Canine detection of explosives under adverse environmental conditions with and without acclimation training

Authors: Sarah A. Kane, Lauren S. Fernandez, Dillon E. Huff, Paola A. Prada-Tiedemann, Nathaniel J. Hall

Abstract

Canines are one of the best biological detectors of energetic materials available; however, canine detection of explosives is impacted by a number of factors, including environmental conditions. The objectives of this study 1) determine how canine detection limits vary when both the canine and odorant are tested in varying temperature and humidity conditions (canine and odor interactive effects); and 2) determine if an acclimatization plan can improve detection limits in an adverse environmental condition. Eight working line canines were trained to detect four prill ammonium nitrate (AN), Composition 4 (C4), trinitrotoluene (TNT) and double base smokeless powder (SP). In Experiment 1, canines completed a 3-alternative forced choice 3-down-1-up staircase threshold assessment in five environmental 40°C and 70% relative humidity (RH), 40°C and 40% RH, 0°C and 90% RH, 0°C and 50% RH and 21°C and 50% RH. Canines showed a 3.5-fold detection limit increase (poorer detection) for C4 in 40°C and 70% RH compared to their detection limit at 21°C and 50% RH. In Experiment 2, the eight canines were split into two groups (n = 4), control and acclimation groups. The control group completed the threshold assessment for C4 at 21°C and 50% RH each day for 20 days, with 5 minutes of petting prior to testing. The acclimation group completed the same assessment daily starting at 21°C and 50% RH but temperature and RH were incremented daily over the course of 6 days to the 40°C and 70% RH condition. After the initial six days, the acclimation group completed daily assessments at 40°C and 70% RH condition for the remainder of the experiment. All acclimatization group canines started their session with 5 minutes of toy or food retrieves. Detection limits for C4 for all dogs were tested in 40°C and 70% RH on day 11 and day 22. The acclimatization plan improved detection limits in the 40°C and 70% RH condition for C4 compared to the non-acclimated group. In this set of experiments, canine detection limits for four explosive odorants were found to vary based on environmental condition and were mostly driven by impacts on the canine rather than odor availability. The acclimatization plan did result in lower detection limits (i.e., increased performance). Future work should determine what factor (exercise or environmental exposure) is more effective in acclimatization for odor detection work.

Introduction

Explosive detection canines (EDCs) are critical for public and military safety. Currently, specially trained canines are one of the best detectors for explosives because of their mobility, and capacity to deploy in a variety of environments [1]. EDCs are able to search environments which other detection technologies cannot access [1]. In addition, EDCs can be deployed in a variety of environments, both indoor and outdoor [2]. Farr et al. (2021) surveyed EDC handlers who reported they were regularly deployed to search in climates to which they had limited acclimation prior to searching.

Although the versatility of EDCs is a benefit, further research is needed to understand how temperature and humidity can affect their detection abilities [3]. Because EDCs can be rapidly deployed to harsh environments, it is crucial to know how these environmental changes can impact their detection limits. In addition, such research could provide a useful resource for handlers on best practices for working EDCs in adverse conditions.

Much of what we know about environmental effects on detection limits does not come from the EDC literature, but from conservation detection dogs. In one such study, relative humidity and air temperature were not correlated with detection rates of the desert tortoise [4]. This lack of a result could be procedural as canine teams were rested once the dog showed behavioral signs of heat stress, or their rectal temperature was greater than 40°C. Because dogs were only worked when they were in optimal physiological conditions, fatigue or stress caused by the environment did not impact their detection rates. This study did note that canine fatigue was a limiting factor in duration of search [4]. It was unclear if that fatigue was due to the physical demands of the task, the environmental conditions in which the canines worked, or a combination of both. Another study, however, found that temperature did affect detection rates; higher air temperatures and higher windspeeds led to greater detection distances (better detection) [5]. Unfortunately, due to the lack of climactic variation in this study, further conclusions on the effect of climate (specifically relative humidity) on detection were limited.

