environment · health · food-safety · family: it worked in the lab
precise untilit matters
Chemical sensors systematically fail outside the laboratory
Problem statement
Laboratory chemical and biological sensors can detect trace volatile compounds with high sensitivity, but making them accurate and reliable in real-world environments is still an open challenge. Sensors must be miniaturized without losing sensitivity, operate at low power for extended deployment, function in complex and changing environments, including humidity and temperature, whose negative effects on sensor responses a 2009 review of electronic-nose technologies lists among the disadvantages of e-nose sensing, and against complex chemical backgrounds with fewer false alarms, and analyze signals in real time. The solicitation asks funded teams to create standardization and benchmarking methods for calibrating devices, and to improve how well results transfer from one device to another; for air sensors, the U.S. EPA reported in 2021 that there were no standard testing protocols, metrics, or targets for evaluating performance uniformly, which makes it difficult for users to understand how well any given device performs. This gap holds back progress in environmental monitoring, food safety, homeland security, medical point-of-care diagnostics, and agricultural management.
Why this matters
Chemical threats — toxic gases, water contaminants, explosives, narcotics, food-borne pathogens — are dynamic and difficult to detect outside controlled conditions. First responders need portable toxic gas detection. Agriculture and food production need sensors that work in air, water, and soil. Clinics need point-of-care diagnostics and field-ready devices for early-stage infectious disease; a 2021 review of diagnostics for neglected tropical diseases says point-of-care tests can be administered in the field in resource-limited settings without the need for a laboratory, working around underequipped laboratories and shortages of healthcare staff. The opioid crisis calls for miniaturized opioid sensors that can also catch novel synthetic compounds: a 2022 note in the CDC's Morbidity and Mortality Weekly Report on overdose deaths in Tennessee in 2019–2021 describes nitazenes as a novel group of powerful illicit synthetic opioids, tests for which are often not included in standard toxicology panels. Climate monitoring requires large-scale sensing of methane and other trace-gas emissions. In February 2024 NSF funded 16 Phase 1 teams under Track L, with $10.4 million in total, to work on applications like these.
What’s been tried and why it hasn’t worked
The solicitation does not review earlier sensor designs one by one. The working alternative it describes is trained animals. Dogs are used routinely at airports to screen for illicit drugs and explosives, and have been used in medical diagnostics to tell patients with some cancers or COVID-19 from healthy controls. The solicitation says that using animals is not cost-effective, cannot be applied to all situations, and does not lend itself to automation or use at large scale. It names electronic nose technologies, bio-hybrid systems, high-resolution mass spectrometry, optical spectrometry, electrochemistry, machine learning, and neuromorphic computing as fields to build on, without saying where each has fallen short in the field. Other sources do: a 2009 review of electronic-nose technologies lists among the disadvantages of e-nose sensing problems with reproducibility and recovery, negative effects of humidity and temperature on sensor responses, and the inability to identify individual chemical species, and a 2026 study of lab-calibrated metal-oxide gas sensors reports that domain shifts such as sensor-to-sensor variation between production batches and aging or poisoning over time can substantially degrade model accuracy, so that recalibration typically requires time-consuming laboratory procedures. It expects funded teams to create standardization and benchmarking methods for device calibration, open datasets, and models that improve transfer between devices, which suggests these shared resources are not yet in place.
What would unlock progress
Standardized calibration and benchmarking protocols for field-deployed chemical sensors would be foundational. Beyond that: computing and AI methods that reduce false alarms in complex and changing environments; energy-efficient sensor designs for field deployment; and neuromorphic computing approaches inspired by biological olfactory processing that can handle noisy, multivariate chemical signatures.
Entry points for student teams
A student team could design and execute a benchmarking study comparing 2-3 commercially available gas sensors (e.g., MQ-series semiconductor sensors, PID detectors) under systematically varied environmental conditions (temperature, humidity, interfering gases) using a controlled chamber, producing a standardized performance dataset that currently does not exist. This is a feasible instrumentation project. Alternatively, a team could build a multi-sensor fusion prototype that combines cheap commercial sensors with environmental compensation algorithms, testing whether software calibration can improve field reliability.
