The detection dog is the oldest instrument in the counter-IED inventory and the one people are least willing to characterize honestly. I spent my Army career as a Counter-IED and Electronic Warfare Officer, and the professional habit that job builds is asking every sensor the same rude question: not "how sensitive is it," but "under what conditions does it stop being that sensitive, and would I be able to tell?" We ask that question relentlessly of electronics. We ask it much less of the dog, partly because the dog is genuinely extraordinary and partly because the failure mode is embarrassing to talk about.
The framing you meet everywhere is some version of "a dog's nose is millions of times more sensitive than yours." Set aside whether the multiplier is meaningful. The deeper problem is that it describes the animal as a fixed specification, like a receiver with a published noise figure. A detection dog is not a fixed specification. It is a biological platform whose performance depends on its body temperature, its breathing, and its willingness to keep working, and all three are things the weather acts on directly.
The experiment that removes the excuse
Here is why this pair of papers is worth your time rather than the hundredth blog post about scent cones.
When a working dog performs badly in the heat, there is always an available explanation that lets the dog off the hook: the odor changed. Vapor pressure rises with temperature, wind redistributes the plume, humidity alters how odorants partition onto surfaces. All of that is real chemistry, and it means a field trial can never separate "the environment changed the scent" from "the environment changed the dog."
The Canine Olfaction Research and Education Lab at Texas Tech separated them with a piece of equipment. In Fernandez et al. 2024, published in PLOS ONE on 25 September 2024 and funded by the U.S. Army Research Office, eight dogs worked in a 3 m by 3 m environmental chamber that could be driven below 0 °C and above 43 °C. But the explosive itself never sat in that chamber. Each sample lived in a glass vial in a water bath held at its own fixed temperature, 30 °C to 40 °C depending on the material, and its headspace was pulled off and serially diluted by mass flow controllers before being delivered to a steel sniffing port. As the abstract puts it:
The air dilution olfactometer controlled concentrations independent of environmental condition.
So the dog experienced the weather. The odor did not. Any change in the measured threshold has to be attributed to the animal, because the only thing that varied was the animal's environment. This is the same logic as putting a receiver in a thermal chamber while feeding it a calibrated signal from outside: you are characterizing the instrument, not the propagation path.
Threshold was measured with a descending staircase, three correct responses in a row dropping the concentration by half a log step, an error raising it back, continuing until the dog produced eight reversals. Threshold is reported as the geometric mean of the last six reversals, on a log scale of proportion of vapor saturation. Lower is better. The four energetics were double-base smokeless powder, Composition C4, prill ammonium nitrate and flaked TNT. The five conditions were a 21 °C / 50% RH baseline plus 40 °C at 70% and 40% RH, and 0 °C at 90% and 50% RH.
What they found
The single cleanest sentence in the results is this one:
Importantly, performance was not better in an environmental condition in comparison to the standard condition.
Every extreme condition degraded at least one odor, and no extreme condition improved anything. Smokeless powder, the odor the dogs were best at, was degraded in all four extreme conditions, worst in heat with humidity: mean log threshold moved from −2.594 at baseline to −2.156 at 40 °C and 70% RH. That is roughly 0.44 log units, meaning the dogs needed on the order of two and a half times the concentration to hit the same threshold. Ammonium nitrate degraded significantly in both hot conditions. C4 degraded in cold with high humidity. Six of the eight dogs individually showed their largest decrement in the hot, humid condition.
TNT showed no significant effect, and the authors are refreshingly direct about why it is not good news. The dogs' TNT threshold sat near the highest concentration the olfactometer could produce even under ideal conditions, so there was no room left in the measurement to detect a decrement. That is a ceiling in the apparatus, not a resilience in the dog. It is the kind of limitation a weaker paper would have quietly reported as a null result.
Two physiological threads run alongside the thresholds. Subcutaneous temperature rose highest in heat plus humidity, significantly higher than at the same 40 °C with dry air, which is what you would expect when evaporative cooling is the primary mechanism and panting is how dogs do it. And that temperature tracked performance: for smokeless powder, a 1 °C increase in mean subcutaneous temperature was associated with a 0.19 change in log threshold. Separately, latency to approach the odor port at the start of a trial rose in the heat, and longer latency predicted worse thresholds. The dog that is slow to get to work is already telling you its nose is off.
The number that should end the conversation
Buried in the study's limitations is the most operationally significant figure in either paper. Sessions were terminated early if a dog refused to search for five consecutive trials or showed distress, on welfare grounds.
