ArticlePDF Available

Feasibility of Using Computer-Assisted Software for Recognizing Individual Rio Grande Cooter (Pseudemys gorzugi)


Abstract and Figures

Mark-recapture methods used in population demography studies involve marking of animals, such as tagging, notching, and tattooing. These techniques are invasive and potentially harmful to the animals. Photo-identification using natural animal markings is less invasive and has become more widely used for a range of taxa including invertebrates, fishes, reptiles, amphibians, and mammals. During 2016 and 2017, we studied the demographics of the Rio Grande Cooter (Pseudemys gorzugi) using traditional mark-recapture techniques (i.e., shell notching and toe clipping). However, P. gorzugi displays plastral marks that could potentially be used for individual recognition. Because the photo-identification process 'by-eye' is time consuming, we tested the efficiency of three pieces of software, I 3 S Pattern þ , Wild.ID, and APHIS, for individual identification of P. gorzugi using plastron pattern. Matching results of each program were generated into ranks with the 1 st rank being the most likely match. Within the top 20 ranked images, Wild.ID yielded the highest number of correct matches (83.87%), followed by APHIS (ITM; 69.35%), APHIS (SPM; 67.74%), and I 3 S Pattern þ (61.29%). We found the quality of photos significantly contributed to the software effectiveness; however, turtle age and plastron wear did not affect the accuracy of the photo-identification software. We concluded that Wild.ID can be used as a non-invasive photo-recognition technique for P. gorzugi in a short-term population study.
Content may be subject to copyright.
Feasibility of Using Computer-Assisted Software for Recognizing Individual
Rio Grande Cooter (Pseudemys gorzugi)
Thanchira Suriyamongkol
and Ivana Mali
Mark-recapture methods used in population demography studies involve marking of animals, such as tagging,
notching, and tattooing. These techniques are invasive and potentially harmful to the animals. Photo-identification
using natural animal markings is less invasive and has become more widely used for a range of taxa including
invertebrates, fishes, reptiles, amphibians, and mammals. During 2016 and 2017, we studied the demographics of the
Rio Grande Cooter (Pseudemys gorzugi) using traditional mark-recapture techniques (i.e., shell notching and toe
clipping). However, P. gorzugi displays plastral marks that could potentially be used for individual recognition. Because
the photo-identification process ‘by-eye’ is time consuming, we tested the efficiency of three pieces of software, I
, Wild.ID, and APHIS, for individual identification of P. gorzugi using plastron pattern. Matching results of each
program were generated into ranks with the 1
rank being the most likely match. Within the top 20 ranked images,
Wild.ID yielded the highest number of correct matches (83.87%), followed by APHIS (ITM; 69.35%), APHIS (SPM;
67.74%), and I
S Pattern
(61.29%). We found the quality of photos significantly contributed to the software
effectiveness; however, turtle age and plastron wear did not affect the accuracy of the photo-identification software.
We concluded that Wild.ID can be used as a non-invasive photo-recognition technique for P. gorzugi in a short-term
population study.
STUDYING the abundance and density of wildlife
populations is the first step to developing practical
wildlife management regimes (Thompson et al., 1998;
Dutton et al., 2005; Anderson et al., 2010). Capture-Mark-
Recapture (CMR) is a technique used to identify individuals
within a target population to estimate population size and
other vital rates (i.e., Sreekar et al., 2013; Sannolo et al.,
2016). Marking can be done using several methods such as
visual implant elastomer (VIE) tag, passive integrative
transponder (PIT) tag, toe-clipping, branding, and tattooing
(i.e., Mellish et al., 2007; Morrison et al., 2016; Mali et al.,
2018; Peterson et al., 2018). However, these techniques
involve physical handling and modifying the bodies of
animals, which is considered invasive and could affect
animal behavior or survival (Bloch and Irschick, 2005;
Schmidt and Schwarzkopf, 2010). In turtles, one of the
oldest and most common marking techniques is notching
marginal scutes, but this technique is not recommended for
hatchlings due to the potential of causing shell deformity
and infections (Cagle, 1939; Cooley et al., 2013). An
alternative, non-invasive approach to individual identifica-
tion is to photograph natural markings (i.e., spot, color
pattern, scars), which can reduce stress and injuries associ-
ated with traditional marking techniques (Bolger et al., 2012;
Sacchi et al., 2016).
Photo-identification of an animal’s natural marks has been
used in a wide range of taxa including fishes (Speed et al.,
2007; Dala-Corte et al., 2016), reptiles (Cooley et al., 2013;
Sreekar et al., 2013; Carpentier et al., 2016), amphibians
(Caorsi et al., 2012; Morrison et al., 2016; Matth´
e et al.,
2017), mammals (Anderson et al., 2010; Bolger et al., 2012),
and invertebrates (Huffard et al., 2008). However, this
technique can only be used if organisms have unique
external body patterns that are distinguishable among
individuals and the pattern remains stable over time (Dala-
Corte et al., 2016). Although photographic matching of
individuals using the naked eye generates high accuracy, the
process is often time consuming (Elgue et al., 2014; Matth ´
al., 2017). To reduce image matching time, many computer-
based programs have been developed, such as AMPHIDENT
e et al., 2008), APHIS (Moya et al., 2015), I
S Pattern
(Van Tienhoven et al., 2007), Manta Matcher (Town et al.,
2013), MYDAS (Carter et al., 2014), and Wild.ID (Bolger et al.,
2012). Nonetheless, during development, software is usually
tested on a particular species and the applicability to other
species must be validated (Sacchi et al., 2016). Recent studies
have used photographs to identify individual turtles based on
carapace and plastron patterns in Painted Turtle (Chrysemys
picta belli; Cooley et al., 2013) and Eastern Box Turtle
(Terrapene carolina carolina; Cross et al., 2014) as well as facial
patterns of Green Sea Turtle (Chelonia mydas; Carpentier et
al., 2016). Here, we sought to evaluate the usefulness of using
plastron patterns for individual identification in Rio Grande
Cooter (Pseudemys gorzugi).
The Rio Grande Cooter is a relatively large freshwater turtle
native to the lower Rio Grande and Pecos River drainages in
New Mexico, Texas, and Mexico (Pierce et al., 2016). Their
plastron consists of a black pattern on a yellow background
that appears to be unique to individuals (Fig. 1). In 2016, a
monitoring program was established for the species on the
Black River, New Mexico (Mali et al., 2018). The surveys
generated more than 600 plastron images over a two-year
period. Using these images, we tested the recognition ability
of three pieces of different photo-recognition software,
Wild.ID, I
S Pattern
, and APHIS, for individual recognition
based on plastron photographs taken under field conditions.
Furthermore, we investigated biases in population size
estimation using mark-recapture data based on photo
recognition ability of each piece of software.
Data collection.—The study was conducted on the Black
River, New Mexico from May to August of 2016 and 2017.
Turtles were captured using single-opening, wide mouth
hoop-net traps with diameter of 76.2 cm (Memphis Net and
Twine Company, Memphis, TN) baited with canned sardines
and shrimp (Mirabal et al., 2018). We measured, photo-
Department of Biology, Eastern New Mexico University, 1500 S AVE K Station 33, Portales, New Mexico 88130; Email: (TS) Thanchira.; and (IM) Send reprint requests to TS.