Gazit and Terkel (2003) studied how physiological factors, specifically exercise induced fatigue, affected canine detection of explosives. Gazit and Terkel (2003) compared canines’ ability to detect C4 in indoor and outdoor search areas under two after rest and after strenuous physical activity. They found that canines were always able to detect the explosive regardless of condition, in the indoor search, however after strenuous activity, canines missed explosives in their outdoor search. Search time was also increased, both indoor and outdoor, after exercise. Gazit and Terkel (2003) suggested that the decreased ability to detect C4 after activity was because of changes in the canines’ sniffing to panting ratio. In order to maintain homeostasis during and after exercising, canines need to pant to cool via evaporative cooling [6]. Gazit and Terkel (2003) posited that as panting increased, sniffing deceased, and thus detection of C4 decreased. Canines did show acclimatization throughout the study to detect C4 after exercise [7].

Past research in our laboratory has studied how canine physiological factors alone affect detection limits of narcotic signatures and explosives (Brustkern et al., 2023; Fernandez et al., in review). Fernandez et al. (in review) worked canines through a go/no-go threshold assessment with four different explosive targets. The canines worked in an environmentally controlled chamber; however, the odorant was kept outside of the research area in a temperature-controlled water bath. The purpose of this study was to determine how canine factors alone affected detection limits for explosives. This study found that canines had lowest detection limits in standard (22°C and 50% RH) conditions compared to environments with higher or lower temperatures and RH levels. A similar study which used methyl benzoate as the target material, also kept the odorant in a consistent environment while the dog worked in a variety of temperature and humidity conditions [8]. This study found that canines had poorer detection in high temperature conditions compared to standard, but did not see a performance decrement in low temperature conditions. This study also found that canines had poorer detection immediately upon entering the harsh environment from a controlled condition, i.e. there was a “startup” decrement [8].

In addition to canine factors, odor availability is also affected by the environment [1]. In high temperature and relative humidity conditions more volatiles of an explosive enter the headspace. This should allow for greater availability for canine detection. Conversely, colder conditions decrease volatiles in the headspace and lead to decreased odor availability [9]. Both odor and canine physiological state fluctuate with environmental conditions, and these interactions determine detection limits. Huff et al. (2022) studied how explosive odors behave in a variety of temperature and relative humidity conditions using SPME and GC-MS techniques. In general, when sampling occurred in hot conditions, volatiles in the headspace increased as expected and decreased in colder conditions (9).

It remains unclear, however, how odor availability and canine behavior interact when in adverse environmental conditions together. It is possible that any challenges dogs experience in high temperatures could be mitigated by increased odor availability in these temperatures. Conversely, it is possible that despite elevated temperatures, and increased odor availability, physiological effects on the canine may decrease detection, as seen by Gazit and Terkel (2003). In cold conditions, odor is less available, but this effect may be lessened if the dogs are not under physiological stress (i.e. panting) when working in this condition.

It is evident in the literature that there is a lack of conclusive support on how climate, odor and canine detection interact. The objective of the first experiment of this study was to determine how detection limits of canines are affected by environmental condition when both odorant and canine are held in the same environmental conditions. The second experiment objective was to develop an acclimatization plan to improve detection in adverse conditions.

Experiment Characterizing canine detection sensitivity in varying environments

Materials and methods

Participants

Eight dogs (6M, 2F; 4 Labrador retrievers, 4 German Short-Haired Pointers), former candidates of a government detection program, participated in this project. The dogs had various amounts of prior training with explosives.

