Training
dogs to distinguish between black bear (Ursus
americanus) and grizzly bear (Ursus
arctos) feces
The University of Montana
IBS-CORE Program
Nearly one-eighth of a dog’s brain is dedicated to
olfaction. A dog’s nose contains up to 220 million olfactory sensory cells,
whereas man has about 5 million (Syrotuck 1972). Consequently, trained scent
detection dogs are utilized in many professional applications. Trained dogs are
used to locate narcotics, lost or buried people, and explosives. Additionally
dogs have been used to locate termites, spoiled wine, truffles, cancerous skin
areas, and wildlife feces (scat).
In the last decade scientists have explored the
limits and possibilities of canine olfaction (Brisbin and Austad 1991), and the
use of dogs as tools for wildlife researchers and managers has grown. Dogs have
been trained to locate wolf sign (tracks, scat, urine) for the monitoring of
pack locations and interactions (Sturdvian 1993); as well as the highly
publicized use of dogs for the aversive conditioning of habituated bears (Hunt
1997). Dogs have also been used to find black bear, grizzly bear, and wolf scat
(Wasser and Davenport unpublished) and kit fox scat (Smith in progress) in
order to analyze collected scat for genetic information.
We aimed to advance the use of dogs in wildlife
research by training them to distinguish between the scat of black bears and
grizzly bears. Currently, scat samples are identified for species only in the
laboratory, as such a differentiation cannot be made by visual inspection. If
dogs show the capacity for this level of distinction then they can be used as
an inexpensive litmus test to assist wildlife biologists and managers in their
efforts to monitor wildlife populations.
Dogs were
selected for their ball driven (object-oriented) nature. Dogs who were living as household pets were
used as well as kenneled dogs who were candidates for narcotic detection dog
training. Nine dogs were selected for training, 6 pets and 3 kenneled narcotic
candidates, ranging in scent experience from none to 5 weeks of a 6-week
narcotic detection training course. Ages ranged from 1year to approximately 5
years. One dog was a yellow Labrador Retriever (Kelly), one an Australian
Shepherd (Zach), two Golden Retrievers, and all others were mixed breeds
(Maverick). Training and testing was conducted in August through October 2000,
at two separate indoor locations.
The training method was based on the scent box, a
widely used training apparatus for the training of narcotic detection dogs. At
the first location, the scent box was a box of untreated ¾”plywood (43” long,
10” wide, 6.5” deep), separated into 5 compartments by a single hinged lid.
When the lid was closed the holes (3” holes spaced 9” apart center to center)
allowed access to each compartment. At the second location, the scent box was
made of ¾” melamine and was 4 individual boxes (8” long, 7” wide, 7” deep) that
could latch together to act as a single box. Each box had a 2 ½” hole in the
lid. The dog was trained to display a passive (“Sit”) response at recognition
of the grizzly scat (target odor) by waiting for the ball reward and not pawing
the sample.
Initially one grizzly sample—contained in a plastic
vial with an open lid—was placed in one of the holes of the scent box. The
other holes contained only empty plastic vials, identical to the vial that
contained the scat. The dog was led on leash to the correct hole and encouraged
by hand presentation and voice command to sniff that hole. Immediately the ball
was thrown across the dog’s line of vision and the dog was allowed to retrieve
the ball and was played with and verbally praised for a minute or so. This step
acquainted the dog with the scent box and began the process of associating
grizzly scat odor with the ball.
Next the dog was encouraged to check other holes
before being led to the correct hole and rewarded. This allowed the dog to
understand the concept of searching the entire box. Additionally he learned to
associate the odor with the ball
reward, rather than with the act of smelling a hole. The correct hole was then
moved to other locations within the dog’s search, in order to discourage the
expectation that the target scat would be in a specific location.
The use of the scent box allowed the dog to develop
a good pinpoint (nose in or directly over hole) as it allowed for the use of a
large quantity of scat. Additionally, the hole was an obvious visual cue for
the dog. However, further training required the ability to move small amounts
of scat in and out quickly. The scent box was not amenable to this due to its
tendency to retain odor (thus contaminating subsequent trials by lingering
odor). Therefore, at this stage B.D changed the training apparatus to half-pint
canning jars set within the holes of the scent-box so that the lip remained
above the surface of the box. Jars were fitted with cardboard “washers” to
attain a proper fit in the hole. Two 5-holed scent boxes were placed end to end
and a jar was placed every other hole, so that a total of 5 jars were each
placed 18” apart. Empty holes were covered with cardboard. At the second
training location the jar apparatus was created using two 2x4 pieces of lumber
nailed together with wooden spacers to allow the jars to slide between them and
remain with the lip above the boards. Again the jars were set at 18” apart and
the entire space between them was covered with cardboard.
