APPLICATION OF DQO TO BENTHOS DATA
Application of the Great Lakes
National Program Office's Data
Quality Objective to Benthos Data
Generated by the Annual Water
Quality Survey
Richard P. Barbiero
CSC
1359 West Elmdale Avenue
Suite #2
Chicago, Illinois 60660
Prepared for:
United States Environmental Protection Agency
Great Lakes National Program Office
77 West Jackson Boulevard
Chicago, Illinois 60604
Louis Blume, Project Officer
February 2003
-------
APPLICATION OF DQO TO BENTHOS DATA
-------
APPLICATION OF DQO TO BENTHOS DATA
Acknowledgements
This report was prepared under the direction of Louis Blume, Project Officer, Great Lakes Na-
tional Program Office (EPA Contract No. 68-C-01-091).
-------
APPLICATION OF DQO TO BENTHOS DATA
Summary
A Data Quality Objective (DQO) has been developed by the Great Lakes National Program Office
(GLNPO) to ensure that data collected from their Water Quality Surveys are of suitable quality to
provide decision makers with sufficient certainty to make educated ecological management deci-
sions. The current GLNPO DQO states that data quality should be sufficient for there to be an
80% chance of detecting a 20% change, at the 90% confidence level, between current and historical
measurements of a variable made in a particular lake during a particular season.
This report assesses the extent to which benthos data collected in summer, 1999 comply with the
GLNPO DQO. The most important findings are summarized below:
Sufficient inter-site variation exists between offshore stations within each lake basin to
consider each station representative of a separate statistical population, rather than a repli-
cate of a larger, basin-wide statistical population.
When densities of the most dominant species/taxonomic groups were examined, the tar-
get contained in the DQO was by and large not met by the present sampling effort.
Sample sizes required to meet the DQO when just the most dominant species/taxonomic
groups are considered are unfeasibly large (> 12 in almost all cases; > 35 in the majority of
cases).
Variation was in some cases higher at nearshore than offshore stations. However, because
of the higher overall densities of organisms at nearshore stations, sample sizes required to
satisfy the DQO tended to be slightly higher at deeper stations.
The current detection target of a 20% change is much lower than actual interannual differ-
ences seen in Diporeia densities between 1997 and 2001, even where no consistent trends
(e.g., declines) were noted. Changes this small are therefore likely to be within the range of
natural fluctuation for most benthic organism populations, and as such are probably of
limited inherent ecological interest or use.
In spite of the inability to meet the DQO, the current level of replication was sufficient to
detect interannual changes in Diporeia densities from 1997 to 2001 at almost all sites for
which data were available.
In general, the current DQO is ambiguous with regard to specifically which differences are
of interest. Several interpretations are possible, each requiring a different statistical ap-
proach.
The current DQO does not address which biological variables are of interest, and there-
fore which should be subject to its specifications.
4
-------
APPLICATION OF DQO TO BENTHOS DATA
1 Introduction
1.1 Benthos program
The Great Lakes National Program Office of
the US EPA has been involved in regular sur-
veillance monitoring of the open waters of the
Laurentian Great Lakes since 1983. This sur-
veillance monitoring is meant to satisfy the
provisions of the Great Lakes Water Quality
Agreement (International Joint Commission
1978), which calls for periodic monitoring of
the lakes to evaluate water quality trends over
time. In 1997, a benthic invertebrate monitor-
ing program was added to complement
GLNPO's existing open water program. This
program differs from the open water surveys
in a number of important ways. Given the
homogeneity of most of the open waters of
the Great Lakes, the open water survey is de-
signed to detect changes on a fairly large spa-
tial scale. The statistical populations which
the sampling program of this survey were de-
signed to estimate correspond to lake basins,
thus the half dozen or so sampling stations es-
tablished within each basin serve as replicates.
In contrast, the benthic program originally
employed a sampling strategy designed to
characterize communities from two habitat
types, the nearshore (<50 m) and offshore
(>50 m). The rationale behind this design was
that the offshore benthic communities would
serve as integrators of conditions on a larger,
basin-wide scale, while nearshore locations
would exhibit a stronger dependence on local
conditions and offer a better indication of
relatively short-term responses to local stress-
ors. While representative stations were estab-
lished in both offshore and nearshore loca-
tions, coverage was not sufficient to provide
replication within even the larger, basin-wide
areas. Instead, replication was introduced at
the level of each station, and therefore the sta-
tistical populations in question coincide
roughly with the immediate area of each sam-
pling station. This proved to be particularly
fortuitous since subsequent data has indicated
that substantial biological changes can and do
occur in the offshore at spatial scales consid-
erably smaller than whole lake basins.
1.2 Objectives of study
The primary goal of this study was to deter-
mine if GLNPO's DQO is being met with the
current level of sampling effort in the benthic
program. Specifically, the goals of this study
were several fold:
To assess whether inter-site variability
at offshore stations precludes replica-
tion on a basin-wide basis;
To determine the minimum detectable
differences under the current sampling
regime;
To determine the sample sizes re-
quired to meet the DQO;
To assess the current DQO in relation
to the magnitude of variability seen in
benthos data.
In addition, different possible interpretations
of the GLNPO DQO, and its general suitabil-
ity for biological data will be discussed.
2 GLNPO's Data Quality
Objectives
2.1 DQO for GLNPO water quality sur-
vey
In order to assess lake health using data gener-
ated from the benthic program, or from any
aspect of the water quality survey, sufficient
data quality must be obtained to permit detec-
tion of 'significant' changes in these variables.
For the purposes of the water quality surveys,
GLNPO has defined a significant change as a
20% difference between current and historical
measurements, made for a particular variable
-------
APPLICATION OF DQO TO BENTHOS DATA
in a particular lake during a particular season.
Data quality should be sufficient for there to
be an 80% chance of detecting such a change
at the 90% confidence level.
2.2 Application of DQO to benthic data
As currently formulated, GLNPO's DQO
might not be strictly applicable to the benthic
program since, as noted above, data generated
from nearshore stations were only intended to
be representative of local conditions, while
even offshore stations might capture too
much site-specific variation to be usefully
pooled to represent broader, basin- or lake-
wide areas. In this case, changes might be
more profitably assessed on a station by sta-
tion basis, rather than on a whole-lake, or even
basin-wide, basis.
