Biological Conservation 186 (2015) 326336

Contents lists available at ScienceDirect

Biological Conservation
journal homepage: www.elsevier.com/locate/biocon

Evaluation and optimisation of underwater visual census monitoring

for quantifying change in rocky-reef sh abundance
Timothy Jones a,, Robert J. Davidson b, Jonathan P.A. Gardner a, James J. Bell a
a
b

School of Biological Sciences, Victoria University of Wellington, PO Box 600, Wellington 6140, New Zealand
Davidson Environmental Ltd, PO Box 958, Nelson, New Zealand

a r t i c l e

i n f o

Article history:
Received 23 December 2014
Received in revised form 22 March 2015
Accepted 28 March 2015
Available online 18 April 2015
Keywords:
Power analysis
Fish counts
Strip transects
Monte-Carlo simulation
Condence intervals
Monitoring design optimisation

a b s t r a c t
Monitoring is essential for effective conservation and management but the ability of monitoring to deliver accurate and precise measures of changes in abundance is often not evaluated, particularly in marine
studies. Here we use long-term datasets from three New Zealand marine reserves to evaluate the capacity
of underwater visual census monitoring to quantify trends in abundance for four reef-sh species.
Simulations parameterized by the observed data were used to evaluate multiple monitoring congurations based on statistical power and trend-estimate precision and accuracy. These results were then
used to identify optimal monitoring designs that maximized power, precision or accuracy within budgetary constraints. Power and trend-estimate accuracy and precision were highest for abundant species
and lowest for species exhibiting low and/or highly variable abundances. For the least abundant species,
trend estimates were less accurate and precise for negative compared to positive trends, highlighting a
reduced ability to identify ongoing declines in depleted populations. Optimal monitoring congurations
varied amongst species, locations and whether assessments were based on power, precision or accuracy.
In general, higher within-site replication was required for the least abundant species, whereas greater
site replication was required for more spatially heterogeneous species/locations. In addition, we found
that for some species the optimal monitoring approach changes through time, highlighting the need
for an adaptive approach to monitoring. Finally, we recommend that future monitoring evaluations focus
on assessing precision and accuracy, rather than power, as this places greater emphasis on the assessment of biological rather than statistical signicance.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
Monitoring of marine and terrestrial ecosystems is essential for
effective conservation and management (Lindenmayer and Likens,
2010; Reynolds et al., 2011). Monitoring can provide information
on long-term changes in abundance, demographic parameters
and ecosystem condition (Seavy and Reynolds, 2007;
Lindenmayer and Likens, 2010) and can inform successful conservation and management strategies (Bart et al., 2004). For a
monitoring program to be effective it should provide both accurate
(small bias) and precise (low uncertainty) estimates of trends in
the monitored metric. However, estimating change (e.g. in population abundance) from survey data poses several logistical and
statistical challenges (McDonald-Madden et al., 2010; Molloy
et al., 2010). Because quantifying trends in ecological studies is
so inuenced by temporal and spatial variability (Sims et al.,
Corresponding author.
E-mail address: timothy.t.jones@gmail.com (T. Jones).
http://dx.doi.org/10.1016/j.biocon.2015.03.033
0006-3207/ 2015 Elsevier Ltd. All rights reserved.

2006; Molloy et al., 2010) designing monitoring programs to maximize their capability to quantify these trends requires considerable
planning, eld testing and statistical evaluation before implementation (Reynolds et al., 2011; Lebuhn et al., 2012). However,
a priori, researchers rarely know the degree of variability that
affects abundance measures, which is important considering that
decisions pertaining to the distribution of sampling effort depend
on the relative magnitudes of the different components of variation
(Urquhart et al., 1998; Sims et al., 2006). In many cases, resource
limitations (e.g. time, cost, and availability of trained observers)
may lead to survey designs that meet some of the requirements
of an effective monitoring program by emphasising replication
on specic spatial or temporal scales to the detriment of other
monitoring requirements. For example, resources used to achieve
high spatial coverage may mean that monitoring can only be carried out infrequently, thus giving a poor indication of changes
through time (Field et al., 2005). Given the limited resources dedicated to monitoring and the expense of performing large-scale surveys, particularly in marine habitats, there is a danger that sub-

T. Jones et al. / Biological Conservation 186 (2015) 326336

optimal monitoring may be performed (McDonald-Madden et al.,

2010; Reynolds et al., 2011; Guillera-Arroita and Lahoz-Monfort,
2012).
Underwater visual census (UVC) methodologies are used extensively in marine studies for assessing the abundance of invertebrates, macroalgae and reef sh (Samoilys and Carlos, 2000;
Denny and Babcock, 2004; Stuart-Smith et al., 2008; Edgar and
Barrett, 2012; Eddy et al., 2014). Fish abundance data collected
by UVC are typically characterised by high variability (Samoilys
and Carlos, 2000; Willis et al., 2003). However, few studies have
examined the effectiveness (in terms of accuracy and precision)
of monitoring data collected by UVC to determine long-term trends
in sh abundance, let alone to optimise monitoring designs (however, see Molloy et al., 2010), something that is vital considering
the reliance on these techniques globally when quantifying the status of sh populations (Denny and Babcock, 2004; Babcock et al.,
2010; Eddy et al., 2014). This may be due to the perceived complexity of performing assessments for data types that are not suited to simple statistical tests (Reynolds et al., 2011). Nonetheless,
the widespread use of the UVC methodology makes such assessments a valuable resource for researchers and will help to better
inform management decisions.
In this study we describe an approach that may be used to
design optimum monitoring programs to identify abundance
trends of reef sh species as assessed by UVC. Monitoring should
provide accurate and precise measures of changes in abundance.
Although power analysis (and by proxy analyses of precision, as
power and precision are directly related) is commonly used in
monitoring assessments (Seavy and Reynolds, 2007) it does not
guarantee that monitoring will be accurate (Bart et al., 2004;
Nakagawa and Cuthill, 2007) and so we directly incorporate analyses of monitoring accuracy, as well as assessments of power and
precision. Using the example of four temperate New Zealand reef
sh species we quantify the power of statistical tests for trends,
in addition to trend-estimate accuracy and precision, for multiple
monitoring congurations incorporating different numbers of
sites, transects within sites and monitoring frequencies. We also
quantify the nancial costs of each design and identify the most
cost-effective approaches by identifying the best-performing
designs for specic monetary budgets. We aim to evaluate the current monitoring conguration and to determine optimal monitoring designs for different locations and for species that display
different characteristics in terms of abundance and spatio-temporal variability and whether this varies amongst assessments of
power, precision and accuracy. Finally, we aim to provide a general
methodology for the identication of a cost-effective monitoring
design and provide results and recommendations that could be
applied to temperate reef sh species with similar characteristics
(abundance and variability) in other locations.
2. Methods
Here is given a brief description of the methods and for more
details see Appendices AD of the Supplementary Material.
2.1. Datasets
Datasets from three New Zealand marine reserves were examined: Long Island-Kokomohua Marine Reserve (established 1993,
hereafter LIMR), Tonga Island Marine Reserve (established 1993,
hereafter TIMR) and Horoirangi Marine Reserve (established
2005, hereafter HMR) (Fig. 1). Datasets consisted of the observed
abundance of reef sh collected using the same UVC protocol
and set of trained divers (three divers were responsible for collecting the data over the entire study period at each reserve). Sites

