As an Amazon Associate, Akermin earns from qualifying purchases. Product links on this page are affiliate links.
The range tells you what the display will show. The accuracy band tells you where the manufacturer has characterised the instrument and is prepared to state a maximum error. Those two spans are almost never the same, and the gap between them is where most over confident measurement happens. An infrared camera that will display a temperature well above the boiling point of water may only have a stated accuracy across a much narrower window, and often only while the camera body itself sits in a stated ambient band. Outside that, the instrument keeps reading and the manufacturer has said nothing about how far off it might be.
Range is a display limit, accuracy is a promise with conditions
A range is an engineering fact about the front end and the display. It says that the input stage will not saturate and the digits will not overflow between two limits. Nothing about a range implies that the readings near its extremes have been compared against anything. For many instruments the ends of the range are determined by what the converter can express rather than by any characterisation work at all.
An accuracy statement is a different kind of claim entirely. It is a bounded commitment, and it only means anything when its conditions come with it. Those conditions typically cover the portion of the range it applies to, the ambient temperature the instrument is sitting in, sometimes the temperature of the subject, and in the case of radiometric measurement an assumption about the surface being looked at. A figure quoted without its conditions is not a tighter specification, it is an incomplete one. The explanation of measurement uncertainty covers how these conditions are supposed to be combined.
The band inside the range, and what happens outside it
Picture the range as a wide corridor and the accuracy band as a shorter stretch of it with handrails. Walk inside the handrails and you know how far you can stray. Step past them and the floor is still there, the corridor still leads somewhere, and nobody has measured the drop. That is the correct mental model for a reading taken outside a stated band: usable as an indication, not defensible as a figure.
Three kinds of narrowing show up repeatedly. The first is a portion of the span, where a humidity sensor rated to read across almost the whole scale carries its accuracy statement only across a middle section. The second is ambient conditions, where an instrument is characterised only while its own body sits in a comfortable temperature window. The third is subject dependent, where a non contact measurement assumes a surface behaves in a particular way and quietly stops being accurate on anything shiny. For how range interacts with the fineness of the display, the piece on range against resolution separates those two ideas.
Percent of reading against percent of full scale
The convention a manufacturer chooses changes the meaning of the same number completely. A percent of reading specification scales with what you measure, so the relative error is constant down the range. A percent of full scale specification is a fixed absolute error dressed up as a percentage, and it becomes enormous in relative terms at the bottom of the scale. Take an instrument with a span to just under 2000 units. A hypothetical full scale specification of one percent is a fixed twenty units everywhere.
| Reading | Error from a one percent of full scale specification | That error as a percent of the reading | Error from a three percent of reading specification |
|---|---|---|---|
| 1500 | Plus or minus 20 | About 1.3 percent | Plus or minus 45 |
| 1000 | Plus or minus 20 | 2 percent | Plus or minus 30 |
| 500 | Plus or minus 20 | 4 percent | Plus or minus 15 |
| 200 | Plus or minus 20 | 10 percent | Plus or minus 6 |
| 50 | Plus or minus 20 | 40 percent | Plus or minus 1.5 |
The crossing point is around a reading of 670, and below it the full scale specification is the worse of the two by a widening margin. This is why an instrument with a very wide range is not automatically a more capable one. A wide span pushes the full scale error upward in absolute terms, so unless the manufacturer switches convention or offers sub ranges, the small readings get worse as the headline range gets bigger. The walk through of carbon dioxide monitor accuracy shows the same arithmetic on a specification that combines both conventions.
What these four listings publish, and where the band is missing
| Instrument | Range stated | Accuracy stated | Conditions stated |
|---|---|---|---|
| testo 860i thermal camera | Minus 4 to 662 degrees F | Not stated in the listing bullets | Not stated in the listing bullets |
| TOPDON TC004 thermal camera | Minus 4 to 1022 degrees F | Not stated in the listing bullets | Not stated in the listing bullets |
| DANOPLUS solar irradiance meter | 0.1 to 1999.9 W/m2 and the equivalent in British thermal units per square foot hour | Drift stated as under 1.5 percent per year | Response stated over 340 to 1100 nm, sampling 2.5 per second |
| ZLPLSUPEX ME435 power logger | 10 A to 3000 A with 150 mm coils | Accuracy class 0.5 under IEC 62053-22 | The reference conditions defined by that standard |
The two cameras publish impressively wide spans and no error figure at all. The testo 860i states a 256 by 192 detector described as 49,152 pixels, a wireless connection to a phone application, and analysis features for temperature differences, with no accuracy statement in the bullets. The TOPDON TC004 states the same detector resolution upscaled to 512 by 384, a 3.5 inch display, a laser pointer to mark the measured spot and 64 GB of internal storage, again with no accuracy statement. Both are useful instruments and neither listing lets you put a number on a reported temperature.