Animal welfare considerations

This project was conducted at Texas Tech University and was overseen by the Institutional Animal Care and Use Committee (Protocol #21051–07) and approved by the US Army Medical Research and Development Command Animal Care and Use Office (#78018-ST-H.e001). In addition to their participation in this project, all the dogs had additional enrichment activities every day and received food enrichment. Throughout training and testing, all dogs always had access to water. During testing, if canines refused to search for 5 trials in a row or laid down and did not get up when called, testing was stopped, and they were immediately removed from the environmental chamber. This termination criterion was used to ensure participant safety; this criterion never occurred under standard conditions and was indicative of lack of performance.

Odorants

Four odorants were used in this project; C4, double-based smokeless powder, ammonium nitrate prill (explosives grade), and TNT (see Table 1 for amounts and brands). Use of all energetic materials was approved by the Institutional Laboratory Safety Committee on Energetic Materials (Protocol ILSC#2110E1). There were no distractor odorants, all other valves of the olfactometer (described below) were connected to empty borosilicate glass vials matching the type used to contain the target odorants.

Data collection equipment

Three Bluetooth enabled olfactometers were used in this project, which were a modification of our prior automated line up equipment (see Aviles-Rosa et al., 2021). The odor generation design was identical, but instead of having a horizonal panel of odor ports, the odor port was integrated into the olfactometer box itself. Additionally, the programing was modified to allow Bluetooth control of the olfactometers.

Each olfactometer was an independent replicate containing the same odor delivery system and odorants. Each olfactometer had six one vial contained the target odorant (C4, TNT, AN, or SP) while the remaining five vials were blank controls. These control vials were cleaned and prepared in a manner identical to the target vial.

Odor was generated from a 4L/minute oil-less air pump. Air from the pump first passed through a charcoal filter to scrub the air of odorants. The air was then split to an odor line and a dilution line via a T-connector. The odor line was regulated by a 0–1 lpm rotameter and the dilution line was regulated by a 0–4 lpm rotameter. From that regulating rotameter, the odor line passed to a manifold with one-way valves. Upon activation of a valve, the odor line flow was passed into a vial containing the odorant (or a clean distractor vial). This displaced the headspace of the vial through a one-way check valve into a PTFE mixing manifold where it mixed with clean air from the regulated clean air. Air from this manifold flowed to the port for the dog to sniff.

The olfactometer was controlled by a custom program (https://github.com/njhall1/CanineOlfactometer). Odor dilutions were controlled by manual manipulation of the flowmeters controlling the odor line and dilution line. Thus, dilution was created as an air dilution by changing the ratio of air from the odor and dilution lines. The flowmeters allowed a dilution from 80 to 3% (see Table 2). For threshold testing, a series of dilutions was 80%, 50%, 25%, 12% and 3%. The flow rate, or the amount of odor being pushed into the port was consistent for each concentration, except for 80%. To reach to the 80% concentration, the flow rate had to be lower, however all three olfactometers delivered the same flow rate.

Each olfactometer contained an infrared sensor beam pair that measured all canine entries into the odor ports. These IRs also recorded when the beam was broken so that duration between nose pokes in each box could be calculated. We also calculated latency, or the duration from start of the trial to the time the dog began search (as indicated by a nose poke breaking the IR beam).

This study utilized an alternative forced choice test design, in which the target odor was always present in one of the three olfactometers. The other two olfactometers were pushing the same volume of air from one of the five control (empty) vials.

Temperature sensor

Each dog had a subcutaneous microchip over their shoulder which read their subcutaneous body temperature. At the start of every testing session, approximately every 5 minutes during the session, at the end of the session and 5-minutes post session, an experimenter scanned the 134.2 kHz ISO microchip with a HomeAgain™ Universal Worldscan™ Reader Plus (by MERCK animal health) to record the subcutaneous temperature of the dog [10].

Environmental conditions

There were five temperature and humidity conditions used in this experiment (see Table 3). Experimenters allowed the standard, low-temperature high humidity, high temperature low humidity and high temperature high humidity to vary by 3°C and 5% relative humidity (RH) from the stated condition to allow for natural fluctuations as airflow changed when the door of the environmental chamber was opened, or the AC unit began a thaw cycle. The low temperature low humidity condition was allowed to vary by 3°C and 10% RH due to ambient humidity levels.