Genes
Structural cousins: stuck the same way, in other fields
Sources
"NSF Convergence Accelerator Phases 1 and 2 for the 2023 Cohort - Tracks K, L, M," Program Solicitation NSF 23-590, section "Track L: Real-World Chemical Sensing Applications." (accessed 2026-02-10). Supplemented with NSF investment announcement: Wilson AD, Baietto M, "Applications and Advances in Electronic-Nose Technologies," Sensors (Basel), 9(7):5099–5148, 2009, accessed 2026-09-29; Duvall R, Clements A, Hagler G, Kamal A, Kilaru V, Goodman L, Frederick S, Johnson Barkjohn K, VonWald I, Greene D, Dye T, "Performance Testing Protocols, Metrics, and Target Values for Fine Particulate Matter Air Sensors: Use in Ambient, Outdoor, Fixed Site, Non-Regulatory Supplemental and Informational Monitoring Applications," U.S. EPA Office of Research and Development, EPA/600/R-20/280, 2021, accessed 2026-09-29; Roberts A, Korona-Bailey J, Mukhopadhyay S, "Notes from the Field: Nitazene-Related Deaths — Tennessee, 2019–2021," MMWR Morbidity and Mortality Weekly Report, 71(37):1196–1197, 2022, accessed 2026-09-29; Schauer J, Morsch J, Arendes D, Schütze A, Bur C, "Interpretable Calibration Transfer and Drift Compensation for MOS Gas Sensors in Complex Gas Mixtures," Sensors (Basel), 26(14):4595, 2026, accessed 2026-09-29; Bharadwaj M, Bengtson M, Golverdingen M, Waling L, Dekker C, "Diagnosing point-of-care diagnostics for neglected tropical diseases," PLoS Neglected Tropical Diseases, 15(6):e0009405, 2021, accessed 2026-09-29 go to source 1 ↗ go to source 2 ↗ go to source 3 ↗ go to source 4 ↗ go to source 5 ↗ go to source 6 ↗ go to source 7 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
- NSF Convergence Accelerator Track L represents a significant investment specifically in bridging the lab-to-field gap for chemical sensing.
- Cross-domain connection: this brief shares deep structure with the ocean fiber sensor field deployment problem (both involve sensors that work in labs but fail in real environments due to environmental confounds) and with the underwater IoT energy problem (both involve unattended sensor deployment challenges).
- The lack of calibration standards is a coordination failure, not just a technical one — no single group has incentive to create standards without an ecosystem to adopt them.
- The neuromorphic computing angle is interesting: biological olfactory systems solve exactly this problem (identifying chemicals in noisy, variable environments), suggesting bio-inspired architectures may be more promising than traditional signal processing.
- Related NSF programs: Convergence Accelerator Track K (equitable water solutions) also involves field-deployable sensing.
Reconciliation 2026-09-29: Fact repair after a site review read this brief against both of its sources. The solicitation (NSF 23-590) was fetched on 2026-09-29 from https://www.nsf.gov/funding/opportunities/nsf-convergence-accelerator-phases-1-2-2025-cohort/506015/nsf23-590/solicitation and the NSF announcement of February 13, 2024 was fetched the same day from https://www.nsf.gov/tip/updates/nsf-spurs-technology-development-biological-chemical-sensing-applications. Both were read in full. (1) The Source line gave the Track L heading as the title of the solicitation. The page's title is "NSF Convergence Accelerator Phases 1 and 2 for the 2023 Cohort - Tracks K, L, M," and Track L is one section of it. The Source line now says that. (2) The Problem Statement said that laboratory sensors "systematically fail" in the field, that sensors must cope with "temperature, humidity, wind" and "fuse data from multiple modalities," and that there are "no established calibration standards or benchmarks." The solicitation does not say these things. It says chemical signatures are hard to detect because they are dynamic, at very low concentrations, and embedded in complex chemical backgrounds, and it asks funded teams to reduce false alarms and to create standardization and benchmarking methods for device calibration. The paragraph now follows the solicitation. (3) In Why This Matters, four details were not in either source: "agricultural workers," clinics "in resource-limited settings," "novel synthetic compounds," and greenhouse gas sensors "across vast geographies." The sentences now follow the list of funded projects in the announcement. The closing sentence, which said current technology fails "because the lab-to-field transition remains unsolved," was replaced with what the announcement reports: 16 Phase 1 teams and $10.4 million. (4) What's Been Tried made six technical claims that are in neither source: that electronic noses are confounded in the field, that bio-hybrid sensors are fragile, that miniaturized mass spectrometers are too expensive and power-hungry, that machine learning classifiers fail when temperature and humidity shift, that each research group uses its own calibration protocol, and that data interoperability is "essentially nonexistent." These may be true, but the cited sources do not say them, so the section now reports what the solicitation does say, including its account of why trained animals cannot do this work at scale. (5) What Would Unlock Progress listed sensor fusion across modalities, energy harvesting, and materials that stay selective across temperature and humidity. None of these is in the sources. The first two were reworded to match the solicitation and the third was removed. (6) The entry points called MQ-series sensors "electrochemical." The maker's page for the MQ-2 (Winsen, https://www.winsen-sensor.com/sensors/combustible-sensor/mq2.html, fetched 2026-09-29) lists the sensor type as semiconductor, with tin dioxide as the sensitive material. Winsen's list of its semiconductor gas sensors (https://www.winsen-sensor.com/semiconductor-gas-sensor/, fetched 2026-09-29) includes its other MQ models, and its list of electrochemical sensors includes none. Confirmed against the sources and left as written: toxic gas detection for first responders, the opioid crisis, point-of-care diagnostics, narcotics, explosives and food-borne pathogens as applications; the five application areas in the Problem Statement; neuromorphic computing and biological olfaction as approaches the track builds on; Track K as "Equitable Water Solutions" in the same solicitation. NOT changed, and worth knowing: the title's words "systematically fail" are this write-up's own summary and do not appear in either source. The entry-point claim that a standardized performance dataset "currently does not exist" was not checked, because no single fetch can show that something does not exist. Note on the working notes above: the Track K section of the solicitation does not mention sensing, so "also involves field-deployable sensing" is not supported by it. The notes on coordination failure and on cross-domain links are the collection's own commentary, not statements from NSF. Genes untouched.
