Latitude
Number of
% Native
% Non-native
% Unknown
Details of the number of sites, and composition of the earthworm species for each 5 degree
latitudinal band. Based on the geographical coordinates, each site was classified into 5 degree
latitudinal bands. The number of species is based on the binomial names given by the original
data collector, then revised for consistency. For each named species in each band the latitudinal
range (the difference between the minimum latitude and the maximum latitude, based on all sites
within the dataset that the species occurred) was calculated, and the average taken from all
species within the band. Morphospecies are individuals that were identified to genus level, and
identified by the data collectors as morphologically distinct from other (morpho-) species, but
were not identified to species-level. The percentage of native and non-native species is based on
information provided by the original data collectors, and is therefore often incomplete (depicted
in the ‘% unknown’ column).
References and Notes
1. R. D. Bardgett, W. H. van der Putten, Belowground biodiversity and ecosystem functioning.
Nature 515, 505–511 (2014). doi:10.1038/nature13855 Medline
2. N. Eisenhauer, P. M. Antunes, A. E. Bennett, K. Birkhofer, A. Bissett, M. A. Bowker, T.
Caruso, B. Chen, D. C. Coleman, W. de Boer, P. de Ruiter, T. H. DeLuca, F. Frati, B. S.
Griffiths, M. M. Hart, S. Hättenschwiler, J. Haimi, M. Heethoff, N. Kaneko, L. C. Kelly,
H. P. Leinaas, Z. Lindo, C. Macdonald, M. C. Rillig, L. Ruess, S. Scheu, O. Schmidt, T.
R. Seastedt, N. M. van Straalen, A. V. Tiunov, M. Zimmer, J. R. Powell, Priorities for
research in soil ecology. Pedobiologia 63, 1–7 (2017). doi:10.1016/j.pedobi.2017.05.003
3. L. Tedersoo, M. Bahram, S. Põlme, U. Kõljalg, N. S. Yorou, R. Wijesundera, L. Villarreal
Ruiz, A. M. Vasco-Palacios, P. Q. Thu, A. Suija, M. E. Smith, C. Sharp, E. Saluveer, A.
Saitta, M. Rosas, T. Riit, D. Ratkowsky, K. Pritsch, K. Põldmaa, M. Piepenbring, C.
Phosri, M. Peterson, K. Parts, K. Pärtel, E. Otsing, E. Nouhra, A. L. Njouonkou, R. H.
Nilsson, L. N. Morgado, J. Mayor, T. W. May, L. Majuakim, D. J. Lodge, S. S. Lee, K.-
H. Larsson, P. Kohout, K. Hosaka, I. Hiiesalu, T. W. Henkel, H. Harend, L. D. Guo, A.
Greslebin, G. Grelet, J. Geml, G. Gates, W. Dunstan, C. Dunk, R. Drenkhan, J.
Dearnaley, A. De Kesel, T. Dang, X. Chen, F. Buegger, F. Q. Brearley, G. Bonito, S.
Anslan, S. Abell, K. Abarenkov, Global diversity and geography of soil fungi. Science
346, 1256688 (2014). doi:10.1126/science.1256688 Medline
4. M. Delgado-Baquerizo, A. M. Oliverio, T. E. Brewer, A. Benavent-González, D. J. Eldridge,
R. D. Bardgett, F. T. Maestre, B. K. Singh, N. Fierer, A global atlas of the dominant
bacteria found in soil. Science 359, 320–325 (2018). doi:10.1126/science.aap9516
5. M. Bahram, F. Hildebrand, S. K. Forslund, J. L. Anderson, N. A. Soudzilovskaia, P. M.
Bodegom, J. Bengtsson-Palme, S. Anslan, L. P. Coelho, H. Harend, J. Huerta-Cepas, M.
H. Medema, M. R. Maltz, S. Mundra, P. A. Olsson, M. Pent, S. Põlme, S. Sunagawa, M.
Ryberg, L. Tedersoo, P. Bork, Structure and function of the global topsoil microbiome.
