The Stratification That Fools a Single Reading

Why this matters

Two techs stand in the same room with the same instrument and get different numbers, and both are right. A thermostat holds setpoint all day while the customer works in a space that is five degrees colder. A tank sensor says the water is hot and the fixture runs cold. In every one of those cases the reading was accurate and the conclusion drawn from it was wrong, because a temperature is a sample taken at a location inside a field that varies, and the location is part of the number.

Before you go chasing a profile

Getting readings at several heights is where techs create hazards that have nothing to do with temperature.

Do not climb to take a reading you can take from the floor. A telescoping pole, a logger hoisted on a line, or a non-contact reading of a surface at height removes the fall exposure entirely, which beats managing it. Where you do have to climb, set the ladder on level footing at the correct angle, tie or foot it, and carry the instrument in a holster so both hands stay on the ladder; ladder duties sit at 29 CFR 1910.23 in general industry and 29 CFR 1926 Subpart X where the work is construction.

An attic or roof space in summer is a heat stress hazard before it is a measurement problem. Set a work and rest schedule, carry water, tell someone where you are, and come out on the schedule rather than on how you feel, because impaired judgment is one of the early signs.

An enclosed plenum, tank, pit or duct interior may meet the definition of a permit-required confined space. Evaluate it under 29 CFR 1910.146 before anyone puts a head inside; the hazard there is atmospheric and a body harness does not address it.

Duct and equipment surfaces at operating temperature burn on contact, so strap or clamp probes rather than holding them, and do not steady yourself on a hot casing while you reach for a reading.

What a single reading cannot tell you

This is the useful list, because everything a stratified field does to you comes from one of these five blind spots.

It cannot tell you which way the gradient runs. Warm air above cold and cold air above warm produce identical single readings at the sample height. The direction of the gradient is usually the thing you needed.

It cannot tell you whether you are in the mainstream or a dead pocket. A reading in a stagnant corner, behind a rack, above a dropped ceiling, or in the lee of a large obstruction is a true temperature of a place that is not thermally connected to anything. It will not move when you change the system, and that lack of response reads as a system that is not working.

It cannot tell you whether the number is central or extreme. A reading of 67 degrees F is a different fact if the field spans 66 to 68 than if it spans 62 to 86. Nothing about the reading itself distinguishes those.

It cannot tell you whether the field is stable. A stratified column is a fragile arrangement held up by the absence of mixing energy. A door opening, a fan starting, an exhaust turning on, or a person walking through will collapse part of it, and a reading taken before that is not comparable to one taken after.

It cannot tell you whether a difference between two readings is a real difference. Two numbers taken at different heights, in different corners, at different distances from a wall, differ for reasons that have nothing to do with the equipment.

Where the gradient comes from

Air sorts itself by density, and density falls as temperature rises, so warm air ends up on top wherever nothing forces it to mix. That gives four practical sources.

Buoyancy plus height. Sorting needs vertical room. A 22 ft bay stratifies; an 8 ft office barely can. The taller the volume, the bigger the spread available.

Absent mixing energy. Stratification is what happens when the system is off, when air change is low, or when supply air is delivered above the occupied zone and short-circuits back to a high return without ever coming down. A running mixing fan flattens the profile; the profile rebuilds when it stops.

Surface sources at one end. A hot roof deck, a cold slab, a glazed wall, a large cold or hot appliance. These load one end of the volume directly, and they keep loading it whether or not any air is moving.

Geometry that traps. A mezzanine, a deep truss space, a shelved aisle, a closed loft. Each becomes its own small volume with its own profile.

The profile is the measurement

Replace the single reading with a set. The unit of analysis is a column of readings at stated heights, taken within a few minutes of each other, with the same instrument, at a stated position on the floor plan, with a note of what was running.

Five heights is usually enough in a tall space: near floor, occupant height, mid, upper, and near ceiling. In a room, three is enough: ankle, seated head height, standing head height. Record the height of every reading. A profile with unrecorded heights is five single readings.

Take a second column somewhere else in the space if you suspect horizontal variation, and take it under the same running conditions. Two columns is how you separate "this space is stratified" from "that corner is dead."

Worked example: a high bay, five heights

Unit heaters overhead, a 22 ft ceiling, outdoor 20 degrees F, thermostat on a column at 5 ft. Complaint is cold at floor level, and the thermostat is satisfied.

Overall gradient: 86 - 62 = 24 degrees F over 22 ft, about 1.09 degrees F per foot. The per-foot rate across the four bands runs 1.0, 1.25, about 1.17, and 1.0 degrees F per foot, steepest between 4 and 8 ft, so it is close to linear with no sharp layer boundary.

