Ana səhifə

An examination of the performance of Dyer’s woad (Isatis tinctoria) detection dogs


Yüklə 23.64 Kb.
tarix15.06.2016
ölçüsü23.64 Kb.
An examination of the performance of Dyer’s woad (Isatis tinctoria) detection dogs

Kim Goodwin

Land Resources and Environmental Sciences Dept

Montana State University

8 Dec 2009

Duration: 2 yrs

Cost: $176,400

Project summary

Invasive plants have devastating impacts on ecosystems and pose economic challenges that early intervention can prevent. Early control can minimize impacts but eradication is difficult to attain due to limitations posed by the low density and detectability of individuals. Past research indicates specially-trained domestic dogs (Canis familiaris L.) are more accurate than humans at detection of rare spotted knapweed plants (Centaurea stoebe L.). Studies evaluating the use of dogs to detect other rare plant species have not been investigated to our knowledge. Requested funds will compare the performance of detection dogs to human surveyors in locating rare Dyer’s woad (Isatis tinctoria L.), an invasive plant with extensive infestations in northern Utah, southern Idaho, and western Wyoming. Results will be published and, if the dog method is successful, training and deployment guidelines will be included to support domestic eradication programs and field operations by which the use of detection dog technology is expanded as a functional model.


Introduction

Invasive plants can displace native species, affecting ecosystem processes (Dukes and Mooney 2004) and threatening biodiversity (Wilcove et al. 1998). Although early control of new populations can minimize impacts, successful eradication is difficult to attain1 due to imperfect detection rates 1for juvenile and small adult plants (Tomley and Panetta 2002). Plants that escape detection can reproduce and increase the longevity of the seed bank and the amount of time needed for eradication. Domestic dogs (Canis familiaris L.) have recently been trained to search for and locate new invasions of spotted knapweed (Centaurea stoebe L.) with greater overall accuracy (81%) compared to human surveyors at 59% (Goodwin et al. 2009). Better dog accuracy was due to their (olfactory) ability to detect small knapweed plants (< 0.053 m3) at greater accuracy (67%) than human surveyors (34%) at 81% probability and better.

Invasive plant monitoring using detection dogs might provide greater overall accuracy of detection for spotted knapweed and potentially other invasive plants that are even more difficult to control. To better understand the application and deployment of detection dogs in rare plant monitoring, our first objective is to test the hypothesis canines will detect rare Dyer’s woad (Isatis tinctoria L.) with better accuracy compared to human surveyors. 1Dyer’s woad was chosen as the experimental model based on its State-listed Category 3 noxious weed designation (which requires eradication by landowners) and availability in southwestern Montana. Our second objective will be based on our first objective whereby the results, if successful, will guide the development of training and deployment protocol for operational agencies interested in using detector dog technology to monitor plant populations.
Literature review

Based on the results of our previous study and given all plant species produce blends of volatile compounds with distinctive odors, it seems likely dogs could be trained to search for and detect the odor of Dyer’s woad. In field settings, detection dogs with associated handlers have been used to locate buried landmines (McLean 2003), human remains (Killam 1990), cadaver (Rebmann et al. 2000), desert tortoises (Gopherus agassizii C.; Cablk and Heaton 2006), San Joaquin kit fox scat (Vulpes macrotis mutica Merriam; Smith et al. 2003), and the presence of black-footed ferrets (Mustela nigripes Audubon and Bachman1; Reindl-Thompson et al. 2006). Detection dogs typically search an area by working back and forth in a serpentine pattern while continuously sampling small amounts of odor. When the dog detects a trace of the target scent it follows the odor gradient to its source and gives a trained alert using a reward-based system.


Approach

The intent of this study is to test additional plant species by comparing the 1performance of Dyer’s woad detection dogs to human surveyors, the standard detection method. Favorable results will guide the development of training protocol including canine candidate selection and deployment guidelines to improve surveillance and eradication activities for state and federal noxious weeds.