327

were characterised by boulder or rocky reef substratum devoid of

a macroalgal canopy (i.e. rocky barrens), in a depth range of 5
12 m. At each site, pairs of divers swam along 30 m transects haphazardly placed at the same designated sites and recorded all sh
in a 2 m wide and 2 m high corridor. In addition to sh abundance,
size was visually estimated by the divers, trained in sh size
estimation, in order to classify individuals into legal and sub-legal
size classes. Underwater visibility was at least 4.5 m horizontal distance to ensure that visibility did not affect data collection. Twelve
transects were performed at each site with ve sites at LIMR, seven
at TIMR and eight at HMR surveyed annually (see Davidson et al.,
2014).
We limited our analysis to four species that were observed at all
three MRs; blue cod (Parapercis colias), spotty (Notolabrus celidotus), blue moki (Latridopsis ciliaris) and tarakihi (Nemadactylus
macropterus). Legal-sized (length > 30 cm) blue cod abundance
was also analysed as it has been used as an indicator of a MR effect
(Pande et al., 2008) and because it allows a comparison between
analyses based on total versus legal-sized abundance. To aid with
comparing these results with other species/systems, a range of
summary statistics related to abundance and variability are
included in Supplementary Material.
2.2. Modelling and parameter estimation
Monitoring designs are evaluated using a Monte-Carlo simulation methodology with simulations parameterized such that
they are representative of the observed data, but with a known
trend through time for assessment purposes (Sims et al., 2006). A
total of 15 initial datasets (ve species/size classes  three
reserves) were collated. Three of the initial datasets displayed
non-linear trends through time; total blue cod and legal-sized blue
cod at LIMR and spotty at TIMR (Appendix B Supplementary
Material). However, for these datasets trends were approximately
linear either side of an inection point at 1999/2000 for legal-sized
blue cod at LIMR, 2003/2004 for total blue cod at LIMR and 2005/
2006 for spotty at TIMR. Because we are interested in monitoring
efcacy for estimating linear trends in abundance and to avoid tting a linear model to non-linear trend data, these datasets were
split into two time periods (legal-sized blue cod 19931999,
20002010; total blue cod 19932003, 20042010; spotty
19992005, 20062010). This also allows us to examine whether
power and trend-estimate accuracy and precision of monitoring
designs changes through time.
To estimate abundance and variance parameters to inform simulations, poisson generalized linear mixed effects models (GLMMs)
and negative-binomial GLMMs (both with log-link function), were
tted to each dataset using the lme4 package in R (Bates et al.,
2014). Four variance components were identied as potential contributors to the overall variance of the sh counts: (1) betweensite; (2) within-site; (3) synchronous temporal variation (same
across sites); and (4) temporal variability unique to each site (survey-specic variation). Consequently, random effects of site, year
and survey were trialled to model potential between-site, synchronous, and survey-specic variation, respectively. To account
for potential overdispersion relative to the poisson distribution
and to model within-site variation, an observation-level random
effect was trialled in the poisson GLMMs. This was not necessary
for negative-binomial models because overdispersion is inherently
modelled by a dispersion parameter, m. All possible combinations
of random-effects plus a xed effect for modelling continuous
trends through time were tted for poisson (32 model combinations) and negative-binomial GLMMs (16 model combinations)
and model parameter estimates (intercept, slope and random
effects standard deviations) and Akaike Information Criterion
(AIC) were extracted from each model. The observed data and

328

T. Jones et al. / Biological Conservation 186 (2015) 326336

30 km

35S

40S

45S

170E 175E

Nelson

5 km

5 km

5 km
Tonga Island
Marine Reserve

Horoirangi
Marine Reserve

Long Island
Marine Reserve

Fig. 1. Locations of marine reserves used in this study from the South Island of New Zealand.

model t is visualized by plotting year-specic mean density

(individuals/60 m2, plus bootstrapped 95% condence intervals)
as a function of time for each species/reserve combination, and
overlaying the best-tting GLMM model predictions for mean density (plus associated 95% condence intervals) based on xed
model terms only, as well as model predictions including random
effects estimates. Subsequently, models were ranked according to
AIC and parameter sets of models with DAIC 6 2 were used to
recreate the observed datasets (i.e. observed trend and survey
design) as part of a validation procedure. The nal parameter sets
selected to perform the simulations were those that produced a
good t to the data (based on AIC), and reproduced the characteristics of the observed datasets (see Appendix CD for more details).

2.3. Simulation and testing procedure

Simulations were parameterized by the abundance and variance parameters estimated from the observed datasets, a known
temporal trend in abundance, and by the monitoring conguration
(number of sites, transects per site, monitoring frequency). For
each parameter set, trend and monitoring conguration, 500 randomly generated datasets were created. Two GLMMs, with appropriate distribution and model factors, were tted to each simulated
dataset. The rst GLMM contained only the appropriate random
effects as a suitable null model, whilst the second GLMM included

an additional xed effect to model non-zero trends in abundance.

Power was estimated as the proportion of the 500 simulated datasets where the null model is rejected with p 6 0.05 based on a likelihood-ratio test between the two models. Precision and bias were
quantied based on trend estimates extracted from the second
GLMM t to each simulated dataset and were taken to be the median 95% condence interval (CI) width and median absolute bias
(absolute difference between estimated and true trend) across
the 500 datasets, respectively. All analyses and simulations were
performed in R version 3.0.0 (R development core team, 2014).

2.4. Simulated scenarios and optimal monitoring design evaluation

Monitoring congurations investigated included a range of sites
(310 per MR), transects (4, 8, 12, 16, 20 per site) and monitoring frequency (annual, biennial) with data simulated with moderate
(b = 4.5% year 1 such that population density is 50% higher/33%
lower after ten years) and large (b = 7.7% year 1 such that population density doubles/halves after ten years) positive and negative
trends in abundance. A time period of ten years was used as the
minimum required for assessing the status of populations because
studies have demonstrated direct MR effects in this time frame
(Pande et al., 2008; Babcock et al., 2010). All possible combinations
of sites/transects/frequencies were investigated for each trend and
each parameter set, and specic results are presented. To evaluate

