Chapter 2 Human and Environmental Factors Influence Soil Faunal Abundance–Mass Allometry and Structure

Abstract

We examined structural variation in soil faunal communities from 146 agroecosystems in The Netherlands, using a unique database of taxonomically highly resolved soil samples gathered by uniform methods. For each site, data included measurements of average body mass (M) and population density (N) of each detected taxon and environmental and human‐use factors. We used three descriptors of soil faunal community structure: abundance–mass slope, which is the slope of the regression line through all faunal taxa in a site plotted on log(N)‐versus‐log(M) coordinates (all logarithms were base 10); the taxonomic diversity of each community's fauna (number of animal taxa at the finest available level of taxonomic resolution); and the total biomass of all fauna. The goal of the study was to account for variation in these descriptors and to develop causal hypotheses.These structural descriptors varied systematically. More than half of the variation in each descriptor was explained by external human, environmental, and biotic influences. Few predictors were needed to explain structural variation: above‐ground ecosystem type (ET, describing the kind of human management); soil bacterial biomass; and a measure of precipitation. ET was the most important predictor of below‐ground faunal community structure.Abundance–mass slopes ranged from –0.85 to –0.07 with mean –0.51; only four slopes were more negative than –3/4 (i.e., the log(N)‐versus‐log(M) regression line was steeper than –3/4). Slopes less negative than –1 (respectively, –3/4) indicated that, on average, taxon biomass (respectively, taxon energy consumption) increased with taxon body mass. Abundance–mass slope was more negative in more disturbed sites than in less disturbed sites. Disturbance may have produced this pattern by affecting populations of large‐M taxa, which are slower to reproduce, more than small‐M taxa. Across some types of site (super‐intensive farms and possibly intensive farms), greater soil bacterial biomass was associated with less‐negative abundance–mass slope, suggesting top‐down control of bacterivorous taxa. ET and soil bacterial biomass were sufficient to explain most of the variation in the whole abundance–mass allometric relationship, including slope and intercept.Total faunal biomasses were higher in recently fertilized sites. Greater soil bacterial biomass was associated with the same increase in log faunal biomass between sites, on average, for all ET. Taxonomic diversity differed in sites of different ET in a way related to human disturbance. Precipitation was associated oppositely with diversity in sites of different types. Testable mechanistic hypotheses for the patterns observed here are discussed.

Full text

Page 1

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:11 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
B.Log Faunal Biomass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C. Faunal Diversity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D. Abundance–Mass Intercept and Expected Log Population
Density of Smallest Taxa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
E. Log Faunal Population Density . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Human and Environmental Factors
Influence Soil Faunal Abundance–Mass
Allometry and Structure
DANIEL C. REUMAN, JOEL E. COHEN AND CHRISTIAN MULDER
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Soil Faunal Descriptors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.Abundance–Mass Slope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
B.Faunal Diversity and Total Biomass. . . . . . . . . . . . . . . . . . . . . . . .
C. Abundance–Mass Intercept and Expected log Population Density
of Smallest Taxa. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A. Data on Taxonomy, Average Body‐Mass, and Population Density
B. Environmental Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C. Human‐Use Data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D.Carbon Resource Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A. Classification of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
B.Stepwise Regression. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
C. Testing Assumptions of Linear Models. . . . . . . . . . . . . . . . . . . . . .
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A. Models of Soil Faunal Community Structure. . . . . . . . . . . . . . . . .
B.Relative Importance of Variables . . . . . . . . . . . . . . . . . . . . . . . . . .
C. Interpreting Variation in Structure . . . . . . . . . . . . . . . . . . . . . . . . .
D. Testing for Artifacts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A. Food Web Descriptors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
B. Relative Importance of Variables . . . . . . . . . . . . . . . . . . . . . . . . . .
C. Limitations of This Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
D. Future Directions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Appendix I. Stepwise Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Appendix II. Testing Assumptions of Linear Models . . . . . . . . . . . . . . . . . . .
Appendix III. Detailed Statistical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.Abundance–Mass Slope. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
47
49
50
52
I.
II.
52
53
53
53
54
55
56
56
56
56
57
57
57
60
65
68
69
70
70
71
73
73
75
77
77
78
78
III.
IV.
V.
VI.
79
81
81
ADVANCES IN ECOLOGICAL RESEARCH VOL. 41
# 2009 Elsevier Ltd. All rights reserved
0065-2504/09 $35.00
DOI: 10.1016/S0065-2504(09)00402-4