Nature 560, 233–237 (2018). doi:10.1038/s41586-018-0386-6 Medline
6. H. Hillebrand, On the generality of the latitudinal diversity gradient. Am. Nat. 163, 192–211
(2004). doi:10.1086/381004 Medline
7. E. K. Cameron, I. S. Martins, P. Lavelle, J. Mathieu, L. Tedersoo, M. Bahram, F. Gottschall,
C. A. Guerra, J. Hines, G. Patoine, J. Siebert, M. Winter, S. Cesarz, O. Ferlian, H. Kreft,
T. E. Lovejoy, L. Montanarella, A. Orgiazzi, H. M. Pereira, H. R. P. Phillips, J. Settele,
D. H. Wall, N. Eisenhauer, Global mismatches in aboveground and belowground
biodiversity. Conserv. Biol. 33, 1187–1192 (2019). doi:10.1111/cobi.13311 Medline
8. N. Fierer, M. S. Strickland, D. Liptzin, M. A. Bradford, C. C. Cleveland, Global patterns in
belowground communities. Ecol. Lett. 12, 1238–1249 (2009). doi:10.1111/j.1461-
0248.2009.01360.x Medline
9. J. van den Hoogen, S. Geisen, D. Routh, H. Ferris, W. Traunspurger, D. A. Wardle, R. G. M.
de Goede, B. J. Adams, W. Ahmad, W. S. Andriuzzi, R. D. Bardgett, M. Bonkowski, R.
Campos-Herrera, J. E. Cares, T. Caruso, L. de Brito Caixeta, X. Chen, S. R. Costa, R.
Creamer, J. Mauro da Cunha Castro, M. Dam, D. Djigal, M. Escuer, B. S. Griffiths, C.
Gutiérrez, K. Hohberg, D. Kalinkina, P. Kardol, A. Kergunteuil, G. Korthals, V.
Krashevska, A. A. Kudrin, Q. Li, W. Liang, M. Magilton, M. Marais, J. A. R. Martín, E.
Matveeva, E. H. Mayad, C. Mulder, P. Mullin, R. Neilson, T. A. D. Nguyen, U. N.
Nielsen, H. Okada, J. E. P. Rius, K. Pan, V. Peneva, L. Pellissier, J. Carlos Pereira da
Silva, C. Pitteloud, T. O. Powers, K. Powers, C. W. Quist, S. Rasmann, S. S. Moreno, S.
Scheu, H. Setälä, A. Sushchuk, A. V. Tiunov, J. Trap, W. van der Putten, M. Vestergård,
C. Villenave, L. Waeyenberge, D. H. Wall, R. Wilschut, D. G. Wright, J. I. Yang, T. W.
Crowther, Soil nematode abundance and functional group composition at a global scale.
Nature 572, 194–198 (2019). doi:10.1038/s41586-019-1418-6 Medline
10. T. Decaëns, Macroecological patterns in soil communities. Glob. Ecol. Biogeogr. 19, 287–
302 (2010). doi:10.1111/j.1466-8238.2009.00517.x
11. C. A. Edwards, Ed., Earthworm Ecology (CRC Press, ed. 2, 2004).
12. M. Blouin, M. E. Hodson, E. A. Delgado, G. Baker, L. Brussaard, K. R. Butt, J. Dai, L.
Dendooven, G. Peres, J. E. Tondoh, D. Cluzeau, J. J. Brun, A review of earthworm
impact on soil function and ecosystem services. Eur. J. Soil Sci. 64, 161–182 (2013).
13. D. Craven, M. P. Thakur, E. K. Cameron, L. E. Frelich, R. Beauséjour, R. B. Blair, B.
Blossey, J. Burtis, A. Choi, A. Dávalos, T. J. Fahey, N. A. Fisichelli, K. Gibson, I. T.
Handa, K. Hopfensperger, S. R. Loss, V. Nuzzo, J. C. Maerz, T. Sackett, B. C.
Scharenbroch, S. M. Smith, M. Vellend, L. G. Umek, N. Eisenhauer, The unseen
invaders: Introduced earthworms as drivers of change in plant communities in North
American forests (a meta-analysis). Glob. Change Biol. 23, 1065–1074 (2017).
14. See supplementary materials.
15. M. Rutgers, A. Orgiazzi, C. Gardi, J. Römbke, S. Jänsch, A. M. Keith, R. Neilson, B. Boag,
O. Schmidt, A. K. Murchie, R. P. Blackshaw, G. Pérès, D. Cluzeau, M. Guernion, M. J. I.
Briones, J. Rodeiro, R. Piñeiro, D. J. Diaz Cosín, J. P. Sousa, M. Suhadolc, I. Kos, P. H.
Krogh, J. H. Faber, C. Mulder, J. J. Bogte, H. J. van Wijnen, A. J. Schouten, D. de Zwart,
Mapping earthworm communities in Europe. Appl. Soil Ecol. 97, 98–111 (2016).
