See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/339023177

An empirical evaluation of camera trap study design: how many, how long,
and when?

Article  in  Methods in Ecology and Evolution · February 2020

DOI: 10.1111/2041-210X.13370

CITATIONS READS

0 1,087

28 authors, including:

Roland Kays Brian Arbogast

North Carolina State University University of North Carolina at Wilmington
241 PUBLICATIONS   9,025 CITATIONS    29 PUBLICATIONS   1,718 CITATIONS   

SEE PROFILE SEE PROFILE

Christopher Beirne Hailey Boone

University of British Columbia - Vancouver North Carolina State University
27 PUBLICATIONS   172 CITATIONS    2 PUBLICATIONS   8 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

TEAM Network View project

Bird telemetry monitoring to evaluate loss of habitat in mining area in the northeastern Amazon. View project

All content following this page was uploaded by Fernanda da Silva Santos on 29 April 2020.

The user has requested enhancement of the downloaded file.

 |
Received: 2 August 2019    Accepted: 15 January 2020

DOI: 10.1111/2041-210X.13370

RESEARCH ARTICLE

An empirical evaluation of camera trap study design: How

many, how long and when?

Roland Kays1,2,3  | Brian S. Arbogast4 | Megan Baker-Whatton5 | Chris Beirne6 |

Hailey M. Boone2 | Mark Bowler7 | Santiago F. Burneo8 | Michael V. Cove9 |
Ping Ding10 | Santiago Espinosa11,12 | André Luis Sousa Gonçalves13 |
Christopher P. Hansen14 | Patrick A. Jansen15,16  | Joseph M. Kolowski9 |
Travis W. Knowles17 | Marcela Guimarães Moreira Lima18  | Joshua Millspaugh14 |
William J. McShea9 | Krishna Pacifici2 | Arielle W. Parsons1,2  | Brent S. Pease2 |
Francesco Rovero19,20 | Fernanda Santos21 | Stephanie G. Schuttler1 | Douglas Sheil22 |
Xingfeng Si23  | Matt Snider2 | Wilson R. Spironello13
1
North Carolina Museum of Natural Sciences, Raleigh, NC, USA; 2Department of Forestry and Environmental Resources, North Carolina State University,
Raleigh, NC, USA; 3Smithsonian Tropical Research Institute, Balboa, Panama; 4Department of Biology and Marine Biology, University of North Carolina,
Wilmington, NC, USA; 5The Nature Conservancy, Arlington, VA, USA; 6Nicholas School of the Environment, Duke University, Durham, NC, USA; 7University
of Suffolk, Ipswich, Suffolk, UK; 8Museo de Zoología, Pontificia Universidad Católica del Ecuador, Quito, Ecuador; 9Smithsonian Conservation Biology
Institute, Front Royal, VA, USA; 10College of Life Sciences, Zhejiang University, Hangzhou, Zhejiang, China; 11Facultad de Ciencias, Universidad Autónoma de
San Luis Potosí, San Luis Potosí, México; 12Escuela de Ciencias Biológicas, Pontificia Universidad Católica del Ecuador, Quito, Ecuador; 13Grupo de Pesquisa
de Mamíferos Amazônicos, Instituto Nacional de Pesquisas da Amazônia, Manaus, Brazil; 14Wildlife Department, University of Montana, Missoula, MT, USA;
15
Department of Environmental Sciences, Wageningen University, Wageningen, The Netherlands; 16Centre for Tropical Forest Science, Smithsonian Tropical
Research Institute, Balboa, Panama; 17Department of Biology, Francis Marion University, Florence, SC, USA; 18Laboratório de Ecologia e Conservação, Brazil
Instituto Nacional de Pesquisas da Amazônia – IN, Universidade Federal do Pará, Belém, Pará, Brazil; 19Tropical Biodiversity Section, MUSE – Museo delle
Scienze, Trento, Italy; 20Department of Biology, University of Florence, Florence, Italy; 21Departament of Mastozoology, Museu Paraense Emílio Goeldi, Belém,
Pará, Brazil; 22Faculty of Environmental Sciences and Natural Resource Management, Norwegian University of Life Sciences, Norway and 23Zhejiang Tiantong
Forest Ecosystem National Observation and Research Station, School of Ecological and Environmental Sciences, East China Normal University, Shanghai, China

Correspondence
Roland Kays Abstract
Email: [email protected] 1. Camera traps deployed in grids or stratified random designs are a well-established
Funding information survey tool for wildlife but there has been little evaluation of study design
National Natural Science Foundation of parameters.
China, Grant/Award Number: 31572250
and 31872210; Riverbanks Zoo and 2. We used an empirical subsampling approach involving 2,225 camera deployments
Gardens; Satch Krantz Conservation Fund; run at 41 study areas around the world to evaluate three aspects of camera trap
Francis Marion University Professional
Development Fund and International study design (number of sites, duration and season of sampling) and their influ-
Collaboration Grants; University of North ence on the estimation of three ecological metrics (species richness, occupancy
Carolina-Wilmington Charles L. Cahill
Grant; US National Science Foundation; San and detection rate) for mammals.
Diego Zoo Global; Gordon and Betty Moore 3. We found that 25–35 camera sites were needed for precise estimates of species
Foundation
richness, depending on scale of the study. The precision of species-level estimates
Handling Editor: Diana Fisher of occupancy (ψ) was highly sensitive to occupancy level, with <20 camera sites
needed for precise estimates of common (ψ > 0.75) species, but more than 150
camera sites likely needed for rare (ψ < 0.25) species. Species detection rates were
more difficult to estimate precisely at the grid level due to spatial heterogeneity,

© 2020 North Carolina Department of Natural and Cultural Resources. Methods in Ecology and Evolution © 2020 The British Ecological Society.

Methods Ecol Evol. 2020;00:1–14. wileyonlinelibrary.com/journal/mee3     1 |

 |
2      
Methods in Ecology and Evolu on KAYS et al.

presumably driven by unaccounted habitat variability factors within the study area.
Running a camera at a site for 2 weeks was most efficient for detecting new spe-
cies, but 3–4  weeks were needed for precise estimates of local detection rate,
with no gains in precision observed after 1 month. Metrics for all mammal commu-
nities were sensitive to seasonality, with 37%–50% of the species at the sites we
examined fluctuating significantly in their occupancy or detection rates over the
year. This effect was more pronounced in temperate sites, where seasonally sensi-
tive species varied in relative abundance by an average factor of 4–5, and some
species were completely absent in one season due to hibernation or migration.
4. We recommend the following guidelines to efficiently obtain precise estimates of
species richness, occupancy and detection rates with camera trap arrays: run each
camera for 3–5 weeks across 40–60 sites per array. We recommend comparisons
of detection rates be model based and include local covariates to help account for
small-scale variation. Furthermore, comparisons across study areas or times must
account for seasonality, which could have strong impacts on mammal communities
in both tropical and temperate sites.

KEYWORDS

camera traps, community ecology, detectability, mammals, relative abundance, species

richness, study design, wildlife surveys

1 |  I NTRO D U C TI O N lengthening survey duration to improve estimate precision. The

influence of seasonality on camera trapping data has been largely
Monitoring wildlife populations is more important than ever, given neglected to date, probably because most tropical data come from
the imperiled status of many species around the world, and the rapid cameras run during the dry season to reduce damage to cameras
changes to the earth's climate and landcover (Dirzo et al., 2014). (TEAM Network, 2008), while most temperate zone data come from
Camera traps are now a standard method for monitoring a variety the growing season. A few studies have noted seasonal differences
of species over relatively large areas (Steenweg et al., 2017; Wearn including migratory ungulates in the Serengeti (Palmer, Swanson,
& Glover-Kapfer, 2017), being used to quantify species diversity, and Kosmala, Arnold, & Packer, 2018) and a peak of coyote detections
estimate occupancy and relative abundance. These estimates can be in California when individuals were dispersing (Larrucea, Brussard,
compared across space and time to monitor changes in populations and Jaeger, & Barrett, 2007).
test hypotheses about the effects of landscape and human factors on The surge in popularity of sampling with camera traps, combined
species relative abundance, distribution and interspecific interactions. with improvements in technology and data management, has facili-
The growth of camera trap surveys for monitoring biodiversity tated the accumulation of large centralized datasets from around the
has led to substantial variation in study designs across projects, with world. These datasets can be examined to evaluate the consequences
sample size ranging from 1 to >1,000 camera sites (Burton et al., of survey study design across species and locations allowing for gen-
2015). The foundational questions a researcher must address when eral recommendations to researchers and practitioners. We analysed
designing a camera trap study include the following: how many cam- 41 camera trap datasets including 1,771 sites to evaluate three as-
era traps to run, for how long and at what time of year? Although pects of camera trap study design: (a) how many camera traps should
there have been strong theoretical assessments of camera trap researchers run? (b) how long should they run each individual camera
survey design (Gálvez, Guillera-arroita, Morgan, & Davies, 2016; trap? and (c) does seasonality affect results? For each question, we
Guillera-Arroita, Ridout, & Morgan, 2010), unfortunately, empiri- subsampled large datasets to investigate how sample effort and de-
cal evaluations using real-world data have been fewer. For exam- sign affect estimates of species richness, individual species detections
ple, Si, Kays, and Ding (2014) found that adding camera trap sites and occupancy probabilities. To facilitate comparisons of sites from
was a higher priority than increasing survey length, and Shannon, around the world, we used no habitat covariates in our occupancy or
Lewis, and Gerber (2014) found that precise estimates could be relative abundance models. These covariates would improve the preci-
made with surveys of 30–50 sites for common animals, but those sion of estimated parameters (occupancy and detection rate) by help-
species with low probabilities of detection required extended sur- ing to explain variation caused by different environments. Therefore,
vey lengths, and that adding camera sites was more effective than our results on recommended sampling effort to reach a given level of
KAYS et al. Methods in Ecology and Evolu on |
      3

precision are conservative in comparison to the precision local studies and detection rate. We compare species richness of subsampled data
might achieve when including habitat covariates. Our results allow us to the total number of species detected at a study area by the cam-
to offer general recommendations for camera trap survey design to eras, not the total number of species theoretically possible based on
achieve the most precise estimates of species richness, occupancy and geographical range maps. Likewise, we compare estimates of occu-
relative abundance across a wide range of species around the world. pancy and detection rate from subsamples of data to values calcu-
lated with the full dataset, although we have no way of knowing the
‘true’ value for a given species at a site. We make direct comparisons
2 | M ATE R I A L S A N D M E TH O DS of occupancy estimates from camera arrays set at different spatial
scales, but assume that the occupied site is the small detection zone
We analysed camera trap data from 41 studies from the eMammal directly in front of the camera and not some larger area defined by
(http://www.eMamm​al.org) and TEAM (http://www.teamn​etwork. camera spacing (Efford & Dawson, 2012). To simplify our analyses, we
org) databases and co-authors (Figure 1; Table S1). Subsets of 28, 16 consider number of samples and sampling effort separately, although
and 5 datasets were used to examine spatial, temporal and seasonal there can be an interaction between these factors (Shannon et al.,
questions, respectively. We used 1 year of data from each TEAM site. 2014). Although detection rate can be a misleading measure of rel-
All studies used similar camera trap protocols in that sites were un- ative abundance if field protocols are not standardized (Sollmann,
baited and were set in grids or stratified-random design (i.e. not on well- Mohamed, Samejima, & Wilting, 2013), it has been tied mechanisti-
established trails), making data directly comparable (Jansen, Ahumada, cally to abundance (Rowcliffe, Field, Turvey, & Carbone, 2008), and
Fegraus, & O'Brien, 2014). Cameras were set at ~0.5  m above the has been shown to reflect animal density in a number of studies using
ground on a tree, or high in the tree canopy (Peru), and set to trig- grids or stratified random sampling (e.g. Palmer et al., 2018; Parsons
ger with passive infrared motion sensor at high sensitivity with no rest et al., 2017), and we therefore refer to it here as an index of relative
period between triggers. The camera brands used (Reconyx, Bushnell, abundance.
ScoutGuard) had fast trigger times (<0.5  s) to minimize missing fast-
moving species and used infrared flashes at night to avoid disturbing
animals. Cameras were set to take multiple pictures at each trigger and 2.1 | Number of camera points
continued to retrigger as long as animals were in sensor range. Triggers
within 60 s of each other were considered one detection. We subsampled from existing datasets (Table S1) using program R
Our evaluation of study design focuses on the precision of three (version 3.3; R Core Development Team, 2018) to evaluate the num-
metrics frequently used by ecologists: species richness, occupancy ber of sample points needed for precise estimates of species richness