Submitted: 30 July 2018. Accepted: 22 October 2018. Associate Editor: D. S. Siegel.
Ó2018 by the American Society of Ichthyologists and Herpetologists DOI: 10.1643/CH-18-101 Published online: 29 November 2018
Copeia 106, No. 4, 2018, 646–651
graphed, and marked each turtle using one of the three
marking techniques depending on the turtle size: toe-
clipping (juveniles), PIT tagging (subadults only due to
limited budget), and shell notching (adults). While PIT
tagging and shell notching give turtles unique identification
numbers, toe-clipping was done in cohort. Therefore, toe-
clipped recaptures were later identified by experienced
researchers who were familiar with the species using ‘by-
eye’ photo-identification.
Over the course of our study, we photographed 635
plastron images from 529 individuals using an Olympus
Tough TG-870 16.0 MP Compact Digital Camera (Olympus
Co. Shinjuku, Tokyo, Japan). The photographs were taken in
situ on the river bank or on a boat, which generated real-time
images under field conditions (i.e., natural lighting, various
angles, and different levels of turtle cleanliness). Due to
variations in the quality of the photographs, we standardized
the orientation and extent of all images to display only the
ventral region of the turtles by cropping and rotating using
Adobe Photoshop Express (Adobe Systems, Inc., San Jose,
CA). We then categorized images into good-quality photos
and poor-quality photos and created two databases based on
the visibility of the plastron pattern to minimize the source
of bias during photo-matching analyses (Fig. 2).
Performance of photo-identification software.—We evaluated
the performance of three pieces of freely available, open-
source photo-identification software: Wild.ID (ver. 0.9.28),
S Pattern
(ver. 4.1), and APHIS (ver. 1.0). Wild.ID uses the
Scale Invariant Feature Transform (SIFT) algorithm for
pattern extraction and key points detection (Lowe, 2004;
Bolger et al., 2012). By contrast, I
S Pattern
uses a Speed Up
Robust Feature (SURF) algorithm, which detects the key
points based on pixels representing pattern and non-pattern
areas selected by users (I
S Pattern
manual, version 4.1,
2016). Finally, APHIS contains two approaches for image
matching, Spot Pattern Matching (SPM) and Image Template
Matching (ITM) (Moya et al., 2015). The SPM approach treats
and matches images using the exhaustive search version of
SURF algorithm implemented in I
S series, but APHIS allows
users to pre-process all images before proceeding to match
calculations of the whole set (Moya et al., 2015). This
approach requires users to pre-process each image by
defining reference points and marking spots within the area
for comparison (Moya et al., 2015), which is nearly as time
consuming as using by-eye technique. The ITM approach
uses the matchTemplate function of the Open Computer
Vision Libraries, which divides the pattern area into patches
and overlaps separate images to find the best matches (Moya
et al., 2015; Matth´
e et al., 2017).
To evaluate the performance of the photo-identification
software, we created a test set of known matching images
collected in 2016 and 2017. We created a good-quality
database (i.e., photos with highly visible plastron pattern) of
368 photos which included 62 recaptures (124 photographs)
and 244 single captures (244 photographs) for the compar-
ison. To test the software on photo quality, we also generated
a poor-quality database (i.e., photos with plastron pattern
covered in mud, algae, hand of photographer, or bad
lighting) of 267 photos that included 44 recaptures (88
photographs) and 179 single captures (179 photographs). We
ran each data set (i.e., good quality and poor quality) using
the three selected software programs. Because we knew the
correct matches of recaptured individuals, the performance
of the software was expressed as the percentage of correct
matches that fell within the top 20 highest ranks. We chose
the first 20 ranks as the maximum number of acceptable
ranks of the correct matches generated by each piece of
Fig. 1. Examples of the Rio Grande Cooter (Pseudemys gorzugi) plastron pattern of different size classes: 32 mm (left), 123 mm (middle), and 246
mm (right) plastron length. Based on our observations, the pattern starts to fade when a turtle’s plastron length reaches 90 mm.
Suriyamongkol and Mali—Photo-identification of Rio Grande Cooter 647
software. Wild.ID only generates matching results up to the
maximum of 20 ranks, while I
S Pattern
and APHIS generate
results up to 50 and 100 ranks, respectively.
Implications for population size estimation.—We developed a
set of capture histories over six sampling events that reflect
our survey efforts in the field. For this test, we used only
individuals captured in 2017, where individuals with poor-
quality photos were excluded. Any individual previously
captured in 2016 was considered a new individual in 2017.
To compare biases in estimating population size based on
ability of the software to identify recaptured individuals, a
dataset was developed for each piece of software based on its
performance in recognizing recaptures of P. gorzugi. For
example, if an individual was a recapture and the software
failed to place a correct match within the first 20 ranks, that
individual was treated as a new capture.
Capture histories were analyzed in program R (R Core
Team, 2016) using package Rcapture developed to provide
estimation of abundance for closed populations, open
populations, and robust design for capture-recapture models
using Poisson regressions (Baillargeon and Rivest, 2007). For
simplicity, and given the short sampling period, we assumed
that births, deaths, immigration, and emigration did not
occur during our study; therefore, the population was
assumed to be closed. We ran models that included different
combinations of factors that can affect capture probabilities:
temporal effect (t), heterogeneity (h), and behavioral effect
(b), and chose the best fit model based on the Akaike
Information Criterion (AIC; Burnham and Anderson, 1998).
Performance of photo-identification software.—We found that
Wild.ID had the best image-matching performance among
the tested software for both the good-quality database and
the poor-quality database (Table 1). For the good-quality
database, Wild.ID generated 66.13% and 83.87% correct
matches as the top ranking (i.e., the first rank) and among
the top-20 ranking, respectively. I
S Pattern
had the poorest
performance, generating 41.94% and 61.29% correct match-
es as the top ranking and among the top-20 ranking,
respectively. APHIS (SPM) generated 43.55% and 67.74%
correct matches as the top ranking and among the top-20
ranking, respectively, while APHIS (ITM) generated 61.29%
Fig. 2. Examples of the (A) good-quality images and (B) poor-quality images taken of the plastron pattern of the Rio Grande Cooter (Pseudemys
gorzugi). Good-quality images have highly visible plastron pattern, and poor-quality images have the plastron pattern covered in mud, algae, or the
photographer’s hand.
Table 1. Percentage of correct matching pairs of Rio Grande Cooter (Pseudemys gorzugi) out of 62 pairs for good-quality database and 44 pairs for
poor-quality database generated by Wild.ID, I
S Pattern
, APHIS (ITM), and APHIS (SPM) that fell within the top rank and the top-twenty ranks.