Additionally, application of the DQO requires
an explicit statement of how the benthos data
are to be used to assess differences between
current and historical measurements. As de-
tailed in the next section, 'historical measure-
ments' can be interpreted in a number of dif-
ferent ways. While perhaps these differences
might seem to be largely semantic, the ramifi-
cations of the different interpretations can
dramatically change the fundamental questions
being asked by the monitoring program, and
the statistical techniques used to answer those
questions. By referring to difference between
current and 'historical' measurements, rather
than, say, year to year differences, the current
DQO implies that historical measurements
refer to the pooling of all past data. In this
case, changes in a given variable in the past
would contribute to variability in the historical
measurements, and hinder the detection of
further changes without continual increases in
sample size. If changes are not unidirectional
(e.g., are cyclical), then pooling historical data
could completely preclude detection of further
changes. Two other possible interpretations
of the DQO include the ability to detect dif-
ferences between any two (or more) years, and
the ability to detect directional changes
(trends). The statistical implications of each
of these interpretations is discussed in the next
section.
2.3 Choice of response variables
While a precise definition of historical meas-
urements is necessary before the data can be
assessed for its ability to satisfy the stated
DQO, so is a precise definition of the re-
sponse variable. Unlike most chemical data,
biological data, including that generated from
the benthos program, is multivariate, and
therefore offers a number of potential re-
sponse variables. For instance, changes in the
densities of individual species can be assessed,
potentially limited to either specific indicator
species or to dominant species. In this case,
determination of the ability of the data to sat-
isfy the DQO would require multiple analyses,
one for each species of interest. The total
density of the benthic community could be
also be used, or the total densities of individu-
als within more broadly defined taxonomic
categories (e.g., oligochaetes, chironomids,
etc.). Alternatively, community-level metrics,
such as diversity or species richness, could be
used. Finally, specific indicators of benthic
community structure, such as the Milbrink
(Milbnnk, 1983), Goodnight and Whitley
(Goodnight and Whitley, 1960), or Brinkhurst
(Brinkhurst, 1967) oligochaete indices, could
be employed. Presumably the variables to be
subjected to the DQO criterion should be
ones which are conceptually tractable, and for
which changes would have some understood
ecological meaning.
-------
APPLICATION OF DQO TO BENTHOS DATA
3 Possible Interpretations of
GLNPO's DQO
Determination of the adequacy of the current
benthos sampling regime in satisfying
GLNPO's DQO involves an assessment of
the precision in estimating the mean (fj,) of the
statistical population in question. The preci-
sion of a given statistical estimate is affected
by the natural variation, or variance (o?\ of
the population under study, and the sampling
variability. Since natural variation is not under
investigator control, increases in precision can
only be effected through decreases in sam-
pling variability, which is mainly accomplished
by increasing sample size (»), although theo-
retically improvements in sampling methodol-
ogy can also result in reduced sampling vari-
ability. If a desired precision in estimating a
parametric mean is known in advance, as is
the case with a DQO, and the desired prob-
ability of attaining that precision for a given a
is also specified, then the number of samples
needed to achieve the desired precision can be
calculated according to the statistical test to be
used.
GLNPO's DQO addresses the detection of a
change in a population mean, relative to an
historical value. The precision with which a
change can be detected will be a function not
only of the number of observations in the cur-
rent sample, but also of the number of obser-
vations making up the historical sample. The
adequacy of a given sample size to detect such
a change will therefore depend upon the con-
stitution of the historical sample, and this in
turn depends upon the exact definition of the
'historical sample', and on the statistical test
which is to be used to assess differences be-
tween it and the current sample. A number of
possible interpretations of the DQO exist, and
three of the most likely are outlined below,
along with the statistical considerations in-
volved in their assessment.
3.1 Detection of differences between
current year and all previous years
combined
In the simplest scenario, the current sample is
compared to all past samples pooled. This
corresponds to the most literal interpretation
of GLNPO's DQO, and it assumes that all
past samples estimate the same population
mean, i.e., that there have been no changes in
the variable of interest prior to the current
year. In this case, the appropriate formula for
determining sample size is:
n = -
where: Sp2 =
d =
sample estimate of pooled
population variance; and
the minimum detectable differ-
ence specified by the DQO.
The assumption that there have been no
changes in past years would require statistical
testing to assess, and therefore a de facto testing
for interannual differences between all years
would be required under this interpretation of
the DQO, whether or not that testing were,
strictly speaking, being used to assess differ-
ences of interest (i.e., differences between the
current year and all previous years combined) .
Furthermore, the assumption that all historical
data are statistically the same contains an ele-
ment of self-contradiction, since by definition
each current year's data becomes historical
data with the collection of the next year's data.
3.2 Detection of changes between any
year
A second, and more likely, interpretation of
the DQO is that the detection of any interan-
nual differences that have occurred over the
course of the monitoring program is of inter-
est. In this case, each year's data would con-
7
-------
APPLICATION OF DQO TO BENTHOS DATA
stitute a different 'treatment', and a one-way
ANOVA, or equivalent, would be used to as-
sess differences between 'treatments' (i.e.,
years). The sample size necessary to detect a
specified difference (8) in mean values can be
calculated using the following equation:
where: k =
in
8 =
number of treatments
(in this case years)
number of observations
each treatment
within groups variance
the minimum detectable differ
ence, specified by the DQO
(f> is a parameter related to the power of the
performed test. This equation is solved itera-
tively, inputting successive estimates of n until
a (p corresponding to the desired 1-|3 is
achieved.
Alternatively, the following equation can be
used to test for the minimum detectable dif-
ference with a specified n:
2 J.2
5 =
2ks2>
n
Under this scenario, the target minimum de-
tectable difference of 20% would apply to
changes seen between any two years, even
non-consecutive years. Of course with an
analysis of variance procedure, exactly which
treatments (in this case years) are different
from which others are not specified; what can
be concluded from the analysis is only
whether or not any of the treatments are dif-
ferent from any other treatments.