Legend
Number of cameras
19–58
59–97
98–137

138–176

177–215

F I G U R E 1   Map showing locations of the 41 datasets used for analyses of the importance of survey duration, survey size and seasonality
on estimates of biodiversity variables using camera traps
 |
4      
Methods in Ecology and Evolu on KAYS et al.

(using data from 28 study areas), occupancy and detection rate (using detect 95% of the total number of species detected with all cam-
data from 20 study areas) within a given protected area. We estimated eras. We stratified the studies based on latitude (tropical and tem-
richness as the number of individual species detected in each study perate) and scale of the study (small was ≤0.2 km camera spacing,
area across all cameras. We estimated detection rate as the total large was ≥1.0  km spacing) when considering how many camera
count of detections of each species divided by the number of cam- sites were needed to meet each threshold. We summarized the
era nights (i.e. detections/day) at each study area. We ran occupancy effect of camera number on richness in each stratification level by
models using the single species occupancy framework (MacKenzie fitting an asymptotic curve to the resampled data points using a
et al., 2002) and estimated detection probability (p), defined as the nonlinear least squares method.
probability of detecting an occurring species during a 7-day period at
a camera site, and occupancy (ψ), defined as the expected probability
that a given camera site is occupied, for each species. Although daily 2.2 | Sample duration
periods are sometimes used for occupancy analyses of camera trap
data for common species, these weekly intervals allowed us to use the We subsampled long camera deployments (~60 sampling days) to
same protocol for all species. We recognize that occupancy estimates evaluate the effect of sample duration on detection rate and species
for wide-ranging animals detected with camera traps is more analo- accumulation across each deployment day for selected species to
gous to ‘use’ than true occupancy, though no less informative in terms determine how survey duration affected precision. We selected 24
of habitat use (Burton et al., 2015; Efford & Dawson, 2012). Although representative species of large (>100 kg, n = 6), medium (5–100 kg,
some of the cameras from our fine-scale grids might not be spatially n = 9) and small (<5 kg, n = 9) mammals including three trophic levels
independent for wide-ranging species, we note that the independence (carnivores, omnivores and herbivores) from surveys with a minimum
assumption is not necessary (though it is helpful) since the MLE of ψ of 10 camera sites, excluding deployments with less than 20 sam-
is consistent even if occupancy statuses are not independent (Royle pling days (range = 10–77 deployments per study) across five conti-
& Link, 2006). We constructed models using the package unmarked nents (Table S2). These groupings represented potential differences
(Fiske & Chandler, 2011) in R. We estimated detection rate and oc- in movements/home-ranges across the taxa but also accounted for
cupancy for two species at each of 20 study areas, the most com- some inevitable variation in detection probability associated with
mon species and the least common with at least 100 detections (and the sizes and trophic guilds of target species (Cove, Spínola, Jackson,
therefore enough to have a reasonable chance of models converging). Sàenz, & Chassot, 2013). We calculated the daily detection rate of
The one exception was for the UCSC Grid study area, where only one each species across all sites with detections within a selected study
species had >100 detections. area and examined how the cumulative daily detection rate changed
We subsampled sites by randomly selecting from one to the maxi- with increasing camera deployment. We considered the detection
mum number of cameras in the study, using the data collected at sub- rate at the end of the sampling period as the true result for a site
sampled cameras to generate estimates. A minimum of five camera and set a threshold of precision as the number of sampling days that
sites was used for occupancy analysis due to model convergence is- were required to obtain an estimate within 10% of this value.
sues. For each of 50 subsamples (without replacement) at each camera We used negative binomial generalized linear models to deter-
sample size, we recorded the mean, standard deviation, standard error mine how this threshold duration varied with three species traits—
and 95% confidence interval for richness, detection rate and occu- species-specific trophic guild (e.g. carnivores, omnivores or herbivores),
pancy probability. We used the total sample for each camera, which size (small, medium and large), climatic zone (tropical or temperate)—
was variable, but should not affect our results. For detection rate, we maximum number of days per deployment and final detection rate. To
assume animal movement is constant over the study period so camera determine which species traits or survey effects most influenced the
deployment length should not affect these estimates or recommen- detection rate bias, we compared eight a priori hypotheses, including
dations (i.e. a camera running 5  days captures 10 deer and if it ran a null model, in an information-theoretic framework, based on their
10 days instead it would capture 20). For occupancy, we also did not Akaike weights (Burnham & Anderson, 2002). These analyses allowed
standardize the number of trap days per camera; however, we tested us to consider whether relationships were general to all mammals or
whether there was a relationship between survey effort and the es- specific to guilds or study regions. Furthermore, we plotted the mean
timated occupancy probability but found no significant relationship per cent detection rate error across all species to visualize an inflection
(T37 = 0.867, p = .39). point in the per cent error patterns, such that increasing effort beyond
To determine how many cameras were adequate to obtain pre- a threshold point resulted in diminishing returns in terms of accuracy
cise estimates of occupancy and detection rate, we set two thresh- (sensu Lashley et al., 2018).
olds: (a) when the width of the 95% confidence interval changed To assess species accumulations, we selected five tropical and
<1% and remained <1% regardless of added cameras, represent- five temperate study areas and randomly selected 30 camera sites
ing the point of diminishing return on investment and (b) when to represent each study area. We determined the accumulations for
the standard error was <10% of the mean and remained <10% re- each individual camera site in addition to the total accumulation for
gardless of added cameras, representing a precise measurement. each study area. For both tropical and temperate regions, the totals
For richness, we assessed how many cameras were necessary to of each study area were averaged together to represent how long it
KAYS et al. Methods in Ecology and Evolu on |
      5

can take to reach a high percentage of total species per region, as 3.1 | Number of cameras
determined by the actual camera data.
On average, large-scale (≥1.0  km spacing), tropical study areas re-
quired more cameras (M = 35 ± 1 SE) for species richness to reach
2.3 | Seasonality an asymptote than did smaller-scale (≤0.2 km camera spacing) sites
(Figure 2; Table S3). With much lower total species counts, small-
To evaluate the importance of seasonality in camera trap surveys, we scale sites detected 95% of total species with less effort, requiring
calculated monthly and seasonal estimates of mammal detection rates, 22 ± 3 sites in the more diverse tropics and 17 ± 4 in the temperate
occupancy and species richness for five locations that experience sea- studies. The number of cameras necessary for richness to level off
sonal differences in either temperature or precipitation and that had at varied with camera spacing (β = 70 ± 23, t = 3, p = .0005), with large-
least 1 year of continuous camera data (Table S1). For two temperate scale camera grids requiring more sites. Only two large-scale tem-
climates with seasons determined by changes in temperature (Montana perate study areas were available and levelled off at 17 or 18 species
and North Carolina), we defined seasons as summer (April–September) (Table S3) but were not included in the graph due to low sample size.
and winter (October–March). For three tropical or subtropical climates We plotted the improvement in precision of detection rate
with seasons determined by changes in precipitation, we identified sea- and occupancy estimates with increasing number of cameras for
sons as wet and dry, where the wet seasons used were May–October 39 species from 20 study areas (Figure 3; Table S1). The example
(China), November–April (Peru) and July–December (Panama). plots in Figure 3 illustrate how we determined when an estimate
To consider the effect of seasonality on estimated species richness was precise (10% or 20% SE from final value) and when adding
at each site, we calculated the total number of species observed each additional cameras did not lead to significant improvements in
month. We present observed species richness for sites with equal ef- the estimate (confidence intervals improve by <1%). We found
fort between months (Montana, North Carolina, China and Panama) a strong negative relationship between the occupancy level of a
and where sampling effort was unequal (Peru), we present estimated species and the number of cameras needed to get a precise esti-
richness from the iNEXT package (Hsieh, Ma, & Chao, 2016). To facili- mate (Figure 4c) and the point of diminishing returns (Figure 4d).
tate visual assessment of seasonal trends in species richness and detec- Furthermore, occupancy could only be estimated with certainty
tion rates, we fit a nonparametric LOESS smoother line to the points. for species with occupancy values >0.7 (Figure 4c; Table S4). The
We calculated the detection rate of each species for each study number of cameras needed to get precise occupancy estimates
area. To identify whether season was a significant predictor of de- was marginally correlated with detection rate (linear regression:
tection rate at a study site, we included season as a predictor and β = −13.3 ± 7.1, t = −1.89, p = .08).
species-specific detection rate as the response in a quasi-Poisson There was also no relationship between the detection rate of a
generalized linear model, adjusted for overdispersion (Wedderburn, species and the number of cameras needed for precise estimates
1974). We considered season to have a significant effect on detec-
tion rate if the 95% confidence interval around the season coeffi-
cient did not overlap with zero.
We also quantified the effect of season on species occupancy
and detection. We created species-specific detection histories by
splitting continuous camera deployments into intervals representing
seven camera trap nights. For the Panama dataset (BCI plots), cam-
eras were not deployed continuously, rather they were moved every
7–12 days within each 1 of 10 1-ha survey blocks. In this case, each
7- to 12-day interval represents a separate deployment within the
same survey block. For each species, we used season of deployment
as a predictor, calculated the season coefficient in occupancy and
detection models using package unmarked in R, and if the 95% con-
fidence interval around the season coefficient did not overlap with
zero we considered it a significant effect.