Page 2

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:12 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by: ProjectA-
CORRECTED PROOF
was associated oppositely with diversity in sites of different types. Testable
mechanistic hypotheses for the patterns observed here are discussed.
SUMMARY
We examined structural variation in soil faunal communities from 146
agroecosystems in The Netherlands, using a unique database of taxonomi-
cally highly resolved soil samples gathered by uniform methods. For each
site, data included measurements of average body mass (M) and population
density (N) of each detected taxon and environmental and human‐use
factors. We used three descriptors of soil faunal community structure:
abundance–mass slope, which is the slope of the regression line through all
faunal taxa in a site plotted on log(N)‐versus‐log(M) coordinates (all loga-
rithms were base 10); the taxonomic diversity of each community’s fauna
(number of animal taxa at the finest available level of taxonomic resolution);
and the total biomass of all fauna. The goal of the study was to account for
variation in these descriptors and to develop causal hypotheses.
These structural descriptors varied systematically. More than half of the
variation in each descriptor was explained by external human, environmen-
tal, and biotic influences. Few predictors were needed to explain structural
variation: above‐ground ecosystem type (ET, describing the kind of human
management); soil bacterial biomass; and a measure of precipitation. ET was
the most important predictor of below‐ground faunal community structure.
Abundance–mass slopes ranged from –0.85 to –0.07 with mean –0.51;
only four slopes were more negative than –3/4 (i.e., the log(N)‐versus‐
log(M) regression line was steeper than –3/4). Slopes less negative than –1
(respectively, –3/4) indicated that, on average, taxon biomass (respectively,
taxon energy consumption) increased with taxon body mass. Abundance–
mass slope was more negative in more disturbed sites than in less disturbed
sites. Disturbance may have produced this pattern by affecting populations
of large‐M taxa, which are slower to reproduce, more than small‐M taxa.
Across some types of site (super‐intensive farms and possibly intensive
farms), greater soil bacterial biomass was associated with less‐negative
abundance–mass slope, suggesting top‐down control of bacterivorous taxa.
ET and soil bacterial biomass were sufficient to explain most of the variation
in the whole abundance–mass allometric relationship, including slope and
intercept.
Total faunal biomasses were higher in recently fertilized sites. Greater soil
bacterial biomass was associated with the same increase in log faunal bio-
mass between sites, on average, for all ET. Taxonomic diversity differed in
sites of different ET in a way related to human disturbance. Precipitation
46
DANIEL C. REUMAN ET AL.

Page 3

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by: ProjectA-
CORRECTED PROOF
Loehle, 1991; Martinez, 1991; Paine, 1992; Pimm et al., 1991; Wootton,
1997). Quantitative measurements would better characterize webs for
which trophic links have highly uneven magnitude (e.g., many weak links
and a few strong ones; Reuman and Cohen, 2005). Bersier et al. (2002)
proposed descriptors of web structure that used quantitative link
I. INTRODUCTION
For decades, studies have examined the influence on food web structure of
factors such as environmental variability (Briand, 1983); three‐ versus two‐
dimensionality of habitat (Briand and Cohen, 1987); amount of primary
productivity (Briand and Cohen, 1987; Vander Zanden et al., 1999); acidifi-
cation in lakes (Locke and Sprules, 1994); habitat duration patterns in
temporary ponds (Schneider, 1997); habitat size (Spencer and Warren,
1996; Vander Zanden et al., 1999); and disturbance or ‘‘stress’’ (Havens,
1994; Jenkins et al., 1992). Differences in food web (henceforth web) struc-
ture by habitat type (e.g., pelagic versus terrestrial; Bengtsson, 1994; Cohen,
1994; Havens, 1997) and over time and space using habitats of the same type
(Carney et al., 1997; Clarke, 1998; Closs and Lake, 1994; Schoenly and
Cohen, 1991) have also been investigated. Differences in web structure
have typically been measured with a suite of descriptors based on the binary
predation matrix (P ¼ [pij], where pijis 1 if taxon j eats taxon i, regardless of
how often this occurs, and 0 otherwise), such as: link density; various kinds
of connectances (Warren, 1994, gives definitions); proportions of top, inter-
mediate, and basal taxa; and minimum, mean, modal, median, and maximal
food‐chain length (where the length of a food chain is the number of trophic
links that comprise it).
However, web studies have been criticized for using data of poor quality,
based on insufficient sampling of ecosystems (Bersier et al., 1999; Cohen
et al., 1993; Hall and Raffaelli, 1991; Martinez, 1991; Polis, 1991; Polis and
Strong, 1996; Winemiller, 1990). Statistics such as connectance and modal
food‐chain length depend sensitively on sampling effort (Goldwasser and
Roughgarden, 1997; Martinez et al., 1999; Winemiller, 1990). Incomplete
sampling limits even modern webs (Woodward et al., 2005, their Figure 3).
Descriptors of web structure based on detailed trophic data are especially
vulnerable to undersampling because ‘‘the detection of trophic links system-
atically lags behind the detection and inclusion of species, which may render
the accurate measurement of many web properties inherently problematic’’
(Goldwasser and Roughgarden, 1997).
In response to the sensitivity of traditional web descriptors to sampling
effort, several authors called for quantitative measures of trophic links, so
that the intensity or frequency of each link is used in place of a binary
indicator of whether the link occurs (Cohen et al., 1990; Kenny and
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
47