3.3 Detection of trends overtime
A third possible interpretation of the DQO
would be to enable detection of a trend in the
variable of interest, in other words, to test for
a change over time. This can be done in one
of two ways: through the use of a regression
analysis, with the variable of interest regressed
against time (i.e., year), or by using a correla-
tion analysis, again with the variable of interest
correlated with time. In the first case, the mag-
nitude of change over time can be assessed,
specifically by examination of the regression
coefficient, b. Simple linear regression, how-
ever, has at least two drawbacks in the context
of the current application. First, it assumes a
cause and effect relationship between the in-
dependent and dependent variables. Strictly
speaking, it is unlikely that time in and of itself
would be the causative factor in increases or
decreases in benthos densities. Second, and
more importantly, it assumes that the depend-
ent variable (e.g., benthos density at a site) has
a constant, linear relationship to the independent
variable (time). If the detection of any change
in a variable over time is of interest, and not
just a strictly linear one, then linear regression
analysis is an inappropriate tool.
Alternatively, correlation analysis can be used
to assess whether changes in the magnitude of
one variable are associated with changes in the
magnitude of a second variable. In correlation
analysis, no cause and effect relationship is as-
sumed. However, in simple linear correlation
analysis, a linear relationship between the two
variables of interest is assumed. This restric-
tion can be circumvented by use of a rank-
based correlation procedure, such as the
Spearman rank correlation procedure. With
this procedure, data are converted to ranks
prior to conducting the analysis, so that only
trends in the two variables are assessed, and
not quantitative changes of one in relation to
the other variable. So in this case, the magni-
tude of the change is not assessed, rather the
strength of association between the two variables
is. For this reason, the GLNPO DQO as cur-
rently formulated cannot be applied to an
analysis of this nature, since the DQO ad-
8
-------
APPLICATION OF DQO TO BENTHOS DATA
dresses a specific magnitude of change (i.e., a
20% change over some historical value).
For the purposes of this report, it was as-
sumed that the GLNPO DQO requires the
ability to detect a 20% change in a variable be-
tween any two (or more) years for which data
are available. The approach adopted here,
therefore, follows that outlined above in sec-
tion 3.2.
4 Overview of Current
Benthos Methods
In 1999, the most current year for which full
benthos data is available, a total of 53 stations
were visited during the summer survey (Figure
1). Between seven (Lake Erie) and fourteen
(Lake Michigan) stations were visited in each
lake. At each station, three replicate samples
were collected, using a Ponar grab sampler.
Samples were sieved through a 500-|J,m sieve
in the field, preserved, and transported to the
laboratory. In the laboratory, all animals were
removed from any remaining sediment under
a dissecting microscope and sorted by major
taxonomic group. All oligochaetes and chi-
ronomids were mounted and identified under
a compound microscope; all other inverte-
brates were identified and counted under a
dissecting microscope. Counts were con-
verted into areal units (#/m2) by multiplying
by 19.12, a factor which takes into account the
area sampled by the Ponar.
5 Approach
In all cases, data from 1999 were used, since
this is the most recent year for which com-
plete benthos data are available. For the pur-
poses of this study, offshore stations are de-
fined as those with a depth > 64 m. This is a
slightly narrower definition than the one origi-
nally employed in the design of the benthic
survey, and was adopted largely because of the
high percentage of stations with depths within
a few meters of 50 m, the original demarcation
between near- and offshore stations. The
variables considered for offshore stations in
this study were areal densities (#/m2) of the
total benthos community, the total oligochaete
community (excluding fragments, but includ-
ing immatures), and the two common profun-
dal organisms Diporeia and Stylodrilus. Because
of greater species richness, additional variables
were examined at nearshore stations. These
included total areal densities of the Naididae,
the Tubificidae, the Chironomidae, and the
Sphaeriidae. Sites were excluded from analysis
where numbers were extremely small, or
where organisms were completely absent from
one or more replicate.
Since the original rationale of the sampling de-
sign was that offshore stations within a lake
would be estimating the same statistical popu-
lation, an initial analysis was undertaken to as-
sess within lake differences in each of the
above four variables. These potential differ-
ences were assessed using a one way analysis
of variance (with stations as a random factor).
Where assumptions of homoscedasticity or
normality were not met, a Kruskal-Wallis one
way analysis of variance on ranks was per-
formed.
Minimum detectable differences were deter-
mined for all variables tested at all sites using
the equation in section 3.2. Data were not
tested for homoscedasticity or normality prior
to computation of minimum differences, nor
were any transformations used. It is recog-
nized that in some cases these assumptions
were probably not met. However, given the
number of analyses, it was not deemed feasi-
ble to assess the need for transformations in
each case, particularly given the robustness of
a one way ANOVA, the statistical test upon
which most calculations were based. Mini-
mum detectable differences were determined
-------
APPLICATION OF DQO TO BENTHOS DATA
Offshore Benthos Sites
Nearshore Benthos Sites
Fig. 1. Locations of GLNPO's water quality survey (WQS) benthos sampling stations for summer,
1999 survey. Sites considered offshore are indicated in blue; sites considered nearshore are indi-
cated in red. .
for n (i.e., number of replicates taken) - 3, 4
and 5, and for k (i.e., number of years com-
pared) = 2 through 9. In all cases, |3 was taken
to equal 1-0.80 and a was taken to equal 0.05.
The most recent GLNPO DQO assumes an a
= 0.10; however, tabled values for (f) given a =
0.10 and varying degrees of freedom could not
be found, so the more commonly used a of
0.05 was used instead. Therefore, estimates of
minimum detectable differences will be some-
what more conservative than are required by
the current DQO. General trends, however,
should not be affected by the difference in a.
The adequacy of the current sampling regime
in satisfying the GLNPO DQO was examined
assuming that any between-year differences
are of interest. Natural variation at each site
was estimated using sample variance calculated
from 1999 data. The number of replicates re-
quired to satisfy the stated GLNPO DQO was
calculated for each variable according to the
equation in section 3.2, using the mean value
of the variable in 1999 as the value in relation
to which a 20% change should be detectable,
and assuming a comparison of 5 years, a =
0.10 and p = 1-0.80.
6 Results
6.1 Within Lake Variation in Offshore
Sites
ANOVA analyses were conducted on off-
shore sites within each lake to determine if
10
-------
APPLICATION OF DQO TO BENTHOS DATA
Table 1. One way analysis of variance results examining between site differences in Lakes Huron,
Michigan, Ontario and Superior for offshore stations. For all tests, a = 0.05. Where assumptions
of normality or homos cedasticity were not met, the Kruskall-Wallis one way analysis of variance on
ranks was used.
