Content uploaded by Morand Alain
Author content
All content in this area was uploaded by Morand Alain on Apr 03, 2022
Content may be subject to copyright.
Content uploaded by Julian Pichenot
Author content
All content in this area was uploaded by Julian Pichenot on Jan 12, 2021
Content may be subject to copyright.
Design and test of a semi-automated system based on time-lapse camera
trapping for the monitoring of wildlife overpass use by amphibians
J. Pichenot1, C. Muller1, S. Aravecchia2, C. Pradalier2, G. Tekielak1& A. Morand1
1Cerema, Division Biodiversité Eau et Aménagement –Metz (France)
2Georgia Tech, CNRS, UMI2958 –Metz (France)
Studied species
•Common toad (Bufo bufo)
•Migrating from Wintering to Breeding sites
in March-April
•5 000 - 17 000 adults counted each year
by volunteers
Study site
•Road D657, Moselle (France)
•>2700 vehicles / day
•A 1.6 km long road section where 14
tunnels were constructed in 2013 for
amphibian migration
•We studied the most used tunnel in
previous years (~700 toads)
CONTEXT
☞Camera trapping is a widely used technique for wildlife crossings monitoring.
☞For small ectothermic animals like amphibians, time-lapse camera trapping has
recently been used and appeared promising.
☞We evaluated the effectiveness of this method compared to classic pitfall
trapping for counting migrating toads in a tunnel.
MATERIAL & METHODS
•Pitfall trapping were planned from 08 March to 11 April 2018.
•During the same period, an infrared time-lapse camera (Reconyx® Hyperfire
PC900) was installed to record an image every 10 s from 20:00 to 08:00.
•The comparative analysis was performed on a data sample of 25 days.
•To reduce the time of image analysis (N =108 000), we trained a Convolutional
Neural Network to detect toads on images.
•Identified toads were automatically marked with a bounding box, which
facilitate the counting by an operator on a reduced sample of images.
1
2
3
PRELIMINARY RESULTS
AKNOWLEDGEMENTS : We thank Vincent ASSELOT (Communauté de
Communes Mad ett Moselle), Manon PELLICORI (Master Degree,
Université de Lorraine), Laurent GODE (Parc Naturel Régional de
Lorraine) and all the volunteers involved of this monitoring.
1 2 3
☞Pitfall and time-lapse counts were significantly
correlated, suggesting that time-lapse camera
trap could be used to estimate the actual
number of toads crossing.
r= 0.93
p-value = 3.34e-11
☞Working with more species and during a longer monitoring
period.
☞Improving the detection algorithm to further facilitate the study
of amphibian species and individual behavior during crossings.
MAIN PERSPECTIVES
☞Number of counted toads varied according to
temperature and precipitation, with a significant
drop below 4°C (gray areas).
☞Movements to the ponds were observed
throughout the night, with a peak around 00:00
whereas return movements (to the wood) were
mainly observed in the morning (at sunrise).