Page 4

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by: ProjectA-
CORRECTED PROOF
trophic data, but reflected patterns of energy flow through the web and
indirectly reflected web structure (Section II). We found that variation in
abundance–mass slopes and other descriptors was largely explained in terms
of environmental, human‐use, and biotic factors. We developed testable
hypotheses of mechanisms.
measurements. Their proposed descriptors, based on information‐theoretic
concepts, were intended to improve upon, but serve the same conceptual
purpose, as classical descriptors such as the proportions of top, intermediate,
and basal taxa, and link density. The quantitative‐link‐based descriptors of
Bersier et al. (2002) may be less sensitive to undersampling if undersampling
affects primarily measurements of weak links. However, the highly detailed
web data needed to calculate descriptors such as those of Bersier et al. (2002)
are difficult to obtain; they include an empirical estimate of the strength of
each trophic link. We took the opposite approach here: we used descriptors
expected to be more robust to moderate undersampling because they do not
use trophic data. Vander Zanden et al. (1999) took a similar approach using
stable‐isotopeindicatorsofwebstructurethatdidnotrequiredetailedtrophic
data. We explain how the descriptors we used reflect web structure without
using trophic data (Section II). The work of Cyr et al. (1997) was similar to
ours, but used lakes. Some of their results are comparable to ours.
Most studies of the relationship between population density (N) and
average body mass (M) of species (e.g., Damuth, 1981; Peters, 1983; Russo
et al., 2003; reviews include Blackburn and Gaston, 2001; Kerr and Dickie,
2001; Leaper and Raffaelli, 1999) focused on species from a single broad
taxon or trophic level (e.g., birds or herbivorous mammals). When such data
were gathered globally or regionally, log(N)‐versus‐log(M) scatter plots
often showed a linear relationship and had regression slope (here called
abundance–mass slope) about ?3/4 (e.g., Damuth, 1981, 1987; Greenwood
et al., 1996; Nee et al., 1991). This macroecological relationship has been
explained using metabolic theory (Brown et al., 2004; West et al., 1997) and
other mechanisms (e.g., Blackburn and Gaston, 1993). For all taxa in a local
web (an ecological context very different from that in which data on only one
clade are gathered, be it locally, regionally, or globally), recent studies found
that log(N) was often linearly related to log(M) but abundance–mass slopes
varied widely from web to web (Cohen et al., 2003; Cyr et al., 1997; Jonsson
et al., 2005; Leaper and Raffaelli, 1999; Marquet et al., 1990; Mulder et al.,
2005a, Reuman et al., this volume; Woodward et al., 2005). The faunal
abundance–mass slope of a web was one of the descriptors used in this
study. Total biomass of all fauna and taxonomic diversity of all fauna were
also used.
This study examined the structural variation in 146 soil agroecosystems in
The Netherlands. Our descriptors of soil faunal communities did not use
48
DANIEL C. REUMAN ET AL.

Page 5

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
Percent bacterial cells dividing
Mean bacterial cell length
II. SOIL FAUNAL DESCRIPTORS
Names and units of soil faunal community descriptors are listed in Table 1
and explained here.
Table 1
gories: environmental, human‐use, and non‐faunal biotic variables, and soil faunal
descriptors
Names, abbreviations (if defined), and units for all variables in four cate-
Name of variable Units and measurement information
Environmental variables
Mean daily air temperature
Mean noon air temperature
Mean daily precipitation
Maximum daily precipitation
Modified Julian date
Phosphate content
(after water extraction)
Phosphorus content (after
acetate–lactate extraction)
Soil pH
Latitude
Longitude
Area of site
?C, mean over 21 days before sampling
?C, mean over 21 days before sampling
mm, mean over 21 days before sampling
mm, max over 21 days before sampling
a linear function of Julian Day; see text
mg P2O5/l
mg P2O5/kg dry soil
Power of H in KCl
m south from Amersfoort
m west from Amersfoort
Hectares
Human‐use variables
Above‐ground
ecosystem type (ET)
Forest, pasture, winter farm, organic farm,
conventional farm, intensive farm, or
super‐intensive farm
Animal units that excreted an average of 161 kg
N/(ha yr) and 41 kg P/(ha yr)
%
%
%
kg/(ha y), proportional to livestock density
Standardized livestock density
% of site on which maize grew
% of site on which grass grew
% of site used for other crops
Phosphorus in‐flux from manure
Carbon resource variables
Soil bacterial biomass
Log soil bacterial biomass
Soil organic matter
Bacterial diversity
Shannon–Wiener index of
bacterial diversity
mg C/g dry soil
log mg C/g dry soil
% of dry soil
Band count after DGGE
Index of DGGE band patterns
%, measured one week after sampling
mm
Soil faunal descriptors
Faunal diversity
Total faunal biomass
Abundance–mass slope
Abundance–mass intercept
Expected log(N) of smallest taxa
Number of taxa
mg/m2
–
Individuals/m2, log scale
Individuals/m2, log scale
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
49