F I G U R E 2   Species accumulation curves with increasing camera

3 | R E S U LT S sites showing average (lines) and SE (shading) across 19 tropical and
9 temperate study areas. Vertical lines show where a class of sites
reached 95% of all the total number of species recorded, which was
We examined camera trap sampling design patterns for 106 mammal
least for small-scale (≤0.2 km camera spacing) tropical study areas
species and used common names when describing results in the text,
(17 ± 4 SE sites), followed by small-scale temperate (22 ± 3 sites)
but present taxonomy in Table S2. Animal detections not identified and large-scale (≥1.0 km spacing) tropical studies (35 ± 1 sites).
to the species level were not included. Results per site are in Table S3
 |
6      
Methods in Ecology and Evolu on KAYS et al.

F I G U R E 3   Examples of changes in detection rate (a) and occupancy (b) with increasing number of cameras at a site for two species: white-
tailed deer (a) and red acouchi (b). The black line represents the average of 50 subsamples while the blue dashed line shows standard error and
the red dashed line shows the 95% confidence intervals. For each species, we calculated the number of cameras needed to reach a precise
estimate with standard error that was 10% of the estimate (solid blue line representing point of high precision) and the point of diminishing
returns where the change in the confidence interval between timesteps was <1% (red solid line representing point of diminishing return)

F I G U R E 4   Number of camera sites

needed to estimate detection rate (a, b)
and occupancy values (c, d) to a point of
precision (SE <20% A or <10% C) or to the
point of diminishing returns (confidence
intervals improve by <1% with additional
cameras, b, d). Each dot on the graph
represents a species, with example plots
showing how these were derived in Figure
3. Only two species reached high (<10%
SE) confidence estimates for detection
rate so we show the results for a lower
(20%) confidence level in A. No trend line
is provided in A due to the low coefficient
of determination. Values and species
names are reported in Tables S4 and S5

of detection rate, although only two species reached our 10% SE rapidly, with 89.7% ± 5.60 SE of the species at temperate sites detected
cut-off (Table S5), so we had to evaluate this relationship at 20% SE in 2 weeks and 100% in 30 days. For tropical sites, 84.9% ± 5.05 SE of
(Figure 4a). When considering the point of diminishing returns, there species was detected in 2 weeks, 97.1% ± 1.82 SE in 30 days.
was a positive relationship, indicating that species with higher de- Since detections typically accumulate slower than days, the
tection rates required more camera locations before the confidence ratio of detections/day is erratic and highly erroneous at smaller
intervals reached an asymptote. (<2  weeks) sampling intervals, but then rapidly improve. Across
species, the mean time needed to obtain an estimate within 10% of
the final estimate was 34.4 ± 2.5 SE days (Figure 6). The variation in
3.2 | Sample duration time needed to get a precise measure of detection rate was best ex-
plained by final detection rate (Akaike weight = 0.41, Table S7), with
At a single camera site, species richness increased such that after some additional model support suggesting a difference between
the first 30  days, each camera detected 96.67%  ±  0.01  SE or temperate and tropical study areas (Akaike weight = 0.15, Table S7).
85.62% ± 0.02 SE of the species it would detect over 60 days in temper- We used those two covariates to make predictions of survey length
ate or tropical sites, respectively (Figure 5). However, when all cameras requirements across detection rates that we observed in temperate
at a study area were considered together, species richness plateaued and tropical study areas (Figure 7).
KAYS et al.       7
Methods in Ecology and Evolu on |

F I G U R E 5   Example species accumulation curves at one temperate (a, Albany Area) and one tropical study site (b, Thailand Reserves)
illustrating the rate of species detection by individual cameras (grey lines) and for the whole survey (black lines). Eight additional sites are
graphed in Figure S1. For five temperate (c) and tropical (d) sites combined, curves are shown in proportion to the total number of species
they detected at the end of the 60-day survey

F I G U R E 6   Change in the accuracy of detection rate estimates

with increased sample duration showing mean per cent daily
detection rate error, with 95% confidence intervals (red dashed
lines) and standard error (blue dashed lines). These are estimated F I G U R E 7   Model predictions of the number of days of camera
from 24 representative species at 11 study areas across five trap sampling before the per cent error of the estimated detection
continents. The vertical line (34 days) represents the mean number rate decreases below 10% based on negative binomial regression
of days at which the per cent error decreased below 10% for the models with the final species detection rate and tropical versus
remainder of the survey (Tables S6 and S7) temperate zones as predictors
 |
8      
Methods in Ecology and Evolu on KAYS et al.

TA B L E 1   Summary of significant seasonal variation in occupancy and detection rate models for 70 species across five sites. ‘0’ indicates
no variation, ‘+’ and ‘−’ indicate higher or lower values in the winter (Montana and North Carolina) or wet season (China, Panama and Peru).
Models that did not converge have blank cells. Species indicated with * had so few detections in a season that statistical models did not
converge, but clearly showed strong seasonal patterns such as complete absence due to hibernation. Model parameter values are available
as supplemental material (Figure S2; Table S9)

  Model occupancy Model detection Detection rate Notes

Montana
Columbian Ground Squirrel     −* Hibernates
Brown Bear     − Hibernates
Moose     0  
Striped Skunk     0  
White-tailed Deer 0 − − Elevational migrant
White-tailed Jackrabbit     0  
Gray Wolf     0  
Elk − 0 0  
American Badger     0  
North American Porcupine     0  
Bobcat 0 0 0  
Coyote + − 0  
Snowshoe Hare 0 0 0  
Puma 0 0 0  
Mountain Cottontail 0 0 0  
American Red Squirrel     0  
Mule Deer + + 0 Elevational migrant
Prairie Ridge, NC
Woodchuck     −* Hibernates
Coyote 0 − 0  
Virginia Opossum + − 0  
Northern Raccoon     0  
White-tailed Deer 0 0 0  
Eastern Cottontail 0 0 0  
Bobcat 0 0 0  
Eastern Gray Squirrel 0 − 0 Seasonal seed hoarder
Gray Fox + 0 +  
Red Fox     0  
China
Wild Boar     0  
Reeves's Muntjac 0 − −  
Edwards's Long-tailed Giant Rat     0  
Hairy-fronted Muntjac 0 0 0 Elevational migrant
Pallas's Squirrel     0  
Hog Badger 0 0 0  
Chinese White-bellied Rat + + + Strong seasonal breeder
Masked Palm Civet 0 + 0  
Panama
Collared Peccary 0 − −  
Northern Tamandua     −  
Red Brocket Deer 0 − −  

(Continues)
KAYS et al. Methods in Ecology and Evolu on |
      9

TA B L E 1   (Continued)

  Model occupancy Model detection Detection rate Notes

White-nosed Coati 0 0 −  
Common Opossum     0  
Agouti     0  
Tayra     0  
Paca 0 + 0  
Baird's Tapir     0  
Red-tailed Squirrel     0  
Ocelot 0 0 0  
White-faced Capuchin     0  
Nine-banded Armadillo 0 0 0  
Peru Canopy
Tayra     +*  
Brazilian Porcupine 0 − 0  
Western Woolly Opossum − 0 0  
Olingo 0 0 0  
Hairy Saki Monkey − 0 0  
Tamandua 0 0 0  
Microsciurus sp.     0  
Dwarf Porcupine + − 0  
Black-mantled Tamarin 0 0 0  
White-fronted Capuchin 0 0 0  
Spix's Night Monkey − + 0  
Common Woolly Monkey 0 0 0  
Kinkajou     0  
Scurius sp.     0  
Common Squirrel Monkey 0 0 0  
Two-toed Sloth 0 + 0  
Yellow-handed Titi Monkey     0  
Red Howler Monkey 0 0    

3.3 | Seasonality Peru and Panama, respectively; while for North Carolina it was 50.0%
(Table 1; Figure S2). However, when considering the magnitude of
We examined the effects of seasonality on animal communities at the effect in terms of the average size of the change in detection
five study areas with varying latitudes and species compositions rate (higher season/lower season) for species where detection or oc-
(Table S8). All five sites showed seasonal effects for some but not all cupancy models indicated a significant seasonal effect, the two tem-
species (Table 1 ; Table S9; Figure S2). Although there was no clear perate sites had relatively larger changes (Montana: 480% change,
seasonal trend in species richness for the site in China (Figure 8a), North Carolina: 390% change) than the two tropical sites (Peru: 230%,
overall detection rates increased in the wet season (Figure 8b), and Panama: 180%), while the subtropical Chinese site average change
there was species-specific variation in the detection rates by season was 470%.
(Figure 8c,d). For example, we found much higher detection rates in
masked palm civet, hog badger and Chinese white-bellied rat dur-
ing the wet season, weaker pulses of Reeve's muntjac and wild boar 4 | D I S CU S S I O N
during the dry season, and little seasonal variation in the detection
rates for other species. In the most comprehensive empirical study on camera trap sur-
The proportion of species with significant seasonal changes in vey design to date, we evaluated how the number of cameras
occupancy or detection probability was surprisingly consistent across deployed, the length of time they are run, and the seasonality of
four sites, being 37.5%, 37.5%, 37.5% and 38.4% in Montana, China, the site affect estimates of mammal species richness, occupancy
 |
10      
Methods in Ecology and Evolu on KAYS et al.

F I G U R E 8   Seasonal variation shown as monthly averages for (a) estimated species richness, (b) total mammal detection rate, (c) species-
specific detection rate (for species with >20 records) and (d) seasonal averages for mammals surveyed by camera traps (for species with
>20 records, blue shows wet season) from 2009 to 2011 in Gutianshan Plot, China. Points are average monthly observed richness (a) and
detection rates (b); lines are smoothed observed richness (a) and smoothed average detection rates (b, c); shaded areas are approximate
95% confidence intervals around smoothed averages. The species in plot (c) follow the colour coding in (d), and both only show species with
>20 records. The raw data for all sites are presented in Table S9, seasonal graphs for sites in Montana, North Carolina, Panama and Peru are
presented as Figure S2

and detection rates. Our samples include 41 study areas across 4.1 | Number of cameras
20 countries on five continents, making our results broadly rel-
evant to others designing camera trap studies. All cameras were We found that more camera sites (~35) were needed to detect a
run in systematic or stratified random designs, off major trails and high percentage of species richness for larger-scale studies. A likely
without bait, providing relatively unbiased measures of the ani- explanation is that larger-scale studies sample a greater diversity of
mal community and offering comparable data across a variety of habitats and have a larger component of β-diversity, which serves
habitats. as a good reminder that species lists from small-scale surveys will
KAYS et al. Methods in Ecology and Evolu on |
      11