Reconciliation 2026-09-29 (re-sourcing): The editor asked for the claims removed in round 5 to be re-sourced, so that anything with a primary source behind it could go back. Each source below was fetched and read on 2026-09-29. Restored: (1) The Problem Statement now says that the changing environments sensors must handle include humidity and temperature, whose negative effects on sensor responses the review lists as a disadvantage. This rests on Wilson and Baietto's 2009 review in Sensors, which says that "some disadvantages of e-nose sensing include problems with reproducibility, recovery, negative effects of humidity and temperature on the sensor responses, and inability to identify individual chemical species," read at https://www.ebi.ac.uk/europepmc/webservices/rest/PMC3274163/fullTextXML. The same statement is now quoted in What's Been Tried, where it is the only claim about where electronic noses fall short; it is about e-noses, not all chemical sensors. (2) The Problem Statement also says that, for air sensors, the U.S. EPA reported in 2021 that there were no standard testing protocols, metrics, or targets for evaluating performance uniformly, which makes it difficult for users to understand how well any given device performs. This rests on the EPA Science Inventory record for Duvall et al., EPA/600/R-20/280, which says "There are currently no standard testing protocols, metrics, or targets to evaluate the performance of air sensors uniformly. Without these standard procedures, it is difficult for consumers to understand the performance of any given device," read at https://cfpub.epa.gov/si/si_public_record_Report.cfm?Lab=CEMM&dirEntryId=350785. The original text said there were no standards for field-deployed chemical sensors in general and that results could not be compared; the source is about air sensors in the United States as of 2021, does not use the words "low-cost" or "compare," and that report itself proposed protocols and targets for PM2.5 sensors, so the sentence now carries that scope. (3) Why This Matters now says point-of-care tests can be administered in the field in resource-limited settings without a laboratory. This rests on Bharadwaj et al., PLoS Neglected Tropical Diseases 2021, which says "PoC diagnostics can circumvent issues such as shortage of healthcare staff and underequipped laboratories ... PoC tests can be field deployable, i.e., they have the ability to be administered in the field within resource-limited settings without the need for a laboratory," read at https://www.ebi.ac.uk/europepmc/webservices/rest/PMC8211285/fullTextXML. The original said "medical clinics in resource-limited settings"; the source is a review of diagnostics for neglected tropical diseases, and the sentence now says so. (4) The opioid sentence now says sensors must also catch novel synthetic compounds, resting on a note by the Tennessee Department of Health in the CDC's MMWR (Roberts et al., 2022) which says "Nitazenes are a novel group of powerful illicit synthetic opioids" and that they are "an emerging group of highly potent psychoactive substances, tests for which are often not included in standard toxicology panels," read at https://www.ebi.ac.uk/europepmc/webservices/rest/PMC9484803/fullTextXML. The source covers overdose deaths in Tennessee in 2019–2021, and the sentence names that scope. (5) What's Been Tried now says that models calibrated in the lab lose accuracy under domain shifts. The original said machine learning classifiers fail when environmental conditions shift; the source found, Schauer et al., Sensors 2026, says the shifts are "sensor-to-sensor variation among different production batches" and "time-related changes in sensor response like poisoning and aging," that "such domain shifts can lead to a substantial performance degradation and are critical for reliable field deployment," and that "recalibration typically requires time-consuming and labor-intensive laboratory calibration procedures," read at https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi?db=pubmed&id=42515477&retmode=xml. The sentence follows the source's list of shifts, not the original's. Searched for and left out: that sensors "systematically fail" in the field and that wind is one of the conditions they must handle (no source found saying either); that bio-hybrid sensors are fragile and hard to maintain outside labs (Europe PMC searches for bioelectronic nose and olfactory biosensor reviews on stability found nothing that says this); that miniaturized mass spectrometers are too expensive and power-hungry for distributed deployment (the open-access papers found describe individual instruments and do not make that claim; the reviews that might are behind publisher paywalls and were not read); that each research group uses its own calibration protocol and that data interoperability is "essentially nonexistent" (no source found); that sensor fusion across modalities, energy harvesting, and materials that stay selective across temperature and humidity would unlock progress (no source found); that agricultural workers need soil and air sensors and that climate monitoring needs sensors "across vast geographies" (no source found); and the closing sentence that none of these applications is well served because the lab-to-field transition is unsolved (a causal claim no source was found to make). Genes untouched.