Page 6

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
taxa. Departures of abundance–mass slopes from the benchmark value –3/4
register departures from energetic equivalence or a failure of the metabolic
assumption that all taxa absorb energy at a rate proportional to NM?
(Figure 1).
If webs are size‐structured so that large taxa eat primarily small ones, then
differences between webs in abundance–mass slope reflect differences in how
A. Abundance–Mass Slope
The abundance–mass slope of a site is the coefficient b in the linear model
logðNÞ ¼ blogðMÞ þ a þ e
ð1Þ
fitted by ordinary least squares regression to data on animal taxa in the
soil samples from the site. Bacteria were excluded because their level of
taxonomic resolution (a single node) was not comparable to that of other
taxa. Only the total biomass of fungal mycelia and organic detritus was
quantified (Mulder et al., 2005a), so fungal mycelia were excluded from
calculations of abundance–mass slopes. Protists were not quantified, but
are virtually absent from sandy soils such as those used in this study.
Abundance–mass slope provides information about how faunal taxon
biomass varies with taxon M. Because the biomass B of a taxon is its
abundance N times its average body mass M, B ¼ NM, the abundance–
mass slope plus one indicates how B changes for taxa of increasing body mass
(Cohen et al., 2003). If b ¼ –1, then the trend is for all taxa to have equal
biomass. If the abundance–mass slope is less negative than –1, for example,
if b ¼ –3/4, then biomass tends to increase with increasing body mass; if the
abundance–mass slope is more negative than –1, for example, if b ¼ –5/4,
then biomass decreases with increasing body mass.
Abundance–mass slope also describes how taxon energy consumption
varies with taxon M, reflecting how energy flows through the web. The
energetic equivalence hypothesis assumes that taxa absorb energy from the
environment in amounts that do not depend systematically on M (Damuth,
1981, 1987). If all taxa absorb energy at constant rate R, then since the
metabolic rate E of an individual organism is approximately a power law
of its body mass, E ¼ kM?
indiv(e.g., Peters, 1983), then (henceforth neglecting
variation of individuals’ Mindivfrom the mean M of their respective taxa) we
can write R ¼ kNM?, so that N ¼ R/(kM?), and
logðNÞ ¼ ??logðMÞ þ logðR=kÞð2Þ
Thus the energetic equivalence hypothesis predicts an abundance–mass slope
b ¼ –?. The value of ? is often claimed to be close to 3/4 (Brown et al., 2004;
Peters, 1983; West et al., 1997). Observing an abundance–mass slope less
negative than (respectively, more negative than) –? suggests that larger taxa
absorb more (respectively, less) energy from the environment than smaller
50
DANIEL C. REUMAN ET AL.

Page 7

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
log(a)/log(b)–?, where b ¼ Mc/Mris the consumer‐to‐resource body mass
ratio. Several authors have elaborated this theory to relate aspects of food
webstructuresuchasaverageconsumer‐to‐resourcebodymassratioandmass‐
specific taxonomic diversity to the abundance–mass slope of this study and
other types of mass–abundance allometry (Brown et al., 2004, especially their
Eq. 13; Damuth, 1994; Jonsson et al., 2005; Reuman et al., 2008, this volume).
energy filters through trophic links, which is related to web trophic structure.
In a two‐species food chain, if the population production of the resource r is
kNrMr?and the population production of the consumer, c, of r is kNcMc?,
assuming all production of r is consumed by c, then
akNrM?
r¼ kNcM?
c
ð3Þ
where a is the efficiency by which c converts what it eats into its own
production. Taking logs of this equation and rearranging gives
logðNcÞ ? logðNrÞ ¼ logðaÞ ? ?½logðMcÞ ? logðMrÞ?ð4Þ
Dividing Eq. 4 by log(Mc)–log(Mr) shows that the slope of the line connec-
ting r and c on log(N)‐versus‐log(M) axes (the abundance–mass slope) is
18
ABC
16
14
12
10
8
6
4
−202
−20
Log(M)
Log(N)
2
−202
Figure 1
of taxa in hypothetical webs; solid lines are ordinary least squares regressions whose
slopes are the abundance–mass slopes. The communities in (A, C) have more‐negative
(steeper) abundance–mass slopes than that in (B). The community in (C) has higher
abundance–mass intercept than those in (A, B). Dashed lines have slope –1: two taxa
onthe same dashed line havethe same biomass; taxaon higherdashed lines havemore
biomass. Dotted lines have slope –3/4: two taxa on the same dotted line consume the
same amount of energy; taxa on higher dotted lines consume more energy.
Differences in abundance–mass slope and intercept. Dots are populations
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
51