16. P. F. Hendrix, P. J. Bohlen, Exotic earthworm invasions in North America: Ecological and
policy implications. Bioscience 52, 801–811 (2002). doi:10.1641/0006-
17. T. G. Piearce, The calcium relations of selected lumbricidae. J. Anim. Ecol. 41, 167 (1972).
18. D. J. Spurgeon, A. M. Keith, O. Schmidt, D. R. Lammertsma, J. H. Faber, Land-use and
land-management change: Relationships with earthworm and fungi communities and soil
structural properties. BMC Ecol. 13, 46 (2013). doi:10.1186/1472-6785-13-46 Medline
19. J. Mathieu, T. J. Davies, Glaciation as an historical filter of below-ground biodiversity. J.
Biogeogr. 41, 1204–1214 (2014). doi:10.1111/jbi.12284
20. P. Lavelle, C. Lattaud, D. Trigo, I. Barois, Mutualism and biodiversity in soils. Plant Soil
170, 23–33 (1995). doi:10.1007/BF02183052
21. R. R. Dunn, D. Agosti, A. N. Andersen, X. Arnan, C. A. Bruhl, X. Cerdá, A. M. Ellison, B.
L. Fisher, M. C. Fitzpatrick, H. Gibb, N. J. Gotelli, A. D. Gove, B. Guenard, M. Janda,
M. Kaspari, E. J. Laurent, J.-P. Lessard, J. T. Longino, J. D. Majer, S. B. Menke, T. P.
McGlynn, C. L. Parr, S. M. Philpott, M. Pfeiffer, J. Retana, A. V. Suarez, H. L.
Vasconcelos, M. D. Weiser, N. J. Sanders, Climatic drivers of hemispheric asymmetry in
global patterns of ant species richness. Ecol. Lett. 12, 324–333 (2009).
doi:10.1111/j.1461-0248.2009.01291.x Medline
22. H. Kreft, W. Jetz, Global patterns and determinants of vascular plant diversity. Proc. Natl.
Acad. Sci. U.S.A. 104, 5925–5930 (2007). doi:10.1073/pnas.0608361104 Medline
23. C. Fragoso, P. Lavelle, Earthworm communities of tropical rain forests. Soil Biol. Biochem.
24, 1397–1408 (1992). doi:10.1016/0038-0717(92)90124-G
24. K. J. Gaston, T. M. Blackburn, Pattern and Process in Macroecology (Blackwell, 2000).
25. J. Davison, M. Moora, M. Öpik, A. Adholeya, L. Ainsaar, A. , S. Burla, A. G. Diedhiou, I.
Hiiesalu, T. Jairus, N. C. Johnson, A. Kane, K. Koorem, M. Kochar, C. Ndiaye, M.
Pärtel, Ü. Reier, Ü. Saks, R. Singh, M. Vasar, M. Zobel, Global assessment of arbuscular
mycorrhizal fungus diversity reveals very low endemism. Science 349, 970–973 (2015).
doi:10.1126/science.aab1161 Medline
26. M. Maraun, H. Schatz, S. Scheu, Awesome or ordinary? Global diversity patterns of oribatid
mites. Ecography 30, 209–216 (2007). doi:10.1111/j.0906-7590.2007.04994.x
27. T. Decaëns, D. Porco, S. W. James, G. G. Brown, V. Chassany, F. Dubs, L. Dupont, E.
Lapied, R. Rougerie, J. Rossi, V. Roy, DNA barcoding reveals diversity patterns of
earthworm communities in remote tropical forests of French Guiana. Soil Biol. Biochem.
92, 171–183 (2016). doi:10.1016/j.soilbio.2015.10.009
28. D. C. Coleman, D. A. Crossley, P. F. Hendrix, Fundamentals of Soil Ecology (Elsevier, ed. 2,
2004), pp. 271–298.
29. J. W. Spaak, J. M. Baert, D. J. Baird, N. Eisenhauer, L. Maltby, F. Pomati, V. Radchuk, J. R.
Rohr, P. J. Van den Brink, F. De Laender, Shifts of community composition and
population density substantially affect ecosystem function despite invariant richness.