The dog was led across the jars—one of which
contained approximately 50mL of grizzly scat—until it was clear that the change
in apparatus was not a hindrance to the dog. When the dog displayed a
satisfactory recognition of several grizzly scats we began conflict training.
Generally, conflict training is the introduction of odors for which dog is not
rewarded. For this training the conflict odor was black bear scat. One jar was
filled with approximately 25mL of black bear scat—a lesser amount than the
grizzly so that the grizzly odor was the overwhelming odor. The black bear jar
was located 2 jars away from the grizzly jar. The dog was led on leash to the
box so that he would encounter the black first; if he continued to smell the
black bear or give a positive response he was given a correction (snap of the
collar) and told “no”. The dog was then led along the box and when he smelled
the grizzly he was rewarded with the ball and abundant verbal praise. This was
repeated moving the scats (target and conflict) around between each approach.
The amount of black bear in each jar was increased to equal the grizzly, about
50mL. Various scats were presented and complexity increased until all 5 holes
were filled with scat, 1 grizzly and 4 black bear.
When the dog showed proficiency at hitting on the
target scat the assistant then moved the scat around and rounds were completed
without the handler knowing which jar contained the target scat. This removed
the possibility that the handler was cueing the dog. At this advanced stage of training, the dog was occasionally
given rounds in which there was no grizzly scat present. When the dog
accurately found nothing he was rewarded with strong verbal praise but no
ball. Different grizzly scats were used
throughout training to encourage the dog to generalize grizzly odor and
recognize it in all forms. The assistant contaminated the jars with human odor
when picking them up to switch them in and out, and was therefore careful to
handle all jars equally so that the dog did not associate human odor with a
reward. Each dog was trained for no
more than 20 minutes at a time before taking a break, and more often for less
than 15 minutes.
At all stages the dog was encouraged to sit upon
smelling the correct hole so that he presented a clear sign to the handler that
he had located the scat. He was encouraged to re-smell the correct hole with a
verbal “show me” command. A dog who did not know the sit command was not forced
to sit, as it was inappropriate from a training standpoint to insert obedience
lessons into the middle of a scent training session which must remain fun and
brief. When lacking a sit, the dog
naturally developed another signal (looking at handler, staring into hole with
ear-set moved forward) that the handler learned to read.
The use of leash, the proximity of handler to dog,
and the extent of hand presentations did not remain constant throughout
training. Instead these were dictated primarily by the working style of the
dog, through which preferences were displayed by improved concentration,
confidence, or interest in working.
At the first location B.D handled both Kelly and
Maverick during training and testing. A.H handled Zach at the second location.
For testing, 25 rounds were randomly configured with 4 jars containing black
bear and one containing grizzly for 20 rounds, and 5 jars holding black bear
and no grizzly for 5 rounds. These “blank” rounds were nearly randomly
distributed, selected only so that they did not occur consecutively. From a
training aspect, the blank rounds were potentially demoralizing for the dog
because he was essentially working for no reward. An experienced detection dog
could withstand this, but with novice dogs care was taken to temper the impact
of no reward. The assistant placed out the scat according to the random
assignment, so the handler was unaware of the location of the target scat. The
handler led the dog to the jar-box starting at the first jar and working down
to the fifth jar. The handler had full discretion in how many times to lead the
dog back and forth along the box before deciding that the round was complete.
The dog indicated on a hole and the handler verified with the assistant that
the hole was correct before rewarding the dog, to avoid rewarding in error. If
the dog was incorrect then the assistant said no, and the handler and dog
continued down the box. The false response was recorded by the assistant for
that round as well as the subsequent “hit” or “miss” depending on whether or
not the dog ended up choosing the correct scat. If the handler declared the
round complete and the target scat had been missed the assistant told the
handler the location of the target scat so that the dog could be brought to it
and rewarded. The dog was given a several minute break in between rounds. Each dog encountered the same 25
configurations in a different order.
All training and testing was videotaped. All scats
were frozen as soon after collection as possible. They were transferred from
collection bags into vials using latex gloves.
