Reading a sensor datasheet

Summary

A datasheet number is a fact about the sensor measured one particular way: at one temperature and humidity, over one averaging time, under one (usually unstated) statistical definition. Any of those hidden conditions can move the printed number by a factor of three to ten without the hardware changing. Lining up two vendors’ sheets and buying the better number is how gas-sensing procurement goes wrong.

This page walks line by line through three real, currently published datasheets (Alphasense and Membrapor electrochemical cells, a Figaro metal-oxide sensor) and translates each spec into what it measured and what that means for a Hestia field program. Four takeaways recur: “limit of detection” can mean three incompatible things; “typical” and “warranted” are different promises; “nd” (not determined) is not zero; and CO cross-sensitivity varies by more than 30× between two products both labeled “electrochemical H2 sensor.” Read this before opening another datasheet.

Context numbers used throughout

Atmospheric H₂ background ≈ 0.53 ppm (marine boundary layer mean 552.8 ppb in 2021, NOAA Global Monitoring Laboratory network) [1]; atmospheric He background ≈ 5.24 ppm [2]; geologic H₂ soil-gas exploration anomalies span roughly 1 ppm to percent level; co-evolved CH₄/CO₂/H₂S/He/N₂ are the standard interferent suite (see Helium & co-tracers); downhole environments add high temperature/pressure; H₂ LEL = 4 vol% (40,000 ppm).

What a spec line actually claims

A sensor’s raw output is a noisy, drifting, environmentally modulated function of the true concentration:

$$R(t) \approx R_0(t) + S(t)\,C_{H_2}(t) + \sum_i \kappa_i(t)\,C_i(t) + n(t)$$

where $R_0$ is the (drifting) baseline, $S$ the (drifting) sensitivity, $\kappa_i$ the cross-sensitivity coefficients to interferents, and $n$ stochastic noise. Every datasheet line is a claim about one term of this equation under one set of conditions:

• “Range”: the domain over which $S$ is claimed linear
• “Resolution”: a property of $n(t)$ and the readout electronics
• “Zero drift” and “sensitivity drift”: $dR_0/dt$ and $dS/dt$
• “Cross-sensitivity”: the $\kappa_i$
• “Operating range”: where the vendor claims the whole equation holds at stated accuracy

For every line, ask: which term is this, and at what $(T, RH, P, t_{avg})$ was it measured? The field behavior of these terms is treated on Environmental covariance; this page is about how vendors report them.

Limit of detection: three definitions behind one phrase

Datasheets almost never say which convention they use.

Convention	Formula	Relative size	Who uses it
IUPAC 3σ (blank-based)	$\mathrm{LOD}=3\sigma_{blank}/S$	Baseline (1×)	Analytical chemistry, rigorous sensor papers [3][4][5]
SNR = 1	$\mathrm{LOD}=\sigma_{blank}/S$	~3× smaller (more optimistic) for identical hardware	Marketing “detectable level” claims; ambiguous datasheets
Allan-deviation floor at optimum $\tau^*$	$\min_\tau \sigma_A(\tau)/S$	Depends on $\tau^*$; can beat or lose to a 1 s 3σ figure	TDLAS/laser spectroscopy, some GC/optical work [6][7]
“Resolution” (readout increment)	ADC/display least-significant digit	Unrelated to the noise floor	Nearly every datasheet, frequently conflated with LOD

The IUPAC definition sets the threshold at $\bar{x}_{bl} + 3\,s_{bl}$ of repeated blank measurements [3][4]; at $k=3$ the false-positive rate is ≈0.135% under Gaussian noise. Alphasense’s technical director, in a 2024 peer-reviewed methodology paper, measured cell noise equivalent to 1.6 ppb in a controlled-temperature, controlled-humidity chamber, giving LOD = 4.8 ppb [5]; the chamber conditions account for most of why that number is small.

Noise also depends on averaging time. White noise averages down as $\sqrt{\tau_0/\tau}$, but at long $\tau$ drift dominates and precision worsens again. The Allan variance, adapted from clock metrology to trace-gas work by Werle, Mücke and Slemr [6],

$$\sigma_A^2(\tau) = \frac{1}{2(M-1)}\sum_{i=1}^{M-1}\left(\bar{y}_{i+1}(\tau) - \bar{y}_i(\tau)\right)^2$$

locates the true noise floor and optimal integration time $\tau^*$ [7]. This is the convention that answers the field question: what is my real floor at a 30 s or 5 min station dwell? A datasheet LOD with no stated averaging time is not comparable to a field requirement.