T. Jones et al. / Biological Conservation 186 (2015) 326336

the current monitoring, power, precision and accuracy of the monitoring design at each MR (annual monitoring with 58 sites and 12
transects per site) is presented across species for each trend magnitude/direction. In addition, power and trend-estimate accuracy and
precision are plotted as a function of the number of sites for monitoring designs that are performed annually with 12 transects per site
(b = 0.077) to explore whether additional sites are warranted.
To aid in future development of monitoring programs we identify optimal monitoring designs by considering the estimates of
power and trend-estimate precision and accuracy associated with
each monitoring conguration relative to their nancial cost.
Based on current monitoring, 12 transects are performed per
one-hour dive, with fewer transects requiring a shorter dive, but
more transects requiring additional dives per site, and we assumed
three dives per day. A surface interval of two hours between dives
was also assumed for safety purposes and to allow relocation
among sites. We estimated an hourly rate of $100 (NZD) for scientic divers and $1500 (NZD) for daily boat/skipper hire. The estimated nancial costs were then calculated based on the time
requirements for each monitoring design. Although costs are not
specic to any agency they are indicative of the costs required
for this kind of work and the calculated costs permit comparison
amongst designs relative to their time and logistic requirements.
We identify optimal monitoring designs by identifying the
best design (i.e. the one that maximizes power, precision or accuracy of trend estimates) for a specic budget. This approach is
most relevant to managers and researchers who want to maximize the ability to discern trends in abundance whilst staying
within budgetary constraints. Budgets of 640,000, 70,000 and
100,000 (NZD) for a ten-year period were chosen to reect a
range of constraints (additional budgets were also evaluated
see Supplementary Material). Optimal monitoring designs were
independently assessed for power and trend-estimate precision
and accuracy for each species/reserve and trend magnitude/direction combination. This was to identify whether there were
systematic differences in optimal monitoring design amongst
response variables as well as amongst species, locations and
trends. An alternative approach to identifying optimal monitoring
designs, based on identifying the least-cost monitoring design for
specic targets of power and trend-estimate precision and accuracy (appropriate when a specic precision/bias is required to
meet stakeholders requests or to inform a modelling exercise,
Gerber et al., 2005) was also performed and results are presented
in the Supplementary Material.

3. Results
3.1. Variance parameter estimation
GLMM parameters were obtained from all datasets with the
exceptions of legal-sized blue cod and blue moki at HMR and tarakihi at LIMR as models either failed to converge or because model
parameters were not reective of the data. These datasets were
omitted from further analysis. For remaining datasets, the best-tting model (minimal AIC) was a poisson GLMM with an observation random effect to model overdispersion. Observed densities
and model tted values are presented in Fig. 2. In all but three
cases either the parameter set belonging to the best-tting model
or a model with DAIC 6 2 was chosen to represent the observed
datasets after carrying out parameter validation. For the remaining
three datasets the poisson model parameters produced unrealistic
simulated datasets. In these cases the top-ranking negative-binomial parameter set resulted in better approximations to the actual
dataset and so these parameters were utilized (Table 1 and
Appendices CD).

329

3.2. Evaluation of current monitoring

Response variables (power, absolute bias as a measure of accuracy and 95% CI width as a measure of precision) varied considerably amongst species and reserves, both in terms of
magnitude (Table 2) and relationship with increasing number of
sites (Figs. 2 and 3). In all cases, power and trend-estimate accuracy and precision was lower for negative trend simulations (i.e.
lower power, greater bias and larger 95% CI) than for positive
trends, with the difference being largest for species with low abundance; legal-sized blue cod (LIMR 19931999 and TIMR), blue cod
(TIMR and HMR) and blue moki (Table 2). For large effects, 80%
power was attained for blue cod at LIMR (19932003), spotty at
HMR and TIMR (19992005) and legal-sized blue cod at LIMR for
both time periods for positive trends, but only the 20002010 period for negative trends (Table 2). For moderate effects, power was,
on average, half that of large effects and 80% was only achieved for
blue cod at LIMR (19932003) and spotty at HMR (Table 2). For
bias and precision, values were largely invariant with regards to
trend magnitude/direction for spotty and blue cod at LIMR
(Table 2). However, for legal-sized blue cod and blue moki at all
reserves and total blue cod at TIMR and HMR, bias and 95% CI
width decreased with increasingly positive effect sizes, with up
to a 30% reduction in bias and 95% CI width for large positive
effects compared to large negative effects (Table 2).
Power and trend-estimate precision and accuracy was predicted
to be lower at TIMR than at LIMR for legal-sized blue cod (Fig. 3).
For legal-sized blue cod at LIMR trend-estimate precision and accuracy for 19931999 was lower than for 20002010, with the
exception of analyses based on large positive trends where values
were approximately equal (Table 2). Trend-estimate precision,
accuracy and power was higher for total blue cod compared to
legal-sized blue cod based analyses at TIMR and in the rst ten
years at LIMR (Table 2). However, for the latter period at LIMR,
trend-estimate precision and accuracy was higher for legal-sized
abundance than total abundance (Table 2). Judged on the same
monitoring design, overall power and trend-estimate precision
and accuracy for blue cod was highest at LIMR, followed by TIMR
and HMR, respectively (Fig. 3). However, for monitoring positive
trends with seven or more sites, values were similar between
TIMR and LIMR for the period 20042010 (Table 2, Fig. 3). This convergence is due to a slower rate of increase in power and trendestimate accuracy and precision with increasing number of sites
at LIMR (20042010), than at TIMR or HMR, potentially indicating
alternate limiting factors amongst locations (Fig. 3).
For spotty, power and trend-estimate precision and accuracy
was highest at HMR and lowest at TIMR for the 20062010 period
(Table 2, Fig. 3). For four or fewer sites trend-estimate precision
and accuracy was similar between monitoring at LIMR and TIMR
(19992005), but for larger sample sizes values were higher at
LIMR (Fig. 3). The rate of increase of power and trend-estimate precision and accuracy with increasing number of sites was lower for
TIMR based on the 19992005 period compared to LIMR or the
later period at TIMR, suggesting alternate limiting factors amongst
locations and time periods (Fig. 3). In addition, for higher number
of sites, values were almost equivalent between the two time periods at TIMR, indicating that between-site variation was more limiting in the later period than in the early period (Fig. 3).
For blue moki, power and trend-estimate accuracy and precision was higher at TIMR than at LIMR and was overall lower than
for blue cod or spotty (Table 2). There was no indication of alternate limiting factors amongst locations for blue moki, as power,
precision and accuracy change at a similar rate amongst locations
relative to the number of sites surveyed (Fig. 4). For tarakihi, power
and trend-estimate accuracy and precision was the lowest of all
species with even extensive sampling (ten sites, 12 transects per

330

T. Jones et al. / Biological Conservation 186 (2015) 326336

0 1 2 3 4 5 6

20
0 5 10
1995

2000

2005

2010

2008

2009

2010

2006

2007

2008

2009

2010

2006

2007

2008

2009

2010

0
0.0 0.1 0.2 0.3 0.4

12
0 2 4 6 8
1995

2000

2005

2010

Legend

0.4

0.8

2000 2002 2004 2006 2008 2010

Mean density 95% CI

0.0

Legal-sized blue cod

density (ind/60m2)