A treatment-control experimental design will be used to evaluate dog and human surveyor performance. Six human surveyors with at least one year of survey experience for Dyer’s woad will be selected to participate in this study. In addition, six dogs will also be selected for high play drive, object orientation, and appropriate temperament. The number of dogs selected is based on cost considerations (about $2,000 per dog with preliminary training; 1U.S. Congress, Office of Technology Assessment 1992) and is twice that from the number of dogs over the previous study (n = 3). The dogs will be trained to discriminate the odor of Dyer’s woad with techniques described in our previous study. In brief summary, the dogs will learn to associate the target scent (fresh clipped stem and leaf material) with a reward like play and praise or food. The dogs will then be trained to indicate the presence of the odor to the handler with a trained response such as sitting or scratching at the source of the odor, i.e., the Dyer’s woad plant. Additional odors including clipped material of other plant species will be added as controls to ensure the dogs are responding to the target odor. Once the dogs are competent at discriminating the target odor, search training will be conducted in field settings with fresh clipped plant material and natural plants.

The study area will be located in southern Beaverhead County near Lima, MT USA in a semiarid shrub-steppe site, which historically contained a large Dyer’s woad infestation (85 ha). Population density has declined since 1985 when local eradication efforts were initiated. Over the last four years, about 682 plants (SE = 144) are annually removed from this site. We will delineate sampling units (1.0 ha) known to contain low density Dyer’s woad plants. Ground truth or the true number of plant targets on the sampling unit will be obtained to aid in the interpretation of the detection data. Target location and sampling unit corners will be delineated with GPS1 (geographic positioning system). Target size, density, and grass height and percent cover will be characterized one time prior to the trials and ambient temperature, relative humidity, and wind speed data will be collected every hour to determine the range of weather conditions and average among trials.1

The testing procedure of human surveyors and dogs will be similar to the methods described in our previous study. Briefly, line transect sampling for stationary populations (Anderson et al. 1979) using a census-based approach will be performed. Parallel transect lines will be established at 5 m intervals and defined with survey flags. Dogs will be off-leash and perform the search on their own ahead of the handlers (~10 to 15 m). Handlers will be instructed not to look for targets so dog performance will not be biased. 1Dog and human surveyors will independently search the sampling unit in the early morning (0600 – 1000 hrs) to control for daily weather cycles.

Measurement and statistical methods

The four possible response outcomes for the detection task includes true positive (TP) or hit, false positive (FP; Type I error) or false alert, false negative (FN; Type II error) or miss, and true negative (TN) or correct rejection to target absence. We will record the TP and FN locations with GPS and calculate the detection rate (also called TP rate) as the proportion of TP divided by the sum of TP and FN and false alarm rate (also called FP rate) as the proportion of FP divided by the sum of FP and TN. To compute the TP rate and FP rate, the sampling unit will be divided into a large number of cells with geographic information system technology. The area and number of cells will be based on the density of targets and will range from five m2 (2,000) to 10 m2 (200) cells. A cell will be considered to contain a 1) detection (TP) if a target plant is present and detected, 2) false alert (FP) if a target is absent and an erroneous detection is made, 3) miss (FN) if a target plant is present and not detected, and 4) correct rejection (TN) if a target is absent and a correct non-detection is made.

Detection is a binary task involving hypotheses, H1 (absent or non-target) and H2 (present or target), and independent trials of success and failure (either the target is detected or not) which results in a binomial distribution. However, we will not have prior probabilities and thus, we will compare dog and human surveyor performance with receiver operating characteristic (ROC) analysis (Egan 1975). ROC curves are commonly used in detection theory and incorporate two performance indicators (TP rate and FP rate) to make allowances for chance agreement. For proportions the 95% CI will be determined for each point in ROC space to evaluate the precision of the point estimates. We will generate ROC curves using MatLab version 7.4 (The Math Works, Natick MA, USA).
Timetable

Spring 2010: select canine candidates and begin scent discrimination and search training

Summer/Fall 2010: scout the study area and delineate and ground truth sampling units