Resolution is a different quantity: the smallest increment the readout registers, set by the ADC/display. A field evaluation of low-cost CO₂/CH₄ sensors measured precision (3.54 ppm) far better than the quoted resolution (50 ppm) for the same instrument [8][9]. The reverse case is more dangerous: a display resolution finer than the true LOD gives false confidence that a near-background reading is meaningful.

Response time: t50, t90, t10–90

Response and recovery are reported as $t_x$, time to $x\%$ of final value after a step [10][11]. For a single first-order system with time constant $\tau$ (a fair model for diffusion-limited electrochemical cells, a poor one for MOX/catalytic surface kinetics):

$$C(t) = C_{\text{final}}\left(1 - e^{-t/\tau}\right) \implies t_x = -\tau\ln\!\left(1 - \frac{x}{100}\right)$$

so $t_{50} \approx 0.693\,\tau$, $t_{90} \approx 2.303\,\tau$, $t_{10\text{-}90} \approx 2.197\,\tau$. IEC 60079-29-1 uses $t_{50}$/$t_{90}$ as certification pass/fail parameters [10][11][12]. Two uses: if a sheet quotes both $t_{50}$ and $t_{90}$ and their ratio is far from $\ln 2/\ln 10 \approx 0.30$, the kinetics are not single-pole and the raw response curve should be requested; and recovery is desorption-governed and slower, seriously enough for MOX that dedicated correction models exist [13].

Accuracy, precision, repeatability, resolution

Four words, one bias–variance decomposition:

$$\hat{x} = x_{\text{true}} + \underbrace{b(T,RH,P,t)}_{\text{bias} \Rightarrow \text{accuracy}} + \underbrace{\varepsilon,\ \varepsilon\sim\mathcal{N}(0,\sigma_{\text{rep}}^2)}_{\text{random error} \Rightarrow \text{precision/repeatability}}$$

Accuracy is the systematic bias $b$ against a certified reference. Precision is the spread of repeated measurements, with no claim about centering; an instrument can be precise and badly biased [14][15]. Repeatability is precision under fixed conditions; reproducibility is the spread across units/operators and is almost never on a component datasheet. Resolution is the readout increment. A sheet quoting only “resolution: 0.1 ppm” says nothing about accuracy or repeatability; “±2% repeatability” says nothing about bias without a traceability statement. See Calibration, drift & QA/QC for the uncertainty-propagation treatment.

Drift: two processes, not one

Zero drift and span drift have independent physical origins and must be reported separately:

$$R_0(t) = R_0(0) + d_Z\,t + w_Z(t), \qquad S(t) = S(0)\left(1 + d_S\,t\right) + w_S(t)$$

A vendor reporting only one of $d_Z$ (ppm-equivalent/yr) or $d_S$ (%/yr) leaves half the drift budget unconstrained, a common omission (see the walkthroughs below). Both run per unit time, not per unit use: the clock runs while the sensor sits in a warehouse.

Cross-sensitivity and compensation

Vendor tables report cross-sensitivity as a percentage of equivalent target-gas reading at one interferent concentration: “CO <2% @ 400 ppm CO” is a single-point snapshot of $\kappa_{CO}$, not a guarantee at other concentrations, and simultaneous exposure to two interferents is essentially never characterized. The combustible-gas world’s response (correction) factors [16][17] share the same limitation: calibration-point snapshots read as universal constants.

“Compensated” output means $C_{\text{reported}} = C_{\text{raw}} \cdot f_T(T)\cdot f_{RH}(RH)\cdot f_P(P)$ with corrections fitted over a stated compensated range, usually narrower than the operating range the bare sensor survives [18]. Outside it, the accuracy spec is void while the device keeps producing numbers, with no signal in the output that the correction stopped applying.

TRL (the nine-level NASA maturity scale [19][20]) is the datasheet-adjacent metric vendors never print: a device can have an impressive paper LOD and still be TRL 4–5 with no field reliability record. Infer it from product age, independent evaluations, and warranty language.

Industry practice is converging slowly: Saffell & Martin 2024 is a peer-reviewed acknowledgment from inside the industry that LOD has lacked a reproducible methodology, with a proposed 3σ chamber protocol [5]; the AQ-SPEC and JRC programs exist because vendor self-reported specs were judged insufficient for procurement decisions [27][28]. Allan characterization remains rare outside academic literature [6][7]. Until that changes, integration-time-aware validation is work the buyer does [21][22].