Good-quality Poor-quality
Top rank Top twenty Top rank Top twenty
Wild.ID 66.13% 83.87% 61.36% 77.27%
S Pattern
41.94% 61.29% 18.18% 38.64%
APHIS (ITM) 61.29% 69.35% 65.91% 75.00%
APHIS (SPM) 43.55% 67.74% 34.09% 69.91%
648 Copeia 106, No. 4, 2018
and 69.35% correct matches as the top ranking and among
the top-20 ranking, respectively. Poor-quality images reduced
the number of correct matches recognized by the software
within the top-20 ranking by 2–23%, with I
S Pattern
the most affected software (Table 1).
Implications for population size estimation.—Based on tradi-
tional marking techniques (i.e., toe-clip, PIT tag, and shell
notch), there were 309 captures of 281 individuals. Similarly,
for Wild.ID, there were 309 captures of 281 individuals
because Wild.ID correctly recognized all recaptures in our
sample dataset (i.e., good-quality photos taken within the
same season). Based on I
S Pattern
, there were 309 captures
of 287 individuals (i.e., the software failed to recognize six
recaptures). Based on APHIS (SPM), there were 309 captures
of 285 individuals, and based on APHIS (ITM), there were 309
captures of 282 individuals. Heterogeneity and behavioral
effect model (M
) was the most appropriate model for this
study. Among the tested software, Wild.ID resulted in the
most accurate individual recognitions for estimating popu-
lation abundances in comparison with the traditional
marking techniques (Fig. 3). For traditional marking tech-
niques and Wild.ID, population size was estimated to be
974.2640.9. For I
S Pattern
, population size was estimated
to be 1,246.3631.7. For APHIS (SPM), population size was
estimated to be 1,297.0631.9, while for APHIS (ITM),
population size was estimated to be 982.2633.4 (Fig. 3).
Photo-identification software appeared to be effective in
recognizing plastron marks of individual P. gorzugi even
though none of the software was free of error. Wild.ID
exhibited the highest success of generating correct matches
among the tested software. Our results showed that the
performance of software varied depending on photo-match-
ing algorithm and quality of images. Wild.ID was the
simplest and the most user-friendly software, because it did
not require users to pre-analyze images (i.e., manually
indicate pattern and background areas for each photo).
Therefore, Wild.ID reduces the error due to inconsistency of
identifying the area for matching especially with inexperi-
enced users. Moreover, we found that Wild.ID always
suggested the correct matching candidates within the top-
five ranks, most often the first rank; however, if Wild.ID
could not recognize an individual within the first five ranks,
the matching candidate would not be listed within the first
20 matches.
As expected, the good-quality database yielded higher
image-matching accuracy by all software tested. For poor-
quality photos, the plastron was covered in mud, algae, or
sometimes photographer fingers, and some photos also
contained glares or shadow (Fig. 2). Due to the pattern being
covered, it was nearly impossible to use photo-identification
software that requires pre-processing of photos (i.e., indicat-
ing pattern and background area) such as I
S Pattern
APHIS (SPM). Surprisingly, Wild.ID still yielded a high
number of correct matches. However, we noticed that
Wild.ID was likely recognizing individuals based on the
mud and algae pattern on the plastron rather than the
plastron pattern itself, which is not appropriate for a long-
term study as these patterns can change with time.
The main strength of computer-assisted software is the
time efficiency and low probability of falsely identifying a
unique turtle as a recapture. Because of the requirement for
users to visually confirm the correct matches, a researcher
familiar with the species can conclusively confirm whether
the match is contained within the first 20 ranks. However,
false negatives (i.e., when a correct matching image is not
within the first 20 ranks) can still be a common source of
error for the software (Morrison et al., 2016). This is
important when estimating population size (Elgue et al.,
2014). In particular, false negatives can result in overestima-
tion of population size and underestimation of survivorship
(Bolger et al., 2012). According to our findings, when using
the sample dataset with only good-quality images within a
single season of surveys, inability to recognize only four
individuals as recaptures resulted in an increase of over 300
individuals in a population, as seen with APHIS (SPM). In
contrast, Wild.ID showed no error in recognizing recaptures
over the short period of time, and therefore, was comparable
to traditional marking techniques.
It is important to note that the plastron pattern of Rio
Grande Cooter changes over time. We have observed that the
pigmentation patterns fade in larger turtles (.90 mm in
plastron length). The change of pattern in P. gorzugi differed
from other turtle species such as Painted Turtle (Cooley et al.,
Fig. 3. A comparison between five
individual-recognition techniques:
traditional marking (i.e., shell notch-
ing, PIT tag, toe-clip), Wild.ID, APHIS
(ITM), I
S Pattern
, and APHIS (SPM)
to estimate population abundances
of Rio Grande Cooter (Pseudemys
gorzugi) using mark-recapture meth-
od in Rcapture package based on the
sample dataset containing only good-
quality photos collected within a
single season of surveys. Bars repre-
sent one standard deviation from the
estimated abundances.
Suriyamongkol and Mali—Photo-identification of Rio Grande Cooter 649
2013) and Eastern Box Turtle (Cross et al., 2014), where the
pattern will not fully develop until later in life. This finding
indicates that the distinct plastron pattern cannot be used to
recognize the same individual in a long-term study. The
pattern characteristics of different size classes did not affect
the performance of photo-assisted software in a short time
period (i.e., within a season). However, we noticed that the
ability to find correct matches was reduced for individuals
that were captured more than one year apart even though
the pattern similarities were still recognizable by eye. Due to
the fading of plastron pattern over time, it is necessary to
incorporate other marking techniques when conducting a
long-term study of this species. Although Rio Grande Cooter
might not be the most suitable turtle for using computer-
assisted programs because of this issue, this technique is
useful for recognition of hatchlings, which cannot be marked
using traditional marking techniques (i.e., shell notching)
until they reach an appropriate size. Nonetheless, the success
rate obtained in a single-season study by Wild.ID suggested
that this computer-assisted photo-identification software can
be used reliably for the Rio Grande Cooter and probably
other species of turtles with similar plastron markings.
This study was supported by the Share with Wildlife Program
at the New Mexico Department of Game and Fish and State
Wildlife, Grant T-32-4, #11 under NMDGF Authorization
#3621 and ENMU IACUC 03-02/2016. Moreover, we thank
the Bureau of Land Management, private land owners,
undergraduate and graduate students, and researchers for
helping with surveys: A. W. Letter, K. J. Waldon, A.
Villamizar, A. Parandhaman, S. Sirsi, J. Mirabal, J. L. Curtis,
and M. V. Vandewege.
Anderson, C. J., N. Da Lobo Vitoria, J. D. Roth, and J. M.
Waterman. 2010. Computer-aided photo-identification
system with an application to polar bears based on whisker
spot patterns. Journal of Mammalogy 91:1350–1359.
Baillargeon, S., and L. P. Rivest. 2007. Rcapture: loglinear
models for capture-recapture in R. Journal of Statistical
Software 19:1–31.
Bloch, N., and D. J. Irschick. 2005. Toe-clipping dramati-
cally reduces clinging performance in a pad-bearing lizard
(Anolis carolinensis). Journal of Herpetology 39:288–293.
Bolger, D. T., T. A. Morrison, B. Vance, D. Lee, and H. Farid.