During the testing, dogs were exposed to a total of
33 scats (see table 1), 13 grizzly and 20 black bear. 16 (6 grizzly and 10
black bear) were the scats that were used during training. The 13 grizzly scats
came from 4 different animals, all captive. Dogs were trained with 6 scats from
2 individuals (from same facilities with same diets) and were then tested using
the training scats as well as 7 other scats—3 from the individuals they were
trained on and 4 scats from 2 individuals they had not encountered in training.
The 20 back bear scats were both wild and captive. Dogs were trained on 10
black bear scats from 7 individuals (two black bears had the same diet as the
two grizzlies used for training). Testing included these scats as well as 3
additional captive scats from the animals they had been trained on and 7 new
wild black bear scat that they had not previously encountered.
Hit rates (see table 2) were broken down into
categories depending upon the scat used: scat that was used to train the dogs
(Known); scat that was from individuals that the dogs had never encountered
(Unknown); new scat from known individuals (New from known bears); and multiple
scat from bears which were unknown to dogs prior to testing day, but after dogs
encountered them in training rounds became potentially “known” to a limited
extent (New on testing day). A single grizzly scat appeared from 1-3 times
throughout testing.
The false response rate (hitting on black bear) was
categorized as follows: the number of rounds the dogs gave a false response
when there was grizzly to be found in that round (FR when griz available); the
number of rounds the dog gave a false response when there was no grizzly to be
found (FR when no griz available); the number of individual false responses on
black bear scat the dog had encountered in training (FR on known black bear);
and the number of individual false responses on black bear scat new to the dog
that day (FR on new black bear).
It should be noted that this form of testing did not
allow for complete independence of rounds due to the ability of the dog to
learn from one round and potentially succeed accordingly on subsequent
encounters with that scat. A dog who missed a sample in one round would learn,
by lack of reward, that he was correct to pass it up (negative learning). A dog
who was led back to a sample after missing it and then rewarded, potentially
learned to correctly hit that scat in subsequent encounters (positive
learning). Because the potential for learning existed either way we chose
positive learning. Given this, note that the statistical analysis is not a
perfect examination of the testing. That said, results indicate that dogs
selected grizzly bear at a rate far exceeding random (for p=.85, and p=.65 both significant in the
extreme case of a=.001). Overall hit rates
for Maverick equaled 65% and 85% for Zach and Kelly. When broken down, the most
notable hit rates of 89-100% occurred when dogs hit on known scat. The hit
rates for new on training day scat ranged from 50-100%. There were only two
unknown scats, Maverick missed both of them, while Kelly and Zach correctly
identified one of the two. Of the three scats from known bears, Maverick
selected one, Kelly two, and Zach all three. The number of rounds in which the
dogs gave at least one false response were dramatically different in rounds
where there was grizzly available (15, 15, and 20%) versus rounds where there
was no grizzly to find (40, 80, and 80%). Maverick showed a tendency to false
respond on familiar black bear (71%) more often than unfamiliar black bear
(29%). Zach and Kelly false responded more often on new black bear (64 and 67%)
than on known black bear (36 and 33%).
Training and testing also resulted in a qualitative
assessment of characteristics of detection dog candidates. Of the nine dogs
that began training, three were tested (Maverick, Kelly, and Zach). Three dogs
were progressing in training but had not yet reached a testable level at the
time of this writing. Three dogs had been dropped from training. Of the dropped
dogs, one housepet had become bored with the repetitive nature of the scent box
and began pawing at each hole and lying down or going to the door. His aptitude
for learning was great; he had learned the odor quickly and maintained high
ball drive throughout. The second dog, also a housepet, displayed great ball
drive while at home but was unable to maintain drive when at the training
facility. The third dog was a narcotic detection dog candidate who maintained
ball drive, achieved odor recognition easily and remained interested in the
scent box. While her aptitude was sufficient, her attitude hampered her
success. She displayed an unwillingness to work with the handler consistently
in a methodical progression along the scent box.
Differing characteristics were also apparent among
the dogs who completed training. Maverick had previously completed 5 weeks of
narcotic training but did not finish due to injury. Kelly had been through
several weeks of narcotic training but was partnered with handlers who did not
build her working confidence. Though Maverick achieved 65% hit rate in our
testing and Kelly 85%, Maverick was the more confident worker and much easier
to read. Kelly required a higher level of handling skills because she was prone
to guess or give unreadable responses. Given equal amounts of training we
expect that Maverick would have achieved or surpassed Kelly’s success rate
based on their working styles. Zach had no experience with scent work, but
received the most training for this testing. His personality aided his progress
in that it was in his nature to please people and he developed rituals easily
and enjoyed sticking to a routine. Thus he remained eager to approach the scent
box, and worked methodically.