Page 8

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
log(M) ¼ –1.5. Expected log(N) of smallest taxa has a clear biological
interpretation and is independent of the units used to measure M. If units
of M were chosen so that the smallest taxa had M ¼ 1, then expected log(N)
of smallest taxa would equal abundance–mass intercept. In a size‐structured
system, expected log(N) of smallest taxa is directly affected by the energy
available to basal species and by their consumers.
B. Faunal Diversity and Total Biomass
Soil faunal taxonomic diversity, or simply faunal diversity, was defined as the
number of faunal taxa in each site. For clarity, we emphasize that this count
excluded bacteria, fungal mycelia, plant roots, and detritus. Many previous
studies used the number of web nodes (often denoted S) to describe or model
web structure (e.g., Cohen, 1990; Williams and Martinez, 2000). Other
studies examined spatial variation in the diversity of species within a clade
such as birds or nematodes (Jetz and Rahbek, 2002; Mulder et al., 2003).
Variation of single‐clade diversity may be easier to study than variation in
whole‐web diversity because taxonomic expertise is often specific. However,
organisms live in physical sites which include taxa from many clades, so
variationinsitediversityisalsoecologicallyimportant,andthelocaldiversity
of clades depends in part on the site’s entire biotic structure. Thinking of
diversity at a site as a vector of the numbers of taxa with one entry for the
number of taxa of each major clade is a useful way to combine clade‐specific
and whole‐site or whole‐web approaches in future research. The total faunal
biomass of each site was computed by summing MN over all faunal taxa.
C. Abundance–Mass Intercept and Expected log Population
Density of Smallest Taxa
We defined the abundance–mass intercept of a site as the coefficient, a, in Eq. 1,
fittedtofaunalMandNdatafromthesite.Abundance–massinterceptdepends
ontheunitsofMbecausetheverticalaxis(thelog(N)‐axis)occursatM¼1unit
(log(M) ¼ 0).ThesameunitsforM (microgramsdrymass)wereusedthrough-
outthestudybuttheabundance–massinterceptisnoteasytointerpretbecause
log(M) ¼ 0 (M ¼ 1 mg) occurred in the middle of the body‐mass range.
We used a web descriptor that contained the same information as
abundance–mass intercept and was easier to interpret biologically: the
expected log(N) of the smallest taxa. The smallest faunal taxa in each site
were about the same size. The minimum log(M) occurring in each site had
mean value –1.5 (M ¼ 10–1.5¼ 0.032 mg, corresponding to the average body‐
mass values of soil nematodes such as Aphelenchoides and Metateratocepha-
lus) and ranged from –1.4 to –1.6. An expected log(N) of smallest taxa was
defined for each site to be the value log(N) of the best‐fitting line Eq. 1 at
52
DANIEL C. REUMAN ET AL.