Datasheet genres and spec provenance

Document type	What it is	Trust level	Example
One/two-page product spec sheet	Condensed, single-condition summary aimed at procurement	Medium: numbers are real but conditions are compressed or omitted	Alphasense H2-AF Technical Specifications v1.0 [21]; Membrapor H2/M-1000 Specification Sheet [22]
Multi-page engineering guide	Operation principle, characteristic curves, cautions, shared across a product family	High for physics/caveats; curves are explicitly “typical,” not warranted	Figaro “General Information for TGS Sensors” [23]; Figaro TGS2600 Product Information [24]
Certification / standard test report	Pass/fail against a named standard with defined gases and procedures	High but scope-limited to what the standard tests, usually safety/alarm performance	ISO 26142:2010 hydrogen detection apparatus standard [25][26]
Independent / peer-reviewed evaluation	Third-party measurement of vendor claims	Highest, but rare and product/batch specific	Saffell & Martin 2024 [5]; low-cost CO₂/CH₄ field evaluation [8][9]; AQ-SPEC [27]; Palmisano et al. [28]
Vendor application note	Narrow single-topic guidance, written to sell the platform	Medium	Honeywell LEL correction-factor note [16]; SGX pellistor-poisoning note [29]

Spec provenance, in decreasing evidentiary weight:

• Independently reproduced. A third party measured the same spec [5][8][9][27][28]. Rare.
• Warranted. A number the vendor stands behind, usually narrower than the surrounding figures (Alphasense’s range: “2,000 ppm H₂, limit of performance warranty”) [21].
• Typical. Central tendency, explicitly not guaranteed. Figaro states it outright: “The only characteristics warranted are those in the Specification table” [24].
• “nd” / “N.D.” (not determined). Appears on the Alphasense sheet for sensitivity drift and H₂S filter capacity, and on the Membrapor sheet for pressure coefficient [21][22]. Not zero: unmeasured. Filling in “presumably small” is an assumption the vendor declined to make.

Datasheet walkthroughs: three real spec sheets, line by line

Alphasense H2-AF (electrochemical)

Fetched directly from the current Alphasense technical specifications document [21]:

Datasheet line (as printed)	What it literally measures	What it means for Hestia field use
Sensitivity: 10–40 nA/ppm in 400 ppm H₂ at 23°C	Slope $S$ at one calibration point, one temperature, ±4× unit-to-unit spread	Every individual sensor must be calibrated on receipt; the spread is the dominant early-life uncertainty source
Response time: t90 < 35 s in one rendered table, < 45 s in a second, differently formatted table later in the same PDF	Time to 90% of final value, zero→400 ppm	The single fetched file contains two internally inconsistent spec tables (also disagreeing on sensitivity: 10–40 vs. 10–25 nA/ppm), carrying two different document references (“H2-AF/FEB23” and “H2-AF/JUL16”): an update applied to one table but not fully propagated. Cross-check every number against the rendered page a human would read
Zero current: −25 to +15 ppm equivalent in the first table vs. ±15 ppm in the second	Baseline offset in clean zero air	A multi-tens-of-ppm zero uncertainty sits on top of a 1 ppm-scale exploration anomaly; establish zero on-site immediately before use
Resolution: <0.7 ppm RMS noise	Short-term (unstated $\tau$) noise floor	No averaging time given; treat as best-case short-$\tau$ and re-derive an Allan-type floor for your actual dwell time
Range: 2,000 ppm H₂ (limit of performance warranty)	The only number in the table explicitly warranted	Everything else on the sheet is a “typical” claim; the warranty boundary is what the vendor stands behind
Linearity: −200 to −500 ppm error at full scale	Absolute error band at top of range	Up to 25% error at 2,000 ppm; irrelevant near background, material for percent-level readings
Zero drift: <20 ppm/yr in lab air	$d_Z$ only	”In lab air”: dry, clean, temperature-stable. Field drift with real humidity/temperature swings is not this number
Sensitivity drift: nd	$d_S$ explicitly not determined	Half the drift budget is unconstrained on this sheet
Operating life: >24 months to 80% signal (24-month warranted)	Typical time to 20% sensitivity loss	Budget a hard replacement cadence; do not assume linear decline
Sensitivity @ −20°C: 10–25% of 20°C output; @ 50°C: 220–275% of 20°C output	Multiplicative $f_T$, uncompensated	A >10× sensitivity swing across a realistic field temperature range; the sheet itself says correction requires ±0.5°C co-located temperature measurement [21]
Cross-sensitivity: CO <2%, NO<40%, C₂H₄<25%, others <1–4%	$\kappa_i$ snapshots at one concentration each	Low CO cross-sensitivity is this part’s main selling point; NO and C₂H₄ are far from negligible if present
T/P/RH range: −30 to 50°C / 80–120 kPa / 15–90% RH	Operating envelope, not necessarily the compensated envelope	Downhole and desert-diurnal extremes can exceed the operating envelope outright

