How to Take a Light Level Reading Somebody Else Can Repeat
Why this matters
Two techs read the same office a week apart and report 41 fc and 29 fc. Neither made a mistake. One read at 30 in above finished floor with the blinds shut at night, the other read on the floor at four in the afternoon with the perimeter blinds up, and both wrote down one number. When that argument reaches a customer, the shop loses it, because the shop has a number and the customer has a number and neither is evidence. What makes a light reading defensible is not a better meter. It is a record that carries every convention the number depends on, and there are more of them than most people expect.
The record is the deliverable
Build the record first and fill it in as you go. Every field below exists because leaving it out lets a second person get a legitimately different answer.
| Field | Why it is on the form |
|---|---|
| Plane and height | Horizontal at 30 in is not horizontal at grade, and neither is vertical at 5 ft |
| Grid: spacing, count, and where it stops | Decides which parts of the pattern get sampled |
| Basis: initial or maintained | A maintained value already contains its light loss factors |
| Time, date, and daylight state | Whether any of the reading came from the sky |
| Electric lighting state and time energized | Warm-up and dimming preset |
| Surface condition and room finishes | Reflectances feed the reading whether you name them or not |
| Meter make, calibration date, accuracy basis | Decides how much of the reading is the room |
| Occupancy and obstruction state | Furniture, stock height, doors |
| Operator position | Where the body was, relative to the cell |
A number without those fields is an anecdote. A number with them is reproducible, and it is what ends the dispute in your favour when yours is the reading that was recorded properly.
Fix the space before you touch the meter
Exclude daylight, or measure it separately. The clean method is to read after dark with the blinds closed. Where that is not possible, read the grid twice, once with the electric lighting on and once with it off, and subtract. Switch it off at the wall switch or the lighting control station, never by operating a circuit breaker as a routine switching means; a breaker is an overcurrent device and repeated switching duty is not what it was listed for. Any work inside the panel itself is de-energized and locked out under 29 CFR 1910.333(b)(2), with the live-dead-live proving sequence of NFPA 70E-2021, 120.5, in the edition your employer's electrical safety program adopts.
Let the source stabilize. A solid-state luminaire drifts as its heat sink comes up to temperature and settles within minutes. A high-intensity discharge source takes far longer to reach full output from cold and will not restrike at all until it has cooled, so a survey that begins the moment somebody flips the switch is reading a source on its way up. Record the elapsed time energized as a field, not as a memory.
Fix the control state. A daylight sensor, an occupancy sensor holding a partial-output setback, or a scene preset will each quietly move your reading. Put the controls in a known state and write down which state.
Leave the surfaces alone. Do not clean a lens, wipe a fixture or change a lamp before a survey you intend to compare against a design value. The dirt is part of the maintained condition you are measuring, and removing it converts the reading to something between maintained and initial with no way to say where.
What the meter contributes to the number
Three separate properties of the instrument land in your reading, and they behave differently in arithmetic.
Cosine correction. Illuminance is flux per unit area of the receiving plane, so light arriving at a shallow angle must count for less in proportion to the cosine of its incidence angle. A meter's diffuser is what does that, and a poorly corrected cell under-reads exactly where the geometry is widest, which is between fixtures and near walls, which is where your minimum lives. A cosine-corrected head is not optional for uniformity work.
Spectral correction, and its character. A silicon cell does not have the eye's spectral response. The correction filter closes most of that gap and leaves a residual mismatch that depends on the source spectrum. Its character is a fixed multiplicative factor for a given source type, not a random spread and not an additive offset, and that has three consequences worth memorizing. It divides out of a ratio taken with one meter on one source type, so your max-to-min is nearly free of it. It does not cancel against an absolute target. And it does not cancel in a before-and-after retrofit comparison, because a retrofit is precisely a change of source spectrum, which is the one case where people assume it cancels.
Accuracy basis. Percent of reading shrinks as the reading shrinks. Percent of full scale does not, and on a coarse range it dominates a small reading completely. Read the basis off the specification before the number enters any arithmetic.
Grid, plane, and the operator's own body
Set the grid before the first reading and draw it on the plan. Bound it where the specification bounds it, and if the specification is silent, state where you bounded it. Then hold three things constant across every station:
- Level the cell. A hand-held head tilted ten degrees on a horizontal plane is reading a plane that does not exist. A small tripod removes the largest source of station-to-station scatter.
- Get your body out of it. A person standing over a cell at desk height shadows it and can take a double-digit percentage off the reading. Step off with the meter on a tripod and read the display from a distance.
- Account for what you are wearing. A high-visibility vest is a bright vertical surface a foot from the cell and it reflects. That matters more outdoors at low levels than in an office at 40 fc, and outdoors is where you need the vest, so use the tripod and step away.
If a reading requires a machine to be in its normal running state, take it from outside the guarded area on a tripod rather than reaching a hand toward a moving part. If the machine has to be stopped instead, it is isolated and its stored energy released under 29 CFR 1910.147 before anyone puts a meter where the guard was.