LAKE
Z Benthos Z Oligochaetes
Diporeia
Stylodrilus
HURON
MICHIGAN
ONTARIO
SUPEROR
5.08
94.6
10.6*
28.7
0.01
<0.001
NS
<0.001
4.35
24.7
13.5*
26*
0
<0
0
0,
.01
.001
.02
.002
5,
55,
38,
26,
.32
.2
.1
.5*
0,
<0.
<0
0,
.01
.001
.001
.002
17.6*
14.2
15*
23.3*
0.01
<0.001
0.01
0.01
Kruskal-Wallis One Way Analysis of Variance on Ranks used: statistic - H.
these sites should serve as replicates within
each lake. Variables tested were densities of
total benthos, total oligochaetes, Diporeia and
Stylodrilus. For all variables tested, with the
sole exception of total benthos densities in
Lake Ontario, significant differences were
found between sites at a = 0.05 (Table 1).
Therefore all sites were considered separately
in subsequent analyses.
6.2 Comparison of variability between
offshore and nearshore sites
To determine if within site variability was re-
lated to depth, i.e., if nearshore sites were
more or less variable than offshore sites, the
standard deviation of total benthos estimates
for individual sites was plotted against depth
for all sites. The more extreme standard de-
viations were associated with shallower sites,
although not all shallow sites exhibited a large
standard deviation (Figure 2). Since the
GLNPO DQO is formulated in terms of per-
cent, rather than absolute, change, though,
high standard deviations would not necessarily
result in large required sample sizes if those
standard deviations were associated with high
values of the variable of interest. Benthos
densities do in fact tend to be higher at shal-
lower sites, and this might counteract the
higher variances seen at shallow sites.
The coefficient of variation:
X
provides a measure of the variance relative to
the mean, and when this was plotted against
depth, no relationship was seen (Figure 2).
This indicates that the higher standard devia-
tion of some shallow sites was due to higher
total numbers.
When the natural logarithm of the sample size
required to satisfy the DQO for the estimation
of total benthos at each site was plotted
against site depth, an apparent positive rela-
tionship was seen (Figure 3). This relationship
was just statistically significant at a = 0.10, as
determined by least squares regression (F =
2.77, P = 0.10). A more significant relation-
ship was found between the sample size re-
quired to estimate total numbers of oli-
gochaetes and depth (F = 5.59, P = 0.02). In
both cases depth explained a relatively low
percentage of the variance in required sample
size (total benthos: R2 = 0.03; total oli-
gochaetes: R2 = 0.09). This suggests that, on
the whole, deeper sites tend to require a
greater degree of replication to enable detec-
tion of a 20% change for these variables than
11
-------
APPLICATION OF DQO TO BENTHOS DATA
4000
3000 -
c
o
> 2000
Q
CD
T3
C
JD
CO
1000 -
Ontario
Erie
O Michigan
O Huron
O Superior
oo o
oo .6*
o
50
100
^50
Depth (m)
100
80-
o
"CD
-£ 40 -
CD
O
200
-O
250
300
O
0
0
O
O
O O
o
0
(D
50
100
^50
Depth (m)
200
250
300
Figure 2. Standard deviation (top panel) and coefficient of variation (bottom panel) of estimates of
total benthos densities at each site, graphed against site depth. Variance estimates based on 1999
values.
do shallower sites, although this tendency is
not marked. As noted, this is due in large part
to the smaller densities of organisms seen at
deeper sites. Such a relationship was not
found, however, between sample sizes re-
quired to satisfy the DQO for densities of Di-
poreia and station depth (Figure 3). The prob-
able reason for this is that, unlike total ben-
thos and total oligochaete densities, Diporeia
numbers tend to increase with depth.
12
-------
APPLICATION OF DQO TO BENTHOS DATA
1000
Total Oligochaetes
Ontario
Erie
O Michigan
O Huron
O Superior
300
300
50
100
150
200
250
300
Depth (m)
Figure 3. Relationship between sample size required to satisfy DQO and site depth for estimation
of total benthos, total oligochaete and Dipomaareai densities. In all cases, k = 5, a = 0.10 and |3 =
1 - 0.80.
13
-------
APPLICATION OF DQO TO BENTHOS DATA
6.3 Minimum detectable differences for
interannual comparisons at offshore
sites
Average lake-wide minimum detectable differ-
ences for the current level of replication, as-
suming a comparison of 5 years of data and a
= 0.05, were less than 1,000 individuals/m2 in
all cases except in Lake Huron, where the av-
erage minimum detectable difference for Dio-
reia was 1,692/m2 and for the total benthos
was 2,343/m2 (Table 2). This is reflective of
the generally higher variances, and hence
higher minimum detectable differences, seen
at the offshore sites in that lake compared to
the other lakes. Average minimum detectable
differences for the four variables were broadly
similar in Lakes Ontario and Michigan, and
somewhat lower in Lake Superior, due to the
lower densities of organisms in that lake.
Since the GLNPO DQO is stated in terms of
a percent, rather than an absolute, change, a
more relevant measure of the adequacy of the
current sampling program in fulfilling the re-
quirements of the DQO is minimum percent
detectable difference. In most cases, the
Table 2. Maximum, minimum and average minimum detectable differences (#/m2) and percent
detectable differences for offshore (z > 64 m) stations in Lakes Michigan, Huron, Ontario and Su-
perior, under the current sampling regime. In all cases, n = 3, k = 5, a = 0.05 and |3 = 0.80.
LAKE Minimum Detectable
Variable Max Min
MICHIGAN
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
HURON
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
ONTARIO
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
SUPERIOR
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
1,144
1,109
1,331
1,040
3,055
2,289
2,093
5,407
1,020
1,570
1,954
2,084
1,513
1,396
1,454
1,791
314
157
157
595
491
104
68
579
343
39
39
245
68
104
39
68
Difference
Ave
794
580
589
864
1,692
651
965
2,343
741
443
443
951
437
539
531
855
Percent Detectable Difference
Max Min Ave
89%
617%
178%
81%
175%
310%
206%
166%
72%
168%
392%
127%
356%
592%
617%
432%
34%
46%
27%
24%
22%
21%
14%
15%
40%
62%
21%
32%
30%
44%
35%
45%
52%
188%
88%
41%
108%
165%
152%
95%
61%
123%
143%
71%
168%
236%
253%
169%
14
-------
APPLICATION OF DQO TO BENTHOS DATA
I
I
!§
T3
.0
8
CD
T3
E
c
2000
1800 \
1600
1400
1200
1000
800
Diporeia
n = 3
n = 4
n = 5
2600
2400 J
2200
2000
1800
1600
1400
1200
Total Individuals
2 4 6 8 10
Number of years compared
2468
Number of years compared
10
CD
O
c
.