2012. A computer-assisted system for photographic mark–
recapture analysis. Methods in Ecology and Evolution 3:
Burnham, K. P., and D. R. Anderson. 1998. Model Selection
and Inference: A Practical Information-Theoretic Ap-
proach. Springer Verlag, New York.
Cagle, F. R. 1939. A system of marking turtles for future
identification. Copeia 1939:170–173.
Caorsi, V. Z., R. R. Santos, and T. Grant. 2012. Clip or snap?
An evaluation of toe-clipping and photo-identification
methods for identifying individual Southern Red-Bellied
Toads, Melanophryniscus cambaraensis. South American
Journal of Herpetology 7:79–84.
Carpentier, A. S., C. Jean, M. Barret, A. Chassagneux, and
S. Ciccione. 2016. Stability of facial scale patterns on green
sea turtles Chelonia mydas over time: a validation for the
use of a photo-identification method. Journal of Experi-
mental Marine Biology and Ecology 476:15–21.
Carter, S. J., I. P. Bell, J. J. Miller, and P. P. Gash. 2014.
Automated marine turtle photograph identification using
artificial neural networks, with application to green turtles.
Journal of Experimental Marine Biology and Ecology 452:
Cooley, C., S. Smith, C. Geier, and T. Puentes. 2013. The use
of photo-identification as a means of identifying western
painted turtles (Chrysemys picta bellii) in long-term demo-
graphic studies. Herpetological Review 44:430–432.
Cross, M. D., G. J. Lipps, Jr., J. M. Sapak, E. J. Tobin, and K.
V. Root. 2014. Pattern-recognition software as a supple-
mental method of identifying individual eastern box
turtles (Terrapene c. carolina). Herpetological Review 45:
Dala-Corte, R. B., J. B. Moschetta, and F. G. Becker. 2016.
Photo-identification as a technique for recognition of
individual fish: a test with the freshwater armored catfish
Rineloricaria aequalicuspis Reis & Cardoso, 2001 (Siluri-
formes: Loricariidae). Neotropical Ichthyology 14:e150074.
Dutton, D. L., P. H. Dutton, M. Chaloupka, and R. H.
Boulon. 2005. Increase of a Caribbean leatherback turtle
Dermochelys coriacea nesting population linked to long-
term nest protection. Biological Conservation 126:186–
Elgue, E., G. Pereira, F. Achaval-Coppes, and R. Maneyro.
2014. Validity of photo-identification technique to analyze
natural markings in Melanophryniscus montevidensis (Anura:
Bufonidae). Journal of Herpetology 13:59–66.
Huffard, C. L., R. L. Caldwell, N. DeLoach, D. W. Gentry, P.
Humann, B. MacDonald, B. Moore, R. Ross, T. Uno, and
S. Wong. 2008. Individually unique body color patterns in
octopus (Wunderpus photogenicus) allow for photoidentifi-
cation. PLoS ONE 3:e3732.
Lowe, D. G. 2004. Distinctive image features from scale-
invariant keypoints. International Journal of Computer
Vision 60:91–110.
Mali, I., A. Duarte, and M. R. Forstner. 2018. Comparison of
hoop-net trapping and visual surveys to monitor abun-
dance of the Rio Grande cooter (Pseudemys gorzugi). PeerJ 6:
e, M., M. Sannolo, K. Winiarski, A. Spitzen-van der
Sluijs, D. Goedbloed, S. Steinfartz, and U. Stachow.
2017. Comparison of photo-matching algorithms com-
monly used for photographic capture–recapture studies.
Ecology and Evolution 7:5861–5872.
e, M., T. Sch¨
onbrodt, and G. Berger. 2008. Computer-
utzte Bildanalyse von Bauchfleckenmustern des
Kammmolchs (Triturus cristatus). Zeitschrift f¨
ur Feldherpe-
Mellish, J. A., D. Hennen, J. Thomton, L. Petrauskas, S.
Atkinson, and D. Calkins. 2007. Permanent marking in
an endangered species: physiological response to hot
branding in Steller sea lions (Eumetopias jubatus). Wildlife
Research 34:43–47.
Mirabal, J., A. W. Letter, K. J. Waldon, and I. Mali. 2018.
Pseudemys gorzugi (Rio Grande Cooter). Attractions to trap
baits. Herpetological Review 49:323–324.
Morrison, T. A., D. Keinath, W. Estes-Zumpf, J. P. Crall,
and C. V. Stewart. 2016. Individual identification of the
endangered Wyoming toad Anaxyrus baxteri and implica-
tions for monitoring species recovery. Journal of Herpetol-
ogy 50:44–49.
650 Copeia 106, No. 4, 2018
Moya, ´
O., P. L. Mansilla, S. Madrazo, J. M. Igual, A. Rotger,
A. Romano, and G. Tavecchia. 2015. APHIS: a new
software for photo-matching in ecological studies. Ecolog-
ical Informatics 27:64–70.
Peterson, D., R. B. Trantham, T. G. Trantham, and C. A.
Caldwell. 2018. Tagging effects of passive integrated
transponder and visual implant elastomer on the small-
bodied white sands pupfish (Cyprinodon tularosa). Fisheries
Research 198:203–208.
Pierce, L. J. S., J. N. Stuart, J. P. Ward, and C. W. Painter.
2016. Pseudemys gorzugi Ward 1984—Rio Grande Cooter,
Western River Cooter, Tortuga de Oreja Amarilla, Jicot´
del R´
ıo Bravo. Chelonian Research Monographs 5:100.1–
R Core Team. 2016. R: a language and environment for
statistical computing. R Foundation for Statistical Com-
puting, Vienna, Austria.
Sacchi, R., S. Scali, M. Mangiacotti, M. Sannolo, and M. A.
L. Zuffi. 2016. Digital identification and analysis, p. 59–72.
In: Reptile Ecology and Conservation: A Handbook of
Techniques. C. K. Dodd (ed.). Oxford University Press,
Oxford, U.K.
Sannolo, M., F. Gatti, M. Mangiacotti, S. Scali, and R.
Sacchi. 2016. Photo-identification in amphibian studies: a
test of I3S Pattern. Acta Herpetologica 11:63–68.
Schmidt, K., and L. Schwarzkopf. 2010. Visible implant
elastomer tagging and toe-clipping: effects of marking on
locomotor performance of frogs and skinks. The Herpeto-
logical Journal 20:99–105.
Speed, C. W., M. G. Meekan, and C. J. Bradshaw. 2007. Spot
the match—wildlife photo-identification using informa-
tion theory. Frontiers in Zoology 4:2.
Sreekar, R., C. B. Purushotham, K. Saini, S. N. Rao, S.
Pelletier, and S. Chaplod. 2013. Photographic capture-
recapture sampling for assessing populations of the Indian
gliding lizard Draco dussumieri. PloS ONE 8:e55935.
Thompson, W. L., G. C. White, and C. Gowan. 1998.
Monitoring Vertebrate Populations. Academic Press, San
Town, C., A. Marshall, and N. Sethasathien. 2013. Manta
Matcher: automated photographic identification of manta
rays using keypoint features. Ecology and Evolution 3:
Van Tienhoven, A. M., J. E. Den Hartog, R. A. Reijns, and V.
