As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
Temperature range is a statement about limits, the coldest and hottest absolute values an instrument will report. NETD is a statement about differences, the smallest change in temperature that rises above the instrument own noise. They are not two grades of the same quality, and neither can be inferred from the other: a camera can carry a span of several hundred degrees while resolving four hundredths of a degree between adjacent pixels, and the two figures never meet because one describes where the scale starts and stops and the other describes how finely the scale can be subdivided at any point along it.
One number is about differences, the other about limits
Picture a ruler. Its length is the range: it will not measure anything longer than itself. The finest division marked on it is something else entirely, and a two metre ruler divided into millimetres is not a contradiction. Thermal instruments work the same way, except that they carry two different notions of fineness, which is where most of the confusion comes from.
A temperature range is a hard gate. A target outside it produces no reading at all, or a saturated one, and no processing recovers it. That makes range the first thing to check, because it is binary.
A noise floor is a soft limit. A difference smaller than it still produces a reading, the reading is simply dominated by noise, so the information is present but not separable. That makes it a question of degree rather than a gate, and it is why the same instrument can be excellent for one job and useless for another inside the same range.
The reason these two get conflated is that both are quoted in temperature units. One is quoted as a pair of absolute values and the other as a single interval, which is the tell. If the number has a sign and a reference point, it is a limit. If it is an interval with no reference point, it is a difference. The same distinction applies across every instrument class, which is why the general treatment of measurement range against resolution is worth reading before any specification sheet.
Four specification families that get collapsed into accuracy
In practice a thermal instrument publishes figures from four independent families, and buyers routinely treat all four as accuracy.
| Family | What it constrains | Typical form | Example from these listings |
|---|---|---|---|
| Range or span | The absolute limits that will be reported at all | A pair of absolute temperatures | The AMPBANK H128 states minus 4 to 842 degrees Fahrenheit; the TOPDON TC004 states minus 4 to 1022 degrees Fahrenheit |
| Accuracy or tolerance | How far an absolute reading may be from the truth | Plus or minus a percentage, or a fixed temperature, or whichever is greater | The H128 states readings within plus or minus 3 percent |
| Sensitivity or noise floor | The smallest difference distinguishable from noise | A single interval in millikelvin | The H128 states 40 mK; the TC004 does not state a figure in its bullets |
| Display or temperature resolution | The increment the readout can show | A step in degrees | The H128 states a temperature resolution of 0.1 degrees Celsius |
The AMPBANK H128 is a useful specimen because it publishes one figure from each family in the same bullet set, along with a 240 by 240 image reconstructed from a 128 by 128 detector, a 25 Hz frame rate, IP65 protection, a 6.56 foot drop rating and 24 hours of battery. The TOPDON TC004 publishes a wider span and a reconstructed 512 by 384 image from a 256 by 192 detector, but states no sensitivity figure, which means its ability to resolve small differences is not stated in the listing bullets. Two cameras in the same category, and the specification families they choose to publish do not overlap completely.
Notice what the four families do to each other. Nothing in the range constrains the noise floor. Nothing in the noise floor constrains the tolerance. The display step constrains only what you can read off the screen. An instrument can be excellent in one family and silent in another, and a buyer who reads only the largest printed number learns almost nothing.
Where percent of reading quietly changes the answer
Range and accuracy interact in a way that catches people out, because a percentage tolerance behaves very differently depending on what the percentage is taken of. The H128 states readings within plus or minus 3 percent without saying whether that is 3 percent of the reading or 3 percent of full scale, and that omission matters more than the number.
Take its stated top of range, 842 degrees Fahrenheit, which is 450 degrees Celsius, and run both interpretations.
| Reading | Plus or minus 3 percent of that reading | Plus or minus 3 percent of a 450 degree top of scale |
|---|---|---|
| 20 degrees Celsius | 0.6 degrees | 13.5 degrees |
| 60 degrees Celsius | 1.8 degrees | 13.5 degrees |
| 120 degrees Celsius | 3.6 degrees | 13.5 degrees |
| 250 degrees Celsius | 7.5 degrees | 13.5 degrees |
| 450 degrees Celsius | 13.5 degrees | 13.5 degrees |
At the top of the scale the two readings agree, which is why the ambiguity is easy to miss. At the bottom they differ by more than twenty times, and a full scale interpretation would make a room temperature reading almost meaningless. This is the real cost of a wide range: a fixed percentage spread across a large span is a large absolute error at the quiet end, and the instrument that covers everything is rarely the instrument you want for one narrow job.