2007

1
0
0.0 0.4 0.8 1.2

Blue cod density

(ind/60m2)

2000 2002 2004 2006 2008 2010

1995

2000

2005

GLMM fit 95% CI

(fixed effects only)

2010

GLMM fit 95% CI

(including random effects)

0.0

0.4

0.0 0.2 0.4 0.6

0.8

2000 2002 2004 2006 2008 2010

Blue moki density

(ind/60m2)

2006

2000 2002 2004 2006 2008 2010

Tarakihi density
(ind/60m2)

Horoirangi MR

12

Long Island MR

0 2 4 6 8

Spotty density
(ind/60m2)

Tonga Island MR

2000 2002 2004 2006 2008 2010

1995

2000

2005

2010

Year
Fig. 2. Observed mean densities (95% CI) of spotty, tarakihi, blue cod, legal-sized blue cod and blue moki plotted over time for data collected at Tonga Island MR (left
column), Long Island MR (central column) and Horoirangi MR (right column). GLMM model t and associated 95% condence intervals are also shown for predictions based
on xed effects only (intercept and slope), and predictions incorporating random effects terms.

Table 1
Parameters estimated from GLMMs and used in data simulations. q0 refers to the initial population density, inverse log-transformed from model intercept estimates and m refers
to the negative-binomial dispersion parameter. Omitted values indicate terms not present in the associated model.
Species

MR

Start

End

Model type

q0 (ind./60 m2)

Year
Legal blue cod
Legal blue cod
Blue cod
Blue cod
Spotty
Blue moki
Legal blue cod
Blue cod
Spotty
Spotty
Blue moki
Tarakihi
Blue cod
Spotty
Tarakihi

LIMR
LIMR
LIMR
LIMR
LIMR
LIMR
TIMR
TIMR
TIMR
TIMR
TIMR
TIMR
HMR
HMR
HMR

1993
2000
1993
2004
1993
1993
1999
1999
1999
2006
1999
1999
2006
2006
2006

1999
2010
2003
2010
2010
2010
2010
2010
2005
2010
2010
2010
2010
2010
2010

Poisson
Poisson
Poisson
Poisson
Poisson
Neg. Bin.
Poisson
Poisson
Poisson
Poisson
Neg. Bin.
Poisson
Neg. Bin.
Poisson
Poisson

0.32
1.22
2.80
5.16
8.25
0.06
0.04
0.14
1.50
3.32
0.22
0.20
0.07
4.76
0.03

Random effects Std. Dev.

Site

0.20

0.17

1.05

0.39
0.97
0.12
0.44
0.38
0.45
0.22
0.58
0.09

Surv.
0.33
0.18
0.38
0.51
0.67
0.50
0.40
0.31
0.69
0.66
0.77

Obs.
0.78
0.62
0.57
0.54
0.67
0.31
1.41
0.96
0.67
0.71
0.40
1.27
0.75

0.20
0.54

0.37
2.50

39.5
17.1

20.5
13.6

10.5
8.7
11.7
3.8

5.3
10.2
11.4
13.7

8.4
8.0
16.4

38.8
17.3

21.1
14.1

10.5
9.0
11.8
3.8

5.3
10.3
11.9
14.7

8.8
8.1
17.1

41.2
18.3

23.2
15.0

10.7
9.1
12.0
4.0

5.7
10.6
13.3
18.1

10.3
8.7
19.5

40.2
18.8
72.4
27.0
71.2
25.5

24.7
15.7
41.3
26.1
45.5
26.9

10.6
9.1
11.9
4.1
17.0
18.6
19.5
6.6
18.3
16.6
20.7
6.8

5.9
10.4
14.0
19.9
10.2
17.9
17.7
24.5
8.9
18.8
19.3
27.3

10.9
8.9
20.8
14.0
14.8
28.2
14.3
14.6
29.8

0.0455
b=
0.077

70.0
24.7

45.3
26.3

18.6
15.8
21.1
8.0

10.0
19.7
23.0
31.6

17.4
15.1
32.9

b=

0.0455

b = 0.0455

b = 0.077

b=

95% CI width (10

63.4
29.1
13.6
40.0
11.4
15.6

50.5
29.4
24.0
47.2
14.4
23.2

18.1
16.5
20.5
7.7
79.8
80.6
71.6
100
35.4
44.0
32.4
99.2

10.2
18.3
24.0
31.6
100
70.2
75.2
53.8
88.8
35.4
31.2
21.6

48.6
58.0
14.4

95.4
94.8
37.0

20.2
15.9
36.3

0.077
b=

9.8
15.6
11.4
31.0
TIMR
HMR

8.2
18.8
17.0
40.0
LIMR
TIMR

39.2
42.6
30.8
98.4
77.4
81.4
71.4
100
LIMR
TIMR 9905
TIMR 0610
HMR

84.8
35.0
24.6
16.6
100
72.0
54.6
27.0
LIMR 9303
LIMR 0410
TIMR
HMR

40.6
51.0
18.0
76.6
87.4
29.2
LIMR 9399
LIMR 0010
TIMR

0.077

b=

0.0455

b = 0.0455

b = 0.077
b=

site and annual monitoring) achieving only 40% power for large
positive effects at HMR (Table 2). In addition, although increased
sampling had some benet at HMR, there was little benet of
increased sampling at TIMR (Fig. 4).
3.3. Optimal monitoring design xed cost

Bias (10
Power (a = 5%)
Data

331

For each cost level, four monitoring designs were optimal (i.e.
maximizing power, precision or accuracy) in most cases, with each
design either maximizing sites, transects per site or monitoring
frequency within budgetary limits (Fig. 5). For budgets of
$40,000, $70,000 and $100,000 the monitoring designs with 5 sites
(5s), 20 transects per site (20t), and biennial monitoring (2f), 6s
12t1f and 9s12t1f were most frequently the best performing
monitoring designs (Fig. 5). However, within budgets there was
considerable variation among species/locations, trend direction
and response variables. When optimal designs differed amongst
response variables it was most often that designs maximizing precision and power had greater within-site replication compared to
those designs that maximize accuracy, which tended to have
higher monitoring frequency ($40,000 and $70,000) or number of
sites ($100,000) rather than higher within-site replication
(Fig. 5). Amongst trends, optimal designs for negative trends
tended to have greater spatial replication at the expense of monitoring frequency when compared to optimal designs for positive
trends (Fig. 5).
Differences in optimal monitoring conguration amongst species are best illustrated by categorizing the designs by how sampling effort was prioritized. The rst group consisting of 5s20t
2f ($40,000), 9s20t2f ($70,000) and 6s20t1f ($100,000) prioritizes within-site replication, with number of sites as a secondary
consideration and was most often optimal for species with low
abundances; legal-sized blue cod at LIMR (19931999) and TIMR,
total blue cod and tarakihi at HMR and blue moki at LIMR
(Fig. 5). The second group consisting of 4s8t1f ($40,000), 7s
8t1f ($70,000) and 10s8t1f ($100,000) prioritizes monitoring
frequency, with number of sites a secondary consideration. This
group was most often optimal for spotty at LIMR and TIMR
(20062010) and also for total blue cod abundance at LIMR
(20042010) (Fig. 5). The third group consisting of 3s12t1f
($40,000) and 4s20t1f ($70,000) prioritizes monitoring frequency, with within-site replication a secondary consideration.
This group was optimal, primarily with regards to accuracy, in several instances for spotty, blue moki and blue cod (Fig. 5). The fourth
group consisting of 7s12t2f ($40,000), 6s12t1f ($70,000) and
9s12t1f ($100,000) was characterised by an equal emphasis on
monitoring frequency, number of sites and within-site replication.
Monitoring designs within this group were most often optimal and
performed the best for blue cod at TIMR and LIMR (19932003),
spotty at HMR and TIMR (19992005), blue moki at TIMR and
legal-sized blue cod at LIMR (20002010) (Fig. 5). However, in
some instances sampling priorities changed depending on cost
level with, for example, optimal monitoring designs for blue cod
at TIMR and LIMR (19932003) characterised by high within-site
replication at $40,000, but by congurations with an equal emphasis on all aspects of survey design for higher cost programs (Fig. 5).