M. Peddemors. 2007. A computer-aided program for
pattern-matching of natural marks on the spotted ragged-
tooth shark Carcharias taurus. Journal of Applied Ecology
Suriyamongkol and Mali—Photo-identification of Rio Grande Cooter 651
... For example, photo identification of individuals has been used for a wide range of species such as sharks (Van Tienhoven et al. 2007;Holmberg et al. 2009), rays (González-Ramos et al. 2017, insects (Caci et al. 2013), reptiles (Knox et al. 2012;Moro and MacAulay 2014;Bauwens et al. 2017), amphibians (Hoque et al. 2011) and mammals (Kelly 2001;Hiby et al. 2009;Halloran et al. 2015). Although data from photo identification has been used to estimate population trajectory (Holmberg et al. 2009) and survival estimates (Morrison et al. 2011), accurate identification of individuals is key for monitoring endangered species as misclassifications can result in inflated population estimates (Suriyamongkol and Mali 2018;Johansson et al. 2020). ...
... Reliable marking techniques are especially critical when monitoring long-lived species such as freshwater turtles. Photo identification has been used to identify individuals with fairly good success in freshwater turtle species such as Chrysemys picta belli (Cooley et al. 2013), Terrapene carolina carolina (Cross et al. 2014), Trachemys scripta elegans (Janzen et al. 2000a, b), Chelydra serpentina (Kolbe and Janzen 2001), and Pseudemys gorzgui (Suriyamongkol and Mali 2018); however, success was limited to adult turtles (e.g., Cross et al. 2014) or within a season for species whose pattern changes or fades over time (Janzen et al. 2000a, b;Kolbe and Janzen 2001;Suriyamongkol and Mali 2018). In Clemmys guttata, the plastron pattern appeared unique to turtles studied in Pennsylvania while changes to the spot pattern on the carapace of some individuals were observed (Gray 2008). ...
... Reliable marking techniques are especially critical when monitoring long-lived species such as freshwater turtles. Photo identification has been used to identify individuals with fairly good success in freshwater turtle species such as Chrysemys picta belli (Cooley et al. 2013), Terrapene carolina carolina (Cross et al. 2014), Trachemys scripta elegans (Janzen et al. 2000a, b), Chelydra serpentina (Kolbe and Janzen 2001), and Pseudemys gorzgui (Suriyamongkol and Mali 2018); however, success was limited to adult turtles (e.g., Cross et al. 2014) or within a season for species whose pattern changes or fades over time (Janzen et al. 2000a, b;Kolbe and Janzen 2001;Suriyamongkol and Mali 2018). In Clemmys guttata, the plastron pattern appeared unique to turtles studied in Pennsylvania while changes to the spot pattern on the carapace of some individuals were observed (Gray 2008). ...
Full-text available
The ability to uniquely identify individuals is critical to estimating and monitoring trends in population sizes, one of the key metrics used to evaluate a species' conservation status and success of mitigation strategies. For freshwater turtles, shell notching and/or passive integrated transponder (PIT) tags are commonly used to mark individuals. However, because notch codes and PIT tags can be lost over time and require more invasive procedures, we explored if photographs offer an effective method to reliably identify individuals. The Blanding's turtle (Emydoidea blandingii) is a globally endangered species with distinct black and yellow markings on its plastron. We used the I 3 S Pattern software with custom parameters to classify patterns on Blanding's turtle plastrons and to identify individuals. We MARKLE et al. 48 analyzed 826 plastron images from 707 individual Blanding's turtles taken between 1998 and 2019 from 12 study areas distributed throughout their Canadian range. When plastron photos were pooled across the sampled range (i.e., all study areas), there was an 84% probability of correctly identifying an individual turtle within the top 3 suggested matches, whereas when identifying Blanding's turtles within a specific study area, identification accuracy was 82% in Central Ontario and 97% in Nova Scotia. Individual identification from plastron markings did not work well in areas where iron staining obscured the plastron pattern or for hatchlings and juveniles whose patterns changed over time. For example, the only misclassification in the Nova Scotia study area was for a turtle with photos through various life stages. In areas without iron staining, plastron photo identification offers a cost-effective, non-invasive method to identify individual adult Blanding's turtles to support population monitoring and community science initiatives, and has the potential to assist with range-wide coordination to counteract illegal wildlife trade.
... Despite their proliferation, little has been published formally comparing software packages. Of the comparisons that have been done (e.g., Morrison et al. 2016a, Cruickshank and Schmidt 2017, Matthé et al. 2017, Suriyamongkol and Mali 2018, all used data sets consisting of photographs of herpetofauna. In all cases, this involved capturing the animals and photographing them with a handheld camera with standardized distance and positioning. ...
... Cruickshank and Schmidt (2017), analyzing high-quality images of yellowbellied toads (Bombina variegata), found that Wild-ID presented a correct, top-ranked match 91.6% of the time and that a correct match was within the top 2 candidates 95% of the time. Suriyamongkol and Mali (2018) studied Rio Grande cooter (Pseudemys gorzugi) and found that Wild-ID offered a correct match as the top choice 61-66% of the time, and that a correct match was somewhere in the top 20 ranks 77-84% of the time. ...
Camera‐trapping is widespread in wildlife studies, especially for species with individually unique markings to which capture–recapture analytical techniques can be applied. The large volume of data such studies produce have encouraged researchers to increasingly look to computer‐assisted pattern‐recognition software to expedite individual identifications, but little work has been done to formally assess such software for camera‐trap data. We used 2 sets of camera‐trap images—359 images of jaguars (Panthera onca) and 332 images of ocelots (Leopardus pardalis) collected from camera traps deployed in 4 study sites in Orange Walk District, Belize, in 2015 and 2016—to compare the accuracy of 2 such programs, HotSpotter and Wild‐ID, and assess the effect of image quality on matching success. Overall, HotSpotter selected a correct match as its top rank 71–82% of the time, whereas the rate for Wild‐ID was 58–73%. Positive matching rates for both programs were highest for high‐quality images (85–99%) and lowest for low‐quality images (28–52%). False match rates were very low for HotSpotter (0–2%) but these were greater in Wild‐ID (6–28%). When lower ranks were also considered, both programs performed similarly (overall 22–24% nonmatches for HotSpotter, 17–26% nonmatches for Wild‐ID). We found that in both programs, images more often matched to other images of the same quality; therefore, including multiple reference images of an individual, of different qualities, improves matching success. These programs do not provide fully automatic identification of individuals and human involvement is still required to confirm matches, but we found that they are effective tools to expedite processing of camera‐trap data. We also offer usage recommendations for researchers to maximize the benefits of these tools. © 2020 The Wildlife Society. We used photographs of jaguars and ocelots collected in 2015–2016 from camera traps in Belize to test and compare the utility of 2 freely available pattern recognition programs—HotSpotter and Wild‐ID—for identifying individuals by unique markings. We found both programs to be effective tools, though not fully automatic, for processing camera‐trap data, and recommend that multiple, different quality images of individuals be included in reference databases to maximize the efficacy of the programs.