The other two are more informative in opposite ways. The DANOPLUS irradiance meter publishes a drift figure rather than a basic accuracy, which is unusual and genuinely helpful, because drift is the part most listings hide. It also publishes the spectral window its response is defined over, which matters a great deal for an irradiance measurement, since a detector responding from 340 to 1100 nm is not seeing the whole solar spectrum. The ZLPLSUPEX ME435 publishes an accuracy class under a metering standard, which is the most formal statement of the four and also the one most often misread, because a class figure applies at reference conditions and the same standard permits a larger error at the bottom of the current range and at poor power factor.
Reading a specification backwards from your own measurement
Start from the reading you expect rather than from the top of the datasheet. Write down the value you think you will see, the ambient temperature you will be standing in, and the part of the range that value sits in. Then go looking for whether the accuracy statement covers all three of those. If it covers the value but not the ambient, your morning measurement in an unheated plant room is outside the band even though the number looks ordinary.
Next, convert every percentage into absolute units at your expected reading, because that is the only form in which specifications from different conventions can be compared. Finally, decide whether the resulting uncertainty is small enough to answer the question. If you need to distinguish two surfaces that differ by a degree and the instrument carries a band several times that wide, the right conclusion is that this measurement cannot settle the question, and it is far better to establish that before the site visit than afterwards. For the electrical equivalent of the same discipline, multimeters with temperature measurement shows how often a secondary function carries a much looser band than the instrument primary one.
Frequently Asked Questions
Does a reading outside the accuracy band mean the number is wrong?
It means the number is unquantified rather than necessarily wrong. The instrument will still behave in some reasonably orderly way outside the stated band, and the error may well be modest. What you have lost is the ability to say how large it might be, which is precisely what an accuracy statement exists to give you. For an indication that something is hot or damp that is acceptable. For a figure you intend to put in a report it is not.
Why do thermal camera accuracy statements mention the camera temperature?
Because the detector in an uncooled thermal camera measures a difference between the scene and its own internal reference, and that reference moves as the body of the camera warms or cools. Manufacturers therefore state a band of ambient temperatures over which the internal compensation has been characterised. Take the camera out of a warm van into freezing air and it will read, but it will be outside the conditions the figure was proven under until it settles.
Is percent of full scale always worse than percent of reading?
At the bottom of the range, yes, dramatically so. A full scale specification is a fixed absolute error expressed as a fraction of the span, so it stays the same size in engineering units no matter how small the reading is, and therefore grows without limit as a proportion of that reading. At the top of the range the two conventions converge. The practical rule is to convert any full scale figure into absolute units first and then compare it with the reading you actually expect to take.
How should I treat a drift figure quoted per year?
As something that accumulates on top of the basic accuracy rather than instead of it. A drift of one and a half percent annually means an instrument that left the factory inside its band can be several percent away after a few seasons with nothing visibly wrong. That is the argument for periodic comparison against a reference, and it is also why a drift figure in a listing is a useful piece of honesty rather than a weakness.
What does an accuracy class under a metering standard actually promise?
A class number is a permitted percentage error under a defined set of reference conditions laid down in the standard, not a blanket figure. The same standard contains a table that widens the permitted error at the bottom of the current range and at reduced power factor, because both make the measurement harder. Quoting the class alone describes the most favourable corner of the operating envelope.
If a listing publishes no accuracy at all, is the instrument untrustworthy?
Not necessarily, but you cannot say anything about it either way, and that is the problem. Plenty of capable instruments have listings written by someone who had the range to hand and not the specification sheet. Treat the absence as a question to resolve from the manual before you rely on the instrument, and record in your notes that the figure was unconfirmed if you could not find it.