Membrapor H2/M-1000 (electrochemical)

Fetched from the current Membrapor specification sheet, revision Dec-25 [22]:

Datasheet line	What it means	Hestia implication
Output signal: 25 ± 10 nA/ppm (footnote: “H₂ sensors lose >10% of their initial sensitivity within the first month after production before the output stabilizes”)	Burn-in/aging disclosed as a footnote	New sensors need a documented settling period and a fresh calibration after it, not at receipt
Resolution: <2.0 ppm (“Electronics dependent”)	Depends on the reader electronics, not the sensor alone	The same cell has a different “resolution” per transimpedance amplifier/ADC; comparing “resolution” across products without the readout chain compares incomparables
Typical baseline range (pure air, 20°C): −10.0 to +10.0 ppm	±10 ppm zero band at one temperature	Comparable in size to a modest geologic anomaly; never trust a single spot reading near background without averaging
Max zero shift (+20 to +40°C): 18 ppm	$\Delta R_0$ over a 20°C swing	Zero is temperature-sensitive even in a part that doesn’t headline it
Repeatability: <2% of signal	Precision, not accuracy	Says nothing about bias; pair with an independent calibration standard
Pressure coefficient: N.D.	Never tested	Do not assume pressure insensitivity; pumped soil-gas sampling creates transient pressure excursions (see Soil-gas surveys)
Humidity effect: “None” (stated flatly)	A strong claim with no supporting data, from a technology whose physics is humidity-sensitive	Treat vendor “None” claims for a physically plausible cross-sensitivity with skepticism; see Electrochemical cells for the electrolyte-drying mechanism this glosses over
Expected long-term output drift: <15% signal loss/year	$d_S$, actually reported (contrast Alphasense’s “nd”)	Vendors in the same category disclose different halves of the drift budget; read both lines before concluding one product “drifts less”
Cross-sensitivity: CO ~60–80%	$\kappa_{CO}$ an order of magnitude larger than Alphasense’s <2%	Not interchangeable products: a CO-rich environment badly corrupts this part’s H₂ reading while barely touching the Alphasense part
Boilerplate: “The data contained in this document is for guidance only… Customers should test under their own conditions”	Standard liability disclaimer	Both vendors carry near-identical language [21][22]; industry-standard risk allocation that places field validation on the buyer

Figaro TGS2600 (metal-oxide)

Fetched from Figaro’s product information sheet and general engineering guide [23][24]:

Datasheet line	What it means	Hestia implication
Target gases: “Air contaminants (hydrogen, ethanol, etc.)”	H₂ is one of several gases this element responds to	Non-selective by design; a reading needs a co-located selective sensor to disambiguate, or acceptance as a general “reducing gas” indicator
Typical detection range: 1–30 ppm H₂	A narrow low-concentration window	Below atmospheric background at the bottom, far below percent-level readings at the top; a screening tool only
Sensor resistance (Rs) in air: 10–90 kΩ	9× unit-to-unit baseline spread	Every unit needs individual $R_0$ characterization; a shared batch calibration curve is not valid
Sensitivity (change ratio): 0.3–0.6 × Rs(10 ppm H₂)/Rs(air)	2× unit-to-unit sensitivity spread	Compounds the calibration burden above
Conditioning period before test: 7 days	Factory burn-in before the quoted numbers were measured	A freshly powered field unit will not match the datasheet curve
”Initial action”: Rs drops sharply for the first few seconds after every power-on, regardless of gas [23]	A cold-start artifact that mimics a gas signal	Filter with a fixed post-power-on dead time, or every session risks a spurious high reading at the first station
”TGS sensors require the presence of around 21% (ambient) oxygen… and cannot properly operate in a zero or low oxygen content atmosphere” [23]	The mechanism is O₂-dependent	Rules out headspace or downhole gas where O₂ is depleted: the reducing, anaerobic conditions often co-located with active seeps
Caution: “Exposure to silicone vapors… will coat the sensing material, irreversibly inhibiting sensitivity” [23]	A named, common-substance poisoning mechanism	Field kit hygiene matters: sunscreen, hand lotion, silicone-sealed enclosures near the inlet are realistic vectors
Long-term stability (~500 days): “suitable for maintenance-free operation” [23]	Vendor interpretation of one internal dataset	Marketing narrative on a chart, not a warranted drift rate like Alphasense’s “<20 ppm/yr”