Tarakihi

Blue moki

Spotty

Blue cod

Legal-sized blue cod

4. Discussion

Species

Table 2
Power, median absolute bias and median 95% CI width for the current monitoring design (annual monitoring with 12 transects per site) at LIMR (ve sites), TIMR (seven sites) and HMR (eight sites).

b = 0.0455

b = 0.077

T. Jones et al. / Biological Conservation 186 (2015) 326336

Identifying an effective monitoring program for a specic ecological goal/aim is a challenge shared by ecologists, conservation
managers and governmental agencies. Our results highlight the
large differences amongst species, and even for the same species
across reserves and time periods, in the ability of UVC monitoring
to quantify trends in abundance. Although we have only presented

332

T. Jones et al. / Biological Conservation 186 (2015) 326336

3 4 5 6 7 8 9 10

0.30
0.10

0.20

LIMR 93-99
LIMR 00-10
TIMR
= - 0.077
= 0.077

0.00

95% CI Width

Absolute Bias

0.00 0.01 0.02 0.03 0.04 0.05

60
40
0

20

Power (5%)

80

100

3 4 5 6 7 8 9 10

3 4 5 6 7 8 9 10

0.30
0.20
0.10

95% CI Width

0.04

3 4 5 6 7 8 9 10

LIMR 93-03
LIMR 04-10
TIMR
HMR
= - 0.077
= 0.077

0.00

0.00

0.02

Absolute Bias

60
40
20

Power (5%)

80

0.06

100

3 4 5 6 7 8 9 10

3 4 5 6 7 8 9 10

3 4 5 6 7 8 9 10

0.15
0.10
0.05

95% CI Width

0.02

LIMR
TIMR 99-05
TIMR 06-10
HMR
= - 0.077
= 0.077

0.00

0.00

0.01

Absolute Bias

60
40
20

Power (5%)

80

0.03

100

3 4 5 6 7 8 9 10

3 4 5 6 7 8 9 10

Number of Sites
Fig. 3. Power, bias and 95% CI width associated with trend estimates (b = 0.077) for monitoring designs with 310 sites per reserve with annual monitoring and 12 transects
per site. Panels correspond to results for legal-sized blue cod (A), total blue cod (B) and spotty (C) from the three reserves and corresponding time-periods.

results for four species, these results may be applied to other temperate reef sh species monitored by UVC in other locations and
we provide both a methodology and the data to do this. We have
also shown that within budgetary constraints, optimal monitoring
designs can be identied through a rigorous evaluation of alternate
monitoring congurations.
4.1. Limiting factors
Previous monitoring design analyses have shown that monitoring can be limited by low abundances (Nicholson and Jennings,
2004; Maxwell and Jennings, 2005; Blanchard et al., 2008). This
is true for the species examined here, whereby power and trendestimate accuracy and precision was highest for the more abundant species, blue cod (at LIMR and TIMR) and spotty, and lowest
for the less abundant species, blue moki and tarakihi. Because
the original surveys were aimed predominantly at recording blue
cod abundance, it is perhaps not surprising that this monitoring
was best suited to them as surveys were performed in their preferred habitat. However, for naturally scarce species, such as blue
moki, and heavily shed species whose populations are depleted,
monitoring in general is likely to be less effective (Blanchard
et al., 2008). In particular, the ability to quantify trends in abundance was lower for legal-sized blue cod than total blue cod in
the early years after establishment of LIMR, due to the initial scarcity of legal-sized individuals. However, when legal-sized

individuals became more abundant, the greater variability of total

blue cod abundance meant that the monitoring of legal-sized
individuals was more effective than monitoring of total abundance.
In addition, our results revealed a low ability to identify negative
trends in abundance for the least abundant species. This highlights
the decreased ability of monitoring to adequately determine the
rate of decline of an already scarce species unless sampling is
highly replicated in space and time (Maxwell and Jennings, 2005).
Variability in abundance is likely to be the greatest determinant
of whether monitoring can effectively identify trends in abundance. Although spotty were always observed in relatively high
abundance, this varied considerably within and amongst sites
and through time. In particular, schools of more than 50 spotties
were regularly encountered, leading to considerable variation
(e.g. counts >50 versus counts 210), potentially limiting the
ability to determine accurate and precise measures of trends in
abundance. However, tarakihi abundance was the most variable.
Tarakihi are predominantly found in soft-bottom habitats, but
are intermittently observed over rocky reefs (Francis, 2001). In
addition, tarakihi were often observed in schools and so were
observed in great abundance on occasion, which contrasts with
many transects with zero counts. As a result, even with considerable sampling effort, the ability of the monitoring program
to demonstrate a trend in abundance is low and therefore not
cost-effective in the areas/habitats examined for tarakihi. In many
situations those species that managers and conservationists are

333

T. Jones et al. / Biological Conservation 186 (2015) 326336

9 10

95% CI Width
3

0.00 0.05 0.10 0.15 0.20 0.25 0.30

Absolute Bias
3

0.00 0.01 0.02 0.03 0.04 0.05

60
40
0

20

Power (5%)

80

100

9 10

LIMR
TIMR
= - 0.077
= 0.077

9 10

9 10

0.3

HMR

0.1

= - 0.077
= 0.077

0.0

0.00

0
3

TIMR

0.2

95% CI Width

0.04

Absolute Bias

0.02

60
40
20

Power (5%)

80

0.06

0.4

100

9 10

9 10

Number of Sites
Fig. 4. Power, bias and 95% CI width associated with trend estimates (b = 0.077) for monitoring designs with 310 sites per reserve with annual monitoring and 12 transects
per site. Panels correspond to results for blue moki (A) and tarakihi (B) from each of the two reserve datasets analysed.