... However, depending on the distinctiveness and stability of natural marks, non-invasive identification may only be applied to part of the population and its accuracy is subject to developmental changes of these features, e.g., accumulation of notches, healing of scars, changes in pigmentation or periodic moult (Gowans and Whitehead 2001;Koivuniemi et al. 2016;Suriyamongkol and Mali 2018; see also various articles in Karczmarski et al. 2022a, b). Misidentification, either by cataloguing different individuals as the same (because of similar features) or recording the same individual as different on different occasions (due to mark change), is a source of error that may introduce noise to the dataset, leading to inaccurate analytical and modelling results (James et al. 2009;Yoshizaki et al. 2009;Morrison et al. 2011). ...
Reliable identification of individuals plays an important role in behavioural studies of free-ranging animal populations. In field studies of elephants, the naturally acquired markings on their ears, such as notches, tears and holes, are frequently used for individual identification. Although not as easily discernible from a distance as ear markings, the facial wrinkle pattern around the eye, temporal gland and ear on both sides of elephant’s head are individually unique and, with application of high-resolution photographs, can also be used for individual identification. In fact, the wrinkle pattern is highly consist- ent and reliable as the primary identifiable feature; it changes little over time, facilitates identification of individuals with non-distinctive ear pattern (e.g., calves), and performs well against several practical challenges to the traditional ear-pattern approach. We used data from a 3-year photo-identification study of African elephant population to examine how the two identification methods, one that uses marks on elephant ears and the other using facial wrinkle pattern, affect the results of basic analyses of social dynamics, such as patterns of associations and social preferences, derived from datasets generated with these two identification methods. Comparative analyses demonstrate that by increasing the identifiability of otherwise poorly marked individuals and minimising identification error, the wrinkle-based method reduces substantially the sample bias, enhances the robustness of datasets, and minimises analytical error. While ear-pattern-based distinctiveness is age- dependent, the wrinkle-based method facilitates a more representative sample of the population, with photo-ID data collected non-discriminately across all age classes. This carries further implications, such as enabling more accurate depiction of elephant sociality, long-term population monitoring, calculation of class-specific population parameters, etc. Adopting the facial wrinkle pattern for elephant individual identification is relatively easy, and we encourage future and ongoing stud- ies to consider incorporating the facial wrinkle approach. Given the advantages of wrinkle-based identification and recent advances in machine learning, we recommend it to be considered for the development of automated matching algorithms; such development would benefit long-term socio-behavioural studies and monitoring of elephant populations.
... For every turtle captured, we took standard measurements (method D in Iverson and Lewis, 2018) to the nearest 1 mm using tree calipers (i.e., straight line carapace length, carapace width, plastron length, plastron width, body depth). We marked new individuals using marginal scute notching, PIT (Passive Integrated Transponder) tags, or toe-clipping, according to the size of the turtle (Cagle, 1939;Buhlmann and Tuberville, 1998;Suriyamongkol and Mali, 2018). For P. gorzugi and T. scripta, we clipped the terminal 0.5 cm of the medial toe claw on the hind foot (either right or left) using sterilized veterinarian nail clippers. ...
Full-text available
Aquatic turtles represent important biotic components of freshwater ecosystems. The Pecos River watershed is inhabited by six freshwater turtle species, including the widespread Trachemys scripta (Red-eared Slider) and a species of conservation concern, Pseudemys gorzugi (Rio Grande Cooter). Here, we assessed isotopic niche widths of Rio Grande Cooter and niche overlap where it co-occurs with Red-eared Slider in the Pecos River tributaries, New Mexico, USA. We used carbon (d 13 C) and nitrogen (d 15 N) stable isotope analyses of two different tissue types: blood and claw. Our results showed niche partitioning among different populations of P. gorzugi and among sex classes within a population. At the sites where both species occur, we documented niche overlap, especially for d 15 N values. Stable isotopes showed similar ellipse area overlap (SEA B) of T. scripta and P. gorzugi among populations (~20% 2), but little to no overlap of standard ellipse areas for small sample sizes (SEA C). The distribution of prey items in the diets of P. gorzugi and T. scripta revealed the differences in resource selection. We observed that differences in the diets of P. gorzugi among populations correspond to local resource availability, suggesting opportunistic foraging behavior of P. gorzugi. Our study aids in understanding the ecology and natural history of P. gorzugi, one of the least studied freshwater turtles in the USA. Moreover, our study provides insights to interspecific relations of T. scripta in their native range.
... An error-free identification was achieved for an online experiment which opens future application in uniquely patterned species in the wild such as zebras and giraffes. The study conducted by Suriyamongkol and Mali (2018) addressed the use of four computer-assisted softwares (wild-ID, APHIS (ITM), APHIS (SPM), and I3S Pattern) for the individual recognition of Rio Grande Cooter. It was established that wild-ID performed better with an accuracy of 83.87% and can be implemented for population count. ...
Full-text available
This work presented a literature review on animal species and individual identification tools, as well as animal monitoring capabilities. We gathered the literature to cover different aspects of technologies that are widely in use for animal identification, from the traditional up to the latest methods. This study includes species and individual animal identification attributes namely body patterns, footprints, facial features, and sound for identification purposes. The large volume of data collected could be automatically processed using machine learning and deep learning techniques to achieve both species and individual animal identification more efficiently as compared to the human workforce. It is a much faster and accurate approach considering the large volume of data, than manual processing, which is extremely expensive, time-consuming, tedious, and monotonous. We established that machine learning and advancements in deep learning hold significant promise to high-accuracy identification of both species and individual animal. Methods used for individual identification are mainly implemented in endangered species by the conservation management. The traditional methods such as the use of footprints, drawings of animal biometrics are integrated into the recent growth of technology to eliminate the human skill needed to achieve species and individual identification through the use of machine learning and deep learning algorithms for automatic identification purposes.
... Manual photoidentification of individual animals was pioneered in the 1970 on orcas (Orcinus orca; Bain, 1990) and since been used in a range of taxa. With the digital revolution in both image and computer processing, automatic identification using biometric algorithms has been attempted on a range of species such as Delta Smelt (Hypomesus transpacificus; Castillo et al., 2019), Rio Grande Cooter (Pseudemys gorzugi; Suriyamongkol and Mali, 2018), Whale Shark (Rhincodon typus; Holmberg et al., 2009), and several mammal species (reviewed in Kumar and Singh, 2016). ...
Traditional tagging methods for fishes can have issues relating to both animal welfare and economic costs. Biometric data such as iris patterns can be captured via digital cameras, which allows for non-invasive tagging and inexpensive and rapid analysis. The purpose of this study was to investigate if the iris of Atlantic Salmon (Salmo salar) is a suitable biometric template for long-term identification of individuals. Atlantic Salmon were individually tagged in the body cavity using PIT tags at the juvenile pre-smolt stage, and the left eye was photographed six times over a 533-day period. Changes in iris stability were assessed both qualitatively and using iris-recognition software. Identification of individual Atlantic Salmon using the iris was not successful over the entire period, as the iris pattern changed significantly with time. Over a shorter time period (four months) with frequent samplings, iris software was able to correctly identify individual fish. The results show that iris identification has potential to replace other methods for Atlantic Salmon over short timeframes.