The filled-in record
Open office, 24 ft by 32 ft, four-year-old recessed luminaires, no cleaning history.
| Field | Recorded |
|---|---|
| Plane and height | Horizontal, 30 in above finished floor |
| Grid | 12 stations on 8 ft centers, bounded 4 ft in from each wall |
| Basis | Maintained (in-service condition, nothing cleaned or relamped) |
| Time and daylight | 9:15 pm, blinds closed, no exterior contribution |
| Lighting state | All luminaires at full, no preset, energized 22 minutes before first reading |
| Meter | Cosine and color corrected, calibrated within the last 12 months, plus or minus 4 percent of reading plus 1 count on the 0 to 200 fc range, resolution 0.1 fc |
| Spectral mismatch, this source type | Within plus or minus 3 percent for a 4000 K white source, per the meter manufacturer |
| Operator | Tripod at each station, operator 6 ft away |
Readings, in station order: 41.2, 38.6, 44.1, 36.9, 42.5, 39.8, 45.3, 37.4, 40.1, 43.6, 38.2, 41.9 fc.
Sum 489.6 fc over 12 stations gives average 40.8 fc, maintained, horizontal at 30 in above finished floor. Minimum 36.9 fc, maximum 45.3 fc. Average to minimum 40.8 / 36.9 = 1.11 to 1; maximum to minimum 45.3 / 36.9 = 1.23 to 1.
Now the error terms, each with its basis and its character before it is used.
Random placement spread. Station 1 was re-read five times: 41.2, 40.8, 41.5, 41.0, 41.3 fc, a spread of about plus or minus 0.9 percent of reading. Character: independent random. Independent random terms average down across stations, so on a 12-station mean it shrinks by the square root of 12, which is 3.46, to 0.26 percent, or 0.11 fc on the 40.8 fc average.
Spectral mismatch. Basis: percent of reading. Character: a fixed multiplicative bound of unknown sign for this source type, so it is written with one inequality and it does not average down with more stations. It stays within 3 percent on the average, which is 1.22 fc.
Resolution floor. 1 count is 0.1 fc, which on a 40.8 fc reading is 0.25 percent. Character: a bound.
The systematic term is larger than the random term by 7 / 0.26 = a factor of about 27, and that is the payload of the whole exercise. Adding stations improves your knowledge of the pattern and does almost nothing for your knowledge of the absolute level. If somebody wants a defensible absolute number, the lever is the meter's calibration and its spectral match to the installed source, not a bigger grid.
Calibration. Basis: percent of reading. Character: a bound of unknown sign, so it adds linearly to the other bound rather than combining in quadrature. Within 4 percent on the average, which is 1.63 fc. With the spectral term the systematic total is within 7 percent, or 2.86 fc.
What the basis would have done. Had this meter specified plus or minus 4 percent of full scale on the 0 to 200 fc range instead of 4 percent of reading, the band on a 40.8 fc reading would be plus or minus 8.0 fc, which is 19.6 percent of the reading, worse by a factor of 19.6 / 4 = 4.9. Same headline percentage on the specification sheet, five times the doubt.
What survives. Report the average as 40.8 fc maintained horizontal at 30 in above finished floor, with the calibration band stated as a bound rather than folded into the value. Report the ratios, 1.11 and 1.23, as effectively free of the multiplicative term because it divides out of a ratio taken with one meter on one source type. Do not add the three error terms together into a single plus or minus; they have three different characters and stacking them linearly produces a number that means nothing.
Checking your own figures
- Average recomputed, not copied. 489.6 fc / 12 stations = 40.8 fc. Twelve values printed, twelve summed.
- Ratios recomputed from the printed extremes. 40.8 / 36.9 = 1.106, reported as 1.11. 45.3 / 36.9 = 1.228, reported as 1.23.
- Random term reduced, systematic term not. 0.9 percent / sqrt(12) = 0.26 percent; the 3 percent spectral bound is carried at 3 percent on both a single station and the mean.
- The terms were not summed. Three error terms with three characters, reported separately. No combined plus or minus appears anywhere.
- Bound written with one inequality. Spectral mismatch is stated as within 3 percent, not as a two-sided interval around a known value.
- Basis comparison uses one reading. The 4.9 factor compares plus or minus 4 percent of reading against plus or minus 4 percent of a 200 fc full scale, both evaluated at the same 40.8 fc.
- Every reported illuminance carries plane, height and basis. Horizontal, 30 in above finished floor, maintained, on the average, the minimum and the maximum alike.
- Nine record fields listed, nine filled. The filled-in table carries plane and height, grid, basis, time and daylight, lighting state, meter, spectral mismatch, and operator position; room finishes and occupancy are folded into the space description above the table rather than dropped.
References
- 29 CFR 1910.333(b)(2), de-energizing and lockout for work on electrical circuits, and NFPA 70E-2021, 120.5, for the live-dead-live verification sequence in the edition your employer's electrical safety program adopts
- 29 CFR 1910.147, control of hazardous energy, where a machine must be stopped and its stored energy released before a measurement is taken inside a guarded area
- Meter manufacturer's specification sheet and calibration certificate, which own the accuracy basis, the cosine correction and the spectral mismatch figure for a given source type
- IES recommended practice for measuring illuminance in the space type concerned, in the edition your specification names, which binds only through that specification
- See related: Why Uniformity Matters More Than Average Level; What a Light Loss Factor Is Accounting For