Dogs were tested once under each condition for each target odor. All dogs were first tested with each odor under standard conditions. Four dogs were randomly assigned to one of two testing orders (A or B; see Table 4 for order). Each odor was tested to completion before moving to the next to reduce the potential of olfactometer contamination by switching odors multiple times per day. Threshold limits were determined first for C4 then SP, AN, and last TNT.

Experimental set up

Research was conducted in a 3.6 x 3.4 m environmental chamber. The room was equipped with a heater, AC unit, dehumidifiers (hOmeLabs and ALORAIR), humidifiers (AILINKE), a fan and a mister system (homenote Misting Cooling System). The misters hung from the ceiling to help the room reach humidity conditions. The room also contained a SensorPush HT.w wireless thermometer/hygrometer sensor which measured the humidity and temperature in the room (rated ± 2% RH and ± 0.2°C). Three Bluetooth enabled olfactometer boxes were arranged as a triangle in the room and presented all odors for canines (Fig 1).

Fig 1: Environmental chamber room set up (overhead view).The humidifier and dehumidifier number varied based on conditions requirements. The Bluetooth enabled olfactometer boxes (3) or BLE olfactometers are pictured.

Throughout the duration of the experiment, dogs were kept in runs with continuous outdoor access. Dogs were walked twice daily for at least 20 min per walk for the duration of this study. The average high outdoor temperature during the duration of this experiment was 33.5°C with average lows of 17.5°C.

Odor availability analysis

A solid phase microextraction headspace sampling was performed on all target explosive odorants while being exposed to each extreme condition inside the environmental chamber. This extraction mode aimed to understand how the chemical odor profile remained during an “active operational” environmental. Each environmental condition and target odorant combination included a total of ten replicate samples.

All chromatographic separations utilized an Agilent 7890A GC Series GC system, equipped with front and rear split/splitless injection ports, fitted with an HP5 30 m, 0.25 mm internal diameter capillary column. Helium was the carrier gas at a flow rate of 1 mL min-1 at an average velocity of 37 cm s-1. The mass spectrometer (MS) operated in electron ionization (EI) full scan mode from 45 to 550 amu, with a 3-min solvent delay. All compounds in the headspace were assigned based on mass spectral matches to the NIST 2017 mass spectral library.

Quantification of target VOC concentrations being extracted by the SPME fiber was performed via external calibration using a standard mix of signature VOCs within each explosive class. For the target VOCs of each explosive class, the calibration curve range between 1–100 mg L^-1^ to demonstrate the fiber capacity to extract target analytes at trace concentrations in addition to higher concentrations quantified above 100ppm without losing linearity as observed with the linear dynamic range coefficients of > 0.98. To approximate the amount of VOCs extracted by the SPME fiber, the slope of the line obtained in the calibration curve was used as a response factor for each of the analyzed compounds.

Canine odor detection testing and training

All canines had been previously trained to detect explosives in an olfactometer during a previous study [11]. Dogs were trained to detect the target odorants in a 3 alternative forced choice test paradigm prior to starting threshold testing. Canines all reached 85% accuracy at the 80% concentration across 20 trials for each odor before beginning threshold testing on that odor.

All training and testing was conducted under double blind conditions such that the handler was never aware of the location of the olfactometer presenting the target odor. Canine responses (i.e. alerts), were scored automatically using infrared beam pairs. All dogs had a nose hold and freeze alert response for the target odorants. If dogs made a correct response, a tone played, and the handler provided a reinforcer to the dog. If the dog made a false alert or failed to respond to the correct port, an “incorrect” tone played. When the dog was incorrect no reinforcer was given. At the end of the trial, there was a 20s inter-trial interval to purge the olfactometers of any residual odor.