... Photographic identification is minimally invasive, can provide a variety of information (e.g., cameras provide data on locations, temperature, and diurnal and nocturnal activity patterns), is relatively inexpensive, and records information on individuals for long-term monitoring of sensitive species (Morrison et al. 2011(Morrison et al. , 2016Crall et al. 2013). Computer-assisted photo identification has been used to identify individuals of a diverse set of taxa, including mammals (Bolger et al. 2012, Crall et al. 2013, Gilman et al. 2016, amphibians (Bendik et al. 2013, Morrison et al. 2016, Kim et al. 2017, fish (Crall et al. 2013), birds (Burghardt 2008, Sherley et al. 2010, reptiles (Treilibs et al. 2016, Suriyamongkol andMali 2018), and insects (Caci et al. 2013, De Gasperis et al. 2017. Conditions for PMR require that: 1) researchers capture photographs of individuals that are either free-ranging or live-captured, with remotely-triggered or hand-held cameras; 2) individuals have unique marks or patterns on their body that allow observers to differentiate among individuals; and 3) individual's patterns are stable over the duration of the study period and are able to be photographed across a range of environmental conditions (Bolger et al. 2012). ...
Full-text available
Successful conservation and management of protected wildlife populations require reliable population abundance data. Traditional capture-mark-recapture methods can be costly, time-consuming, and invasive. Photographic mark-recapture (PMR) is a cost-effective, minimally invasive way to study population dynamics in species with distinct markings or color patterns. We tested the feasibility and the application of PMR using the software Hotspotter to identify Nicrophorus spp. from digital images of naturally occurring spot patterns on their elytra. We conducted a laboratory study evaluating the identification success of Hotspotter on Nicrophorus americanus (Olivier, 1790) and Nicrophorus orbicollis (Say, 1825) before implementation of a mark-recapture study in situ. We compared the performance of Hotspotter using both ‘high-quality’ and ‘low-quality’ photographs. For high-quality photographs, Hotspotter had a false rejection rate of 2.7–3.0% for laboratory-reared individuals and 3.9% for wild-caught individuals. For low-quality photographs, the false rejection rate was much higher, 48.8–53.3% for laboratory-reared individuals and 28.3% for wild-caught individuals. We subsequently analyzed encounter histories of wild-caught individuals with closed population models in Program MARK to estimate population abundance. In our study, we demonstrated the utility of using PMR in estimating population abundance for Nicrophorus spp. based on elytral spot patterns.
... Photo-id software has been developed to facilitate the task of identifying a diverse set of species, such as in [19][20][21][22][23][24][25][26][27][28][29][30][31]. Each of these photo-id systems have been developed on a laptop and personal computer due to the complexity of the proposals developed, such as the work shown in [32] that compares four animal photo-id systems. In the case of the blue whale, an identifying characteristic is the shape of its dorsal fin observed on both the right and left flanks. ...
Full-text available
Photo-identification (photo-id) is a method used in field studies by biologists to monitor animals according to their density, movement patterns and behavior, with the aim of predicting and preventing ecological risks. However, these methods can introduce subjectivity when manually classifying an individual animal, creating uncertainty or inaccuracy in the data as a result of the human criteria involved. One of the main objectives in photo-id is to implement an automated mechanism that is free of biases, portable, and easy to use. The main aim of this work is to develop an autonomous and portable photo-id system through the optimization of image classification algorithms that have high statistical dependence, with the goal of classifying dorsal fin images of the blue whale through offline information processing on a mobile platform. The new proposed methodology is based on the Scale Invariant Feature Transform (SIFT) that, in conjunction with statistical discriminators such as the variance and the standard deviation, fits the extracted data and selects the closest pixels that comprise the edges of the dorsal fin of the blue whale. In this way, we ensure the elimination of the most common external factors that could affect the quality of the image, thus avoiding the elimination of relevant sections of the dorsal fin. The photo-id method presented in this work has been developed using blue whale images collected off the coast of Baja California Sur. The results shown have qualitatively and quantitatively validated the method in terms of its sensitivity, specificity and accuracy on the Jetson Tegra TK1 mobile platform. The solution optimizes classic SIFT, balancing the results obtained with the computational cost, provides a more economical form of processing and obtains a portable system that could be beneficial for field studies through mobile platforms, making it available to scientists, government and the general public.
... The I 3 S algorithm has been embedded in several software interfaces (I 3 S Classic, Spot, Contour, Pattern and Pattern þ ) particularly adapted for surveys of a variety of vertebrate taxa, such as, for example, reptiles (Sreekar et al. 2013;Dunbar et al. 2014;Araujo et al. 2016;Treilibs et al. 2016;Hayes et al. 2017;Steinmetz et al. 2018;Suriyamongkol and Mali 2018;Wessels et al. 2018), amphibians (Ribero andRebelo 2011;Sannolo et al. 2016;Davis et al. 2018;Matos et al. 2018) and fishes (Van Tienhoven et al. 2007;Chaves et al. 2016;Araujo et al. 2017). ...
Full-text available
Context : In recent years, multiple computer algorithms, which allow us to perform photographic capture-recapture analysis, have been developed. Their massive application, also in wildlife demographic and ecological studies, is largely due to the fact that these tools are non-invasive and non-expensive. To maximise the performance of these programs, it is essential to have a good photo-standardisation so as to avoid bias in the results. A lot of 'non-standardised' photos are not usable for capture-mark-recapture (CMR) analysis, entailing the loss of potentially exploitable data. Aims : No study has accurately investigated the effect of the corporal bending of an animal on the performance of the interactive individual identification system (I3S) algorithm. For this reason, we assessed the effect of this photographic standardisation parameter (PSP) on the reliability of this algorithm. Methods : We assessed the effect of the body position of Triturus cristatus between capture and recapture photos on the error rates of a group of standardised pictures, performing a generalised linear model analysis. We have also evaluated the effect of image correction (i.e. straightening of newts' bodies) on the error rates (expressed by false rejection rates, FRRs) of the first (standardised) photo-group (G1) and of a non-standardised photo-group (G2). To perform this, we used I3S Pattern + for the photo-matching analysis and I3S -Straighten for the correction of the pictures. Key results: The difference of body angles between capture and recapture pictures had a significantly increased error rates in G1. Digital correction of body bending reduced the error rates. For the pictures where corporal bending was not digitally corrected, the top 20 FRRs were 0.38 and 0.33 for G1 and G2 respectively. For corrected (straightened) pictures, the top 20 FRRs were 0.026 and 0.15. Conclusions: Our findings showed a high impact of newt corporal bending and photographic treatment on the I3S algorithm reliability. Implications. We provide some recommendations to avoid or minimise the effects of this PSP and improve photo-standardisation during and after CMR studies of species of Urodela. In this way, pictures that would be unusable in photo-matching software under current practice could become usable, increasing the available data to conduct a survey.
