Choosing the right camera traps based on interests, goals, and

Marcella J. Kelly- Professor, Virginia Tech Dept of Fish and Wildlife Conservation

WildLabs Community – Tech Tutors July 15, 2021 Remote-Camera Trapping Background

„ Remote cameras/camera traps/game cameras

„ Been around since the late 1890s. But using trip wires and track pads and gave single shots only. „ „ 1980s deer hunters => scout hunting grounds

„ 1990s biologists expanded techniques using multiple shot film cameras - film

„ 2000s (mid) brought affordable digital camera technology Remote Camera Applications - Today

„ Scientific Studies „ – especially for monitoring of various forest carnivores (e.g. American , , , , , , etc.), but also for big game, and large- movement across highways, prey studies, denning behavior (black ); physical condition of (sun bears) „ Birds – count and monitor ground bird; avian nest predation „ Herps: e.g. monitoring of timber rattlesnakes. But few herp studies.

„ Remote Wildlife Photography

„ Recreational users (e.g. hunters etc.) Camera Types

Cameras now use mostly passive (PIR) infrared sensors

„ PIR –triggers by motion/heat differential when moving object differs in temperature from the environment and moves in front of the sensor

„ Up and coming– remotely download to a base station or satellite uplink Things to consider

„ Do you need protection from wildlife?

„ White flash or infrared?

„ Do you need to lock cameras due to theft?

„ User-friendliness?

„ Do you have a price range?

„ Still photos or video?

„ How long do you need them to last?

„ Protection from the weather?

„ One camera or two per station?

„ How often can you check them?

„ Battery life

„ Memory card size Kelly et al. 2013 Camera set up with 2 cameras and locks Example Brands – between $100 and $250 Bushnell Camera Lens

Browning IR Sensor

Camera Lens

IR Sensor Moultrie

$57 when you buy 10 or more

IR sensor

Camera Lens Reconyx (older model) $400-500 and Camera Lens not so user friendly Info Screen

Operational Buttons

Power Button

IR Sensor SD Card Slot User Friendliness

Reconyx Video – Professional Series ($500 – $600)

Camera Lenses

IR Sensor Things to consider for data analysis and results

„ What species are you interested in?

„ How many cameras for a scientific study?

„ How far apart or what spatial arrangement?

„ What are you interested in learning about?

„ species presence or distribution (occupancy)

„ species diversity

„ species activity levels

„ species use of specific features

„ species population abundance, density and trends,

„ species interactions (co-occurrence)

„ species survival Camera Setup – baited vs. unbaited

Camera PIR set with hanging bait Unbaited trail set with paired PIR cameras

Unbaited PIR set outside of a culvert Baited cubby set triggered by a Long et al. (2008) Noninvasive pressure pad => to get picture Unbaited PIR set along a mouse runway Survey Methods for Carnivores of ear tag General Camera Trapping Uses with increasing complexity

1) Behavior 2) Indices of activity - trap rates 3) Presence/absence (i.e. occupancy) 4) Population Estimation using Mark- Recapture (CMR, SCR, SMR, SPIM) 1) Behavioral Research New Information - Den behavior – emergence times and frequency Activity Patterns

14%

12%

10%

8%

6%

4%

2%

0% 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00

Puma (N=208) (N=172) Nest Predation documented by remote cameras

Nest predation on bird eggs by and crocodile eggs by monitor lizards Underpass Usage

https://www.youtube.com/watch?v=2bICTWNRrGE Scavenging behavior – VT Study Dominance at carcasses Herp studies

„ https://t.co/SIwtuF8vqT

„ SNAKE study using cameras on time lapse 2) Index of Activity –Trap rate or trap success „ Number of photo captures per 100 Trap nights TS = (Caps/TN)*100

Capture = independent events within a 30 minute period Some use 60 minute time periods Indices of Activity - Trap Success

Trap success = the number of trap events per night or per 100 trap-nights

Not as powerful as occupancy or individual ID, but still can be useful

Camera Survey at the Mountain Lake Biological Station, VA; (Kelly and Holub 2008) However, indices are problematic

Smiley J4 male For Example: 2005 Pine Forest Survey in Belize

Total Jaguar “Captures” = 109 12 individuals

dates time x-location y-location place 06/22/01 night 284995 1861441 NM 06/24/01 day 284995 1861441 NM 07/12/01 21:08 284995 1861441 NM Total Captures of J27 male = 75 07/28/01 07:59 284917 1850123 MR 07/28/01 20:07 283538 1851661 RR2 08/14/01 11:14 287315 1860485 NM2 08/23/01 06:55 287315 1860485 NM2 3) Presence/Absence or detection/non- detection (e.g. occupancy)

„ “Occupancy” replaced the term presence when imperfect detection was included.