Ecol. Lett. 20, 1315–1324 (2017). doi:10.1111/ele.12828 Medline
30. N. Eisenhauer, J. Schlaghamerský, P. B. Reich, L. E. Frelich, The wave towards a new
steady state: Effects of earthworm invasion on soil microbial functions. Biol. Invasions
13, 2191–2196 (2011). doi:10.1007/s10530-011-0053-4
31. M. A. Drumond, A. Q. Guimarães, H. R. El Bizri, L. C. Giovanetti, D. G. Sepúlveda, R. P.
Martins, Life history, distribution and abundance of the giant earthworm Rhinodrilus
alatus RIGHI 1971: Conservation and management implications. Braz. J. Biol. 73, 699–
708 (2013). doi:10.1590/S1519-69842013000400004 Medline
32. L. Santini, N. J. B. Isaac, L. Maiorano, G. F. Ficetola, M. A. J. Huijbregts, C. Carbone, W.
Thuiller, Global drivers of population density in terrestrial vertebrates. Glob. Ecol.
Biogeogr. 27, 968–979 (2018). doi:10.1111/geb.12758
33. D. Song, K. Pan, A. Tariq, F. Sun, Z. Li, X. Sun, L. Zhang, O. A. Olusanya, X. Wu, Large-
scale patterns of distribution and diversity of terrestrial nematodes. Appl. Soil Ecol. 114,
161–169 (2017). doi:10.1016/j.apsoil.2017.02.013
34. Intergovernmental Panel on Climate Change, Climate Change 2014 Synthesis Report
Summary for Policymakers (2014);
35. D. K. Hackenberger, B. K. Hackenberger, Earthworm community structure in grassland
habitats differentiated by climate type during two consecutive seasons. Eur. J. Soil Biol.
61, 27–34 (2014). doi:10.1016/j.ejsobi.2014.01.001
36. M. A. Bradford, G. F. Veen, A. Bonis, E. M. Bradford, A. T. Classen, J. H. C. Cornelissen,
T. W. Crowther, J. R. De Long, G. T. Freschet, P. Kardol, M. Manrubia-Freixa, D. S.
Maynard, G. S. Newman, R. S. P. Logtestijn, M. Viketoft, D. A. Wardle, W. R. Wieder,
S. A. Wood, W. H. van der Putten, A test of the hierarchical model of litter
decomposition. Nat. Ecol. Evol. 1, 1836–1845 (2017). doi:10.1038/s41559-017-0367-4
37. A. Rice, P. Šmarda, M. Novosolov, M. Drori, L. Glick, N. Sabath, S. Meiri, J. Belmaker, I.
Mayrose, The global biogeography of polyploid plants. Nat. Ecol. Evol. 3, 265–273
(2019). doi:10.1038/s41559-018-0787-9 Medline
38. A. Shade, R. R. Dunn, S. A. Blowes, P. Keil, B. J. M. Bohannan, M. Herrmann, K. Küsel, J.
T. Lennon, N. J. Sanders, D. Storch, J. Chase, Macroecology to unite all life, large and
small. Trends Ecol. Evol. 33, 731–744 (2018). doi:10.1016/j.tree.2018.08.005 Medline
39. J. M. Anderson, J. S. I. Ingram, Eds., Tropical Soil Biology and Fertility: A Handbook of
Methods (Cambridge Univ. Press, ed. 2, 1993), pp. 88–91.
40. ISO, “Soil quality - Sampling of soil invertebrates - Part 1: Hand-sorting and extraction of
earthworms” (ISO 23611-1:2018);
41. J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch, S. Preibisch,
C. Rueden, S. Saalfeld, B. Schmid, J.-Y. Tinevez, D. J. White, V. Hartenstein, K. Eliceiri,
P. Tomancak, A. Cardona, Fiji: An open-source platform for biological-image analysis.
Nat. Methods 9, 676–682 (2012). doi:10.1038/nmeth.2019 Medline
42. J. Koricheva, J. Gurevitch, K. Mengersen, Eds., Handbook of Meta-Analysis in Ecology and
Evolution (Princeton Univ. Press, 2013).