Cross-cutting comparison

Metric	Alphasense H2-AF (EC)	Membrapor H2/M-1000 (EC)	Figaro TGS2600 (MOX)	Ref
Range	0–2,000 ppm (warranted)	0–1,000 ppm (2,000 overload)	1–30 ppm	[21][22][24]
Response t90	<35 s or <45 s (internally inconsistent across two tables, same PDF)	<35 s	seconds-scale, device-dependent	[21][22][23]
CO cross-sensitivity	<2%	~60–80%	Not separated; CO is a co-target gas by design	[21][22][23]
Zero drift (stated)	<20 ppm/yr (lab air)	included in <15%/yr overall figure	Not separately quantified	[21][22]
Sensitivity drift (stated)	“nd”	<15% signal loss/yr	Not separately quantified	[21][22]
Temperature swing effect	10–25% output at −20°C; 220–275% at 50°C	Zero shifts 18 ppm over 20→40°C	Rs/Ro shown graphically, not tabulated	[21][22][23]
Warranty / operating life	24 months to 80% signal	3-year warranty; 3–5 yr expected life	Not stated on 2-page sheet	[21][22][24]
Ambient O₂ dependence	Tolerant of moderate excursions	Tolerant of moderate excursions	Requires ~21% ambient O₂ to function at all [23]	[23]

Spec-sheet artifacts vs. real covariance

Companion to the quantitative treatment in Environmental covariance.

Lab-to-field gap. A field evaluation of low-cost CO₂/CH₄ sensors found seasonal $R^2$ for the same sensor swinging from 0.65 in one validation window to 0.12 in another, driven by environmental factors absent from manufacturer characterization [30]. AQ-SPEC exists because vendor numbers were found insufficiently predictive for procurement [27]. Humidity is repeatedly singled out as the most underestimated variable [31].

MOX humidity/temperature dependence is visible on Figaro’s own curves: Rs/Ro spreads measurably across 35/65/95% RH at fixed 1500 ppm exposure, and across −10 to 40°C [23]. A MOX reading without simultaneous RH/T logging is qualitative only.

MOX oxygen dependence is built into the ionosorption mechanism (MOX sensors): roughly 21% ambient O₂ required [23]. An O₂-starved MOX sensor does not fail loudly; it produces an uncharacterized resistance easily misread as a large gas response.

Manufacturing variance acts like a cross-sensitivity: the TGS2600’s 9× baseline and 2× sensitivity spreads [24] mean “the response to X ppm” is not a single number before any environmental variable moves.

Electrochemical CO cross-sensitivity varies by design choice invisible from the category name: <2% (Alphasense, filtered) vs. ~60–80% (Membrapor, unfiltered) [21][22]. A 2015 JRC study using ISO 26142’s own poison protocol found most catalytic, MOX, and electrochemical H₂ sensors resisted permanent change from SO₂, H₂S, NO₂, and siloxanes, with the electrochemical platform showing a persistent sensitivity change after NO₂, and flagged the standard’s 10 ppm HMDS siloxane test as possibly unrepresentative [28].

Poisoning is a named, common-substance risk: Figaro lists hair products and silicone rubber/putty for MOX [23]; SGX identifies silicone, lead, and sulfur/phosphorus compounds as causes of immediate, sometimes irreversible pellistor failure, detectable only by bump test or calibration [29][32]. Near drilling mud and well-site chemicals this is operational hygiene, not a theoretical caveat. Arrays of sensors with complementary failure modes are one mitigation; see Sensor arrays & machine learning.

“nd” is not zero

Both electrochemical sheets contain fields marked “nd”/“N.D.” (Alphasense: sensitivity drift, H₂S filter capacity; Membrapor: pressure coefficient) [21][22]. A comparison spreadsheet that treats a blank cell as zero manufactures a false advantage for whichever vendor didn’t test that parameter.

Use for geologic H₂

Field summary

No datasheet LOD, drift rate, or cross-sensitivity number is usable for exploration planning without checking three things: the averaging time behind the LOD, whether the number is typical or warranted, and whether the CO/interferent cross-sensitivity fits your site’s gas sources. The two electrochemical sensors here differ by >30× in CO cross-sensitivity under the same category label, and the MOX sensor’s characterized behavior is void in the O₂-depleted conditions common near seeps. Both facts are visible only in a line-by-line read. Treat every vendor number as provisional until verified against the unit and firmware revision in hand.