Canine odor threshold testing

The threshold testing was conducted as an alternative forced choice task with a 3-down 1-up descending staircase procedure. In an alternative forced choice task, the target odorant was present in one of the three olfactometers in every trial. In the 3- down, 1- up threshold test procedure, after three consecutive correct trials the odor concentration was decreased by one dilution step. If the dog made an incorrect response, the odor concentration was increased by one step. A reversal point was defined as anytime the direction of odorant concentration changed. Threshold testing continued until dogs completed seven reversals. Testing also terminated if dogs made three consecutive correct responses at the lowest concentration step (3%). Additionally, testing also terminated if dogs met a welfare criterion by not searching for five consecutive trials or showed signs of elevated heat stress and refused to search for one trial. Elevated heat stress signals consisted of a dog lying down in the testing chamber, excessively panting, and refusing to continue searching for one trial. The criterion of trials not searching was reduced when accompanied by these additional signs of heat stress.

Dogs remained in the environmental chamber until they had completed testing by achieving 7 reversals, scoring 3 correct responses at 3% concentration, until they had been in the chamber for 40 minutes, or if they met our welfare criterion. Most testing sessions were approximately 25 minutes.

Data analysis

Odor detection threshold was calculated as the geometric mean of all seven reversal points. Threshold was log transformed for all analyses. If dogs’ detection exceeded the maximum dilution capabilities of the olfactometer (3 correct responses at the lowest concentration), the lowest dilution point was used as a reversal point for the remaining reversal points of that session. For example, if a dog reached 3% by reversal 2, the remaining five reversal point concentrations were imputed as 0.03. If a dog terminated a session early due to meeting the welfare criterion, the remaining reversal points were imputed as 80%, indicating failure to complete the task.

To evaluate the effect of environmental conditions on detection limits, a linear mixed effect regression model was fit in which log threshold was predicted by breed, environmental condition, odor and an interaction between environmental condition and odor, with a random effect of individual dog. This model produced a residual plot with some boundary effects, due to the limitations of the highest and lowest concentrations produced by the olfactometers (maximum concentration was 80% and minimum was 3%). Otherwise, the residual plot showed no patterns and so the model was deemed fit for interpretation. Models were fit in R using the lme4 package [12] and significance interpretation was made using the Anova function from the car package [13] and lsmeans from the emmeans package [14].

Results and discussion

The thresholds for each odor in each condition can be seen in Table 5 and Fig 2. The detection limits (log scale) for each odor varied by condition and odor identity, supporting previous work [11]. For smokeless powder, dogs’ detection limits were lower than the dilutions available by the Bluetooth olfactometers. This was expected based on the detection limits observed by Fernandez (2023).

Fig 2: Detection limit (log scale) for each odorant in each environmental condition.The figure indicates that the detection limit for smokeless powder (SP) was not reached. The points on the graph are the average log geometric mean thresholds for all dogs in each condition for each odor. The error bars represent the 95% confidence interval for those geometric means.

Post hoc tests were conducted comparing each environmental condition to standard conditions for each target odor (see Table 6). Fig 2 shows the average geometric means for all dogs in each condition for each odor and the 95% confidence intervals. Post hoc tests showed that detection limits for C4 were higher (poorer) under high temperature high humidity (est = 0.47, t = 2.7, p = 0.03) and high temperature low humidity conditions (est = 0.38, t = 2.3, p = 0.05). Low temperature low humidity and low temperature high humidity conditions did not differ from standard conditions (p>0.05). Smokeless powder thresholds did not differ between environmental conditions (all p>0.05), but this may reflect that dogs’ thresholds was below the limits of dilution of the olfactometer.

For AN, poorest detection limits occurred under the hot temperature and low temperature high humidity conditions, however these differences from standard did not reach the statistical significance criterion (all p>0.05). For TNT, dogs showed detection difficulty in all conditions and no environmental condition differed from standard.