43. M. D. Bartlett, M. J. I. Briones, R. Neilson, O. Schmidt, D. Spurgeon, R. E. Creamer, A
critical review of current methods in earthworm ecology: From individuals to
populations. Eur. J. Soil Biol. 46, 67–73 (2010). doi:10.1016/j.ejsobi.2009.11.006
44. M. J. Crawley, The R Book (Wiley, 2012).
45. M. B. Bouché, Strategies lombriciennes. Ecol. Bull. 25, 122–132 (1977).
46. G. G. Brown, How do earthworms affect microfloral and faunal community diversity? Plant
Soil 170, 209–231 (1995). doi:10.1007/BF02183068
47. J. Seeber, G. U. H. Seeber, R. Langel, S. Scheu, E. Meyer, The effect of macro-invertebrates
and plant litter of different quality on the release of N from litter to plant on alpine
pastureland. Biol. Fertil. Soils 44, 783–790 (2008). doi:10.1007/s00374-008-0282-6
48. M. Blouin, Y. Zuily-Fodil, A. T. Pham-Thi, D. Laffray, G. Reversat, A. Pando, J. Tondoh, P.
Lavelle, Belowground organism activities affect plant aboveground phenotype, inducing
plant tolerance to parasites. Ecol. Lett. 8, 202–208 (2005). doi:10.1111/j.1461-
49. J. Boyer, G. Reversat, P. Lavelle, A. Chabanne, Interactions between earthworms and plant-
parasitic nematodes. Eur. J. Soil Biol. 59, 43–47 (2013). doi:10.1016/j.ejsobi.2013.10.004
50. G. Loranger-Merciris, Y.-M. Cabidoche, B. Deloné, P. Quénéhervé, H. Ozier-Lafontaine,
How earthworm activities affect banana plant response to nematodes parasitism. Appl.
Soil Ecol. 52, 1–8 (2012). doi:10.1016/j.apsoil.2011.10.003
51. G. G. Brown, C. A. Edwards, L. Brussaard, in Earthworm Ecology, C. A. Edwards, Ed.
(CRC Press, ed. 2, 2004), pp. 13–49.
52. M. B. Bouché, F. Al-Addan, Earthworms, water infiltration and soil stability: Some new
assessments. Soil Biol. Biochem. 29, 441–452 (1997). doi:10.1016/S0038-
53. M. Joschko, H. Diestel, O. Larink, Assessment of earthworm burrowing efficiency in
compacted soil with a combination of morphological and soil physical measurements.
Biol. Fertil. Soils 8, 191–196 (1989). doi:10.1007/BF00266478
54. T. Hengl, J. Mendes de Jesus, G. B. M. Heuvelink, M. Ruiperez Gonzalez, M. Kilibarda, A.
Blagotić, W. Shangguan, M. N. Wright, X. Geng, B. Bauer-Marschallinger, M. A.
Guevara, R. Vargas, R. A. MacMillan, N. H. Batjes, J. G. B. Leenaars, E. Ribeiro, I.
Wheeler, S. Mantel, B. Kempen, SoilGrids250m: Global gridded soil information based
on machine learning. PLOS ONE 12, e0169748 (2017).
doi:10.1371/journal.pone.0169748 Medline
55. D. N. Karger, O. Conrad, J. Böhner, T. Kawohl, H. Kreft, R. W. Soria-Auza, N. E.
Zimmermann, H. P. Linder, M. Kessler, Climatologies at high resolution for the Earth’s
land surface areas. Sci. Data 4, 170122 (2017). doi:10.1038/sdata.2017.122 Medline
56. D. K. Hall, G. A. Riggs, MODIS/Terra Snow Cover Monthly L3 Global 0.05Deg CMG,
Version 6 (NASA National Snow and Ice Data Center Distributed Active Archive Center,
2015). doi.10.5067/MODIS/MOD10CM.006
57. R. J. Zomer, A. Trabucco, D. A. Bossio, L. V. Verchot, Climate change mitigation: A spatial
analysis of global land suitability for clean development mechanism afforestation and
reforestation. Agric. Ecosyst. Environ. 126, 67–80 (2008).
58. R. J. Zomer, D. A. Bossio, A. Trabucco, L. Yuanjie, D. C. Gupta, V. P. Singh, Trees and
Water: Smallholder Agroforestry on Irrigated Lands in Northern India. IWMI Res. Rep.
122, 45 (2007).