most interested in may also be those that are least amenable for
monitoring (Sommerville et al., 2011). This information is however
vital prior to monitoring commencement so that realistic targets
and monitoring priorities are established (Seavy and Reynolds,
2007). To achieve better results for monitoring tarakihi (or any
other species that does not predominantly occupy the surveyed
habitat), surveys in alternate habitats and/or depth strata may be
required, which may impair the ability to monitor other species,
highlighting the trade-offs faced in developing a monitoring program that is suitable for all species in a given location. The analysis
of the variable tarakihi data from sub-optimal habitat is also informative because individuals may exist in areas of sub-optimal habitat for much of their lives. Indeed, for heavily targetted species, it
may be that more individuals now exist outside their optimal habitat because of responses to shing pressure. Thus, the ability to
survey individuals in sub-optimal habitats may be as informative
as surveys of the same taxon inside optimal habitats.
4.2. Optimal monitoring design
Identifying an optimal monitoring design is challenging given
the costs and the number of factors that affect monitoring design
performance (Reynolds et al., 2011). Our results highlight that
there can be no one survey ts all because the choice of optimum
monitoring design varies amongst species, locations and through
time. However, a general nding was that 12 transects per site
was most often the most cost-effective design (see also
Supplementary Material regarding optimal monitoring designs
for xed targets of power, accuracy and precision). This is perhaps
not surprising as this number of transects maximized the replication per dive and is the most effective use of underwater time.
However, the possible replication per dive is likely to vary among
locations and diving conditions, but it is likely that in most situations maximizing the replication per dive within those constraints
is likely to be the most cost-effective. The optimal monitoring

designs identied are inherently related to the limiting factors

affecting the ability to quantify trends in abundance for each species. For example, optimal designs for spotty prioritized monitoring frequency with number of sites as a secondary concern,
indicating that inter-annual variability and between-site variation
were the main limiting factors. In contrast, optimal designs for blue
moki and tarakihi focused on within-site replication (12 or 20 transects), which for blue moki is likely a response to their low abundance and for tarakihi has the effect of down-weighting the
importance of individual extreme observations associated with
schooling behaviour. However, in the majority of cases optimal
monitoring designs were those that had an equal emphasis on
within-site replication, number of sites and monitoring frequency.
This is likely to be an indication that there is no particular limiting
factor, such as low abundance or spatial variability, but several factors operating together and therefore a balanced approach to
monitoring design is needed.
The optimum monitoring design also depended on trend direction with negative trend assessments opting for greater spatial
replication and lower monitoring frequency than those for positive
trend assessments. This has important implications because at the
outset of monitoring the direction of future change is unknown
and a focus on positive trends may sacrice the ability to determine negative trends in abundance. From a management perspective it may be more important to demonstrate a negative trend,
and identifying the best monitoring design to achieve this will
result in an enhanced ability to effectively manage populations
with lower rates of unnecessary management action.
Optimal monitoring conguration also varied amongst time
periods for those species examined, suggesting that limiting factors
are likely to shift with time as populations change. As abundance of
legal-sized blue cod at LIMR increased, optimal monitoring designs
shifted from those favouring high spatial replication because of initial low abundances to those with a balanced approach when
abundances were higher in later years. In addition, for total blue

Legal-sized power
blue cod accuracy
LIMR 00-10 precision

Other

8-12-1

9-12-1

6-20-1

Cost < $100,000

10-8-1

Other

4-20-1

6-12-1

7-8-1

9-20-2

Cost < $70,000

Other

3-12-1

7-12-2

4-8-1

Cost < $40,000

5-20-2

Variable

T. Jones et al. / Biological Conservation 186 (2015) 326336

Data

334

= 0.077
9-8-1

= - 0.077

power

Legal-sized
blue cod accuracy
LIMR 93-99 precision

= +/- 0.077

Legal-sized power
blue cod accuracy
TIMR
precision
Blue cod
HMR

power
accuracy
precision
power

Blue cod
LIMR
04-10

accuracy

Blue cod
LIMR
93-03

accuracy

Blue cod
TIMR

accuracy

5-12-1

precision
power

10-12-2*

10-12-2*

precision
power
7-12-1

precision
Spotty
HMR

power

3-16-1*

accuracy

8-20-2

3-16-1*

precision
Spotty
LIMR

power
accuracy

8-4-2

6-8-1

9-8-1

precision
Spotty
TIMR
06-10

power

7-4-1

accuracy

7-4-1

Spotty
TIMR
99-05

accuracy

precision
power
6-8-1

power
Blue moki
accuracy
LIMR
precision

4-20-2

power
Blue moki
accuracy
TIMR
precision
Tarakihi
HMR

9-8-1

precision

power

9-16-2

5-12-1
5-16-2

accuracy

8-20-2

6-16-1

precision
Tarakihi
TIMR

power

5-16-2

10-8-2 & 5-20-2

10-8-2 & 5-20-2

accuracy

3-8-1

6-8-1

7-12-1 & 8-8-1

precision

6-8-1

Fig. 5. Illustration of those monitoring designs that attained the highest power, precision and accuracy for maximum costs of 40,000, 70,000 and 100,000 NZD. Monitoring
design congurations are given as nsitesntransects-frequency with frequencies of 1 and 2 denoting annual and biennial monitoring, respectively. indicates that a power of
100% was achieved and that these monitoring designs were the least expensive designs achieving 100% power.

cod at LIMR and spotty at TIMR optimal monitoring conguration

shifted from an initially balanced approach to one with greater
emphasis on monitoring frequency and number of sites. This is
suggestive of an increase in temporal and large-scale spatial variability with time, with the former potentially due to uctuations
in natural resources, which as populations increase in abundance
towards carrying capacity, have a larger inuence on population
dynamics, and the latter potentially due to differential reservation

effects at different locations (Eddy et al., 2014). Changes in spatial

and temporal variability are likely to occur throughout the course
of most long-term monitoring programs. This is particularly the
case for targetted species within marine reserves as at their establishment shing pressure is likely to have heavily inuenced spatial patterns in abundance (Stuart-Smith et al., 2008), which are
likely to change when shing ceases (Lester et al., 2009).
Consequently, no monitoring conguration is likely to remain