I
Q)
.0
.2
"O
£
"CD
T3
E
'c
800
2 4 6 8 10
Number of years compared
2 4 6 8 10
Number of years compared
Figure 4. Average minimum detectable differences in areal densities of Diporeia, total benthos, total
oligochaete and Stylodrilus at offshore sites in Lake Huron for different numbers of replicates (ri)
when comparing 2-9 years of data. Variance estimates are based on 1999 data.
smallest percent difference that the current
sampling regime permits detection of was sub-
stantially larger than 20%, reaching several
hundred percent in a large number of cases
(Table 2). In only two cases was the current
level of replication adequate to permit detec-
tion of a 20% change: given estimates of vari-
ance from 1999, current levels of replication
permit detection of 14% and 15% changes in
total oligochaete and total benthos densities,
respectively, at station HU 93 in Lake Huron.
Average minimum percent detectable differ-
ences were highest in Lake Superior; this was
due to not to higher variance, but rather to
lower overall densities of organisms in the
lake. Thus, while in general smaller differ-
ences are able to be detected in Lake Superior,
the low densities seen in the lake counteract
this to make minimum percent detectable dif-
ferences larger than in the other lakes.
15
-------
APPLICATION OF DQO TO BENTHOS DATA
CD
O
.0
CO
t5
CD
-t->
CD
T3
E
'c
1300
1200
1100 -
1000 -
900
800 \
700
600
500
Diporeia
n = 3
n = 4
n = 5
1000
900
800
700
600 -
500 -
400
Total Individuals
2 4 6 8 10
Number of years compared
2468
Number of years compared
10
8
(D
.Q
CO
700
400
CD
T3
|
1
i 300
Total oligochaetes
£ 600 -
500 -
700
600
500
400
300
2 4 6 8
Number of years compared
10
Stylodrilus
2468
Number of years compared
10
Figure 5. Average minimum detectable differences in areal densities of Diporeia, total benthos, total
oligochaete and Stylodrilus at offshore sites in Lake Michigan for different numbers of replicates (n)
when comparing 2-9 years of data. Variance estimates are based on 1999 data.
Minimum detectable differences were calcu-
lated for interannual comparisons in which
between 2 to 9 years of data were compared,
and with 3, 4 and 5 replicates per site, for a =
0.05. For any given level of replication, mini-
mum detectable differences increase as the
number of years in the comparison increases
(Figures 4-7). In general, increasing the
number of replicates from 3 to 4 resulted in a
decrease in minimum detectable differences of
approximately 17-23%, while increasing repli-
cation to 5 further reduced minimum detect-
able differences by approximately 10-15%. In
absolute terms, taking 4 replicate samples at
each site reduced the minimum detectable dif-
ference by an average of 170 individuals/m2
for the four variables considered (range: 88 -
470/m2), while 5 replicate samples per site re-
duced the minimum detectable difference by
an average of 259/m2 (range: 133 - 714/m2),
compared to 3 replicates (Table 3).
16
-------
APPLICATION OF DQO TO BENTHOS DATA
0
o
!§
T3
0
-I»
0
T3
E
3
E
1000
900
800
700
600
500
400
300
Diporeia
n = 3
n = 4
n = 5
1200
1100
1000 -I
900
800 -I
700
600 -I
500
400
Total Individuals
2 4 6 8 10
Number of years compared
2468
Number of years compared
10
0
o
c
0
0
m
0
-i»
0
T3
|
E
c
600
550
500
450
400
350
300
250
200
Total oligochaetes
450
150
2 4 6 8
Number of years compared
10
2468
Number of years compared
10
Figure 6. Average minimum detectable differences in areal densities of Diporeia, total benthos, total
oligochaete and Stylodrilm at offshore sites in Lake Ontario for different numbers of replicates (ri)
when comparing 2-9 years of data. Variance estimates are based on 1999 data.
17
-------
APPLICATION OF DQO TO BENTHOS DATA
8 600
Number of years compared
2 4 6 8 10
Number of years compared
CD
o
c
.
CD
t
T3
JD
.Q
fi
£
CD
T3
E
D
E
c
600
550
500
450
400
350
300
250
200
Total oligochaetes
800
750
700
650
600
550
500
450
400
350
300
2468
Number of years compared
10
Stvlodrilus
2468
Number of years compared
10
Figure 7. Average minimum detectable differences in areal densities of Diporeia, total benthos, total
oligochaete and Stylodrilus at offshore sites in Lake Superior for different numbers of replicates (»)
when comparing 2-9 years of data. Variance estimates are based on 1999 data.
18
-------
APPLICATION OF DQO TO BENTHOS DATA
Table 3. Estimated average minimum detectable differences (#/m2) for off-
shore (z > 64 m) stations in Lakes Michigan, Huron, Ontario and Superior, for
samples sizes of 3, 4 and 5. In all cases, k = 5, a = 0.05 and |3 = 1-0.80.
LAKE
Variable
MICHIGAN
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
HURON
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
ONTARIO
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
SUPERIOR
Diporeia
Stylodrilus
Z Oligochaetes
Z Benthos
Sample Size
345
794
580
589
864
1,692
651
965
2,343
741
594
443
951
437
539
531
855
635
464
471
691
1,353
520
771
1,873
593
474
354
760
349
431
424
684
552
403
409
601
1,176
452
671
1,629
515
413
308
661
304
375
369
595
6.4 Required sample sizes for offshore
sites
The number of replicates needed to detect a
20% change in the four variables tested, as-
suming a five year comparison, ranged from 3
for the site in Lake Huron noted above (HU
93), to 702 to detect a 20% change in total
benthos densities at site SU 16 in Lake Supe-
rior (Figure 8). The median number of repli-
cates needed to meet the DQO for all vari-
ables was 39, while fully 80% of the variable/
site combinations tested would require at least
12 replicates to satisfy current the DQO.