„ For species where you can not tell individuals apart. Estimate detection and occupancy simultaneously.

Darryl MacKenzie et al. (2006, 2015) demonstrated p/a information (detection histories) could be incorporated directly into a maximum likelihood estimation framework Landscape occupancy Farris et al. 2015, 2016.

0.6

0.5

0.4

Introduced species 0.3 ) Ψ (

0.2

0.1

Indian probability of occupancy 0 0 2 4 6 8 10 12 14 Distance to Village Multi-season occupancy NATIVE SPECIES

Cryptoprocta ferox fossana goudotii 1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2 Probability of Occupancy of Probability 0 0 0 2008 2010 2011 2012 2013 2008 2010 2011 2012 2013 2008 2010 2011 2012 2013

Galidia elegans fasciata concolor 1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

0.4 0.4 0.4

0.2 0.2 0.2 Probability of Occupancy of Probability 0 0 0 2008 2010 2011 2012 2013 2008 2010 2011 2012 2013 2008 2010 2011 2012 2013

Canis familaris species EXOTIC SPECIES 1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2 Probability of Occupancy of Probability Farris et al. 2017 0 0 2008 2010 2011 2012 2013 2008 2010 2011 2012 2013 ABUNDANCE/DENSITY ESTIMATION

1998: Seminal paper by Karanth and Nichols establishing use of camera traps to study naturally marked carnivores 4) Abundance/density estimation

Individual ID and Capture Histories Compare spot/stripe patterns to identify individuals => get abundance and density estimates

Kelly et al. 2003, Silver et al. 2004 Individual ID of male deer from antlers

Population size estimates for male deer from mark- recapture Black Research

Population size estimates using “marked” bears. Note streamers as “marks”.

Bridges et al. 2004 density and sex ratios in Belize

Satter et al. 2019 Multi-season SCR – growth and survival

Ocelot survival through time

Ocelot population growth rate between time periods

Satter et al. 2019 Ocelot density surfaces

Darker blue indicates areas of higher ocelot density

Satter et al. 2019 Rich et al. 2019 Multiple densities? 25 Mopane Non-mopane Overall

20

15

10 Density(#/100km²)

5

0 Spotted Wild Civet

Rich et al. 2019 Abundance/Density Estimation is rapidly expanding

„ Spatially explicit mark-recapture

„ Maximum likelihood (Program DENSITY)

„ Bayesian techniques (SpaceCap – etc.)

„ Mark-Resight for partially marked populations

„ Spatial Mark-Resight

„ SPIM – Spatial partially identity models for marked pops with categories. Camera Trapping - Strength/Weakness 1. Strengths: y Minimally intrusive y Abundant data with relatively minimal labor y Detects multiple species simultaneously y Photos often important for public outreach y Studies of elusive species possible 2. Weaknesses: y Costly equipment (especially for initial set up) y Equipment malfunctions (e.g. sensitive to weather etc.) y Individual identification (hence abundance estimate) is only possible for species with distinctive marks or pelage (e.g. tigers, jaguars, etc.) y Results may be sex-biased CHALLENGES: Camera Trap Data Management Spreadsheets, Citizen Science, Artificial Intelligence Importance of Data Mgmt

„ Encourage entering data on ALL species and humans the first time around.

„ Often takes longer than collecting data if you enter everything

„ Formatting for various programs and analyses also take time Entering data on species captured

1) Manual data entry using “home-made” spread sheets (Excel, Access).

2) Use Citizen Science existing platform

3) Use an Artificial Intelligence existing platform Kelly et al. 2012 Manual Data entry

Dcoument camera info at each first and last trigger, including when changing out memory cards.

In addition to camera station, date, and camera number, I have now added “time” to our trigger cards in case the time stamp malfunctions (you can correct for that later). Kelly et al. 2012 Manual Data entry!! Helps if you have access to undergrads or interns Manual Data Entry for Individual ID

Kelly et al. 2013 Citizen Science eMammal https://emammal.si.edu/ eMammal - facilitates the sharing of camera trap images and data for research and education purposes. Provides the public a query tool, allowing a user to access and share camera trap information from a variety of projects.

Encourages collaboration with other institutions and individuals. For a fun animal ID game using eMammal favorites, click on eMammal Lite for more! eMammal

License to Use Data. Provider grants to SI the royalty-free, nonexclusive, worldwide right, but not the obligation, to use, reproduce, publish, distribute, or otherwise use all Data (including, without limitation, to aggregate it with other data to create new products, to copy it, to cache it and to incorporate it into other works in any form, media or technology now known or later developed), and to sublicense such rights to third parties for purposes within the scope of the Data, Software and Web Site User Agreement (Attachment).