Page 9

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
The Royal Netherlands Meteorological Institute (www.knmi.nl, De Bilt)
supplied air temperature and precipitation data (minima, maxima, and
averages for the 21 days before sampling) near each investigated location.
The mean of the highest and lowest temperatures on each day were provided,
and the mean of these minima and maxima is calculated over the period of
21 days.
III. DATA
All data for this study were gathered between 21 April 1999 and 4 June 2002
from 146 farms, pastures, and forests on Pleistocene sand in The Nether-
lands. Mulder et al. (2003, 2005a,b) gave complete details on how data were
gathered. We here describe the nature of the data including definitions and
units of measurement. Names, units, and abbreviations of variables are listed
in Table 1.
A. Data on Taxonomy, Average Body‐Mass, and
Population Density
For each site, all soil animals, including nematodes, arthropods (mites,
insects, and myriapods), enchytraeids (potworms), and lumbricids (earth-
worms), were identified to genus or family. Of the 65 nematode, 177 arthro-
pod, and 18 oligochaete taxa identified in any of the 146 sites, 78% of the
nematodes, 88% of the microarthropods, and 100% of the oligochaetes were
identified to genus; the rest to family. For each taxon, average body mass
(M, in micrograms dry mass) and population density (N, in individuals per
square meter of soil surface) were measured. The same M was used for all
sites where a given taxon occurred. Bacteria, fungal mycelia, plant roots, and
detritus were each treated as single ‘‘taxa’’; of these, M and N were obtained
only for bacteria. Protists were ignored.
B. Environmental Data
Modified Julian date started on 17 September with a value of –164 and
increased by one each subsequent day until 16 September, where it stopped
with a value of 200 in a non‐leap year and 201 in a leap year. Both modified
Julian date and standard ‘‘Julian Day’’ (Mulder et al., 2003) must start on
some day of the year, artificially indicating a difference of 365 or 364 between
that day and the previous one (for leap years and non‐leap years, respectively).
We used modified Julian date because the new discontinuity occurred in the
middle of the largest interval with no sampling. Modified Julian date values
ranged from –16 to 122, well away from the discontinuity.
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
53

Page 10

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
Agricultural Economics Research Institute of The Netherlands (LEI‐DLO,
www.lei.wur.nl). Bio‐organic and biodynamic farming techniques were used
on organic farms. Compost and farmyard manure were used for fertilization;
no biocides were used. Pastures were used for both grazing and farming.
They were similar to organic farms, and used specific agronomic practices to
enhance nitrogen fixation by the rhizobia of the clovers Trifolium repens and
Soil phosphorus content was measured using two methods of extraction.
Phosphate content after water extraction used aqueous extraction at water‐
to‐soil ratio 60:1 by volume, after 22 h of pre‐equilibrating soil with water,
and 1 h of gently shaking before filtration (Sissingh, 1971). Phosphate con-
tent after water extraction reflects the maximal possible concentration of
phosphorus in the soil moisture biofilm, a thin layer of water around soil
particles. Phosphorus content after acetate–lactate buffer extraction includes
phosphorus occluded in oxides on the surfaces of soil particles and in water‐
insoluble compounds. Results of the two methods were highly correlated in
our 146 sites (R2¼ 81.4%).
Acidity (pH) of oven‐dried soil samples was measured in 1 M potassium
chloride solution. Latitude and longitude were given in meters, falsely pro-
jected south and west, respectively, from Amersfoort (52?0902200N,
5?2301500E; see Mulder et al., 2005b). This stereographic double‐projection
on the Bessel spheroid is widely used in The Netherlands.
C. Human‐Use Data
1. Above‐Ground Ecosystem Type
Above‐ground ecosystem type (ET) took seven values: forest, winter farm,
pasture, organic farm, conventional farm, intensive farm, and super‐
intensive farm (Mulder et al., 2005a,c, 2006; Schouten et al., 2004). The last
five ETs were called cultivated farms.
Winter farms were defined to be lands not cultivated or grazed at the time
of sampling, but previously and later used for grazing or to grow non‐cereal
crops. Previous land‐use of winter farms included multicropping, intercrop-
ping, crop rotation, and alley cropping. Forests were subjected to the low‐
intensity management of traditional agroforestry; they were typically planta-
tions of Scots pine (Pinus sylvestris), but sometimes also included European
larch (Larix decidua) or naturalized Douglas fir (Pseudotsuga menziesii).
Other sites were cultivated actively at the time of sampling. Management
regime was the most important factor in defining the other ET values.
Organic farms, pastures, and conventional farms were subjected to middle‐
intensity management; intensive and super‐intensive farms were subjected to
high‐intensity management. Organic farms were certified organic by the
54
DANIEL C. REUMAN ET AL.