However, since each variable is not sampled
independently at each site, the number of rep-
licate samples required on a site by site basis
would be determined by the variable requiring
the greatest number of replicates at that site to
satisfy the DQO. Examined on this basis, the
median number of replicates at offshore sites
needed to meet the DQO was estimated to be
97, while 80% of sites would require 19 repli-
cates or more. These estimates obviously rep-
resent an impractical degree of replication,
both from the standpoint of ship time re-
quired to amass such a number of samples,
and from the standpoint of time required for
sample analysis.
19
-------
APPLICATION OF DQO TO BENTHOS DATA
1000
CD 100
^ N
^W
'M
o-o.
CD E
o: ro
w
S Benthos
S Oligochates
Diporeia
Stylodrilus
HU32 HU38 HU 48 HU 54
HU61
HU 93 HU 95
1000
CD 100
N
-------
APPLICATION OF DQO TO BENTHOS DATA
6.5 Minimum detectable differences for
interannual comparisons at nearshore
sites
Minimum detectable differences were calcu-
lated for each nearshore site, assuming a com-
parison of five years and a = 0.05, for the fol-
lowing six variables: areal densities of total
benthos, total oligochaetes, total Tubificidae,
total Chironomidae, Diporeia and Sphaeridae.
In only five cases was the current level of rep-
lication sufficient to detect a 20% difference
when comparing five years of data (Figure 9).
Most values for Lakes Erie, Huron and Michi-
gan were between 50 and 100% (medians =
65%, 59% and 79%, respectively), while mini-
mum detectable differences were somewhat
higher in Lake Ontario (median = 166%). Of
the three nearshore stations in the latter lake,
two exhibited substantial variability in all vari-
ables examined. No substantial differences in
minimum detectable differences were appar-
ent between the different variables.
6.6 Required sample sizes for near-
shore sites
The number of replicates needed to detect a
20% change in the six variables tested, assum-
ing a five year comparison and a = 0.10,
ranged from three for several variables at three
sites in Lakes Erie and Michigan, to 837 repli-
cates to estimate total chironomids at a station
in Lake Ontario (Figure 10). Overall, a me-
dian sample size of 25 was required when all
variables at all sites were considered, with 80%
of site/variable combinations requiring nine
replicates or more. When just the maximum
required sample size at each site for all six
variables was considered, which is a better in-
dication of the required site-wise level of repli-
cation, the median number of replicates
needed to satisfy the DQO at all sites was 77,
with 80% of the sites requiring 51 replicates or
more for the DQO to be met for all variables
tested. Given the greater number of variables
under consideration here, this is a more rigor-
ous test of the sampling program than was
used for offshore stations. Considering just
those variables that were also assessed in the
offshore sites, namely total benthos, total oli-
gochaetes and Diporeia (Stylodrilus does not ap-
pear in substantial numbers at shallower sites),
the median number of replicates needed to
satisfy the DQO at each site was 38, with 80%
of the sites requiring at least 13 replicates.
These numbers are lower than those seen in
the offshore sites. On average, then, near-
shore sites would require less replication to
satisfy GLNPO's DQO, compared to off-
shore sites. Even so, the level of replication
required at nearshore sites for acceptable esti-
mates of total benthos, total oligochaetes and
Diporeia densities, given the current DQO, are
in the vast majority of cases infeasible from a
practical standpoint.
7 Relation of DQO to 1997-
2001 Diporeia data
To determine the relation of the GLNPO
DQO to a benthic variable of ecological inter-
est, and for which data from 1997-2001 are
available, changes in the density of Diporeia
were examined. The amphipod Diporeia has
historically been one of the most abundant
and widespread benthic organisms in the
Great Lakes (Dermott and Corning, 1988) and
is an important link in the Great Lakes food
chain, feeding on pelagic-derived detritus
(Gardner et al., 1990) and in turn providing an
important food source for many fish species
(Scott and Grossman, 1973). Recently its
numbers have been declining in significant
portions of its range in the Great Lakes, with
potentially huge consequences for community
structure and energy flow through the Great
Lakes food web.
21
-------
APPLICATION OF DQO TO BENTHOS DATA
CD
o
I
CD
300 -
o 200 -
.0
8
CD
O
-*->
CD
a
CD
Q_
CD
O
100 -
Z Benthos
Z Oligochates
Z Tubificidae
Z Chironomidae
Diporeia
Sphaeridae
ER15
ER43
ER61
ER78
ER91
ER 93
ER 95
500
CD 40° -
b
0) 300 -
.a
CO
200 -
100 -
CD
Q
CD
Q_
CD
o
I
!§
b
o 200 -
.0
8
HU06
HU96
HU97
HU 98
ON 65
ON 67
ON 69
SU 22
300 -
CD
Q
-*->
CD
a
CD
Q.
100 -
Ml 30 Ml 31
42 Ml 46 Ml 48
Station
49 Ml 50 Ml 52 Ml 53
Figure 9. Minimum detectable differences, as a percent, for estimates of total benthos, total oli-
gochaete and total Tubificidea, total Chironomidae Diporeia and Sphaeridae areal densities. In all
cases, k = 5, a = 0.05 and P = 1 - 0.80. Reference line represents current data quality objective of
20%
22
-------
APPLICATION OF DQO TO BENTHOS DATA
1000
CD 100
o: ro
CO 10
E Benthos
E Oligochates
E Tubificidae
E Chironomidae
Diporeia
Sphaeridae
ER15 ER43
ER61
ER78 ER91 ER 93 ER 95
1000
HU 06 HU 96 HU 97 HU 98 ON 65 ON 67 ON 69 SU 22
1000
CD 100
N
a:
CO 10
Ml 30
52 Ml 53
Station
Figure 10. Sample sizes required at nearshore sites to detect a 20% change in total benthos, total
oligochaete, total Tubificidae, total Chironomidae, Diporeia and total Sphaeridae areal densities
when comparing 5 years, assuming a = 0.10 and |3 = 1 - 0.8. Variance estimates and initial densi-
ties are based on 1999 values. Reference line represents n = 3.
23
-------
APPLICATION OF DQO TO BENTHOS DATA
For this analysis, only those sites for which all
five years of data were available were used.
Within site differences in densities between
any two years were assessed using a one way
ANOVA with year as the factor. This corre-
sponds to the approach adopted above, and is
the interpretation of the DQO used through-
out this report. In accordance with the DQO,
a significance level of ot = 0.10 was used. In
the case of Diporeia, what is of interest is spe-
cifically a [decreasing] trend over time, rather
than any year to year differences. To test for
this, the non-parametric Spearman rank corre-
lation coefficient was calculated for each site.
