Mounting a gas sensor on a drone looks like a weekend integration. Both parts exist, both are cheap, and the mission — find the leak, map the plume — is obvious. Then the first flight produces readings that bear no relationship to anything real, and the team discovers the central problem: the aircraft destroys the thing it is trying to measure. Four propellers pull air from above the sensor, mix it with clean air from tens of feet away, and blow it down at 20 mph. The sensor is sampling the drone, not the environment.
Aerial air quality sensing is solvable, but only through careful sampling-path design, sensor selection matched to the target gas, and honest handling of response time.
Choose the Sensor Technology First
Four technologies cover almost all drone-borne gas work, and they are not interchangeable. Target gas, concentration range, and response time decide which you can use.
| Technology | Detects | Typical response | Strengths | Limits |
|---|---|---|---|---|
| Electrochemical (EC) | CO, H₂S, NO₂, SO₂, O₂, NH₃ | 15–60 s (T90) | Low ppm sensitivity, low power, low cost | Slow, cross-sensitive, consumable, ages out |
| Photoionization (PID) | Broad VOCs | 2–5 s | Fast, very sensitive to hydrocarbons | Non-selective, lamp fouls, humidity sensitive |
| NDIR (infrared) | CO₂, CH₄, refrigerants | 5–30 s | Stable, selective, long life, no consumption | Bulkier, higher power, limited to IR-active gases |
| Metal oxide (MOX) | VOCs, CH₄, CO (broad) | 10–60 s | Very small and cheap | Poor quantification, heavy drift, heater power |
| Optical / TDLAS | CH₄ at range | Near-instant | Standoff — no sampling needed at all | Cost, mass, path-integrated reading only |
That last row deserves attention: tunable diode laser absorption spectroscopy measures gas along a beam path to a reflecting surface, never bringing a sample to a sensor, so it sidesteps the downwash problem entirely for methane survey.
Note too what "air quality" often means to a customer: particulate matter. PM2.5 and PM10 need optical particle counters, and rotor wash re-entrains dust from below, biasing near-ground readings high.
The Downwash Problem
A multirotor in hover induces a flow field reaching roughly a rotor diameter above the aircraft and several below it. Anything in that field measures mixed air, and mixing is what destroys a concentration measurement. Worse, the error changes with throttle, so the data is not even consistently wrong. Countermeasures, in rough order of effectiveness:
- Sample above the rotor plane. A short mast puts the inlet in air that has not been pulled through the disc — the single most effective change, and nearly free.
- Sample on a tether below the aircraft. A pod hung 10 to 20 feet down sits outside the strongest wash and profiles vertically, at the cost of handling complexity.
- Fly, stop, then sample. Hold at the point and let the reading settle before recording, accepting that hover time costs battery.
- Use forward flight. In translation the wake trails behind and a forward-mounted inlet sees relatively undisturbed air.
- Characterize rather than eliminate. Fly a known source at known concentrations and build a correction, so residual mixing is quantified.
Confined spaces are the hard case: in a tank or tunnel there is no ambient wind and wash recirculates off the walls, so sensing and flight profile must be designed together — see drones for confined spaces.
Inlet and Tubing: Where Good Data Goes to Die
Most drone gas payloads pull air through a tube to a sensor in a small manifold, and every element of that path affects the reading.
Material
Sticky gases adsorb onto tube walls and desorb slowly, smearing the measurement across time. PTFE and PFA are the standard choices; silicone tubing is the classic mistake because it absorbs VOCs enthusiastically. Keep the tube short and out of direct sun.
Flow and volume
Sample lag is tube volume divided by pump flow rate. A 4 mm tube one foot long holds roughly 4 mL; a diaphragm pump at 0.5 L/min clears it in under a second. Bigger manifolds and lower flows add seconds of pure delay before the sensor even starts responding. Size the pump so transport delay is small next to the sensor's response time, and log the flow so lag can be corrected afterward.
Protection
The inlet needs a hydrophobic membrane and a particulate filter, or the first rain shower ends the payload. Filters restrict flow, so count them in the lag budget.
Response Time Versus Ground Speed
This is the constraint founders most often miss. If a sensor has a T90 response of 30 seconds and the aircraft flies at 20 mph, the reading at any moment reflects air the drone passed through up to 900 feet ago. The measurement is not wrong — it is smeared over a span far larger than the plume you were trying to find.
Three ways to handle it: fly slower, matching ground speed to the sensor (a 30-second sensor at 30 ft resolution means under 1 mph, in practice a hover-and-step profile); pick a faster sensor, since PID and optical devices respond in seconds and permit real survey speeds; or deconvolve, reconstructing the profile from a characterized step response, which works but amplifies noise and demands accurate timing. Either way the sensor time constant belongs in the mission plan alongside altitude and overlap, exactly as ground sample distance drives a mapping mission in drone photogrammetry and aerial mapping.
Geotagging a Plume
A concentration number without a position is a curiosity. Customers want a map, and producing one means fusing three time series — concentration, position, and wind — with correct alignment.
Practical requirements: timestamp everything from one clock, preferably GPS time; log faster than the sensor responds, typically 1 to 10 Hz; apply transport lag as a fixed offset; and record altitude above ground, because plume structure is vertical. Then add wind — speed and direction let you back-calculate a likely source from downwind measurements, which is what turns a colored track into an answer. All of it depends on disciplined onboard recording, the infrastructure in flight data logging on a drone. Deliver a georeferenced layer the customer's GIS can open: the interpretation is the product.
Calibration Drift Is the Ongoing Cost
Gas sensors do not hold their calibration. Electrochemical cells consume electrolyte and need recalibration every few months, with a service life of one to two years. MOX sensors drift with humidity and age, PID lamps foul, and even NDIR — the most stable of the group — benefits from a periodic zero. What a credible program does about it:
- Bump test before every mission with a span gas cylinder — a 30-second check that the sensor responds at all.
- Full multi-point calibration on a schedule, with certified gas, documented and traceable.
- Temperature and humidity compensation from co-located sensors, since raw gas readings are strongly affected by both; see choosing a temperature and humidity sensor.
- An in-flight zero where possible — a valve that switches the inlet to a scrubbed reference so the baseline can be checked mid-mission.
Build calibration records into the data pipeline so every exported map carries its calibration date and compensation. Environmental compliance customers will ask, and increasingly regulators will too.
Scoping the Program
Sequence it this way: define the target gas and detection limit, then the sensor, then the sampling path, then the flight profile, then integrate. Choosing an aircraft first and asking what sensor fits is how programs end up with data nobody trusts. Ground-truth the system against a reference instrument before it goes to a customer site, and compare scope with hyperspectral cameras on drones, which answers different questions from the same platform.
Projects House develops sensing payloads and the aircraft that carry them — sensor selection, sampling path design, data fusion, and the ground-truth program that makes the output defensible. Tell us what you need to detect through our contact form.