Now set that against the sensitivity figure from the same listing. Forty millikelvin is 0.04 degrees Celsius. On the most generous reading of its tolerance, the H128 can distinguish a difference roughly fifteen times smaller than its own uncertainty about the absolute value at room temperature, and several hundred times smaller at the top of its range. That is not a contradiction and it is not a flaw. Relative discrimination and absolute calibration are genuinely independent, which is exactly why the NETD figure exists as a separate specification.
The same split in loggers, clamp meters and irradiance meters
This is a general instrument pattern rather than a thermal imaging quirk, and seeing it in a simpler device makes it obvious.
The HOBO temperature and humidity logger states a temperature span of minus 4 to 158 degrees Fahrenheit, a relative humidity span of 15 to 95 percent, and a humidity tolerance of plus or minus 3.5 percent from 25 to 85 percent relative humidity. Read those three carefully: the reporting range is wider than the guaranteed range. Between 15 and 25 percent humidity the logger still produces numbers, and those numbers carry no published tolerance. That is standard practice and it is honest, but a buyer who reads only the span will believe the instrument is specified across all of it. The same logger states intervals from one second to eighteen hours, IP50 protection and a one year battery, none of which touches either figure. Where you want devices built for a narrow band around room conditions, the comparison of temperature and humidity monitors works through that trade in detail.
The Triplett SPLT400 irradiance and temperature meter sits at the other extreme of disclosure. Its bullets state a watts per square metre readout, a 300 millisecond sampling time, a cosine corrected silicon photodiode sensor, tilt measurement from minus 90 to 90 degrees and a compass azimuth function. The irradiance range is not stated in the listing bullets, and neither is the irradiance tolerance. What it does publish is the sampling interval, which belongs to a fifth family altogether, how often a new value is produced. A fast sample rate on an unspecified range is a reminder that the number a manufacturer chooses to print is the one it considers a selling point, not necessarily the one that constrains your work. The same assessment applies to contact temperature measurement on a general purpose meter, which the roundup of multimeters with temperature measurement covers.
Choosing which of the two numbers constrains your job
Decide what kind of answer you need before comparing specifications, because that determines which family is binding.
- You need to know whether a target is hotter than its neighbour. The noise floor binds. Range only has to contain both targets, and absolute tolerance is irrelevant because the same error applies to both.
- You need a number for a report or a contract. Accuracy binds, and you need to know what the percentage is taken of. A tolerance quoted without that is incomplete.
- You are measuring something very hot or very cold. Range binds absolutely, and you should also ask which range the published noise floor was measured in.
- You are logging a trend over days. Stability and the guaranteed window bind, not the span. Check that your expected values sit inside the tolerance window rather than merely inside the reporting range.
- You are screening a large area quickly. Sample or frame rate binds, with the noise floor second, because a feature you pan past in one frame has to clear the noise in that frame.
The habit worth building is to write out all four families for any instrument you are considering and mark each one either with its published figure or with a note that it is not stated. A sheet with four numbers is a comparison. A sheet with one number and three blanks is a marketing claim, and the blanks are the part that will decide whether the instrument does your job.
Frequently Asked Questions
If a camera reads to 842 degrees Fahrenheit, can it resolve 0.04 degrees Celsius up there?
Not usually. A noise floor is normally measured near room temperature, and most cameras switch to a different gain setting for their high range. Manufacturers rarely publish a figure for the high range at all. The honest reading of a camera that states a wide span and a low noise floor is that those two claims describe different operating conditions, not one condition.
Does a wider temperature range cost me sensitivity?
It can. Covering a very wide span usually means reducing the gain so the top of the range does not saturate, and lower gain means a given amount of electronic noise represents a larger temperature. This is why instruments offer switchable ranges rather than one enormous one. A listing that publishes a single span and a single noise floor is not telling you which range the floor applies to.
Why is the accuracy window narrower than the measuring range?
Because the sensor remains linear and stable over a smaller interval than it remains responsive over. A humidity logger that reports from 15 to 95 percent relative humidity but guarantees its tolerance only from 25 to 85 percent is being precise rather than evasive. Readings outside the guaranteed window are still produced and still useful as a trend; they simply carry no stated tolerance.
Is a display step of 0.1 degrees Celsius the same as a 40 mK noise floor?
No. The display step is the increment the readout can show, which is 100 mK in that example. The noise floor is how much the underlying value wanders. An instrument can have a display finer than its noise, in which case the last digit is unstable, or coarser than its noise, in which case the readout throws away real information. Both situations are common.
Which of the two numbers should I check first?
Check the range first, because it is a hard gate: an instrument that does not reach your target temperature is unusable regardless of anything else. Then check the smallest difference specification, because that decides whether you can see the thing you are looking for once it is inside the range. Accuracy comes third unless you are producing numbers for a report or a contract.