Page 11

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by: ProjectA-
CORRECTED PROOF
tion by polymerase chain reaction (PCR) using a general probe for bacterial
16S‐ribosomal DNA (Bloem and Breure, 2003). The variable genetic diversi-
ty of bacteria used here was the number of bands appearing after electropho-
resis at fixed denaturant concentration. Band patterns were analyzed by
image analysis using two replicates per site. Bacterial cells were counted in
soil smears by fluorescent staining (Paul et al., 1999). Cell numbers, lengths
Trifolium pratense. On conventional farms, mineral fertilizers were used
to compensate for the much smaller amount of farmyard manure used
compared to organic farms.
On intensive and super‐intensive farms, both organic and mineral fertili-
zers were used in substantial amounts. More biocides were used on super‐
intensive farms than on intensive farms. Biocide and fertilizer use informa-
tion was gathered through farmer interviews. The Dutch Central Bureau of
Statistics (CBS) regularly surveyed the use of chemical pesticides in arable
and horticultural farming (www.cbs.nl). Previous surveys were conducted in
1992, 1995, 1998, and 2000. Livestock density also played an important role
in site classification, as did farm area, crop mixture, the farming regime used
during the 5 years before sampling, and recent harvest or planting (Mulder
et al., 2005d,e). The majority of sites of this study were rural (87.0%); the
majority of farms were no‐tillage (61.4%). Numbers of sites of each ET
are listed by Reuman et al. (this volume, their Table 4).
2. Other Human‐Use Data
Standardized livestock density was measured as the numbers of animal units
(cows, calves, pigs, and poultry) per hectare that excreted an average of
161 kg N ha–1yr–1and 41 kg P ha–1yr–1according to the CBS (www.cbs.
nl, accessed February 2006). The percentages of each site on which grass,
maize, and other crops (mainly potatoes and beets) grew were measured.
Phosphorus in‐flux from animal manure was assumed to be proportional to
standardized livestock density, and was therefore not included as a separate
predictor in models.
D. Carbon Resource Data
Carbon resource data describe important carbon pools that support the
bottom of the soil faunal food web: soil organic matter and bacteria. Soil
bacterial biomass (Mulder et al., 2005a) and its logarithm were predictors in
models (see Section IV). Soil organic matter was measured as a percent of dry
soil. Genetic diversity of bacteria was determined using Denaturing Gradient
Gel Electrophoresis (DGGE; Mulder et al., 2005a,b) after DNA amplifica-
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
55

Page 12

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
data, listed in most elementary statistics texts and reviewed by Cohen and
Carpenter (2005) and Reuman et al. (this volume). The validity of four
assumptions can be tested statistically with our data. Because 110 of the
146 sites satisfied all four testable assumptions at the 1% level (Reuman et al.,
this volume), we used abundance–mass slopes and intercepts of all sites for
subsequent modeling of these descriptors. Including sites that violated the
(in micrometers) and the frequencies of dividing cells (percentages of
bacterial cells dividing one week after sampling) were determined by direct
confocal laser scanning microscopy coupled to a fully automatic image
analysis system (Mulder et al., 2005a).
IV. METHODS
All computations were done in Matlab version 6.5.0.180913a (R13).
A statistical significance level of 1% was used.
A. Classification of Variables
Variables classified (in Table 1) as environmental variables, human‐use
variables, and carbon resource variables were used to predict response
variables classified as soil faunal descriptors.
B. Stepwise Regression
To analyze variation in soil faunal descriptors and to develop hypotheses
about causes of the variation, stepwise linear regression was an appropriate
exploratory technique because of its simplicity. We used the stepwise linear
regression algorithm of Appendix I. Starting from amodel that predicts a soil
faunal descriptor by its mean, the algorithm alternately may add predictors
for greater explanatory power, and remove them for model simplicity.
A backward‐only elimination procedure was not used because we wanted
simple models. A forward‐only procedure was not used because many pre-
dictors of this study are related, and a model without redundancy was desired.
Our stepwise procedure rarely removed predictors, so a forward‐only method
might have produced similar results.
C. Testing Assumptions of Linear Models
The standard linear model Eq. 1 makes five principal assumptions about
56
DANIEL C. REUMAN ET AL.

Page 13

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
ET was always the most influential single predictor and was always chosen
first in stepwise regression.
ET provides information about soil faunal structure beyond that
contained in all other predictors combined. All predictors including ET
(but excluding interaction terms) explained significant variation in faunal
diversity (69.9%) beyond that explained (57.6%) by all predictors except ET
regression assumptions was not statistically problematic because we used only
the point estimates of b and a, not their confidence intervals. Stepwise regres-
sion was also used toconstructlinearmodels of b and a usingonly the 110 sites
for which assumptions were satisfied (Appendix III), with similar results.
We also used the general linear model
y ¼ Xb þ e
ð5Þ
where y is a site descriptor and X encodes predictors. We tested whether
data met the assumptions of each general linear model using seven tests,
described in Appendix II. Results of these tests are in Appendix III. Typically,
assumptions were not violated, or were only mildly violated.
V. RESULTS
A. Models of Soil Faunal Community Structure
Linear models of abundance–mass slope, log faunal biomass, and faunal
diversity explained, respectively, 58.0%, 60.7%, and 73.8% of the variation
in the three response variables (Table 2). Of the 23 potential predictors
(environmental, human‐use, and carbon resource variables, Table 1), step-
wise regression selected only ET, maximum daily precipitation, soil bacterial
biomass, and soil organic matter. Soil organic matter provided only a mar-
ginal increase in total R2, but this may be expected since soil organic matter
was uniformly low for sandy soils such as those of this study, and did not
vary much among our sites compared to differences in soil organic matter
between sandy soils and other types of soil such as clay and peat. A large
percentage of observed variation in all soil faunal descriptors was explained.
Only a few predictors were needed.
B. Relative Importance of Variables
1. ET was the Most Important Predictor
ET alone explained 51.7%, 56.6%, and 66.0% of the variation in abundance–
mass slope, log faunal biomass, and faunal diversity, respectively (Table 2).
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
57