Across the whole study, 57 of 320 sessions ended early. Fifty-six of those 57 happened in the extreme conditions. Exactly one occurred at baseline.
And the dogs were not merely slower before they stopped. Mean accuracy in the run-up to early termination was 47%, against a task where guessing yields 50%. The authors state the implication plainly: early removal highlights the inability of the dogs to perform the task better than chance under those environmental conditions.
A sensor that returns worse-than-chance output for a stretch before it goes offline is a specific and dangerous failure mode. It does not read as failure. It reads as clearance. This is the same structural problem I wrote about in Report 097 and Report 103: the instrument keeps producing confident output while the thing that makes the output meaningful has quietly gone away.
The companion study closes the loophole
You could still object that the olfactometer is artificial, and the researchers agree with you. They say so directly, noting that in an operational search the odor source itself is exposed to the conditions, and that changes in odor availability might partly compensate for the decrement in the dog.
So they ran the other experiment too. Kane et al. 2024, from the same lab in the same journal, let both the dog and the odorant sit in the conditions, and measured the vapor with SPME GC-MS so they would know what was actually available. Dogs showed a 3.5-fold poorer detection limit for C4 at 40 °C and 70% RH compared with 21 °C and 50% RH. The conclusion is the sentence this whole report is built on:
Poorest performance for C4 detection occurred under highest temperature and humidity conditions in which there was highest VOC availability as measured through SPME GC-MS, indicating that performance decrement was likely due to canine factors rather than odor availability
There was more explosive vapor in the air, confirmed by instrument, and the dogs did worse. That removes the last comfortable explanation. Heat does not mainly hurt detection by hiding the scent. It hurts detection by degrading the detector.
This also resolves an oddity in the first paper. C4 performed unexpectedly well in the hot, humid condition there, and rather than leave it hanging the authors went back to the chamber logs and found the temperature and humidity had fallen below target in 7 of 16 C4 sessions after the door was opened. The companion study, with conditions properly held, found the decrement. Two papers, one anomaly, tracked down instead of narrated around.
The part that is actually good news
The second experiment in Kane et al. is the reason this report is not simply a list of things that go wrong.
The eight dogs were split into two groups of four. The control group ran the C4 threshold assessment at 21 °C and 50% RH every day for 20 days, with five minutes of petting beforehand. The acclimation group ran the same daily assessment but with temperature and humidity stepped up over six days until they reached 40 °C and 70% RH, then stayed there, and started each session with five minutes of toy or food retrieves. Both groups were tested in the hot, humid condition on day 11 and day 22.
The acclimated dogs detected better. Within about ten days, a straightforward conditioning plan measurably recovered performance that the environment had taken away.
The mechanism is the interesting part, and the authors are careful not to overstate it. There was no difference in subcutaneous temperature between the groups, so this was not the dogs learning to run cooler. What changed was behavior: the acclimated dogs searched faster, with reduced latency and reduced inter-box interval. The improvement showed up in working tempo, not thermoregulation. The authors also note honestly that they cannot separate the exercise from the heat exposure, since the acclimation group got both, and they do not know whether ten days is the minimum.
What I would not claim from this
Both papers used eight dogs. That is a small sample, and the authors name it as their first limitation. In the threshold study the dogs had all previously been eliminated from working-dog programs, for reasons including distractibility and environmental sensitivity, so they are not a random draw from the deployed population. The olfactometer's dilution range put a ceiling under TNT and limited what could be resolved for some other odors. And a Go/NoGo port in a chamber is not a search: it removes navigation, terrain, distraction and handler interaction, all of which the heat presumably also degrades.
None of that undermines the core finding, because the core finding is a direction, not a coefficient. You should not take "0.19 log units per degree" into the field as a planning factor. You should take the fact that the arrow points at the dog.
It is also worth saying what the studies do not say: they do not show detection dogs are unreliable. The dogs detected four energetic materials at low concentrations under a demanding adaptive procedure. The finding is that their sensitivity is a function of their physiological state, and that the state is observable.
What to ask
If you run, buy, or depend on canine detection, three questions follow directly from this work.
First, ask what the environmental conditions were during the search, and treat them as a recorded parameter rather than context. A negative result at 40 °C and 70% humidity is not the same evidence as a negative result at 21 °C, and the difference is now measured rather than assumed. Every other instrument in a technical inventory has a stated operating envelope. This one has one too, and the operational culture largely does not write it down.