be incomplete representations of the larger area. Given higher spe- below zero, when you have rare species there are so many estimates
cies richness in tropical areas, it was surprising that both tropical right around that lower bound that precision is actually quite high,
and temperate small-scale sites required similar number of cameras whereas with more abundant species we would have a lot more vari-
(17–22) to detect 95% of the total species eventually detected with ation in the count and lower precision.
the full dataset. Given the variation across studies, we recommend We were surprised to not find a relationship between detec-
that studies aiming to quantify the diversity of species in an area tion rate and the number of cameras within a study area. Only two
use at least 35 camera sites for large-scale surveys (≥1.0 km spac- species reached our 10% SE precision cut-off for detection rate,
ing), and 25 for small-scale (≤0.2 km camera spacing) tropical sites. and even when considering a lower point of precision (20% SE),
Occupancy and detection rate are two metrics often used as there was no strong tendency for species with higher detection
indices of abundance to compare across years or sites for moni- rates to be easier to survey with fewer sites (Figure 4a). In fact,
toring population trends or testing ecological hypotheses, and are we found that species with higher detection rates required more
the most common metrics used to study species that cannot be cameras to reach the point of diminishing returns. We suspect this
uniquely identified (Kays et al., 2017). We found that more cam- difference between occupancy and detection rate stems from a
eras were needed to get precise estimates of species with lower higher spatial variation in detection rates within a study site. As a
occupancy values (i.e. less common species) which is in agreement binary variable (each site is either 0 or 1), occupancy probabilities
with recent studies (Beaudrot, Ahumada, O'Brien, & Jansen, 2019; are capped at 1 and inherently less variable than detection rates,
Shannon et al., 2014). Among the 39 species considered, occupancy and thus appear to be easier to estimate precisely from camera
could only be estimated precisely (SE < 10% of occupancy estimate) traps. It is important to note that we used raw detection rate val-
for species with high (>0.7) occupancy probabilities (Figure 4a; ues and made no attempt to account for variation within sites with
Table S4). Extrapolating this linear relationship suggests that 3–4 ecological models. Adding habitat covariates and using model es-
times more cameras (~150 camera sites) would be needed to pro- timates for detection rates would likely improve the precision of
duce precise estimates for species with an occupancy probability detection rate estimates, adding important ecological inference,
of 0.25. This is worrisome given that rare species are the most im- and making comparisons between sites more valid.
portant to monitor, and that low occupancy probabilities are com-
mon in camera trapping studies. For example, almost half (48%) of
the 158 carnivore populations evaluated in a recent global analysis 4.2 | Sample duration
had occupancy levels <0.25 (Rich et al., 2017). Adding ecological
covariates to occupancy models would help increase precision of By subsampling long camera deployments, we were able to evaluate
occupancy estimates, and are therefore especially important for the relationship between sample duration and accuracy and offer
working with species with low occupancy probabilities. These mod- recommendations for optimized study design. In general, the num-
els might allow more precise estimates with fewer camera sites than ber of species detected by a camera rapidly accumulated during the
our results suggest, as we did not include any covariates to enable first 2 weeks of deployment, after which few new species were de-
large-scale comparisons. tected. Only the more diverse tropical sites continued to accumulate
Community-level occupancy models (Dorazio, Kéry, Royle, & species up to about 1 month (Figure 5).
Plattner, 2010) might also help improve the precision of occupancy We found that detection rates—the widely used measure of rela-
estimates for rare species by sharing information across species, but tive abundance—were highly variable for the first 2 weeks of camera
there are extensive assumptions about guild-specific detection proba- deployments: the addition of a single detection could dramatically
bility relationships that need to be considered (Pacifici, Zipkin, Collazo, change the estimate. However, after 2–3 weeks, there was a sharp
Irizarry, & DeWan, 2014). Detecting ecological relationships will improvement in precision for all species, and from about 1  month
always be difficult for species with few detections and we suggest that onwards, estimates were within 10% of the estimate from the full
these situations might warrant the use of lures or baits, or adaptive 60-day samples (Figure 6). The exact sampling duration needed to
sampling (Pacifici, Reich, Dorazio, & Conroy, 2016; Specht et al., 2017). reach this 10% precision level varied depending on the final detec-
We expected that it would require fewer cameras to precisely es- tion rate of the target species, with precise estimates being reached
timate occupancy for species with higher detection rates, since data sooner for species with higher detection rates, with slightly different
would presumably accumulate faster, but we found no relationship relationships for tropical versus temperate areas. The higher error
(Table S5). We suspect that this reflects patchy space use by animals, rates for apparently rare species (based on final detection rates)
with higher detection rates being driven by repeated use of the same showcases that researchers should likely implement species-specific
sites, rather than many sites across the camera array. Alternatively, efforts when targeting rare species but might achieve accurate rel-
this might be simply a principal of the data structure. One possibility ative abundance information for common species within short sam-
is related to pseudo-Poisson patterns of count data, where variation pling windows. Similar to our examinations of detection rate and the
and mean tend to be proportional and the lower bound becomes number of cameras, spatial heterogeneity among sites most likely
less constraining at higher densities (Guisan, Edwards, & Hastie, influences these site-specific detection rates and capture rates
2002). Another possibility could be that since the count cannot go might therefore be subject to more local variation in detectability
 |
12      
Methods in Ecology and Evolu on KAYS et al.