Winter 2010: continue training detection dogs

Spring 2011: select human surveyor participants

Summer/Fall 2011: conduct performance trials

Winter 2011: analyze data and formulate reports
Budget and justification

Personnel: $60,000 salary for two dog trainers at $15,000 per year for 2 years

$30,000 for research assistant at $15,000 per year for 2 years

$6,000 for six human surveyors at $1,000 per person for 1 month

Equipment: $12,000 for six 1young, high-quality German Shepherds or Belgian Malinois with obedience training ($2,000 each)

Travel: $4,000 for research assistant travel from Bozeman 12 times in summer/fall 2010 to delineate and ground truth sampling units and eight times in 2011 ($200 per round trip at $0.55 per mile) to conduct field trials

$2,000 for dog trainers to travel to study site and deploy dogs in field trials

Supplies: $12,000 for1 food and veterinary outlays ($2,000 per dog for 2 years)

Total direct costs: $126,000

Indirect costs at 40%: $50,400



Total project cost: $176,400
Qualifications – student thesis proposal

LRES 590 Master’s thesis (12 cr)

LRES 500 Seminar (2 cr)

BIOL 525 Research methods (3 cr)

STAT 401 Applied methods in statistics (3 cr)

STAT 410 Methods for data analysis (3 cr)

STAT 446 Sampling (3 cr)

PSY 461 Judgment and decision making (3 cr)

PSY 501 Advanced research design and analysis (3 cr)
Literature cited

Anderson, D. R., J. L. Laake, B. R. Crain, and K. P. Burnham. 1979. Guidelines for line transect sampling of biological populations. J Wild Manag 43:70-78.

Cablk, M. E., and J. S. Heaton. 2006. Accuracy and reliability of dogs in surveying for desert tortoise (Gopherus agassizii). Ecol Appl 16:1926-1935.

Dukes, J. S., and H. A. Mooney. 2004. Disruption of ecosystem processes in western North America by invasive species. Rev Chil Hist Nat 77:411-437.

Egan, J. P. 1975. Signal detection theory and ROC-analysis. New York: Academic Press.

Goodwin, K. M., R. Engel, and D. Weaver. 2009. Detection dogs outperform human surveyors in rare spotted knapweed detection, unpublished manuscript, Montana State Univ., Bozeman, MT.

Killam, E. W. 1990. The detection of human remains. Springfield: Charles C. Thomas.

McLean, I. G., editor. 2003. Mine detection dogs: training, operations and odour detection. Geneva: International Centre for Humanitarian Demining.

Rebmann, A., E. David, and M. H. Sorg. 2000. Cadaver dog handbook. Boca Raton: CRC Press.

Reindl-Thompson, S. A., J. A. Shivik, A. Whitelaw, A. Hurt, and K. F. Higgins. 2006. Efficacy of scent dogs in detecting black-footed ferrets at a reintroduction site in South Dakota. Wildl Soc Bull 34:1435-1439.

Smith, D. A., K. Ralls, A. Hurt, B. Adams, M. Parker, B. Davenport, M. C. Smith, and J. E. Maldonado. 2003. Detection and accuracy rates of dogs trained to find scats of San Joaquin kit foxes (Vulpes macrotis mutica). Anim Conserv 6:339-346.

Tomley, A. J. and F. D. Panetta. 2002. Eradication of the exotic weeds Helenium amarum (Rafin) H. L. and Eupatorium serotinum Michx. from south-eastern Queensland. Pages 293-296 in H. Spafford Jacob, J. Dodd, and J. H. Moore, eds. Proc. of the 13th Australian Weeds Conference, Plant Protection Society of Western Australia. Perth, Australia.

1U.S. Congress, Office of Technology Assessment. 1992. Technology against terrorism: structuring security. Washington, DC: U.S. Government Printing Office. Report OTA-ISC.

1Wilcove, D.S., D. Rothstein, J. Dubow, A. Phillips, and E. Losos. 1998. Quantifying threats to imperiled species in the United States. BioScience 48(8):607-15.







Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©kagiz.org 2016
rəhbərliyinə müraciət