T. Jones et al. / Biological Conservation 186 (2015) 326336

optimal over long periods of time, but regular assessments and the
ability to alter monitoring based on those assessments will ensure
that the most appropriate monitoring design is employed
(Lindenmayer and Likens, 2010).
In the majority of cases monitoring designs that maximized
power and precision matched those that maximized accuracy.
However, when monitoring design choices differed, those that
maximized power and precision (which are inherently related)
tended to be those with greater within-site replication and number
of sites, whereas monitoring designs that maximized accuracy
tended to select for frequency and number of sites. In some cases
there is therefore likely to be a trade-off between accuracy and
precision, particularly at lower costs where budgetary constraints
mean that monitoring effort cannot be focused on both high spatial
and high temporal replication. However, by performing an evaluation similar to the one outlined here a monitoring design that performs adequately for both accuracy and precision can be identied.
4.3. Additional considerations
When deciding on a monitoring program there are additional
considerations that are not necessarily captured by a statistical
approach. Identifying suitable sites may be the most difcult monitoring design aspect to implement. Sites must share characteristics,
such as habitat and depth, and there may be a natural limit to the
number of sites that can be monitored within any region. In addition, there are extra costs associated with additional dives to locate
suitable areas. However, surveying too few sites is likely to result in
estimates that are serially biased towards the trends exhibited at
those sites rather than being representative of the entire population (Bart et al., 2004). Individual locations may display very different trends through time as a response to differential habitat
(Miller and Russ, 2014), previous and ongoing shing pressure
(Stuart-Smith et al., 2008; Lester et al., 2009) and for MR monitoring, proximity to reserve boundaries (McClanahan and Mangi,
2000). By monitoring more sites it is possible to reduce this bias
as a greater proportion of the target population is monitored; as
such, it is necessary to consider the degree of spatial coverage that
is deemed suitable to address the purposes of monitoring.
However, focusing on spatial replication may come at the cost of
monitoring frequency, with infrequent monitoring less likely to
capture chance events such as pulses of recruitment, the establishment of invasive species or a rapid decrease in abundance due to
poaching. Monitoring more frequently will give better estimates
of the rates of these events and may aid in management and give
an increased scientic understanding of the ecosystem. Therefore,
monitoring designs which take a balanced approach (equal emphasis on spatial and temporal replication) are likely to provide the
best results with the fewest drawbacks with regards to statistical
properties and these additional concerns. The majority of the species/reserve combinations examined in our work favoured this
approach.
Within the context of MR or marine protected area (MPA) monitoring there are several limitations of the approach we adopt,
regarding the omission of reference location replication and the
potential importance of non-linear trends in long-term studies.
Reference location data was not incorporated into these analyses
because the goal of this study was to have a generalized assessment of reef sh monitoring by UVC (which may not always be
accompanied by reference location data), rather than being specic
to MPAs. Specic to MPAs, monitoring should aim to show that
trends or changes within an MPA differ from those at reference
locations. Indeed, with regards to assessing MPA effects the ability
to identify changes at reference locations may be as important as
identifying within-MPA changes (e.g. ongoing decline at reference
locations versus MPA stability). Therefore, studies similar to this

335

one, but with a focus on MPA and reference site replication, with
a range of MPA effects and different analysis structure (i.e. focusing
on differences between changes at MPA and reference locations),
would be advantageous. However, the results presented herein
regarding the differences among species/locations with respect to
the optimal assignment of monitoring effort, would likely also
apply to studies taking this approach. Some potential differences
may be that greater reference site and within-site replication
may be required relative to MPA sites, due to potentially lower
and more varied abundance (particularly for heavily shed species)
as a result of continued and variable shing pressure among reference sites.
In this study, linear trends (or more appropriately curvilinear,
due to the log-link function) were simulated for several reasons.
Firstly, a linear trend can be summarized by a single reference
value (b), enabling the evaluation of accuracy and precision based
on estimates for b. Secondly, linear trend assessments may be
directly incorporated into monitoring and management plans
(e.g. if trend exceeds X, implement action Y), with the ability to
estimate trends directly related to the ability to effectively implement management actions (Samhouri et al., 2010). Finally, many of
the examined datasets exhibited linear trends over time-scales of
612 years (see Fig. 2), consistent with our evaluation based on a
ten-year timescale. However, in longer MPA studies non-linear
trends may be exhibited due to changes associated with trophic
cascades and the emergence of indirect MPA effects ( Shears and
Babcock, 2003; Babcock et al., 2010). Designing monitoring based
on non-linear trend assessments, however, may be challenging
due to the lack of a single reference value representing the trend,
and also the sensitivity of assessments to the assumed functional
form of the non-linear response. However, studies examining
monitoring based on non-linear trend assessments, particularly
with regards to designs that are suited to particular functional
forms, would be a valuable addition to the eld of study associated
with monitoring program design. In particular, optimising monitoring to identify transitions in MPA response (e.g. from direct to
indirect effects, Babcock et al., 2010), may provide better information regarding the ecosystem wide changes associated with longterm exclusion of shing.

5. Conclusion
In the past, power has been most often used as an indicator of
monitoring design effectiveness (Seavy and Reynolds, 2007).
Although we show here the utility of power analyses (i.e. many
requirements of a precise and accurate monitoring design are also
requirements of a powerful one), we recommend that analyses of
power should be superseded by analyses focused on maximizing
accuracy, which may be overlooked with a strict focus on power
(Bart et al., 2004), and precision. Power analyses are based upon
null hypothesis signicance testing and are often focused on
optimising effect detection. However, upon collecting data something is inevitably detected and monitoring should focus on
quantifying what is detected rather than ruling out null hypotheses
(McBride et al., 1993; Nakagawa and Cuthill, 2007; Gerrodette,
2011). Furthermore, statistical signicance does not equate to
practical or biological signicance (Anderson et al., 2000). This is
often addressed by dening effect sizes that are deemed biologically signicant (Nakagawa and Cuthill, 2007). However, truly biologically signicant changes in abundance are often unknown,
particularly at the beginning of monitoring, and so there is a risk
of monitoring programs being established based on false ideals.
We argue that performing prospective analyses of precision and
accuracy is more appropriate because it lends itself towards gathering and interpreting data in a more meaningful way.