due to microhabitat variation, which should be accounted for in peaked in early summer, white-tailed deer peaked in late summer,
model-based approaches (Kolowski & Forrester, 2017). and mule deer peaked at the transition between winter and summer
In practice, species richness and relative abundance measures (Figure S2c).
for a study area are not taken from single camera points but de-
rived from an array. Thus, our analysis of the performance of single
cameras is most relevant when considering the return on invest- 4.4 | Study design recommendations
ment of time, rather than absolute error rates of each individual
camera. In this case, the stabilization of detection rate between 2 Based on these analyses, we recommend that studies aimed at es-
and 3 weeks suggests that 3 weeks should be the minimum deploy- timating species richness and relative abundance/occupancy of
ment length, while estimates will continue to improve up to about mammal species use arrays of at least 40–60 camera traps run for
1 month. These recommendations agree with earlier studies that 3–5 weeks. Studies targeting rare species will need more camera lo-
point out that increasing the number of locations is generally better cations but could also benefit from the use of target-specific attract-
than sampling longer at the same site for detection rate (Si et al., ants or more adaptive study designs (Pacifici et al., 2016). Studies
2014), but our results draw from a broader dataset of species and aimed at estimating species diversity over small areas (i.e. <1  km2)
localities. may use fewer cameras (~20), but should note this will be an incom-
plete record of diversity over larger scales. Study area averages of
detection rate required more effort to estimate precisely than oc-
4.3 | Seasonality cupancy, due to high variation within camera arrays, so we recom-
mend comparisons of detection rates be model based and include
All five sites we examined for seasonal dynamics had a mixture of local covariates to help explain small-scale variation, which should
species with and without seasonal variation in occupancy and/or rela- also provide additional ecological inference about the target species.
tive abundance. Thus, sampling an area in only one season would not Finally, comparisons across study areas or over time must account
necessarily yield estimates of animal abundance and species richness for seasonality, which had strong impacts on mammal communities
that are representative. The number of species at each site showing in both tropical and temperate sites.
some seasonal trends was high, ranging from 37% to 50% of the mam-
mal community, without obvious latitudinal trends (Table 1). However, AC K N OW L E D G E M E N T S
the magnitude of the effect was roughly two times stronger in the We appreciate support for the original camera trapping field
two temperate study areas we analysed, where species are known work from National Natural Science Foundation of China
to migrate, hibernate or seasonally cache food. Hibernation and sea- (#31572250, #31872210), Riverbanks Zoo and Gardens, Satch
sonal migration are not known for the mammal species that showed Krantz Conservation Fund, Francis Marion University Professional
seasonal variation in Peru and Panama, and we suspect that these Development Fund and International Collaboration Grants,
patterns could be caused by seasonal breeding or shifts in foraging University of North Carolina-Wilmington Charles L. Cahill Grant,
strategies in response to changes in fruit availability. Species detect- US National Science Foundation and San Diego Zoo Global. Data
ability is also known to change from wet to dry season in Panama from the Tropical Ecology Assessment and Monitoring (TEAM)
(Rowcliffe, Carbone, Jansen, Kays, & Kranstauber, 2011), highlighting Network are the result of a collaboration between Conservation
the importance of using an analytical framework that accounts for de- International, the Smithsonian Institution and the Wildlife
tectability. The subtropical montane Chinese study area had similar Conservation Society, and partially funded by these institutions,
effect sizes to the temperate study areas. Some of the specific pat- the Gordon and Betty Moore Foundation and other donors. We
terns for the Chinese species can be explained by known behaviours also acknowledge logistical, permitting and field support from
of reducing activity during the cold (Masked palm civet; Zhou et al., the Ecuador Ministry of the Environment, Sumaco National Park,
2014), elevational migration (Hairy-fronted muntjac; Zheng, Bao, Ge, Jonas Nilsson, Wildsumaco Wildlife Sanctuary, Instituto Nacional
& Zheng, 2005) or seasonal breeding (Chinese white-bellied rat; Bao, de Pesquisas da Amazônia (INPA), Conservation International
1993). Suriname, Organization for Tropical Studies, Uganda Wildlife
While it might not be surprising that all animal communities Authority, Museo Tridentino di Scienze Naturali and Institute of
would have strong seasonal effects, our results illustrate the im- Tropical Forest Conservation. Jen Zhou for maintaining and organ-
portance of taking season into account when planning field surveys izing eMammal data for analysis.
or making comparisons between existing data. Although some of
these changes were predictable based on animal biology, most were AU T H O R S ' C O N T R I B U T I O N S
not, especially in tropical sites (Table 1). Furthermore, the simple R.K., C.B., H.M.B., M.V.C., C.P.H., A.W.P., B.S.P., S.G.S., K.P. and
two-season comparison we used for our analyses belies a more com- M.S. conceived the ideas, designed the methodology and analysed
plicated temporal pattern seen for many species when looking at the data; B.S.A., M.B.-W., M.B., S.F.B., P.D., S.E., A.L.S.G., P.A.J.,
variation in abundance over a finer (i.e. monthly) scale. For example, T.W.K., M.G.M.L., F.S., F.R., D.S., X.S., W.R.S. and W.J.M. con-
at our Montana study area, Columbian ground squirrel detections tributed data; R.K. led the writing of the manuscript. All authors
KAYS et al.       13
Methods in Ecology and Evolu on |
contributed critically to the drafts and gave final approval for Jansen, P. A., Ahumada, J. A., Fegraus, E. H., & O’Brien, T. G. (2014).
TEAM: A standardised camera trap survey to monitor terrestrial
publication.
vertebrate communities in tropical forests. In P. Fleming, P. Meek,
P. Banks, G. Ballard, A. Claridge, J. Sanderson, & D. Swann (Eds.),
DATA AVA I L A B I L I T Y S TAT E M E N T Camera trapping: Wildlife management and research (pp. 263–270).
All camera trap detection data used in this paper are available at Dryad Collingwood, Australia: CSIRO Publishing.
Digital Repository https://doi.org/doi.org/10.5061/dryad.p8cz8​w9m2 Kays, R., Arbogast, B. S., Baker-Whatton, M., Beirne, C., Boone, H. M.,
Bowler, M., … Spironello, W. R. (2020). Data from: An empirical eval-
(Kays et al., 2020).
uation of camera trap study design: How many, how long and when?
Dryad Digital Repository, https://doi.org/doi.org/10.5061/dryad.p8cz8​
ORCID w9m2
Roland Kays  https://orcid.org/0000-0002-2947-6665 Kays, R., Parsons, A. W., Baker, M. C., Kalies, E. L., Forrester, T., Costello,
R., … McShea, W. J. (2017). Does hunting or hiking affect wildlife
Patrick A. Jansen  https://orcid.org/0000-0002-4660-0314
communities in protected areas? Journal of Applied Ecology, 54,
Marcela Guimarães Moreira Lima  https://orcid. 242–252. https://doi.org/10.1111/1365-2664.12700
org/0000-0002-2203-7598 Kolowski, J. M., & Forrester, T. D. (2017). Camera trap placement and the
Arielle W. Parsons  https://orcid.org/0000-0003-1076-2896 potential for bias due to trails and other features. PLoS ONE, 12(10),
e0186679. https://doi.org/10.1371/journ​al.pone.0186679
Xingfeng Si  https://orcid.org/0000-0003-4465-2759
Larrucea, E. S., Brussard, P. F., Jaeger, M. M., & Barrett, R. H. (2007). Cameras,
coyotes, and the assumption of equal detectability. Journal of Wildlife
REFERENCES Management, 71, 1682–1689. https://doi.org/10.2193/2006-407
Bao, Y. (1993). A summary of studies on Rattus niviventer. Journal of Lashley, M. A., Cove, M. V., Chitwood, M. C., Penido, G., Gardner, B.,
Zhejiang Normal University (Natural Science), 16, 50–54. Deperno, C. S., & Moorman, C. E. (2018). Estimating wildlife activity
Beaudrot, L., Ahumada, J., O'Brien, T. G., & Jansen, P. A. (2019). Detecting curves: Comparison of methods and sample size. Scientific Reports, 8,
tropical wildlife declines through camera-trap monitoring: An evalua- 4173. https://doi.org/10.1038/s41598-018-22638-6
tion of the tropical ecology assessment and Monitoring protocol. Oryx, MacKenzie, D. I., Nichols, J. D., Lachman, G. B., Droege, S., Royle, J. A., &
53(1), 126–129. https://doi.org/10.1017/S0030​60531​8000546 Langtimm, C. A. (2002). Estimating site occupancy rates when detec-
Burnham, K. P., & Anderson, R. D. (2002). Model selection and multi-model tion probabilities are less than one. Ecology, 83, 2248–2255. https://
inference. New York, NY: Springer. doi.org/10.1890/0012-9658(2002)083[2248:ESORW​D]2.0.CO;2
Burton, A. C., Neilson, E., Moreira, D., Ladle, A., Steenweg, R., Fisher, J. TEAM Network. (2008). Terrestrial vertebrate (camera trap) protocol imple-
T., … Boutin, S. (2015). Wildlife camera trapping: A review and rec- mentation manual, V. 3.0. Arlington, VA: Tropical Ecology, Assessment
ommendations for linking surveys to ecological processes. Journal and Monitoring Network, Center for Applied Biodiversity Science,
of Applied Ecology, 52, 675–685. https://doi.org/10.1111/1365- Conservation International.
2664.12432 Pacifici, K., Reich, B. J., Dorazio, R. M., & Conroy, M. J. (2016). Occupancy
Cove, M. V., Spínola, R. M., Jackson, V. L., Sàenz, J. C., & Chassot, O. estimation for rare species using a spatially-adaptive sampling de-
(2013). Integrating occupancy modeling and camera-trap data to es- sign. Methods in Ecology and Evolution, 7(3), 285–293. https://doi.org/
timate medium and large mammal detection and richness in a Central 10.1111/2041-210X.12499
American biological corridor. Tropical Conservation Science, 6(6), Pacifici, K., Zipkin, E. F., Collazo, J. A., Irizarry, J. I., & DeWan, A. (2014).
781–795. https://doi.org/10.1177/19400​82913​0 0600606 Guidelines for a priori grouping of species in hierarchical commu-
Dirzo, R., Young, H. S., Galetti, M., Ceballos, G., Isaac, N. J. B., & Collen, nity models. Ecology and Evolution, 4(7), 877–888. https://doi.org/
B. (2014). Defaunation in the Anthropocene. Science, 345(6195), 10.1002/ece3.976
401–406. https://doi.org/10.1126/scien​ce.1251817 Palmer, M. S., Swanson, A., Kosmala, M., Arnold, T., & Packer, C. (2018).
Dorazio, R. M., Kéry, M., Royle, J. A., & Plattner, M. (2010). Models Evaluating relative abundance indices for terrestrial herbivores
for inference in dynamic metacommunity systems. Ecology, 91(8), from large-scale camera trap surveys. African Journal of Ecology, 56,
2466–2475. https://doi.org/10.1890/09-1033.1 791–803. https://doi.org/10.1111/aje.12566
Efford, M., & Dawson, D. (2012). Occupancy in continuous habitat. Parsons, A. W., Forrester, T., McShea, W. J., Baker-Whatton, M. C.,
Ecosphere, 3, 1–15. https://doi.org/10.1890/ES11-00308.1 Millspaugh, J. J., & Kays, R. (2017). Do occupancy or detection rates
Fiske, I., & Chandler, R. (2011). unmarked: An R package for fitting hi- from camera traps reflect deer density? Journal of Mammalogy, 98,
erarchical models of wildlife occurrence and abundance. Journal of 1547–1557. https://doi.org/10.1093/jmamm​al/gyx128
Statistical Software, 43(10), 1–23. https://doi.org/10.18637/​jss.v043. R Core Development Team. (2018). R, A language and environment for sta-
i10 tistical computing. Vienna, Austria: R Core Team.
Gálvez, N., Guillera-arroita, G., Morgan, B. J. T., & Davies, Z. G. (2016). Rich, L. N., Davis, C. L., Farris, Z. J., Miller, D. A. W., Tucker, J. M., Hamel,
Cost efficient effort allocation for camera-trap occupancy surveys S., … Kelly, M. J. (2017). Assessing global patterns in mammalian
of mammals. Biological Conservation, 204, 350–359. https://doi. carnivore occupancy and richness by integrating local camera trap
org/10.1016/j.biocon.2016.10.019 surveys. Global Ecology and Biogeography, 26(8), 918–929. https://doi.
Guillera-Arroita, G., Ridout, M. S., & Morgan, B. J. T. (2010). Design of occu- org/10.1111/geb.12600
pancy studies with imperfect detection. Methods in Ecology and Evolution, Rowcliffe, M. J., Carbone, C., Jansen, P. A., Kays, R., & Kranstauber, B. (2011).
1, 131–139. https://doi.org/10.1111/j.2041-210X.2010.00017.x Quantifying the sensitivity of camera traps using an adapted distance
Guisan, A., Edwards, T. C., & Hastie, T. (2002). Generalized linear and gen- sampling approach. Methods in Ecology and Evolution, 2, 467–476.
eralized additive models in studies of species distributions: Setting Rowcliffe, M., Field, J., Turvey, S., & Carbone, C. (2008). Estimating
the scene. Ecological Modeling, 157, 89–100. https://doi.org/10.1016/ animal density using camera traps without the need for individual
S0304-3800(02)00204-1 recognition. Journal of Applied Ecology, 45, 1228–1236. https://doi.
Hsieh, T. C., Ma, K. H., & Chao, A. (2016). iNEXT: An R package for rar- org/10.1111/j.1365-2664.2008.01473.x
efaction and extrapolation of species diversity. Methods in Ecology and Royle, J. A., & Link, W. A. (2006). Generalized site occupancy models
Evolution, 7, 1451–1456. https://doi.org/10.1111/2041-210X.12613 allowing for false positive and false negative errors. Ecology, 874,
 |
14      
Methods in Ecology and Evolu on KAYS et al.

835–841. https://doi.org/10.1890/0012-9658(2006)87[835:GSOMA​F] Wedderburn, R. W. M. (1974). Quasi-likelihood functions, generalized

2.0.CO;2 linear models, and the gauss-newton method. Biometrika, 61(3),
Shannon, G., Lewis, J. S., & Gerber, B. D. (2014). Recommended survey 439–447. https://doi.org/10.1093/biome​t /61.3.439
designs for occupancy modelling using motion-activated cameras: Zheng, X., Bao, Y., Ge, B., & Zheng, R. (2005). The current distribution
Insights from empirical wildlife data. PeerJ, 2, e532. https://doi. and conservation status of black muntjac in Gutian Mountain Nature
org/10.7717/peerj.532 Reserve. Journal of Natural Resources, 20, 508–513. Retrieved from
Si, X., Kays, R., & Ding, P. (2014). How long is enough to detect terrestrial http://en.cnki.com.cn/Artic​le_en/CJFDT​OTAL-ZRZX2​00504​005.htm
animals? Estimating the minimum trapping effort on camera traps. Zhou, Y., Newman, C., Palomares, F., Zhang, S., Xie, Z., & Macdonald, D. W.
PeerJ, 2, e374. https://doi.org/10.7717/peerj.374 (2014). Spatial organization and activity patterns of the masked palm
Sollmann, R., Mohamed, A., Samejima, H., & Wilting, A. (2013). Risky civet (Paguma larvata) in central-south China. Journal of Mammalogy,
business or simple solution – Relative abundance indices from 95(3), 534–542. https://doi.org/10.1644/13-MAMM-A-185
camera-trapping. Biological Conservation, 159, 405–412. https://doi.
org/10.1016/j.biocon.2012.12.025
S U P P O R T I N G I N FO R M AT I O N
Specht, H. M., Reich, H. T., Iannarilli, F., Edwards, M. R., Stapleton, S. P.,
Weegman, M. D., … Arnold, T. W. (2017). Occupancy surveys with Additional supporting information may be found online in the
conditional replicates: An alternative sampling design for rare spe- Supporting Information section.
cies. Methods in Ecology and Evolution, 8(12), 1725–1734. https://doi.
org/10.1111/2041-210X.12842
Steenweg, R., Hebblewhite, M., Kays, R., Ahumada, J., Fisher, J. T., Burton, How to cite this article: Kays R, Arbogast BS, Baker-Whatton
C., … Rich, L. N. (2017). Scaling up camera traps: Monitoring the plan- M, et al. An empirical evaluation of camera trap study design:
et's biodiversity with networks of remote sensors. Frontiers in Ecology
How many, how long and when? Methods Ecol Evol.
and the Environment, 15, 26–34. https://doi.org/10.1002/fee.1448
Wearn, O. R., & Glover-Kapfer, P. (2017). Camera-trapping for conserva- 2020;00:1–14. https://doi.org/10.1111/2041-210X.13370
tion: A guide to best-practices. Woking, UK: WWF.
SUPPLEMENTAL MATERIAL