59. J. Danielson, D. Gesch, “Global Multi-resolution Terrain Elevation Data 2010
(GMTED2010)” (2011);
60. D. Bates, M. Mächler, B. Bolker, S. Walker, Fitting Linear Mixed-Effects Models Using
lme4. J. Stat. Softw. 67, 1–48 (2015). doi:10.18637/jss.v067.i01
61. A. F. Zuur, E. N. Ieno, C. S. Elphick, A protocol for data exploration to avoid common
statistical problems. Methods Ecol. Evol. 1, 3–14 (2010). doi:10.1111/j.2041-
62. N. Eisenhauer, A. Stefanski, N. A. Fisichelli, K. Rice, R. Rich, P. B. Reich, Warming shifts
‘worming’: Effects of experimental warming on invasive earthworms in northern North
America. Sci. Rep. 4, 6890 (2014). Medline
63. M. Nieminen, E. Ketoja, J. Mikola, J. Terhivuo, T. Siren, V. Nuutinen, Local land use effects
and regional environmental limits on earthworm communities in Finnish arable
landscapes. Ecol. Appl. 21, 3162–3177 (2011). doi:10.1890/10-1801.1
64. A. F. Zuur, E. N. Ieno, A. A. Saveliev, Mixed Effects Models and Extensions in Ecology with
R (Springer, 2009).
65. S. Dray, A.-B. Dufour, The ade4 Package: Implementing the Duality Diagram for Ecologists.
J. Stat. Softw. 22, 1–20 (2007). doi:10.18637/jss.v022.i04
66. L. Breiman, Random forests. Mach. Learn. 45, 5–32 (2001). doi:10.1023/A:1010933404324
67. A. Liaw, M. Wiener, Classification and regression by randomForest. R News 2, 18–22
68. U. Grömping, Variable importance assessment in regression: Linear regression versus
random forest. Am. Stat. 63, 308–319 (2009). doi:10.1198/tast.2009.08199
69. C. Strobl, A. L. Boulesteix, A. Zeileis, T. Hothorn, Bias in random forest variable importance
measures: Illustrations, sources and a solution. BMC Bioinformatics 8, 25 (2007).
doi:10.1186/1471-2105-8-25 Medline
70. B. H. Menze, B. M. Kelm, R. Masuch, U. Himmelreich, P. Bachert, W. Petrich, F. A.
Hamprecht, A comparison of random forest and its Gini importance with standard
chemometric methods for the feature selection and classification of spectral data. BMC
Bioinformatics 10, 213 (2009). doi:10.1186/1471-2105-10-213 Medline
71. G. James, D. Witten, T. Hastie, R. Tibshirani, An Introduction to Statistical Learning
(Springer, 2013).
72. MuMIn, Multi-Model Inference. R Package version 1.42.1 (2018); https://CRAN.R-
73. N. Gorelick, M. Hancher, M. Dixon, S. Ilyushchenko, D. Thau, R. Moore, Google Earth
Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 202, 18–
27 (2017). doi:10.1016/j.rse.2017.06.031
74. R: A Language and Environment for Statistical Computing (2016).
75. P. Lavelle, A. V. Spain, Soil Ecology (Springer, 2001).
76. P. A. Sanchez, S. Ahamed, F. Carré, A. E. Hartemink, J. Hempel, J. Huising, P. Lagacherie,
A. B. McBratney, N. J. McKenzie, Mde. L. Mendonça-Santos, B. Minasny, L.
Montanarella, P. Okoth, C. A. Palm, J. D. Sachs, K. D. Shepherd, T.-G. Vågen, B.
Vanlauwe, M. G. Walsh, L. A. Winowiecki, G.-L. Zhang, Digital soil map of the world.
Science 325, 680–681 (2009). doi:10.1126/science.1175084 Medline

File (1)

Content uploaded by Jean-François Ponge
Author content
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Soil organisms are a crucial part of the terrestrial biosphere. Despite their importance for ecosystem functioning, few quantitative, spatially explicit models of the active belowground community currently exist. In particular, nematodes are the most abundant animals on Earth, filling all trophic levels in the soil food web. Here we use 6,759 georeferenced samples to generate a mechanistic understanding of the patterns of the global abundance of nematodes in the soil and the composition of their functional groups. The resulting maps show that 4.4 ± 0.64 × 10²⁰ nematodes (with a total biomass of approximately 0.3 gigatonnes) inhabit surface soils across the world, with higher abundances in sub-Arctic regions (38% of total) than in temperate (24%) or tropical (21%) regions. Regional variations in these global trends also provide insights into local patterns of soil fertility and functioning. These high-resolution models provide the first steps towards representing soil ecological processes in global biogeochemical models and will enable the prediction of elemental cycling under current and future climate scenarios.