Canine subcutaneous temperature and detection limit

Fig 3 shows average canine subcutaneous temperature readings across sessions. Only the first 25 minutes are shown as most data are captured within the first 25 minutes of a session. Similar to that observed by Fernandez (2023), canine subcutaneous temperatures were highest in the high temperature and high humidity condition. A simple mixed effect model evaluated the impact of environmental condition on the temperature recorded (with a random effect of dog). Condition had a significant impact on subcutaneous temperature (X^2^ = 1082, df = 4, p<0.0001). Post hoc tests indicate subcutaneous temperature significantly varied between all conditions except comparisons between the two low temperature conditions, which were non-significant (est = -0.071, t = -0.75, p = 0.96).

Fig 3: Canine internal temperature change throughout a session, broken down by condition.The maximum session time was 40m, however most dogs finished the detection task within 20 min (which is shown). Error bars show the 95% boot strap estimated confidence intervals.

Subcutaneous temperatures in the two low temperature conditions were not accurate and as such were dropped from further analysis. The readings from the HomeAgain™ Universal Worldscan™ Reader Plus (by MERCK animal health) in low temperature were physiologically impossible for non-hypothermic dogs. These errors were likely from the chip being impacted by the temperature condition in the environmental chamber. These chips were not reading core temperature, which would have been a more accurate measure of body temperature.

Next, we evaluated the impact of mean subcutaneous temperature within a session on measured thresholds in the standard and two high temperature conditions. We conducted a mixed effect model to evaluate whether threshold was predicted by the average subcutaneous temperature across the first 25 minutes of the session (for the two hot and the standard conditions), as well as odor and their interaction. There was no significant interaction between odor and mean temperature on threshold (X^2^ = 2.2, df = 3, p = 0.54). Threshold was however impacted by odor (X^2^ = 122, df = 3, p<0.0001) and mean temperature (X^2^ = 7.7, df = 1, p = 0.005). As mean subcutaneous temperature in a session increased by 1°C, threshold increased (became poorer) on average by 0.21 on the log scale. Higher mean subcutaneous temperature was associated with poorer detection limits (Fig 4). This suggests that canine subcutaneous temperature may be a simple and useful metric to predict or warn handlers of poorer detection limits.

Fig 4: Effect of Subcutaneous temperature on detection limits.Line shows the best fit regression for each odor.

Latency and Inter-Olfactometer-Interval

Two additional behavioral measures were examined for their relationship to detection threshold. First latency was measured as the time from the start of the trial until the dog inserted their nose into an odor port. Latency to start a trial did differ by the environmental conditions (X^2^ = 10.2, p = 0.04). A post hoc test indicates, however, that the only pairwise difference between conditions was that longer latencies were observed in the low temperature high humidity condition compared to standard (t = 2.81, p = 0.04). Latency neared statistical significance between high temperature high humidity compared to standard (t = 2.64, p = 0.069).

Although there was minimal difference in latency between conditions on average, a mixed effect model was fit in which mean session latency, odor, and their interaction predicted log odor detection threshold. Results indicate there was no interaction between odor and latency (X^2^ = 1.248, p = 0.74), but there were main effects of latency (X^2^ = 32.29, p<0.0001) and odor (X^2^ = 169.2, p<0.0001). For every 1s mean increase in latency there was a 0.06 increase in log threshold detection limit (poorer detection; see Fig 5).

Fig 5: Effect of mean session latency on log detection threshold.

Secondly, mean inter-box-interval was calculated for each session and trial. The inter-box-interval reflected the time the dogs took between nose pokes, reflecting time to get from one box to the next. The inter-box-interval did not vary in a statistically significant manner between the environmental conditions (X^2^ = 6.57, p = 0.16) alone. Interestingly, however, the mean session inter-box-interval was related to overall threshold detection. A mixed effect model predicting threshold by the mean session inter-box-interval, odor, and their interaction, indicated that mean inter-box-interval predicted threshold (χ^2^ = 10.89, df = 1, p = 0.001). Odor also predicted threshold as expected (X^2^ = 153.49, df = 3, p<0.0001) but the interaction between inter-box-interval and odor only approached statistical significance (X^2^ = 6.87, p = 0.07). An average increase in the inter-box-interval of 1s led to an average increase in log threshold of 0.22, indicating that slower movement from one box to the next was associated with poorer threshold (see Fig 6).