Page 14

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
Abundance–mass
Table 2
Coefficients of determination (R2) for linear models of site descriptors (first column)
Response
variable
R2, sole predictor
ET (%)
Second predictor
Increment in
R2(%)
Third
predictor
Increment in
R2(%)
Total R2
(%)
slope
51.7
ET ? soil bacterial biomass
(Figure 3)
6.3
NA
NA
58.0
Log faunal
biomass
56.6
Soil bacterial biomass
(Figure 4)
4.1
NA
NA
60.7
Faunal diversity
66.0
ET ? maximum daily
precipitation (Figure 5)
6.3
Soil organic
matter
1.6
73.8
Final models never had more than three predictors. Stepwise regression did not include an additional predictor in a model unless inclusion caused a
significant increase in R2(1% level, F‐tests). ET, maximum daily precipitation, soil bacterial biomass and soil organic matter are defined in Table 1. ET ? y
denoted interaction terms between ET and y. NA ¼ Not Applicable, because no significant third predictor was selected. Total R2may differ from sum of
component R2s due to rounding

Page 15

Comp. by: rmervinRevises10000890258 Date:25/2/09 Time:02:56:13 Stage:1st
Revises File Path://pchns1301/WOMAT/Production/PRODENV/0000000001/
0000012327/0000000016/0000890258.3D Proof by:
cronym:bs:AECR Volume:41002
QC by:ProjectA-
CORRECTED PROOF
For log faunal biomass, models using both human‐use and carbon resource
predictors explained significantly or marginally significantly more variation in
bothdescriptorsthanmodelswithhuman‐usepredictorsalone(R2¼ 58.7%vs.
63.7%, F7,124¼ 2.4, p ¼ 0.022), and also more variation than models with
carbon resource predictors alone (R2¼ 38.4% vs. 63.7%, F10,124 ¼ 8.7,
p < 0.0005).
(F6,111¼ 7.6, p < 0.0005). The same was true with marginal significance for
abundance–mass slope (61.4% vs. 55.3%, F6,111¼ 3.0, p ¼ 0.01), and log
faunal biomass (67.9% vs. 62.9%, F6,111¼ 2.9, p ¼ 0.012).
2. Human‐Use Variables Were More Important Than
Environmental Variables
Human‐use variables substantially influenced soil faunal descriptors even
after controlling for the influence of environmental variables. Environmental
effects on descriptors were mediated by human‐use factors. In detail, human‐
use variables explained more variation in each soil faunal descriptor than did
environmental variables (faunal diversity: 65.5% vs. 45.0%; abundance–mass
slope: 51.4% vs. 34.1%; log faunal biomass: 59.9% vs. 49.5%). Human‐use
and environmental variables together explained significantly more variation
in each soil faunal descriptor than environmental variables alone (F‐tests,
p < 0.0005 for all three), but not significantly more than human‐use variables
alone (F‐tests, p > 0.042 for all three).
For the model including all human‐use and environmental predictors, R2
values for each soil faunal descriptor were: for faunal diversity, 67.3%; for
abundance–mass slope, 58.8%; for log faunal biomass, 63.2%.
3. Comparison Between Human‐Use and Carbon
Resource Variables
Human‐use variables substantially influenced abundance–mass slope and
faunal diversity after controlling for carbon resource variables. The effects
of carbon resource variables on abundance–mass slope and faunal diversity
were mediated by human‐use factors. In detail, human‐use variables
explained more variation in these two descriptors than did carbon resource
variables (abundance–mass slope: 50.7% vs. 37.7%; faunal diversity: 65.4%
vs. 33.0%). Human‐use and carbon resource variables together explained
significantly more variation in each of these two descriptors than carbon
resource variables alone (F‐tests, p < 0.0005), but not significantly more than
human‐use variables alone (F‐tests, p > 0.113). For the model including
all human‐use and carbon resource predictors, R2values were 68.4% for
faunal diversity and 53.1% for abundance–mass slope.
ABUNDANCE–MASS ALLOMETRY IN FOOD WEBS
59