Table S1. Camera trapping data sets used for study design analysis.
Analysis
Camera Spatial Temporal Seasonal
Name Region Location Latitude Longitude Source Deployments Spacing Arrangement (n=28) (n=16) (n=5)
Kenya Wildlife Stratified
-3.372440 39.873430
Service Project Africa Kenya eMammal 30 Large Random x x
Korup Africa Cameroon 5.046442 8.839358 Team 60 Large Grid x
Nouabala Ndoki Africa Congo 2.519552 16.513092 Team 64 Large Grid x
Ranomafana Africa Madagascar -21.22491 47.473436 Team 60 Large Grid x
Udzungwa Africa Tanzania -7.783217 36.868785 Team 61 Large Grid x
Virunga Massif Africa Rwanda -1.454465 29.540277 Team 60 Large Grid x
Bukit Barisan Asia Indonesia -5.658888 104.465402 Team 60 Large Grid x
Carnivore
Intraguild
Interactions in 14.156110 102.475360
Select Thailand Stratified
Reserves Asia Thailand eMammal 30 Large Random x
Nam Kading Asia Laos 18.398413 104.176223 Team 60 Large Grid x
Pasoh Asia Malaysia 3.045081 102.317913 Team 60 Large Grid x
Stratified
29.254127 118.119625
Gutianshan Plot Asia China Other 19 Small Random x x x
Central Stratified
9.155988 -79.843884
BCI Plots America Panama Other 77 Small Random x
Central Stratified
9.175022 -79.148352
Gonzolillo America Panama eMammal 25 Small Random x
 Central
9.159299 -79.808457
Panama Team America Panama Team 60 Large Grid x
Central
10.352668 -84.046701
Volcan Barva America Costa Rica Team 60 Large Grid x
Speulderbos Europe Netherlands 52.253420 5.699852 eMammal 42 Small Grid x x
Czech
48.665098 14.706484
Zofin Europe Republic eMammal 25 Small Grid x
Albany Area
Camera Trapping North 42.723440 -73.862510 Stratified
Project America New York eMammal 30 Large Random x
North
47.073325 -113.25250
Bandy Ranch America Montana eMammal 20 Large Grid x
North Carolina North North Stratified
35.986940 -79.395830
Candid Critters America Carolina eMammal 30 Large Random x
North North Stratified
35.813957 -78.714713
Prairie Ridge America Carolina eMammal 163 Small Random x x
North
48.107133 -112.67768
Roosevelt Ranch America Montana eMammal 215 Large Grid x x x
North
38.893386 -78.147111
SCBI America Virginia eMammal 49 Small Grid x
North
38.894766 -78.146253
SCBI Grid America Virginia eMammal 28 Small Grid x
North
38.891784 -76.559497
SCBI2 America Virginia eMammal 46 Small Grid x
North North
35.817065 -78.726349
Schenck America Carolina eMammal 54 Small Grid x
North
38.517662 -90.557613
TRC America Missouri eMammal 47 Small Grid x
North
37.015170 -122.07271
UCSC Grid America California eMammal 49 Small Grid x
South
-1.771972 -51.512277
Caxiuana America Brazil Team 60 Large Grid x
South
-2.663266 -59.968673
Manaus America Brazil Team 90 Large Grid x
Peperpot Nature South Stratified
5.798960 -55.116270
Park America Suriname eMammal 18 Small Random x
Peruvian Amazon South
-1.779528 -75.640346
Terrestrial America Peru eMammal 23 Large Grid x
Peruvian Amazon South
-3.1 -72.9
Canopy America Peru Other 41 Large Grid x
Sumaco Cloud South Stratified
-0.566570 -77.596560
Forest America Ecuador eMammal 30 Small Random x
South
4.775049 -56.171565
Suriname America Suriname Team 65 Large Grid x
Bayuelin Nature Stratified
29.043012 102.990763
Reserve Asia China eMammal 40 Small Random x
Niuweihe Nature Stratified
32.1744 104.1853
Reserve Asia China eMammal 55 Large Random x
Okaloosa North Stratified
30.632937 -86.615033
S.C.I.E.N.C.E America Florida eMammal 72 Small Random x
Stratified
0.462391 36.971745
Mpala Primary Africa Kenya eMammal 10 Small Random x

HKK ForestGEO Asia Thailand 15.62918 99.22246 eMammal 77 Small Grid x

South
-0.680300 -76.434331
Yasuni America Ecuador Team 60 Large Grid x
Table S2. Family and scientific name for 106 species of mammals presented in this paper.

Family Genus and species Common Name

Aotidae Aotus vociferans Spix's night monkey
Atelidae Alouatta seniculus Red howler monkey
Atelidae Alouatta palliata Mantled howler monkey
Atelidae Lagothrix lagothricha Common woolly monkey
Bovidae Capricornis sumatraensis Sumatran serow
Bovidae Cephalophus leucogaster White-bellied Duiker
Bovidae Cephalophus harveyi Harvey's duiker
Bovidae Philantomba monticola Blue duiker
Bovidae Tragelaphus scriptus Bushbuck
Callitrichidae Cebuella pygmaea Pygmy marmosetd
Callitrichidae Saguinus nigricollis Black-mantled tamarin
Caluromys Caluromys lanatus Western woolly opossum
Canidae Canis lupus Gray wolf
Canidae Canis latrans Coyote
Canidae Urocyon cinereoargenteus Gray fox
Canidae Vulpes vulpes Red fox
Cebidae Cebus capucinus White-faced capuchin
Cebidae Cebus albifrons White-fronted capuchin
Cebidae Saimiri sciureus Common squirrel monkey
Cercopithecidae Macaca nemestrina Southern pig-tailed macaque
Cervidae Alces alces Moose
Cervidae Capreolus capreolus Roe deer
Cervidae Cervus canadensis Elk
Cervidae Cervus elaphus Red deer
Cervidae Mazama americana Red brocket deer
Cervidae Muntiacus reevesi Reeves's muntjac
Cervidae Muntiacus crinifrons Hairy-fronted muntjac
Cervidae Odocoileus virginianus White-tailed eer
Cervidae Odocoileus hemionus Mule deer
Cuniculidae Cuniculus paca Lowland paca
Dasypodidae Dasypus kappleri Greater Long-nosed armadillo
Dasypodidae Dasypus novemcinctus Nine-banded armadillo
Dasyproctidae Dasyprocta punctata Central American agouti
Dasyproctidae Dasyprocta leporina Red-rumped agouti
Dasyproctidae Myoprocta acouchy Red acouchi
Didelphidae Didelphis virginiana Virginia opossum
Didelphidae Didelphis marsupialis Common opossum
Didelphidae Marmosa robinsoni Robinson's mouse opossum
Didelphidae Monodelphis sp. Short-tailed opossum
Didelphidae Philander opossum Gray four-eyed opossum
Elephantidae Loxodonta africana African elephant
Erethinzontidae Coendou ichillus Streaked dwarf porcupine
Erethizontidae Coendou prehensilis Brazilian porcupine
Erethizontidae Erethizon dorsatum North American porcupine
Eupleridae Fossa fossana Malagasy civet
Felidae Leopardus pardalis Ocelot
Felidae Leopardus tigrinus Oncilla
Felidae Leopardus wiedii Margay
Felidae Leptailurus serval Serval
Felidae Lynx rufus Bobcat
Felidae Panthera onca Jaguar
Felidae Prionailurus bengalensis Leopard cat
Felidae Puma concolor Puma
Herpestidae Herpestes naso Long-nosed mongoose
Hystricidae Atherurus macrourus Asiatic brush-tailed porcupine
Leporidae Lepus townsendii White-tailed jackrabbit
Leporidae Lepus americanus Snowshoe hare
Leporidae Sylvilagus nuttallii Mountain cottontail
Leporidae Sylvilagus floridanus Eastern cottontail
Macroscelididae Petrodromus tetradactylus Four-toed elephant shrew
Megalonychidae Choloepus didactylus Linnaeus's two-toed sloth
Mephititdae Mephitis mephitis Striped skunk
Muridae Leopoldamys edwardsi Edwards's long-tailed giant rat
Muridae Niviventer confucianus Chinese white-bellied rat
Mustelidae Arctonyx collaris Hog badger
Mustelidae Eira barbara Tayra
Mustelidae Lontra canadensis River otter
Mustelidae Martes flavigula Yellow-throated marten
Mustelidae Martes martes Pine marten
Mustelidae Meles meles European badger
Mustelidae Mellivora capensis Honey badger
Mustelidae Melogale moschata Chinese ferret-badger
Mustelidae Mustela frenata Long-tailed weasel
Mustelidae Taxidea taxus American badger
Myrmecophagidae Tamandua mexicana Northern tamandua
Myrmecophagidae Tamandua tetradactyla Southern tamandua
Nesomyidae Cricetomys gambianus Northern giant pouched rat
Nesomyidae Nesomys rufus Island mouse
Pitheciidae Cheracebus lucifer Yellow-handed titi monkey
Pitheciidae Pithecia hirsuta Hairy saki
Procyonidae Bassaricyon alleni Eastern lowland olingo
Procyonidae Nasua narica White-nosed coati
Procyonidae Potos flavus Kinkajou
Procyonidae Procyon lotor Northern raccoon
Procyonidae Procyon cancrivorus Crab-eating raccoon
Sciuridae Callosciurus erythraeus Pallas's squirrel
Sciuridae Glaucomys volans Southern flying squirrel
Sciuridae Marmota monax Woodchuck
Sciuridae Microsciurus sp. Squirrel
Sciuridae Sciurus carolinensis Eastern gray squirrel
Sciuridae Sciurus granatensis Red-tailed squirrel
Sciuridae Sciurus sp. Tree squirrel
Sciuridae Tamias amoenus Yellow-pine chipmunk
Sciuridae Tamias striatus Eastern chipmunk
Sciuridae Tamiasciurus hudsonicus American red squirrel
Sciuridae Urocitellus columbianus Columbian ground squirrel
Suidae Phacochoerus africanus Common Warthog
Suidae Sus scrofa Wild Boar
Tapiridae Tapirus bairdii Baird's tapir
Tapiridae Tapirus terrestris Lowland tapir
Tayassuidae Pecari tajacu Collared peccary
Tragulidae Tragulus kanchil Lesser mouse-deer
Ursidae Ursus americanus American black bear
Ursidae Ursus arctos Brown bear
Viverridae Genetta maculata Large-spotted genet
Viverridae Paguma larvata Masked palm civet
Table S3. Estimates of how many cameras are needed to estimate richness within 5% of the
maximum number of species detected in different protected areas around the world representing
both temperate and tropical climates and having both small (≤0.2km) and large (>0.2km) camera
spacing.
# Cameras to detect 95% of
Protected Area species Spacing Type
Arabuko 17 Large Tropical
Bandy Ranch 17 Large Temperate
BCI 42 Large Tropical
Bukit Barisan 35 Large Tropical
Caxiuana 30 Large Tropical
Gonzolillo 12 Small Tropical
Korup 42 Large Tropical
Manaus 52 Large Tropical
Nam Kading 24 Large Tropical
Nouabala Ndoki 43 Large Tropical
Pasoh 37 Large Tropical
Ranomafana 29 Large Tropical
Roosevelt Ranch 18 Large Temperate
SCBI 38 Small Temperate
SCBI Grid 20 Small Temperate
SCBI2 12 Small Temperate
Schenck 28 Small Temperate
Si 14 Small Temperate
Speulderbos 26 Small Tropical
Suriname 29 Large Tropical
TRC 36 Small Temperate
UCSC Grid 23 Small Temperate
Udzungwa 42 Large Tropical
Virunga Massif 41 Large Tropical
Volcan Barva 32 Large Tropical
Yasuni 37 Large Tropical
Zofin 19 Small Tropical

Table S4. Estimates of how many cameras are needed to precisely estimate occupancy in
different protected areas around the world representing both temperate and tropical climates and
having both small (<500m) and large (>500m) camera spacing. We used two metrics to
determine at which number of cameras precision no longer improved: (1) the width of the 95%
confidence interval changed <1% and (2) the standard error was <10% of the mean, regardless of
whether more cameras were added. Protected areas for which the maximum number of cameras
did not appear to be enough for precision to asymptote are marked as “DNLO” (i.e., Did Not
Level Off). Otherwise, the minimum number of cameras necessary for a precise estimate is
listed. We also provide 95% Lower Confidence Limits (LCL) and Upper Confidence Limits
(UCL) for the occupancy estimate.