This experiment showed that trained dogs can
distinguish grizzly bear from black bear scat. From a professional training
standpoint, a dog with a 70% hit rate successfully completed training. A 90%
hit rate is required for Washington state certification. Kelly and Zach were
above 70% (85%) and nearly reached certification level. Maverick nearly reached
70% (65%). This success rate is notable because of two primary factors: short
training time and lack of adequate exposure to a large number of diverse scats.
In the narcotic detection program taught by B.D, dogs are trained for 6 weeks
(240 hours) to locate and alert to 4 major odors and their variations. Each dog completes approximately 600
repetitions using approximately 100 different training aids (target scents) in
which amount and packaging is varied. Maverick and Kelly were trained 1 and 4
days respectively, or approximately 1-2 hours and 4-6 hours. However, both of
them have completed approximately 5 weeks of narcotics training. Thus they did
not have to learn how to use the scent box, but rather concentrated on learning
the odor. Zach, who had no previous scent work, had 11 days of training, or
approximately 15 hours. B.D predicts that it would take 2-3 days of intense
training followed by 4-6 weeks of 3-times-weekly training to ready a dog for
the professional application of this type of odor discrimination. Short
training sessions would be imperative due to the repetitive nature of this
training, in which the scent box is always the item a dog searches.
In addition to a much longer training period,
work-ready dogs would need a great deal more scat diversity than was offered to
the dogs in this trial. Clearly, scat from two grizzlies was not enough
diversity in odor to allow the dog to generalize the scent as these dogs showed
a greater accuracy in detecting the scat they had been trained on than new
scat. Trained narcotic detection dogs—who
undergo a great deal of training and are exposed to a much larger
diversity of target odors—experience increased detection success rates once
they have ample professional working experience. Detection success is proportional to training time and diversity
of scent exposure. This is an expected and accepted relationship among
detection canine professionals, and is of the utmost importance to this study
as well.
Of the new scats encountered during testing, one was
particularly troublesome for all three dogs. Kelly missed it the first time it
appeared, she then hit it the next two times it appeared, indicating that she
learned that scat in just one encounter with it. Maverick and Zach both missed
that scat all three times that it appeared. In the case of Zach, this scat was
the only grizzly he missed. This scat was from a captive individual who had been
fed vegetation. The individuals used during training ate primarily chow with
some vegetation added. The second new individual was hit with a higher success
rate (all 5 times it appeared Zach and Kelly hit it, 3 of the 5 Maverick hit
it). This second individual was a cub that was fed exclusively chow. The lack
of diversity of scat and not having any wild scat in our study doesn’t allow us
to make broad inferences about the importance of diet in the dogs’ ability to
learn grizzly scat odor. However, this observation indicates that diet does
matter to some degree. In the face of limited exposure it may be easier for the
dog to detect a grizzly with a similar diet to the diet expressed in training
samples. We expect that increased exposure to scat diversity would reduce diet
as a selection factor.
The rounds
which contained no grizzly bear scat were problematic for all dogs. Fifty-five
percent of all false responses happened during these “blank” rounds. This is
especially impressive when taken into account that blank rounds accounted for
only 20% of all rounds. It is worth noting that we did not consider false
responses to be as serious of an error as missing grizzly scat. Presumably,
professional distinction dogs would pre-screen scat before being sent to the lab.
The inclusion of a few non-target scats would not significantly burden many
studies.
Maverick was
the only dog who had 3 blank rounds which were correct, he falsely responded on
his other two blank rounds. Kelly and Zach gave at least one false response on
each of the five blank rounds. Kelly and Zach were also similar in that when
they falsely responded they were more likely to chose a black bear scat that
was new to them, indicating that when unsure they selected something entirely
unfamiliar. This is in contrast to Maverick, who when unsure selected black
bear that was familiar to him from training. The disproportionate amount of
false responses on blank rounds made us believe that the problem was one of
training. Blank rounds made the dog work yet didn’t result in a reward, an
inherently disappointing outcome for the dog. Because of this, we did not
introduce the concept of blank rounds until the dogs were about ¾ of the way
through their training, setting up the precedent that there was always grizzly
to be found. Additionally, in the narcotic training that Maverick and Kelly had
previously, there was always target scent in the scent box. In future studies
it is imperative that dogs be introduced to the concept of blank rounds in the
initial stages of training.