... The software is less sensitive to poor image quality compared to fully automatic software and it has E. Kristensen, et al. Fisheries Research 229 (2020) 105622 previously been used to identify both individual newts and land turtles (Suriyamongkol and Mali, 2018). I 3 S Pattern + works by identifying colour differences resulting in a "fingerprint" which can be compared to a reference database (Den Hartog and Reijns, 2016). ...
It is difficult to determine the individual behaviour, growth, and survival of large predatory freshwater fish without stressing or even killing them. In this study, we tested the ability of an image recognition program to identify individual recaptured pike (Esox lucius) based on skin pattern. We and local anglers caught 209 pike individuals, including 45 recaptures, over the course of 1.5 years. The software correctly identified the image of all recaptured fish as already present in the database. Based on recaptures, we predicted a population of 560 pike in the lake and showed a much more comprehensive size structure than seen using standardized gillnet sampling which only yielded four pike. Angling and photographic identification involving reporting by citizen-scientists provided valuable temporal data in a non-invasive manner compared to traditional sampling. This methodology is based on freeware software and removes the need for any tagging equipment. It can potentially be used on other freshwater fish species and larger populations if the skin pattern is distinctive. However, it relies on the willingness of anglers to report their catches.
Full-text available
Abundance estimates play an important part in the regulatory and conservation decision-making process. It is important to correct monitoring data for imperfect detection when using these data to track spatial and temporal variation in abundance, especially in the case of rare and elusive species. This paper presents the first attempt to estimate abundance of the Rio Grande cooter (Pseudemys gorzugi) while explicitly considering the detection process. Specifically, in 2016 we monitored this rare species at two sites along the Black River, New Mexico via traditional baited hoop-net traps and less invasive visual surveys to evaluate the efficacy of these two sampling designs. We fitted the Huggins closed-capture estimator to estimate capture probabilities using the trap data and distance sampling models to estimate detection probabilities using the visual survey data. We found that only the visual survey with the highest number of observed turtles resulted in similar abundance estimates to those estimated using the trap data. However, the estimates of abundance from the remaining visual survey data were highly variable and often underestimated abundance relative to the estimates from the trap data. We suspect this pattern is related to changes in the basking behavior of the species and, thus, the availability of turtles to be detected even though all visual surveys were conducted when environmental conditions were similar. Regardless, we found that riverine habitat conditions limited our ability to properly conduct visual surveys at one site. Collectively, this suggests visual surveys may not be an effective sample design for this species in this river system. When analyzing the trap data, we found capture probabilities to be highly variable across sites and between age classes and that recapture probabilities were much lower than initial capture probabilities, highlighting the importance of accounting for detectability when monitoring this species. Although baited hoop-net traps seem to be an effective sampling design, it is important to note that this method required a relatively high trap effort to reliably estimate abundance. This information will be useful when developing a larger-scale, long-term monitoring program for this species of concern.
Full-text available
Photographic capture–recapture is a valuable tool for obtaining demographic informa- tion on wildlife populations due to its noninvasive nature and cost- effectiveness. Recently, several computer- aided photo- matching algorithms have been developed to more efficiently match images of unique individuals in databases with thousands of images. However, the identification accuracy of these algorithms can severely bias es- timates of vital rates and population size. Therefore, it is important to understand the performance and limitations of state- of- the- art photo- matching algorithms prior to implementation in capture–recapture studies involving possibly thousands of images. Here, we compared the performance of four photo- matching algorithms; Wild- ID, I3S Pattern+, APHIS, and AmphIdent using multiple amphibian databases of varying image quality. We measured the performance of each algorithm and evaluated the perfor- mance in relation to database size and the number of matching images in the database. We found that algorithm performance differed greatly by algorithm and image data- base, with recognition rates ranging from 100% to 22.6% when limiting the review to the 10 highest ranking images. We found that recognition rate degraded marginally with increased database size and could be improved considerably with a higher number of matching images in the database. In our study, the pixel- based algorithm of AmphIdent exhibited superior recognition rates compared to the other approaches. We recommend carefully evaluating algorithm performance prior to using it to match a complete database. By choosing a suitable matching algorithm, databases of sizes that are unfeasible to match “by eye” can be easily translated to accurate individual capture histories necessary for robust demographic estimates.
Full-text available
Photo-identification is used for individual recognition of several animal species. It gives the possibility to take photos of large species from a distance or to avoid invasive marking techniques in small animals. For amphibians , the use of non-invasive marking methods is even more relevant in the light of their global decline. Here we use the photo-identification data from a population of Triturus carnifex to validate the photo-identification software I 3 S Pattern. This recently developed utility has never been applied to amphibians. The software proved to be efficient and accurate for individual recognition for this species. Contrarily to the previous releases of the I 3 S family, I 3 S Pattern is particularly suitable for amphibians characterized by a complex individual pattern of large blotches or irregular spots, which are not readily identified by eye.
One of the greatest limiting factors of studies designed to obtain growth, movement, and survival in small-bodied fishes is the selection of a viable tag. The tag must be relatively small with respect to body size as to impart minimal sub-lethal effects on growth and mobility, as well as be retained throughout the life of the fish or duration of the study. Thus, body size of the model species becomes a major limiting factor; yet few studies have obtained empirical evidence of the minimum fish size and related tagging effects. The probability of surviving a tagging event was quantified in White Sands pupfish (Cyprinodon tularosa) across a range of sizes (19-60. mm) to address the hypothesis that body size predicts tagging survival. We compared tagging related mortality, individual taggers, growth, and tag retention in White Sands pupfish implanted with 8-mm passive integrated transponder (PIT), visual implant elastomer (VIE), and control (handled similarly, but no tag implantation) over a 75 d period. Initial body weight was a good predictor of the probability of survival in PIT- and VIE-tagged fish. As weight increased by 1. g, the fish were 4.73 times more likely to survive PIT-tag implantation compared to the control fish with an estimated suitable tagging size at 1.1. g (TL: 39.29. ±. 0.41. mm). Likewise, VIE-tagged animals were 2.27 times more likely to survive a tagging event compared to the control group for every additional 1. g with an estimated size suitable for tagging of 0.9. g (TL: 36.9. ±. 0.36. mm) fish. Growth rates of PIT- and VIE-tagged White Sands pupfish were similar to the control groups. This research validated two popular tagging methodologies in the White Sands pupfish, thus providing a valuable tool for characterizing vital rates in other small-bodied fishes.
This paper presents a method for extracting distinctive invariant features from images that can be used to perform reliable matching between different views of an object or scene. The features are invariant to image scale and rotation, and are shown to provide robust matching across a substantial range of affine distortion, change in 3D viewpoint, addition of noise, and change in illumination. The features are highly distinctive, in the sense that a single feature can be correctly matched with high probability against a large database of features from many images. This paper also describes an approach to using these features for object recognition. The recognition proceeds by matching individual features to a database of features from known objects using a fast nearest-neighbor algorithm, followed by a Hough transform to identify clusters belonging to a single object, and finally performing verification through least-squares solution for consistent pose parameters. This approach to recognition can robustly identify objects among clutter and occlusion while achieving near real-time performance.