Provider hereby grants to SI the right to obfuscate geographic locations of any images provided by Provider that include IUCN endangered or critically endangered species or species designated by the Provider Zooniverse https://www.zooniverse.org/ People-powered research. This research is made possible by volunteers — more than a million people around the world who come together to assist professional researchers.

Study objects of interest gathered by researchers, like images of faraway galaxies, historical records and diaries, or videos of animals in their natural habitats. Upload your Project

Biology Projects https://www.zooniverse.org/projects?discipline=biology&page=1&status=live Artificial Intelligence

Wildlife Insights https://www.wildlifeinsights.org/home

„ Cloud-based, AI enabled platform

„ Google cloud

„ Species classification Combines field and sensor expertise, using cutting edge technology and advanced analytics to enable people everywhere to share wildlife data and better manage wildlife populations. Anyone can upload their images to the Wildlife Insights platform so that species can be automatically identified using artificial intelligence. Wildlife Insights Artificial Intelligence Models

„ Blank image filtering: filtering out images without animals, while reducing the possibility of removing valuable images of animals.

„ Species Classification: trained AI models to recognize hundreds of species from around the world. High-quality training dataset for AI models, consisting of over 8.7M images. Agreements

Embargoes „ All data from Wildlife Insights core partners will be shared with the public when the platform is released. However, some users may want to keep data private to comply with legal requirements or to publish research. Those users will be able to embargo data for a period of up to 24 months of time. The embargo begins on the date a deployment is created and is measured separately for each deployment. A user may request up to two extensions of up to 12 months each (24 additional months and 48 months total. Project metadata will be available publicly for any embargoed projects and all images will eventually become public, unless images contain sensitive data. Images of Humans „ Any image of a human will not be publicly available on WI, although the user may still store and analyze these images. WI will provide tools to remove any image containing a human from searches and will provide an option for users to delete images of humans. Metadata (e.g. time, location) associated with images of humans will remain publicly available on the platform. Comparing Citizen Science to AI – species ID

• Used 4 data sets in Zooniverse to differentiate among images • Accuracy empty images 91-98% • Accuracy specific species 89-93% • The larger the data set, the better the accuracy • Reduced human effort by 43% What about individual ID? VT Study

„ Bolger et al. (2011) at Dartmouth

„ SIFT algorithm

„ Java program

„ Simple interfaces minimizes computing costs

„ Crall et al. (2013) at Rensselaer Polytechnic Institute

„ Two-algorithm approach to pattern recognition

„ Windows and MacOS versions, source code for Linux

„ Interface has more features Methods „ Ranked all images 3 point scale for quality

Low Medium High Matching Methods Results

Jaguar images Ocelot images * * 1.00 * 0.90 * 0.80 * 0.70 0.60 0.50 0.40 0.30 0.20

Proportion positive IDs positive Proportion 0.10 0.00 all images high medium low all images high quality medium low quality quality quality quality quality HotSpotter jaguars Wild-ID jaguars HotSpotter Ocelots Wild-ID ocelots

* indicates significant difference * indicates significant difference

TAKE HOME – Include multiple quality images since images match more often to images of the same quality level WildBook

http://www.wildme.org/wildbook/doku.php

Wildbook ® is an open source software framework to support collaborative mark-recapture, molecular ecology, and social ecology studies, especially where citizen science and artificial intelligence can help scale up projects. It is developed by the non-profit Detection First

Deep convolutional neural networks (DCNNs) that applies a fully-connected classifier on extracted features. Three separate networks produce: (1) whole- scene classifications looking for specific species of animals in the photograph, (2) object annotation bounding box localizations, and (3) the viewpoint, quality, and final species classifications for the candidate bounding boxes. Individual identification Assigns a name label to each annotation from detection. To do this, SIFT descriptors are first extracted and then compared at keypoint locations. Scores that match the same individual are accumulated to produce a single potential score for each animal. The animals in the database are then ranked by their accumulated scores. To Conclude

„ There seems to be great promise in AI for identifying images to species, but there is concern about data ownership.

„ We are making progress on individual ID – but there is much room for improvement „ Data set too small, not worth the pre-processing time – just using human eye is faster „ Too big, pre-processing and computing may be prohibitive and programs crash.

„ Last issue that no one has really tackled well – what about all those studies out there using 2 cameras per station. We don’t want to double count animals, but we also want to know if the animals are different. Humans are still better at this.