336

T. Jones et al. / Biological Conservation 186 (2015) 326336

Acknowledgements
We thank divers who collected the data (Willie Abel, Laura
Richards, Mike Aviss, Simon Bayly, Dirk deVries, Tim Shaw, Rhys
Barrier, Pete Braggins, Russell Cole and Anjali Pande), Andrew
Baxter, Roy Grose and Martin Rodd and their teams from DOC
(Nelson, Picton and Motueka) and the respective marine reserve
committees for their support. Funding for eld work was provided
by DOC to Andrew Baxter. T. Jones was supported by a Victoria
University of Wellington PhD scholarship.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.biocon.2015.03.
033.
References
Anderson, D.R., Burnham, K.P., Thompson, W.L., 2000. Null hypothesis testing:
problems, prevalence and an alternative. J. Wildlife Manage. 64, 912923.
Babcock, R.C., Shears, N.T., Alcala, A.C., Barrett, N.S., Edgar, G.J., Lafferty, K.D.,
McClanahan, T.R., Russ, G.R., 2010. Decadal trends in marine reserves reveal
differential rates of change in direct and indirect effects. Proc. Natl. Acad. Sci.
U.S.A. 107, 1825618261.
Bart, J., Burnham, K.P., Dunn, E.H., Francis, C.M., Ralph, C.J., 2004. Goals and
strategies for estimating trends in landbird abundance. J. Wildlife Manage. 68,
611626.
Bates, D., Maechler, M., Bolker, B., Walker, S., 2014. lme4: Linear mixed-effects
models using Eigen and S4. R package version 1.1-7, <http://CRAN.R-project.
org/package=lme4>.
Blanchard, J.L., Maxwell, D.L., Jennings, S., 2008. Power of monitoring surveys to
detect abundance trends in depleted populations: the effects of densitydependent habitat use, patchiness, and climate change. ICES J. Mar. Sci. 65, 111
120.
Davidson, R.J., Richards, L.A., Abel, W., Avis M., 2014. Long Island-Kokomohua
Marine Reserve, Queen Charlotte Sound: update of biological monitoring, 1992
2014. Prepared by Davidson Environmental Limited for DOC, Nelson. Survey
and Monitoring Report No. 796.
Denny, C.M., Babcock, R.C., 2004. Do partial marine reserves protect reef sh
assemblages? Biol. Conserv. 116, 119129.
Eddy, T.D., Pande, A., Gardner, J.P.A., 2014. Massive differential site-specic and
species-specic responses of temperate reef shes to marine reserve protection.
Global Ecol. Conserv. 1, 1326.
Edgar, G.J., Barrett, N.S., 2012. An assessment of population responses of common
inshore shes and invertebrates following declaration of ve Australian marine
protected areas. Environ. Conserv. 39, 271281.
Field, S.A., Tyre, A.J., Possingham, H.P., 2005. Optimizing allocation of monitoring
effort under economic and observational constraints. J. Wildlife Manage. 69,
473482.
Francis, M., 2001. Coastal Fishes of New Zealand: An Identication Guide, third ed.
Reed Publishing, Auckland, New Zealand.
Gerber, L.R., Beger, M., McCarthy, M.A., Possingham, H.P., 2005. A theory for optimal
monitoring of marine reserves. Ecol. Lett. 8, 829837.
Gerrodette, T., 2011. Inference without signicance: measuring support for
hypotheses rather than rejecting them. Mar. Ecol. 32, 404418.

Guillera-Arroita, G., Lahoz-Monfort, J.J., 2012. Designing studies to detect

differences in species occupancy: power analysis under imperfect detection.
Methods Ecol. Evol. 3, 860869.
Lebuhn, G., Droege, S., Connor, E.F., Gemmill-Herren, B., Potts, S.G., Minckley, R.L.,
Griswold, T., Jean, R., Kula, E., Roubik, D.W., et al., 2012. Detecting insect
pollinator declines on regional and global scales. Conserv. Biol. 27, 113120.
Lester, S.E., Halpern, B.S., Grorud-Colvert, K., Lubchenco, J., Ruttenberg, B.I., Gaines,
S.D., Airame, S., Warner, R.R., 2009. Biological effects within no-take marine
reserves: a global synthesis. Mar. Ecol. Prog. Ser. 384, 3346.
Lindenmayer, D.B., Likens, G.E., 2010. The science and application of ecological
monitoring. Biol. Conserv. 143, 13171328.
Maxwell, D., Jennings, S., 2005. Power of monitoring programmes to detect decline
and recovery of rare and vulnerable sh. J. Appl. Ecol. 42, 2537.
McBride, G.B., Loftis, J.C., Adkins, N.C., 1993. What do signicance tests really tell us
about the environment? Environ. Manage. 17, 423432.
McClanahan, T.R., Mangi, S., 2000. Spillover of exploitable shes from a marine park
and its effect on the adjacent shery. Ecol. Appl. 10, 17921805.
McDonald-Madden, E., Baxter, P.W.J., Fuller, R.A., Martin, T.G., Game, E.T.,
Montambault, J., Possingham, H.P., 2010. Monitoring does not always count.
Trends Ecol. Evol. 25, 547550.
Miller, K.I., Russ, G.R., 2014. Studies of no-take marine reserves: methods for
differentiating reserve and habitat effects. Ocean Coast. Manage. 96, 5160.
Molloy, P.P., Anticamara, J.A., Rist, J.L., Vincent, A.C.J., 2010. Frugal conservation:
what does it take to detect changes in sh populations? Biol. Conserv. 143,
25322542.
Nakagawa, S., Cuthill, I.C., 2007. Effect size, condence interval and statistical
signicance: a practical guide for biologists. Biol. Rev. 82, 591605.
Nicholson, M.D., Jennings, S., 2004. Testing candidate indicators to support
ecosystem-based management: the power of monitoring surveys to detect
temporal trends in sh community metrics. ICES J. Mar. Sci. 61, 3542.
Pande, A., MacDiarmid, A.B., Smith, P.J., Davidson, R.J., Cole, R.G., Freeman, D., Kelly,
S., Gardner, J.P.A., 2008. Marine reserves increase the abundance and size of
blue cod and rock lobster. Mar. Ecol. Prog. Ser. 366, 147158.
R Development Core Team, 2014. R: A Language and Environment for Statistical
Computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3900051-07-0, <http://www.R-project.org>.
Reynolds, J.H., Thompson, W.L., Russell, B., 2011. Planning for success: identifying
effective and efcient survey designs for monitoring. Biol. Conserv. 144, 1278
1284.
Samhouri, J.F., Levin, P.S., Ainsworth, C.H., 2010. Identifying thresholds for
ecosystem-based management. PLoS One 5, e8907.
Samoilys, M.A., Carlos, G., 2000. Determining methods of underwater visual census
for estimating the abundance of coral reef shes. Environ. Biol. Fish. 57, 289
304.
Seavy, N.E., Reynolds, M.H., 2007. Is statistical power to detect trends a good
assessment of population monitoring? Biol. Conserv. 140, 187191.
Shears, N.T., Babcock, R.C., 2003. Continuing trophic cascade effects after 25 years of
no-take marine reserve protection. Mar. Ecol. Prog. Ser. 246, 116.
Sims, M., Wanless, S., Harris, M.P., Mitchell, P.I., Elston, D.A., 2006. Evaluating the
power of monitoring plot designs for detecting long-term trends in the number
of common guillemots. J. Appl. Ecol. 43, 537546.
Sommerville, M.M., Milner-Gulland, E.J., Jones, J.P.G., 2011. The challenge of
monitoring biodiversity in payment for environmental service interventions.
Biol. Conserv. 144, 28322841.
Stuart-Smith, R.D., Barrett, N.S., Crawford, C.M., Frusher, S.D., Stevenson, D.G., Edgar,
G.J., 2008. Spatial patterns in impacts of shing on temperate rocky reefs: are
sh abundance and mean size related to proximity to sher access points. J.
Exp. Mar. Biol. Ecol. 365, 116125.
Urquhart, N.S., Paulsen, S.G., Larsen, D.P., 1998. Monitoring for policy-relevant
regional trends over time. Ecol. Appl. 8, 246257.
Willis, T.J., Millar, R.B., Babcock, R.C., 2003. Protection of exploited sh in temperate
regions: high density and biomass of snapper Pagrus auratus (Sparidae) in
northern New Zealand marine reserves. J. Appl. Ecol. 40, 214227.