The thermostat at 5 ft sits between the 4 ft and 8 ft readings, which climb at 1.25 degrees F per foot, so it is seeing about 67 degrees F. It is not lying. It is reporting a real temperature at 5 ft in a field that is 62 at the floor and 86 at the roof deck.

Now put the sensor somewhere else and watch the same building produce different behavior. A sensor in a high return at 22 ft would report 86 degrees F and shut everything off with the floor at 62, a 24 degree F gap between the controlling number and the working number. A sensor at ankle height would report 62 and run the equipment continuously, driving the ceiling higher still.

None of those three sensors is faulty. They are three different samples of one field.

What mixing that column would buy, computed

If the whole column were mixed to uniform temperature, the result is the height-weighted average of the profile. Take each band's mean and weight it by the band's thickness:

• 0 to 4 ft: mean 64, thickness 4, contribution 256
• 4 to 8 ft: mean 68.5, thickness 4, contribution 274
• 8 to 14 ft: mean 74.5, thickness 6, contribution 447
• 14 to 22 ft: mean 82, thickness 8, contribution 656

Total 1,633 over 22 ft, which is 74.2 degrees F.

So the same air, redistributed, is 74.2 degrees F everywhere: the floor gains about 12 degrees F and the ceiling loses about 12. A density-weighted average would shift that by a fraction of a degree over a 24 degree F span, so the simple height-weighted figure is fine for field reasoning.

The roof consequence, at the stated outdoor 20 degrees F: roof heat loss tracks the difference between the air against the deck and outdoors. Before mixing that difference is 86 - 20 = 66 degrees F. After, it is 74.2 - 20 = 54.2 degrees F. That is 11.8 degrees F less driving temperature, or about an 18 percent reduction in the roof's share of the loss. Both figures are computed the same way, from the same profile, against the same outdoor temperature, so they compare directly. This is the roof only; walls see a smaller version of the same effect, and nothing here addresses infiltration.

Two honest caveats before anyone quotes the 74.2. Full mixing is an idealization; real destratification equipment gets partway there, and how far depends on its throw and on how much the geometry traps. And the 74.2 will not persist, because once the floor comes up the thermostat at 5 ft is satisfied sooner and the equipment runs less, so the whole column settles lower than 74.2. That settling is the point, not a flaw: the same comfort at the floor for less input.

The sensor's position is a design decision, not an installation detail

Everything above says the sensor location is a choice with consequences, so make it deliberately.

Put the controlling sensor where the condition you are managing actually is. For occupant comfort that is occupied height in the occupied part of the space, not a return grille, not a corridor, not the wall by the door everyone walks past. For a process it is at the process.

Keep it out of anything that loads it directly: direct sun at any hour, not just at the hour you are standing there; an exterior wall with a cold cavity behind it; a supply airstream; the top of an appliance; the path of a draft from a door.

Where the field genuinely varies and one point cannot represent it, the answer is more sensors, not a better one. Averaging sensors across a large open space, and separate sensors per zone where the zones behave differently, both exist for this reason.

The same effect where nobody calls it stratification

A storage tank. Hot water sits on top and cold sits at the bottom, deliberately, because that is what makes a tank deliver its full volume at temperature. A sensor high in the tank reports hot while the lower part is not, so the draw goes cold before the sensor knows anything happened. This is why a single tank temperature is not an inventory reading.

A duct downstream of a mixing point. Two airstreams entering at different temperatures do not mix instantly; they travel side by side for some distance. A single-point sensor in that section can read many degrees away from the true mixed temperature, and where it sits across the cross-section determines whether a freeze protection device sees the cold stream or misses it. Traverse the cross section rather than trusting one probe.

A horizontal fluid line running part full. The top of the pipe is in contact with vapor or air and the bottom with liquid, and their surface temperatures differ accordingly. Reading the top and reporting it as the line temperature is a location error, not an instrument error.

Confirming you took a profile rather than five readings

Re-take the column at the same floor position and the same five heights, with the same running conditions, twenty minutes later. A real profile repeats within a degree or so at each height. If it does not, something in the space is changing on a timescale shorter than your measurement, and until you find out what, no single number from that space means anything.

Then run the honesty check on the record itself: every reading has a height, every column has a floor position, and the note says what was running. Any reading missing one of those three is back to being a sample of an unknown place.

References

• ASHRAE Handbook - Fundamentals, chapters on space air distribution and thermal comfort, for stratification and the occupied zone
• 29 CFR 1910.146 for permit-required confined spaces; 29 CFR 1910.23 and 29 CFR 1926 Subpart X for ladder use in general industry and construction
• Manufacturer documentation for controlling sensor location and averaging sensor application
• See related: How to Read a Surface Temperature Honestly; How Air Temperature and Radiant Temperature Differ