Point of
Occupancy Diminishing Point of
Site Species probability LCL UCL returns Precision
Arabuko Large-spotted genet 0.965 0.588 0.999 22 5
Arabuko Four-toed elephant shrew 0.907 0.700 0.990 13 8
BCI Central American agouti 0.961 0.662 0.998 15 5
BCI Ocelot 0.482 0.265 0.958 24 DNLO
Bukit Barisan Southern pig-tailed macaque 0.859 0.652 1.001 17 DNLO
Bukit Barisan Wild boar 0.967 0.425 1.000 28 DNLO
Caxiuana Greater long-nosed armadillo 0.304 0.145 0.661 20 DNLO
Caxiuana Collared peccary 0.431 0.224 0.877 33 DNLO
Korup Long-nosed mongoose 1.000 0.211 1.000 19 10
Korup Blue duiker 0.850 0.647 0.975 14 29
Manaus Nine-banded armadillo 0.681 0.474 0.995 6 DNLO
Manaus Red acouchi 0.706 0.563 0.833 22 50
Nam Kading Asiatic brush-tailed porcupine 0.374 0.211 0.617 46 DNLO
Nam Kading Wild boar 0.520 0.224 1.000 8 DNLO
Nouabala Ndoki White-bellied duiker 0.153 0.062 0.301 24 DNLO
Nouabala Ndoki African elephant 0.521 0.352 0.684 26 DNLO
Pasoh Southern pig-tailed macaque 0.946 0.622 0.998 6 5
Pasoh Lesser mouse-deer 0.271 0.149 0.448 15 DNLO
Ranomafana Malagasy civet 0.939 0.682 1.000 17 DNLO
Ranomafana Island mouse 0.513 0.294 0.953 13 DNLO
SCBI Grid White-tailed deer 1.000 0.811 1.000 13 5
SCBI Grid American black bear 0.985 0.612 1.000 16 8
Schenck White-tailed deer 1.000 0.892 1.000 6 5
Schenck Eastern gray squirrel 0.965 0.573 0.993 DNLO DNLO
Speulderbos Roe deer 0.852 0.659 0.982 16 32
Speulderbos Wild boar 0.973 0.740 0.999 11 5
Suriname Red-rumped agouti 0.931 0.546 0.997 15 5
Suriname Nine-banded armadillo 0.841 0.314 1.000 14 DNLO
TRC White-tailed deer 0.964 0.797 0.999 15 15
TRC Eastern gray squirrel 0.568 0.405 0.743 52 DNLO
UCSC Grid Mule deer 0.838 0.644 0.962 17 22
 Udzungwa Harvey's duiker 0.888 0.620 0.989 17 11
Udzungwa Honey badger 0.802 0.310 1.000 29 DNLO
Virunga Massif Serval 0.376 0.137 0.995 29 DNLO
Virunga Massif Bushbuck 0.832 0.633 0.950 18 20
Volcan Barva Lowland paca 0.483 0.231 0.990 14 DNLO
Volcan Barva Collared peccary 0.849 0.663 0.981 16 15
Yasuni Tayra 0.421 0.221 0.913 13 DNLO
Yasuni Collared peccary 0.892 0.651 0.990 DNLO 44

Table S5. Estimates of how many cameras are needed to precisely estimate detection rate (DR)
in different protected areas around the world. We used two metrics to determine at which
number of cameras precision no longer improved: (1) the width of the 95% confidence interval
changed <1% and (2) the standard error was <10% of the mean, regardless of whether more
cameras were added. Protected areas for which the maximum number of cameras did not appear
to be enough for precision to asymptote are marked as DNLO (i.e. Did Not Level Off).
Otherwise, the minimum number of cameras necessary for a precise estimate is listed. We also
provide 95% Lower Confidence Limits (LCL) and Upper Confidence Limits (UCL) for the
detection rate estimate.
#
Cameras
for CI # Cameras
width for
Mean change SE/mean
Site Species DR SE LCL UCL <1% <10%
Large-spotted
Arabuko genet 0.15 0.03 -0.1 0.41 DNLO DNLO
Four-toed
Arabuko elephant shrew 2.3 0.45 -1.9 6.49 DNLO DNLO
Central American
BCI agouti 1.75 0.56 -3.53 7.04 DNLO DNLO
BCI Ocelot 0.08 0.01 -0.05 0.2 7 DNLO
Southern pig-
Bukit Barisan tailed macaque 0.19 0.03 -0.05 0.43 18 DNLO
Bukit Barisan Wild boar 0.07 0.01 -0.04 0.19 DNLO DNLO
Greater long-
Caxiuana nosed armadillo 0.14 0.03 -0.11 0.39 DNLO DNLO
Caxiuana Collared peccary 0.93 0.41 -2.95 4.82 DNLO DNLO
Long-nosed
Korup mongoose 0.05 0.01 -0.01 0.12 26 DNLO
Korup Blue duiker 0.36 0.08 -0.34 1.07 DNLO DNLO
Nine-banded
Manaus armadillo 0.08 0.01 -0.01 0.17 DNLO 33
Manaus Red acouchi 0.45 0.16 -1.09 1.98 DNLO DNLO
Asiatic brush-
Nam Kading tailed porcupine 0.14 0.03 -0.13 0.41 DNLO DNLO
Nam Kading Wild boar 0.4 0.2 -1.43 2.24 DNLO DNLO
Nouabala White-bellied
Ndoki duiker 0.06 0.01 -0.02 0.15 25 DNLO
Nouabala
Ndoki African elephant 0.15 0.03 -0.09 0.39 16 DNLO
Southern pig-
Pasoh tailed macaque 0.4 0.07 -0.27 1.07 DNLO DNLO
Lesser mouse-
Pasoh deer 0.15 0.04 -0.2 0.5 DNLO DNLO
Ranomafana Malagasy civet 0.08 0.02 -0.07 0.24 24 DNLO
Ranomafana Island mouse 0.06 0.01 -0.08 0.2 13 DNLO
SCBI Grid White-tailed deer 0.8 0.08 0.06 1.53 DNLO DNLO
American black
SCBI Grid bear 0.07 0.01 -0.01 0.15 DNLO DNLO
Schenck White-tailed deer 1.1 0.1 0.14 2.06 34 31
Eastern gray
Schenck squirrel 0.24 0.05 -0.25 0.73 DNLO DNLO
Speulderbos Roe deer 0.14 0.03 -0.13 0.41 DNLO DNLO
Speulderbos Wild boar 0.61 0.08 -0.18 1.4 DNLO DNLO
Red-rumped
Suriname agouti 0.33 0.06 -0.22 0.89 DNLO DNLO
Nine-banded
Suriname armadillo 0.05 0.01 -0.02 0.13 15 DNLO
TRC White-tailed deer 0.53 0.09 -0.29 1.36 DNLO DNLO
Eastern gray
TRC squirrel 0.24 0.06 -0.3 0.78 DNLO DNLO
UCSC Grid Mule deer 0.4 0.06 -0.16 0.97 DNLO DNLO
Udzungwa Harvey's duiker 0.42 0.12 -0.67 1.51 DNLO DNLO
Udzungwa Honey badger 0.05 0.01 -0.02 0.12 3 DNLO
Virunga Massif Serval 0.05 0.01 -0.02 0.12 15 DNLO
Virunga Massif Bushbuck 0.9 0.18 -0.77 2.57 32 DNLO
Volcan Barva Lowland paca 0.08 0.01 0 0.15 DNLO DNLO
Volcan Barva Collared peccary 0.38 0.06 -0.15 0.91 DNLO DNLO
Yasuni Tayra 0.05 0.01 -0.02 0.11 20 DNLO
Large-spotted
Yasuni genet 0.15 0.03 -0.14 0.45 6 DNLO
Table S6. Effect of deployment duration on site level detection rate for 24 species, including the
study areas, species traits, region, and the number of days to reach <10% error of final detection
rate.
Days to
Final
Trophic reach
Species Study Area Detection Region Size
level <10%
Rate
error
Crab-eating Peperpot Nature
0.02 omnivore tropical medium 51
raccoon Park
Peruvian Amazon
Lowland tapir 0.03 herbivore tropical large 45
Terrestrial
Peruvian Amazon
Jaguar 0.04 carnivore tropical large 48
Terrestrial
Peruvian Amazon
Ocelot 0.04 carnivore tropical medium 55
Terrestrial
European
0.05 carnivore temperate medium 35
badger Speulderbos
Bayuelin Nature
Leopard cat 0.05 carnivore temperate medium 29
Reserve
Yellow-
throated Niuweihe Nature 0.06 carnivore temperate small 37
marten Reserve
Gray four-eyed Peperpot Nature
0.06 omnivore tropical small 58
Opossum Park
Masked Palm Bayuelin Nature
0.07 omnivore temperate small 36
civet Reserve
Pine marten Speulderbos 0.1 carnivore temperate small 23
Niuweihe Nature
Hog badger 0.1 omnivore temperate medium 21
Reserve
Red fox Speulderbos 0.11 omnivore temperate medium 33
Roe deer Speulderbos 0.11 herbivore temperate medium 23
Red deer Speulderbos 0.11 herbivore temperate large 23
Large-spotted Kenya Wildlife
0.12 omnivore tropical small 36
Genet Service Project
Golden
elephant Kenya Wildlife 0.15 carnivore tropical small 24
Shrew Service Project
Wild pig HKK ForestGEO 0.2 omnivore tropical medium 16
Eastern gray
0.22 herbivore temperate small 50
Squirrel SCBI Grid
Warthog Mpala Primary 0.24 omnivore tropical medium 37
North Carolina
Black bear 0.28 omnivore temperate large 40
Candid Critters
African
0.31 herbivore tropical large 30
elephant Mpala Primary
White-tailed Okaloosa
0.31 herbivore temperate large 22
deer S.C.I.E.N.C.E
Kenya Wildlife
Pouched rat 0.41 herbivore tropical small 37
Service Project
Red-rumped Peperpot Nature
0.5 herbivore tropical small 17
agouti Park
Table S7 Model selection results from Poisson regression analyses examining the influence of
size (small, medium, or large), trophic guild (carnivore, omnivore, or herbivore), region
(temperate or tropical), detection rate, and the maximum number of sampling days on the
number of days of camera trap sampling before the percent error of the estimated detection rate
decreases below 10%. Data come from 24 representative species occurring on 5 continents from
the eMammal camera trap database (Table S5).