Re-derive LOD at field dwell time. Every LOD above was measured at an unstated or short averaging time in clean lab air. A protocol spending 30–90 s per station needs the Allan-type floor at that $\tau$; the difference decides whether 1–5 ppm anomalies are resolvable against the 0.53 ppm background. See Environmental covariance.

Match CO cross-sensitivity to site gas sources. A survey near combustion, vehicle exhaust, or mud-gas equipment with a ~60–80% part instead of a <2% part reports CO artifacts as H₂ anomalies, indistinguishable from the H₂ channel alone. This page-4 line item outranks the range and resolution numbers procurement usually compares first. Isotopic fingerprinting is one independent confirmation path (see Isotopes & fingerprinting).

MOX is out for O₂-depleted samples. Buried, anaerobic, or bagged headspace gas sits outside a MOX chemiresistor’s characterized domain [23]. MOX remains defensible as cheap pre-screening in oxygenated ambient air (as scoped in Choosing a sensor), but subsurface sampling without an independent O₂ channel cannot catch this failure mode. The same discipline applies to dissolved-gas instruments; see Dissolved gas in groundwater.

Compensated range, not operating range, is the accuracy boundary. The H2-AF’s raw sensitivity changes >10× between −20 and 50°C, both inside its stated operating range [21]; accuracy then depends on the deployment’s temperature compensation tracking the vendor curve, which the vendor says needs ±0.5°C co-located measurement [21]. Downhole and desert-diurnal deployments (Downhole deployment) are the likeliest to exceed the compensated range silently.

The buyer inherits validation by design. Both sheets carry “for guidance only, test under your own conditions” language [21][22], and one currently published sheet contains two internally inconsistent tables under two document references for the same SKU [21]. Confirm which table is current before trusting any number from it.

TRL is high only for the safety use case. Electrochemical and MOX H₂ sensing is commodity TRL-9 technology for handheld safety monitoring, certified to IEC 60079-29-1 / ISO 26142. That maturity does not transfer to buried multi-year deployment or downhole conditions; see Downhole deployment and Soil-gas surveys for the thin evidence base there, and Standards & regulations for the certification landscape.

Checklist: ten questions to ask before buying an H2 sensor

1. What statistical convention (3σ, SNR=1, Allan floor, or undefined) defines the quoted LOD, and at what averaging time?
2. Is “resolution” the LOD or just the ADC/display increment, and which does the vendor mean by “detectable”?
3. Which numbers are typical and which warranted, and is the warranted range the one that matters for my application?
4. What is the CO (and CH₄, H₂S, humidity) cross-sensitivity, in percent-of-reading at a site-relevant concentration, not as prose?
5. Are zero drift AND span drift both reported? If one says “nd,” what does that imply for my calibration interval?
6. What is the compensated T/RH/P range, as distinct from the operating (survival) range, and does my deployment exceed it?
7. What warm-up/cold-start time applies after realistic storage (weeks unpowered in a field vehicle)?
8. What poisons or exposures void the warranty, and are any plausible at my site (H₂S, silicone, corrosives, condensation, freeze-thaw)?
9. Is there any independent verification of this part, or is the datasheet the only source of truth?
10. What are the calibration interval and operating life, and is total cost of ownership budgeted (replacement cells, cal gas, labor)? (See Calibration, drift & QA/QC.)

Vendors, labs & standards

Entity	Role	Relevant document(s)	Ref
Alphasense (AMETEK)	Electrochemical H₂ cell vendor	H2-AF Technical Specifications v1.0	[21]
Membrapor AG	Electrochemical H₂ cell vendor	H2/M-1000 Specification Sheet, Rev Dec-25	[22]
Figaro Engineering / Figaro USA	MOX chemiresistor vendor	TGS2600-B00 Product Information (Rev 09/13); “General Information for TGS Sensors” (Rev 03/05)	[23][24]
Honeywell	Gas/flame detection vendor and standards-adjacent technical publisher	”Correction Factors for Combustible Gas (LEL) Sensors” technical note	[16]
SGX Sensortech	Pellistor vendor	Application Note AN6: Pellistor Poisoning	[29]
IEC	International standard	IEC 60079-29-1, flammable-gas detector performance requirements (t50/t90 pass/fail criteria)	[10][11][12]
ISO	International standard	ISO 26142:2010, hydrogen detection apparatus (stationary applications)	[25][26]
IUPAC	Chemistry nomenclature body	Gold Book definition of limit of detection	[3][4]
ASTM International	Standards/testing publisher	Journal of Testing and Evaluation: Saffell & Martin 2024 LOD methodology paper	[5]
South Coast AQMD	Independent field/lab sensor evaluation program	AQ-SPEC field and laboratory performance evaluations	[27]
European Commission Joint Research Centre (JRC)	Independent poison-resistance/selectivity evaluation	Palmisano et al. 2015, ISO 26142-protocol poison study	[28]
NASA	Technology-maturity framework originator	Technology Readiness Level scale	[19][20]

Certification text and scope for ISO 26142, IEC 60079-29-1, and UL 2075 are on Standards & regulations; product entries belong on Vendors.