While data from most sites conformed to the
assumptions of normality and homoscedastic-
ity, and in cases where these were not met, de-
viations were relatively small, a non-parametric
test was used for correlations to enable detec-
tion of any, rather than just linear, relation-
ships between year and density. For consis-
tency, a significance level of a = 0.10 was
used.
Both the maximum difference in areal densi-
ties between any two years, and the maximum
difference between any two consecutive years,
were calculated for each site. Maximum dif-
ferences between any two years ranged from
108/m2 to 10,510/m2, with a median value of
1,399/m2 (Table 4). Since the DQO is formu-
lated in terms of a percent, maximum percent-
age differences were also calculated, using a
conservative approach in which the larger of
the two numbers was used as the denomina-
tor. Thus the equation used was:
MaxAnnual - MinAnnual
MaxAnnual
Maximum percentage differences ranged from
43% to 100% (i.e., where the minimum annual
value = 0). The median percent difference
was 72%, and 80% of values were greater than
55%. Therefore, in all cases an interannual
difference substantially higher than the target
detection limit of the DQO was seen. Since
an alternate interpretation of the GLNPO
DQO requires the detection of a 20% change
between values in consecutive years, maxi-
mum differences between consecutive years
were also computed for all sites. In this case,
maximum differences ranged from 108/m2 to
5,487/m2, with a median difference of 1,316/
m2. These corresponded to percentage differ-
ences ranging from 29% to 100%, with a me-
dian of 64%. 80% of sites had a maximum
difference between consecutive years of 49%
or more. Thus all sites experienced interan-
nual changes in Diporeia density greater than
the 20% targeted in the DQO, even when
these differences were limited to those occur-
ring in consecutive years.
Using the criterion for significance stated in
the DQO (a = 0.10), 26 of the 30 sites exam-
ined were found to have statistically significant
interannual differences in Diporeia densities,
when assessed using a one way ANOVA
(Table 4). Since interannual differences can be
a consequence of natural fluctuation in popu-
lation sizes, rather than an indication of a
trend, which is presumably of greater inherent
interest, Spearman rank sum correlation coef-
ficients were also calculated. In this case, 15
of the 30 sites examined exhibited statistically
significant increasing or decreasing trends with
time, indicating that in 11 of the cases for
which interannual differences were detected,
these differences were not directional.
24
-------
APPLICATION OF DQO TO BENTHOS DATA
Table 4. Analysis of Diporeia areal densities at all sites for which five (1997 - 2001) years of data are
available. Maximum differences (A) in density estimates between any two years, and between any
two consecutive years, are shown as both absolute values (#/m2) and percentages. Results of one
way analysis of variance, with year as factor, and Spearman rank correlation analysis, are shown.
Significant results, determined at a = 0.10, are shown in bold.
Site
HU06
HU32
HU38
HU48
HU54
HU61
HU93
HU95
HU96
HU97
Mill
MI 18
MI 27
MI 41
MI 46
MI 48
MI 50
MI 52
ON 25
ON 41
ON 60
SU01
SU10
SUB
SU15
SU16
SU17
SU19
SU21
SU22
Maximum A
(any years")
3,996
3,060
1,358
2,830
4,871
1,759
2,435
2,282
3,295
1,574
1,396
3,811
4,512
5,481
1,294
918
1,434
10,510
778
1,377
1,402
535
255
185
433
153
108
223
166
1,211
100%
66%
73%
83%
85%
48%
79%
71%
86%
49%
63%
84%
71%
95%
43%
100%
100%
100%
57%
71%
55%
48%
63%
81%
68%
80%
85%
81%
49%
54%
Maximum A
(consecutive years")
2,428
1,791
1,358
1,740
2,014
1,128
1,848
2,263
1,402
931
1,396
2,855
4,512
4,640
1,294
918
1,338
5,487
720
1,377
1,134
351
249
178
433
121
108
204
166
1,052
61%
49%
73%
75%
70%
31%
65%
70%
53%
29%
63%
63%
71%
80%
43%
100%
93%
78%
53%
71%
45%
31%
62%
78%
68%
63%
85%
74%
49%
47%
ANOVA
F P
16.6
6.9
12.1
6.7
29.8
5.1
88.0
15.0
5.3
11.8
2.7
6.2
9.1
18.3
3.7
16.0
6.7
84.8
1.8
16.1
4.4
4.9
3.5
2.1
3.7
3.2
2.1
6.2
1.7
16.9
<0.001
0.006
<0.001
0.007
<0.001
0.017
<0.001
<0.001
0.015
<0.001
0.096
0.009
0.002
<0.001
0.042
<0.001
0.007
<0.001
0.214
<0.001
0.025
0.019
0.049
0.161
0.043
0.061
0.152
0.009
0.23
<0.001
Spearman Corr.
r, P
-0.95
1.78
-0.34
-0.46
-0.68
-0.74
-0.86
-0.31
-0.81
-0.35
-0.29
-0.69
-0.13
-0.74
0.53
0.17
-0.82
-0.90
-0.40
-0.18
-0.81
0.05
-0.29
-0.41
-0.22
-0.49
-0.22
-0.59
-0.20
-0.16
<0.001
<0.001
0.209
0.083
0.005
0.001
<0.001
0.257
<0.001
0.194
0.293
0.004
0.629
0.003
0.061
0.578
<0.001
<0.001
0.134
0.514
<0.001
0.842
0.287
0.124
0.410
0.059
0.426
0.020
0.457
0.566
25
-------
APPLICATION OF DQO TO BENTHOS DATA
8 Discussion
In almost all cases examined, the current level
of sampling effort in the benthos program is
inadequate to satisfy GLNPO's DQO, as
presently formulated. Even when just consid-
ering the limited number of variables used in
this report, the great majority of sites would
require over a dozen replicate Ponar samples
to meet the current DQO, and over half
would require more than three dozen repli-
cates. This degree of sampling effort is clearly
infeasible in the content of the GLNPO
monitoring program, and it is in fact unclear
the extent to which it is required to make edu-
cated ecological management decisions using
the benthos data.