Full-text available
Deciphering the global distribution of polyploid plants is fundamental for understanding plant evolution and ecology. Many factors have been hypothesized to affect the uneven distribution of polyploid plants across the globe. Nevertheless, the lack of large comparative datasets has restricted such studies to local floras and to narrow taxonomical scopes, limiting our understanding of the underlying drivers of polyploid plant distribution. We present a map portraying the worldwide polyploid frequencies, based on extensive spatial data coupled with phylogeny-based polyploidy inference for tens of thousands of species. This allowed us to assess the potential global drivers affecting polyploid distribution. Our data reveal a clear latitudinal trend, with polyploid frequency increasing away from the equator. Climate, especially temperature, appears to be the most influential predictor of polyploid distribution. However, we find this effect to be mostly indirect, mediated predominantly by variation in plant lifeforms and, to a lesser extent, by taxonomical composition and species richness. Thus, our study presents an emerging view of polyploid distribution that highlights attributes that facilitate the establishment of new polyploid lineages by providing polyploids with sufficient time (that is, perenniality) and space (low species richness) to compete with pre-adapted diploid relatives.
Full-text available
Aim Although the effects of life history traits on population density have been investigated widely, how spatial environmental variation influences population density for a large range of organisms and at a broad spatial scale is poorly known. Filling this knowledge gap is crucial for global species management and conservation planning and to understand the potential impact of environmental changes on multiple species. Location Global. Time period Present. Major taxa studied Terrestrial amphibians, reptiles, birds and mammals. Methods We collected population density estimates for a range of terrestrial vertebrates, including 364 estimates for amphibians, 850 for reptiles, 5,667 for birds and 7,651 for mammals. We contrasted the importance of life history traits and environmental predictors using mixed models and tested different hypotheses to explain the variation in population density for the four groups. We assessed the predictive accuracy of models through cross‐validation and mapped the partial response of vertebrate population density to environmental variables globally. Results Amphibians were more abundant in wet areas with high productivity levels, whereas reptiles showed relatively higher densities in arid areas with low productivity and stable temperatures. The density of birds and mammals was typically high in temperate wet areas with intermediate levels of productivity. The models showed good predictive abilities, with pseudo‐R² ranging between 0.68 (birds) and 0.83 (reptiles). Main conclusions Traits determine most of the variation in population density across species, whereas environmental conditions explain the intraspecific variation across populations. Species traits, resource availability and climatic stability have a different influence on the population density of the four groups. These models can be used to predict the average species population density over large areas and be used to explore macroecological patterns and inform conservation analyses.
Full-text available
Soils harbour some of the most diverse microbiomes on Earth and are essential for both nutrient cycling and carbon storage. To understand soil functioning, it is necessary to model the global distribution patterns and functional gene repertoires of soil microorganisms, as well as the biotic and environmental associations between the diversity and structure of both bacterial and fungal soil communities1–4. Here we show, by leveraging metagenomics and metabarcoding of global topsoil samples (189 sites, 7,560 subsamples), that bacterial, but not fungal, genetic diversity is highest in temperate habitats and that microbial gene composition varies more strongly with environmental variables than with geographic distance. We demonstrate that fungi and bacteria show global niche differentiation that is associated with contrasting diversity responses to precipitation and soil pH. Furthermore, we provide evidence for strong bacterial–fungal antagonism, inferred from antibiotic-resistance genes, in topsoil and ocean habitats, indicating the substantial role of biotic interactions in shaping microbial communities. Our results suggest that both competition and environmental filtering affect the abundance, composition and encoded gene functions of bacterial and fungal communities, indicating that the relative contributions of these microorganisms to global nutrient cycling varies spatially.
Full-text available
The immense diversity of soil bacterial communities has stymied efforts to characterize individual taxa and document their global distributions. We analyzed soils from 237 locations across six continents and found that only 2% of bacterial phylotypes (~500 phylotypes) consistently accounted for almost half of the soil bacterial communities worldwide. Despite the overwhelming diversity of bacterial communities, relatively few bacterial taxa are abundant in soils globally. We clustered these dominant taxa into ecological groups to build the first global atlas of soil bacterial taxa. Our study narrows down the immense number of bacterial taxa to a “most wanted” list that will be fruitful targets for genomic and cultivation-based efforts aimed at improving our understanding of soil microbes and their contributions to ecosystem functioning.