Open questions

• No published Allan-deviation characterization was found for any commercial H₂-specific electrochemical or MOX sensor at field-relevant dwell times (30 s, 90 s, 5 min). A tractable internal project on Hestia’s incoming sensor stock.
• Ask Alphasense which of the two inconsistent H2-AF tables is authoritative [21], and audit other sheets in procurement history for similar revision artifacts.
• No independent apples-to-apples cross-sensitivity bench test across candidate H₂ cells exists; the CO comparison above is vendor self-report on both sides.
• The ISO 26142 poison-resistance literature is a decade old [28]; check whether current parts have been re-tested and whether the flagged siloxane-concentration concern was addressed in a revision.
• MOX O₂ dependence is qualitative in the sources (“requires ~21% ambient oxygen” [23]); a dose-response curve would set a hard go/no-go O₂ threshold for headspace protocols.
• Revisit this page when Environmental covariance, Calibration, drift & QA/QC, Electrochemical cells, or MOX sensors change, since several numbers here are drawn from those dossiers.

Related pages

• Why H₂ is a hard sensing target
• Choosing a sensor
• Comparison charts
• Environmental covariance
• Calibration, drift & QA/QC
• Soil-gas surveys
• Downhole deployment
• Geologic context
• Helium & co-tracers
• Isotopes & fingerprinting
• Dissolved gas in groundwater
• Electrochemical cells
• MOX (metal-oxide) chemiresistors
• Catalytic pellistors
• Sensor arrays & machine learning
• Standards & regulations
• Vendors
• Glossary

References

Full reference list (32 entries)

[1] Petron, G. et al. “Atmospheric H2 observations from the NOAA Cooperative Global Air Sampling Network.” Atmospheric Measurement Techniques, 17, 4803– (2024). https://amt.copernicus.org/articles/17/4803/2024/

[2] Oliver, B. M. et al. “Helium concentration in the Earth’s lower atmosphere.” Geochimica et Cosmochimica Acta, 48(9), 1759–1767 (1984). https://www.sciencedirect.com/science/article/abs/pii/0016703784900309

[3] “Limit of Detection (LOD) Calculations: 3σ Method vs IUPAC Guidelines.” PatSnap Eureka. https://eureka.patsnap.com/article/limit-of-detection-lod-calculations-3-method-vs-iupac-guidelines

[4] IUPAC. “limit of detection” (L03540). IUPAC Compendium of Chemical Terminology (Gold Book). https://goldbook.iupac.org/terms/view/L03540

[5] Saffell, J. R.; Martin, N. A. “Measurements of the Limit of Detection for Electrochemical Gas Sensors.” Journal of Testing and Evaluation, 52(5), 2675–2684 (2024). https://doi.org/10.1520/JTE20230675

[6] Werle, P.; Mücke, R.; Slemr, F. “The limits of signal averaging in atmospheric trace-gas monitoring by tunable diode-laser absorption spectroscopy (TDLAS).” Applied Physics B, 57, 131–139 (1993). https://link.springer.com/article/10.1007/BF00425997

[7] Giglio, M.; Patimisco, P. et al. “Allan Deviation Plot as a Tool for Quartz-Enhanced Photoacoustic Sensors Noise Analysis.” PubMed 26529758. https://pubmed.ncbi.nlm.nih.gov/26529758/

[8] “Precision and Limits of Detection for Selected Commercially Available, Low-Cost Carbon Dioxide and Methane Gas Sensors.” Sensors, 19(14), 3157 (2019). https://www.mdpi.com/1424-8220/19/14/3157

[9] “Precision and Limits of Detection for Selected Commercially Available, Low-Cost Carbon Dioxide and Methane Gas Sensors.” PMC6679274. https://pmc.ncbi.nlm.nih.gov/articles/PMC6679274/

[10] “Safety alert: Response times of gas detectors” (SA17-10). NSW Resources Regulator. https://www.resources.nsw.gov.au/sites/default/files/documents/sa17-10-response-times-of-gas-detectors.pdf