Model K AICc Delta_AICc AICcWt LL

Final Detection Rate 2 215.13 0 0.97 -105.28
Region 2 222.67 7.54 0.02 -109.05
Herbivores 2 226.37 11.24 0 -110.9
Null 1 228.43 13.3 0 -113.12
Diet 3 228.96 13.83 0 -110.88
Small 2 230.47 15.34 0 -112.95
Max days 2 230.52 15.39 0 -112.97
Size 3 232.91 17.78 0 -112.86

Table S8. Sample size for seasonal analysis.

All mammals
detection rate
Study Area Season Months Trap Nights (n/day) No. Species
China Wet May-October 8,944 0.26 10
Dry November-April 6,786 0.16 9
Peru Wet November-April 3,982 0.15 19
Dry May-October 5,797 0.16 22
Panama Wet July-December 2,839 2.12 14
Dry January-June 3,225 2.54 15
Montana Winter October-March 4,410 0.86 19
Summer April-September 2,377 0.94 18
North Carolina Winter October-March 3,272 3.82 13
Summer April-September 2,886 2.70 10
Table S9. Comparisons of seasonal effects on detection and occupancy of 70 mammal species across five study areas. Detection rates
are per camera day, and model coefficients are bolded when the seasonal effect was significant. These values are also graphed in
Figure S2. Species with a “–” for model coefficients had models that did not converge, usually because of low or uneven sample size.
Detection Detection
Study Rate: Rate: Wet/Winter Wet/Winter Wet/Winter
Area & Wet/ Dry/ Coefficient: Coefficient: Coefficient: Count
Season Species Winter Summer Occupancy (SE) Detection (SE) Model (SE)
China Chinese white-bellied rat 0.12 0.031 1.47 (0.50) 0.68 (0.13) 0.93 (0.41)
Wet/
Dry Masked palm civet 0.013 0.0013 0.77 (0.87) 1.53 (0.69) 1.88 (0.97)
Hairy-fronted muntjac 0.075 0.030 0.06 (0.57) 0.29 (0.15) 0.47 (0.47)
Hog badger 0.013 0.0034 0.02 (0.59) 0.59 (0.41) 1.97 (1.14)
Reeves's muntjac 0.024 0.066 -0.84 (0.47) -0.57 (0.18) -0.98 (0.41)
Pallas's squirrel 0.0027 0.0007 - - 1.54 (1.40)
Wild boar 0.0038 0.025 - - -2.76 (1.52)
Edwards's long-tailed
giant rat 0.0040 0.0034 - - 0.29 (1.26)
Chinese ferret-badger 0.0008 0.0001 - - -
Leopard cat 0.0002 0.00 - - -

Panama Paca 0.1666 0.11 2.05 (1.62) 0.36 (0.18) -0.07 (0.16)
Wet/
Dry White-nosed coati 0.11 0.13 1.44 (1.21) -0.04 (0.21) -0.61 (0.27)
Red brocket deer 0.074 0.12 1.43 (2.66) -0.64 (0.18) -0.68 (0.18)
Collared Peccary 0.35 0.89 0.44 (1.24) -0.39 (0.17) -1.07 (0.24)
Ocelot 0.034 0.024 -1.66 (6.26) 0.43 (0.25) 0.19 (0.18)
Nine-banded armadillo 0.019 0.016 0.05 (0.95) -0.05 (0.40) -0.05 (0.33)
Agouti 1.29 1.17 - - -0.16 (0.11)
Northern tamandua 0.011 0.016 - - -0.99 (0.30)
Red-tailed squirrel 0.080 0.062 - - 0.14 (0.23)
 Tayra 0.0035 0.0074 - - -0.16 (0.50)
Common opossum 0.0092 0.0096 - - -0.35 (0.36)
White faced capuchin 0.0039 0.0019 - - 0.33 (0.71)
Baird's tapir 0.0007 0.0006 - - 0.11 (1.00)
Puma 0.0018 0.0003 - - -
Margay 0 0.0003 - - -
Robinson's Mouse
Opossum 0.0004 0 - - -
Howler monkey 0 0.0003
0 0
Peru Brazilian porcupine 0.0025 0.0074 2.75 (7.29) -2.18 (1.01) -1.21 (1.19)
Wet/
Dry Dwarf porcupine 0.0048 0.0066 1.58 (0.71) -1.27 (0.41) -0.31 (0.95)
Two-toed sloth 0.017 0.0038 0.44 (0.61) 0.97 (0.42) 1.17 (0.82)
Common squirrel
monkey 0.0264 0.017 0.15 (0.55) 0.02 (0.25) 0.69 (0.72)
Olingo 0.0055 0.006 -0.14 (0.75) -0.40 (0.42) -0.68 (0.55)
Common woolly monkey 0.0018 0.0014 -0.21 (0.91) 0.24 (0.80) 0.14 (1.01)
Black-mantled tamarin 0.012 0.012 -0.86 (0.57) 0.47 (0.34) -0.10 (0.73)
Hairy saki monkey 0.0098 0.016 -1.13 (0.55) -0.17 (0.28) -0.56 (0.50)
Western woolly opossum 0.012 0.028 -1.14 (0.51) -0.09 (0.24) -0.82 (0.81)
Red howler monkey 0.0028 0.0021 -1.23 (1.19) 0.81 (0.65)
White-fronted capuchin 0.0065 0.0078 -1.53 (0.84) 0.69 (0.39) -0.07 (0.55)
Spix's night monkey 0.010 0.0081 -1.89 (0.73) 0.83 (0.37) 0.13 (0.71)
Tamandua 0.003 0.0036 -2.18 (1.30) 0.21 (0.74) -0.45 (0.73)
Tayra 0.0003 0.0055 - -
Yellow-handed titi
monkey 0.0028 0.0007 - - 1.97 (1.41)
Kinkajou 0.039 0.0312 - - 0.28 (0.34)
Microsciurus sp. 0.001 0.0045 - - -0.36 (1.17)
 Scurius sp. 0 0.0005 - - 0.60 (1.21)
Oncilla 0.0003 0 - - -
Short-tailed opossum 0.0005 0 - - -
Coati 0 0.0005 - - -
Common opossum 0 0.0014 - - -
Margay 0 0.0005 - - -
Pygmy marmoset 0 0.0002 - - -
0 0
Montana Coyote 0.059 0.070 1.37 (0.45) -0.60 (0.21) 0.23 (0.20)
Winter/
Summer Mule Deer 0.53 0.048 1.25 (0.41) 1.07 (0.29) 0.15 (0.27)
Snowshoe Hare 0.026 0.017 0.45 (0.52) -0.32 (0.44) 0.41 (0.52)
Bobcat 0.0036 0.0042 0.40 (0.96) -0.42 (0.91) -0.17 (0.51)
Puma 0.0025 0.0013 0.28 (1.12) -0.17 (1.18) 0.84 (0.95)
Mountain Cottontail 0.0066 0.0063 0.00 (0.72) -0.18 (0.73) 0.49 (1.16)
White-tailed Deer 0.21 0.69 -0.54 (0.34) -0.95 (0.18) -0.69 (0.30)
Elk 0.026 0.045 -1.71 (0.81) 0.22 (0.52) -1.28 (0.45)
Brown Bear 0.0018 0.012 - - -1.68 (0.71)
American Badger 0.0023 0.0025 - - -0.33 (0.53)
White-tailed Jackrabbit 0.0005 0.0008 - - -0.17 (1.00)
Gray Wolf 0.0016 0.0025 - - -0.79 (0.70)
North American
-
Porcupine 0.0002 0.0004 - -1.78 (1.05)
Moose 0.0007 0.0034 - - -1.85 (1.20)
Striped Skunk 0.0005 0.0025 - - -1.27 (1.04)
American Red Squirrel 0.0007 0.0004 - - 0.93 (1.67)
Columbian Ground
Squirrel 0 0.042 - - -
American Black Bear 0 0.0017 - - -
Yellow-pine Chipmunk 0.0002 0 - - -
 Long-tailed Weasel 0.0002 0 - - -
0 0
North
Carolina Gray Fox 0.21 0.022 1.33 (0.43) 1.04 (0.32) 1.93 (0.65)
Winter/
Summer Virginia Opossum 0.08 0.075 0.80 (0.40) -0.41 (0.21) 0.00 (0.32)
Coyote 0.14 0.19 0.25 (0.56) -0.35 (0.17) -0.33 (0.24)
Eastern Gray Squirrel 0.46 0.13 0.25 (0.34) -0.47 (0.23) 0.98 (0.51)
Eastern Cottontail 0.53 0.34 0.23 (0.33) 0.32 (0.20) 0.47 (0.35)
White-tailed Deer 2.29 1.77 -0.40 (1.24) 0.04 (0.19) 0.15 (0.17)
Bobcat 0.0031 0.0021 -0.84 (1.18) 1.03 (1.13) 0.44 (0.87)
Northern Raccoon 0.11 0.10 - - 0.00 (0.43)
Red Fox 0.004 0.0003 - - 2.22 (2.42)
Woodchuck 0.0006 0.072 - - -4.58 (4.06)
Southern Flying Squirrel 0.0028 0 - - -
River Otter 0.0003 0 - - -
Eastern Chipmunk 0.0006 0 - - -
Figure S1. Species accumulation curves at a four temperate (A-D) and tropical (E-H) study areas
showing increasing diversity detected by individual cameras (grey lines) and for the study area as
a whole (black lines). The combination of these plots are shown in Figure 5. Study areas listed at
the top of each graph are described in Table S1.

A B

C D
E F

G
H
Figure S2. Seasonal variation shown as monthly averages for observed species richness, overall
mammal detection rate, species-specific capture rate (all species > 20 captures), and the % of
species captures stratified by season (all species with > 20 captures) for mammals surveyed for
sites in (I) Montana, (II) North Carolina, (III) Panama, and (IV) Peru. Lines represent smoothed
averages and gray polygons represent 95% confidence intervals. For the stacked bar charts; blue
= winter (A & B) or rainy season (C & D); orange = summer (A & B) or dry season (C& D). All
datasets represent captures from terrestrial cameras, except Peru, where the cameras were
arboreal.

I)Theodore Roosevelt Memorial Ranch, Montana

II) Prairie Ridge, North Carolina
III) Barro Colorado Island, Panama
IV) Maijuna-Kichwa Regional Conservation Area, Peru

Figure S3. The importance of seasonality in models of detection rate (black dots in the count
model), detectability (blue dots in the occupancy model), and occupancy (red dots in the
occupancy model) for sites in Montana (A), Panama (B), China (C), Peru (D), and North
Carolina (E). Coefficient values show whether a species detection rate, occupancy, or detection
probability was higher (positive) or lower (negative) in winter (for temperate sites) or the wet
season (for tropical sites). Error bars represent 95% confidence intervals.
A)
B)

C)
D)
 E)

View publication stats