With an overall successful hit rate of 65% for
Maverick and 85% for Kelly and Zach, it is clear that dogs have the ability to
select grizzly bear scat out of a line up with black bear scat. A comprehensive
training schedule of 4-6 weeks, which includes several dozen different (and
wild) grizzly bear scats and a proportional amount of black bear scat is
recommended for a dog who is to be used professionally in this type of
detection work. It is expected that this diversity in scat exposure would
increase the dog’s ability to generalize grizzly scat odor, thereby increasing
the dog’s ability to hit on unfamiliar scat and reduce any preferences for
certain diet. In real detection work
there would likely be many rounds in which no grizzly scat was in the scent
box. Therefore, it would be imperative to begin training for blank rounds very
early in the training progression. Compared to the variety of work situations
encountered by other professional detection dogs, these dogs would work repeatedly
at one location—the scent box. Dogs selected for this training would need to
possess traits that allowed them to succeed with this type of repetition. Such
characteristics include high ball drive, methodical working style, moderate
energy level, and a willingness to work consistently with the handler. Given
the success of this initial study, we feel confident that these training
suggestions can produce competent scat discrimination dogs to assist
researchers and managers in their worthwhile goals.
This work was funded in part by an IBS-CORE
Undergraduate Research Fellowship to Aimee Hurt through a grant from the Howard
Hughes Medical Institute to the University of Montana. Additional funding was
granted by the Matthew Hansen Foundation. A.H. would like to thank all of the
generous dog owners who entrusted their beloved dogs to her care: Scott and
Colleen, Kathryn and Jason, Carol B., Carol S., George, Rosemary, and Mariah.
Thanks to the busy professionals who took the extra time to gather scat for
this training: Rick Mace, Charlie Robbins, and Ed Cleveland. Thanks to IBS-CORE
staff, Kris Nord, and Brian Steele. Thanks to Meg, Tyler, Karrin, and Deb for
emotional and logistical support. A.H. would like to offer special thanks to E.G.
for vouching for this endeavor and B.D. who was wangled into this project
without even knowing it.
Literature
Cited
Brisbin,
I.L. and Austad, S.N. 1991. Testing the individual odour theory of canine olfaction. Animal Behaviour 42(1):63-69.
Hunt,
C.L. 1997. Proposal for the “Partners in Life” program. Wind River Bear
Institute. Heber City, UT.
Smith,
D.A. 2000. (kit fox research) in press.
Sturdvian,
C. 1993. Locating wolves with a dog trained in tracking and scent- discrimination. (unpublished manuscript).
Syrotuck, W. 1972.Scent and the scenting dog.
American Publications, Rome, NY.
Wasser, S.K. and Davenport, B. 1998. (black bear
research) unpublished.
Table 1.
Scats used in discrimination tests
|
|
Training |
Known
individual, new scat |
New
individual |
TOTAL |
|
Grizzly
captive |
6 |
5 |
2 |
13 |
|
Black
captive |
5 |
3 |
0 |
8 |
|
Black
wild |
5 |
0 |
7 |
12 |
|
TOTAL |
16 |
8 |
9 |
33 |
Table 2. Hit and False Response Rate
|
Dog |
Hit rate |
False response rate |
|||||||
|
|
Known (out of 9) |
Unknown (out of 2) |
New on testing day (out of 6) |
New from known bears (out of 3) |
Total hit (out of 20) |
FR when griz available (out of 20)** |
FR when no griz available (out of 5)** |
FR on new black bear |
FR on known black bear |
Maverick |
9
(100%) |
0
(0%) |
3
(50%) |
1
(33%) |
13
(65%) |
3
(15%) |
2
(40%) |
2
of 7 (29%) |
5
of 7 (71%) |
|
Zach |
9
(100%) |
1
(50%) |
4
(67%) |
3
(100%) |
17
(85%) |
3
(15%) |
4
(80%) |
7
of 11 (64%) |
4
of 11 (36%) |
|
Kelly |
8
(89%) |
1
(50%) |
6
(100%) |
2
(67%) |
17
(85%) |
4
(20%) |
4
(80%) |
6
of 9 (67%) |
3
of 9 (33%) |
** counted per round, thus multiple false responses in any one round counted only once