A number of ambiguities exist in the current
DQO, some of which are specific to its appli-
cation to the biological program, and some of
which are more general. As noted in the in-
troduction, the current DQO does not make
entirely clear what should be compared when
assessing changes in the magnitude of a vari-
able. For the most part in this report it was
assumed that what the DQO required detec-
tion of was a change between any two years of
data. However, as was seen with Diporeia, a
20% fluctuation in density between two years
was the rule, rather than the exception, and
while this was due in part to actual basin wide
declines seen in populations of this organism
in recent years, even those sites which did not
exhibit a trend (i.e., for which rs was not sig-
nificant) showed differences between years
that were several times greater than 20%. This
suggests that the normal range of natural fluc-
tuations in biological populations can be con-
siderably higher than the target difference
specified in the DQO, and therefore that the
DQO is probably unnecessarily stringent for
biological data. In spite of not meeting DQO
requirements with regard to being able to de-
tect changes in Diporeia populations, the cur-
rent sampling program nonetheless found sta-
tistically significant differences in annual Di-
poreia densities at nearly every site for which
data were available. As suggested above, in at
least some cases these differences were proba-
bly just due to natural fluctuations, and would
not necessarily be indicative of an overall
change in the ecological character of the ben-
thos.
More restrictive interpretations of the DQO
would include the ability to detect a 20%
change between consecutive years, and being
able to detect a trend in a variable. In the for-
mer interpretation, as seen with the Diporeia
data, a 20% change might still be too stringent
a requirement for benthos data; nearly every
site examined exhibited a difference between
consecutive years at least twice this great. In
the case of the latter interpretation, unless a
linear trend were assumed (an assumption for
which there would be no apriori support), it
would be difficult, and likely counter produc-
tive, to set a specific criterion for the magni-
tude of the trend to be detected. Presumably
the detection of any trend would be of inter-
est. Instead, a DQO might best be stated in
terms of detecting a relationship between a
variable and time (e.g., year) of a given
strength.
The main deficiency in the current DQO, as it
relates specifically to biological data (including
the benthos data), is that the variable of inter-
est is not specified. When assessing changes
in the chemistry of a lake, determining the
variable of interest is usually relatively straight-
forward (though not necessarily; e.g., instances
involving detection of a large number of con-
geners of an organic pollutant). Such is not
the case with biological data, however, which
is multivariate. As pointed out in the intro-
duction, potential variables include each indi-
vidual species (of which there might be doz-
ens, or even hundreds), groupings of species
under broader taxonomic categories, commu-
nity-level metrics such as species richness or
diversity, and specific indices such as Good-
26
-------
APPLICATION OF DQO TO BENTHOS DATA
night-Whitley (Goodnight and Whitley, 1960)
or Milbrmk's (Milbrmk, 1983). Between 1997
and 1999, a total of 86 taxonomic entities were
recorded from the benthos samples collected
by GLNPO, each one constituting a potential
response variable to which the DQO could be
applied. In many cases these species appeared
in relatively low numbers at the sites at which
they were found, which makes a given percent
difference more difficult to discern, and also
more difficult to interpret. This report limited
the taxonomic groups under consideration to
only those few that contained relatively high
densities of individuals. Even with this restric-
tion, sample sizes required to meet the DQO
were unrealistically high. If the number of
replicates required at each site were deter-
mined by that needed to meet the DQO re-
quirement for all species encountered, the re-
sulting sampling effort would undoubtedly be-
come not only infeasible, but physically im-
possible.
It is also unclear how such changes, if de-
tected, would be interpreted; e.g., what eco-
logical meaning a 20% increase or decrease in
Stempellinella would have. In some cases,
changes in the biological populations of cer-
tain species are of inherent interest. One such
case is Diporeia, as noted above. Large varia-
tions in the population sizes of dominant or-
ganisms might also be of interest, though the
ecological meaning of such variations might
not be readily apparent. Currently, the DQO
does not provide guidance on which species
or taxonomic groups are of interest, and there-
fore on which variables should be subject to
the target contained in the DQO. Ideally, the
benthos data should also be assessed in a way
that enables detection of changes in the entire
benthic community. This would require
adopting a multivariate approach towards data
analysis, and would probably require reformu-
lation of the DQO to specify a probability of
detecting a deviation from a 'baseline' commu-
nity, rather than trying to detect a percent
change in a single variable. This is because of
the difficulty of reducing multivariate data to a
single metric. As such, though, the definition
of a baseline community, as well as what con-
stitutes an ecologically significant deviation
from such a community, would need to be de-
fined. There is some precedent for such an
approach in the Great Lakes (Reynoldson and
Day, 1998); however further consideration of
this alternative is beyond the scope of the pre-
sent report. Ultimately, however, a more de-
tailed statement of the DQO target with re-
gard to the benthos data will depend on an ex-
plicit understanding of what the data will be
used for, and how they will be interpreted.
27
-------
APPLICATION OF DQO TO BENTHOS DATA
9 References
Brinkhurst, R.O. 1967. The distribution of aquatic oligochaetes in Saginaw Bay, Lake Huron. Lim-
nol. Oceanogr. 12:137-143.
Dermott, R. and K. Corning. 1988. Seasonal ingestion rates of Pontoporeia hoyi (Amphipoda) in Lake
Ontario. Can.]. Fish. Aquat. Sd. 45:1886-1895.
Gardner, W.S., M.A. Quigley, G.L. Fahnenstiel, D. Scavia and W.A. Frez. 1990. Pontoporeia hoyi - a
direct trophic link between spring diatoms and fish in Lake Michigan. In 'Large Lakes: Ecologi-
cal Structure and Function, eds. M.M. Tilzer and C. Serruya, pp. 632-644. Springer-Verlag, New
York, NY.
Goodnight, C.I. and T.S Whitley. 1960. Oligochaetes as indicators of pollution. Proc. 15th Industr.
Waste Conf. Purdue Univ. Ext. Eng. 106:139-142.
International Joint Commission. 1978. Great Lakes Water Qualit
tional Joint Commission.
of 1978. Ottawa: Interna-
Milbrink, G. 1983. An improved environmental index based on the relative abundance of oli-
gochaete species. Hydrobiologia 102:89-97.
Reynoldson, T.B. and K.E. Day. 1998. Biological Guidelines for the Assessment of S ediment Quality in the
Laurentian Great Lakes. National Water Res. Inst. Rpt. No. 98-232, Environment Canada,
Burlington Ont.
28
-------
APPLICATION OF DQO TO BENTHOS DATA
29
------- |