Second, watch the tempo, not just the alerts. Both papers found that latency to start searching predicted poorer thresholds, and in the acclimation experiment search speed was what changed when performance recovered. A dog getting slower off the line is a leading indicator of a degraded sensor, available to the handler in real time and free.
Third, ask whether the team trains in the conditions it works in. This is the one place the research offers a lever rather than a caution. Ten days of graduated exposure plus brief exercise produced a measurable improvement in a controlled study. That is a cheap intervention with an evidence base, and it is more useful than any claim about how many times better than a human a dog's nose is.
The general principle is one I keep coming back to on this beat. A sensor's headline sensitivity is measured under conditions chosen to flatter it. What determines whether it works is the envelope, and the envelope is only ever found by someone willing to run the instrument until it fails and then publish the number. Two Army-funded labs did that with the oldest detector we have, and reported that on 56 occasions the detector walked off the job.
Sources
- Fernandez LS, Kane SA, DeChant MT, Prada-Tiedemann PA, Hall NJ, "Environmental effects on explosive detection threshold of domestic dogs," PLOS ONE 19(9): e0306817, published 25 September 2024, DOI 10.1371/journal.pone.0306817. Received 3 October 2023, accepted 24 June 2024. Open access under CC BY. Canine Olfaction Research and Education Lab and the Forensic Analytical Chemistry and Odor Profiling Lab, Texas Tech University. Funded by the U.S. Army Research Office (contract W911NF2120124) and an NSF Graduate Research Fellowship (DGE 2140745). (Primary source. Full 22-page PDF downloaded from PLOS and read end to end. Source of: the block quote that the olfactometer controlled concentrations independent of environmental condition; the 3 m by 3 m chamber and its below-0 °C / above-43 °C range; the per-odor water bath at 30–40 °C that decoupled the sample from the chamber; the four energetics and the five environmental conditions; the three-down-one-up staircase, eight reversals, and geometric-mean-of-last-six threshold definition; the block quote that performance was not better in any environmental condition than standard; the Table 4 means of −2.594 at baseline and −2.156 at 40 °C / 70% RH for smokeless powder and the Table 5 significance pattern for each odor; the six-of-eight individual-dog result; the TNT restriction-of-range explanation; the subcutaneous temperature results including the humidity effect at matched temperature and the 0.19 log-unit-per-°C association for smokeless powder; the latency findings; the 57-of-320 early terminations with 56 in extreme conditions and one at baseline, the 47% mean accuracy before removal, and the authors' statement about inability to perform better than chance; the C4 anomaly and the retrospective finding that conditions fell below target in 7 of 16 C4 sessions; the authors' own limitation that an operational odor source would itself be exposed to the conditions; and the eight-dog sample with prior removal from working-dog programs. Welfare oversight: Texas Tech IACUC protocol 21051-07 and U.S. Army Medical Research and Development Command Animal Care and Use Office 78018-ST-H.e001.)
- Kane SA, Fernandez LS, Huff DE, Prada-Tiedemann PA, Hall NJ, "Canine detection of explosives under adverse environmental conditions with and without acclimation training," PLOS ONE 19(2): e0297538, published 21 February 2024, DOI 10.1371/journal.pone.0297538, PMID 38381723. Received 3 October 2023, accepted 9 January 2024. Open access under CC BY. Same laboratory as the above. (Primary source. Full 22-page PDF downloaded from PLOS and read end to end. Source of: the 3.5-fold poorer C4 detection limit at 40 °C / 70% RH versus 21 °C / 50% RH; the SPME GC-MS measurement of VOC availability; the block-quoted conclusion that the decrement was likely due to canine factors rather than odor availability; the Experiment 2 design with four control and four acclimation dogs, the six-day ramp, the daily assessments, the five minutes of petting versus retrieves, and testing on days 11 and 22; the finding that acclimated dogs improved within 10 days; the absence of a subcutaneous temperature difference between groups and the attribution of improvement to reduced latency and inter-box interval; the humidity-versus-evaporative-cooling reasoning; and the stated limitations of sample size and olfactometer dilution range. Note on an internal inconsistency: the paper's General Discussion refers to the 3.5-fold result as occurring at "21 °C 70% relative humidity," a condition the study did not test. The abstract, the results and the Conclusions section all place it at 40 °C and 70% RH, and this report follows those.)
Onur Oncer
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.