Full-text available
Our basic understanding of plant litter decomposition informs the assumptions underlying widely applied soil biogeochemical models, including those embedded in Earth system models. Confidence in projected carbon cycle-climate feedbacks therefore depends on accurate knowledge about the controls regulating the rate at which plant biomass is decomposed into products such as CO2. Here we test underlying assumptions of the dominant conceptual model of litter decomposition. The model posits that a primary control on the rate of decomposition at regional to global scales is climate (temperature and moisture), with the controlling effects of decomposers negligible at such broad spatial scales. Using a regional-scale litter decomposition experiment at six sites spanning from northern Sweden to southern France-and capturing both within and among site variation in putative controls-we find that contrary to predictions from the hierarchical model, decomposer (microbial) biomass strongly regulates decomposition at regional scales. Furthermore, the size of the microbial biomass dictates the absolute change in decomposition rates with changing climate variables. Our findings suggest the need for revision of the hierarchical model, with decomposers acting as both local- and broad-scale controls on litter decomposition rates, necessitating their explicit consideration in global biogeochemical models.
Full-text available
There has been considerable focus on the impacts of environmental change on ecosystem function arising from changes in species richness. However, environmental change may affect ecosystem function without affecting richness, most notably by affecting population densities and community composition. Using a theoretical model, we find that, despite invariant richness, (1) small environmental effects may already lead to a collapse of function; (2) competitive strength may be a less important determinant of ecosystem function change than the selectivity of the environmental change driver and (3) effects on ecosystem function increase when effects on composition are larger. We also present a complementary statistical analysis of 13 data sets of phytoplankton and periphyton communities exposed to chemical stressors and show that effects on primary production under invariant richness ranged from À75% to +10%. We conclude that environmental protection goals relying on measures of richness could underestimate ecological impacts of environmental change.
Full-text available
High-resolution information on climatic conditions is essential to many applications in environmental and ecological sciences. Here we present the CHELSA (Climatologies at high resolution for the earth's land surface areas) data of downscaled model output temperature and precipitation estimates of the ERA-Interim climatic reanalysis to a high resolution of 30 arc sec. The temperature algorithm is based on statistical downscaling of atmospheric temperatures. The precipitation algorithm incorporates orographic predictors including wind fields, valley exposition, and boundary layer height, with a subsequent bias correction. The resulting data consist of a monthly temperature and precipitation climatology for the years 1979–2013. We compare the data derived from the CHELSA algorithm with other standard gridded products and station data from the Global Historical Climate Network. We compare the performance of the new climatologies in species distribution modelling and show that we can increase the accuracy of species range predictions. We further show that CHELSA climatological data has a similar accuracy as other products for temperature, but that its predictions of precipitation patterns are better. Design Type(s) data integration objective • modeling and simulation objective Measurement Type(s) temperature of air • hydrological precipitation process Technology Type(s) data acquisition system Factor Type(s) Sample Characteristic(s) Earth • planetary atmosphere
Human activities are accelerating global biodiversity change and have resulted in severely threatened ecosystem services. A large proportion of terrestrial biodiversity is harbored by soil, but soil biodiversity has been neglected from many global biodiversity assessments and conservation actions, and our understanding of global patterns of soil biodiversity remains limited. In particular, the extent to which hotspots and coldspots of aboveground and soil biodiversity overlap is not clear. We examined global patterns of overlap by mapping indices of aboveground (mammals, birds, amphibians, vascular plants) and soil (bacteria, fungi, macrofauna) biodiversity. Our analysis indicated that areas of mismatch between aboveground and soil biodiversity covered 27% of the Earth's terrestrial surface. The temperate broadleaf and mixed forests biome had the highest proportion of grid cells with high aboveground biodiversity but low soil biodiversity, while the boreal and tundra biomes had higher soil biodiversity but low aboveground biodiversity. While more data on soil biodiversity is needed, both to cover geographic gaps and to include additional taxa, our results suggest that protecting aboveground biodiversity may not sufficiently reduce threats to soil biodiversity. Given the functional importance of soil biodiversity and the role of soils for human well‐being, soil biodiversity should be further considered in policy agendas and conservation actions by adapting management practices to sustain soil biodiversity and considering soil biodiversity when designing protected areas. This article is protected by copyright. All rights reserved
Macroecology is the study of the mechanisms underlying general patterns of ecology across scales. Research in microbial ecology and macroecology have long been detached. Here, we argue that it is time to bridge the gap, as they share a common currency of species and individuals, and a common goal of understanding the causes and consequences of changes in biodiversity. Microbial ecology and macroecology will mutually benefit from a unified research agenda and shared datasets that span the entirety of the biodiversity of life and the geographic expanse of the Earth.