[11] “IEC 60079-29-1: Gas Detectors for Flammable Gases Performance Requirements.” Studylib. https://studylib.net/doc/27818821/iec-60079-29-1-2026-gas-detectors

[12] “Gas Monitor t90 Response Time.” Newcastle Safety Servicing. https://newcastlesafetyservicing.com/t90-response-time-testing/

[13] “Overcoming the Slow Recovery of MOX Gas Sensors through a System Modeling Approach.” Sensors, 12(10), 13664 (2012). https://www.mdpi.com/1424-8220/12/10/13664

[14] “Sensor Resolution vs Accuracy vs Repeatability.” FUTEK. https://www.futek.com/resolution-vs-accuracy-vs-repeatability

[15] “How to understand the difference between resolution, accuracy, and repeatability / precision.” Balluff. https://www.balluff.com/en-us/blog/resolution-accuracy-and-repeatability-precision

[16] “Correction Factors for Combustible Gas (LEL) Sensors.” Honeywell Technical Note. https://automation.honeywell.com/content/dam/his-sandbox/products/gas-and-flame-detection/documents/Technical20Note2015620Correction20Factors20For20Combustible20Gas20LEL20Sensorsnr20022016.pdf

[17] “Gas Detector Cross-Sensitivity Guide.” ToolGrit. https://www.toolgrit.com/guides/gas-detector-cross-sensitivity-guide

[18] “Compensated Temperature Range.” SensorsONE. https://www.sensorsone.com/compensated-temperature-range/

[19] “Technology Readiness Levels Demystified.” NASA. https://www.nasa.gov/aeronautics/technology-readiness-levels-demystified/

[20] “Technology readiness level.” Wikipedia. https://en.wikipedia.org/wiki/Technology_readiness_level

[21] Alphasense (AMETEK). “H2-AF Hydrogen Sensor, Technical Specifications.” Doc. Ref. H2-AF/FEB23 and H2-AF/JUL16. https://ametekcdn.azureedge.net/mediafiles/project/oneweb/oneweb/alphasense/products/datasheets/alphasense_h2-af_datasheet-v2_en_1.pdf

[22] Membrapor AG. “Hydrogen Gas Sensor H2/M-1000, Specification Sheet.” Rev. Dec-25. https://www.membrapor.ch/sheet/Hydrogen-Gas-Sensor-H2-M-1000.pdf

[23] Figaro Engineering Inc. “General Information for TGS Sensors: Technical Information on Usage of TGS Sensors for Toxic and Explosive Gas Leak Detectors.” Revised 03/05. https://www.electronicaembajadores.com/datos/pdf2/ss/ssga/tgs.pdf

[24] Figaro USA, Inc. “TGS 2600 — for the detection of Air Contaminants, Product Information.” Rev. 09/13. https://www.figarosensor.com/product/docs/TGS2600B00%20(0913).pdf

[25] “ISO 26142:2010 — Hydrogen detection apparatus — Stationary applications.” International Organization for Standardization. https://www.iso.org/standard/52319.html

[26] “ISO 26142 Hydrogen Detection Apparatus – Stationary applications.” H2tools (Pacific Northwest National Laboratory). https://h2tools.org/fuel-cell-codes-and-standards/iso-26142-hydrogen-detection-apparatus-stationary-applications

[27] “Field and laboratory performance evaluations of 28 gas-phase air quality sensors by the AQ-SPEC program.” Atmospheric Environment (2019). https://www.sciencedirect.com/science/article/abs/pii/S1352231019307319

[28] Palmisano, V.; Weidner, E.; Boon-Brett, L. et al. “Selectivity and resistance to poisons of commercial hydrogen sensors.” International Journal of Hydrogen Energy (2015). https://www.sciencedirect.com/science/article/pii/S0360319915005224

[29] SGX Sensortech. “Application Note AN6 — Pellistor Poisoning.” https://www.sgxsensortech.com/content/uploads/2014/08/AN6-%E2%80%93-Pellistor-Poisoning.pdf

[30] “Laboratory and field evaluation of a low-cost methane sensor and key environmental factors for sensor calibration.” PMC10100561. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10100561/

[31] “Gas Sensor Measurement Setup: How to Get Accurate Data.” Amphenol Advanced Sensors Blog. https://blog.amphenol-sensors.com/blog/gas-sensor-measurement-setup

[32] “Catalytic Gas Sensor Poisoning: A Silent Threat to Safety.” JMS Consultants. https://resources.jmsconsultants.co.uk/news/catalytic-sensor-poisoning