Fig 6: Effect of the interval between olfactometer boxes on log detection threshold.

Odor availability and detection limit

One of the goals of this experiment was to determine the interaction of odor availability and canine detection limits in adverse environmental conditions. If odor availability is a greater factor on detection limits than canine physiological response to environmental conditions, it would be expected that the best canine detection would be seen in higher temperatures and relative humidities.

Odor availability, as measured by total VOC accumulation on the SPME fiber, was measured in all environmental conditions except standard conditions. Each odor showed highest odor availability in the high temperature conditions (mean = 501 ppm) compared to the cold conditions (101 ppm) and all odors except smokeless powder showed highest odor availability in the high humidity conditions (mean = 334 ppm) compared to the lower humidity conditions (mean = 302 ppm).

To examine the relationship between odor availability and detection threshold, we evaluated a linear model in which detection threshold was predicted by total VOC accumulated (odor availability), with a separate model for each odor. Environmental conditions were not included due to its collinearity with VOC accumulation. Increases in VOC accumulation were associated with poorer detection thresholds for C-4 (estimate = 0.16, p = 0.04) and AN (estimate = 0.20, p<0.01) and was unrelated to detection threshold for TNT and SP. These results suggests that environmental conditions’ impact on the canine were likely more important to detection because there was either no detected relationship between VOC accumulation and detection threshold or higher VOC accumulation/odor availability (which occurred in higher temperature and humidity conditions) led to poorer detection thresholds.

Experiment 1 conclusion. Overall, the results from this study suggest that decrements in detection performance appear to be largely impacted by effects on the canine. This is exemplified by performance decrements observed with C4 in high temperature conditions, where odor availability was its greatest [9]. Decreased search speed (longer inter-box-interval and search latency) and increased mean subcutaneous temperature (for the standard and high temperature conditions) were significantly associated with poorer detection limits. These behavioral and physiological measures further support that the reduced detection limits are related to canine factors rather than odor availability.

The only significant relationship seen in the data between odorant and environmental condition was between C4 and the high temperature conditions, however the results for AN are similar, although not statistically significant.

One limitation of the current study design was the small sample size (n = 8) with few within-subject replications. It could have been possible to increase the power of this experiment if each dog had been tested in each condition more than once. However, we were concerned that this would lead to acclimatization.

It was not possible to draw conclusions about threshold detection limits and environmental conditions for smokeless powder or TNT. First the dilution range of the olfactometer used was inadequate to provide the range of dilutions necessary to reach threshold for smokeless powder (lowest threshold was still above detection limit), and the highest dilution was too near the detection limit for TNT. We suspect these floor and ceiling effects could have been reduced if an olfactometer with a larger range of dilutions had been available.

An area for future study, based on Experiment 1 would be to explore more extreme environmental conditions. In particular, the cold conditions were not as extreme as what many detection dogs may need to operate in. The low temperature conditions of 0°C was chosen based on the physical limits of the environmental chamber. Further research is necessary in subzero conditions.

The results from Experiment 1 indicate that canine detection is poorer in the high temperature conditions for C4 compared to the standard condition, However, after Experiment 1 it was not evident if this performance decrement could be reduced with specific training. Experiment 2, only studied acclimatization and C4 detection limits, because no other odor in Experiment 1 showed a significant decrement in performance compared to standard conditions. The high temperature and high humidity condition was selected as the testing and training condition because that condition resulted in significantly poorer detection compared to the standard condition.