Replicating Natural Daylight Indoors

Start with a number most lighting specs never mention: the color temperature of a clear blue sky can reach 40,000 K. Direct sunlight sits near 5,500 K and an overcast sky near 7,500 K, but the diffuse blue dome overhead, which is what you see through a skylight with no sun in view, is far bluer than any “daylight” bulb. A room lit only by sky light through a skylight can average tens of thousands of kelvin. If a product cannot produce CCTs well above 6500 K, it cannot look like sky.

Color temperature alone is not enough, because two lights with the same CCT can still look very different. Chromaticity has two dimensions, and the second one, called Duv, measures how far the color sits above (green) or below (pink) the reference curve. Daylight follows its own curve through color space, the CIE daylight locus, which sits slightly green-of-blackbody above 5000 K because of how the atmosphere scatters sunlight. Standards such as ANSI C78.377 allow a Duv of at most 0.005; light outside that band reads as visibly tinted.

Tunable white fixtures blend a warm LED (say 2200 K) with a cool one (6500 K). Mixing two fixed light sources can only ever produce colors on the straight line between them, and as the chart above shows, a straight line cannot follow a curve. In the mid-CCTs the blend lands visibly pink of the daylight locus. Even with ideal endpoints, the deviation reaches roughly double the 0.005 standard around 3500 K, and measured fixtures routinely exceed it several times over. Above 6500 K the line simply ends: sky colors are not in the fixture’s gamut at all. Producing accurate color at every CCT takes three or more LED channels blended by a controller against calibration data, an approach covered in multi-chip color mixing and how color is measured.

Two sources can even share the same chromaticity while having completely different spectra, and the differences matter twice: once for how faithfully objects render (measured as CRI and R9), and once for how much the light drives the circadian system. The newer TM-30 metrics (Rf for fidelity, Rg for gamut) grade the same rendering question across 99 color samples, where CRI uses a handful, so ask for TM-30 data when comparing products. Standard white LEDs pump a narrow 450 nm blue spike into a phosphor and leave a characteristic dip near 480-500 nm, which is precisely where the eye’s circadian photoreceptors are most sensitive. RGB and RGBW panels have failure modes severe enough to deserve their own section below.

RGB panels look like the obvious shortcut: three narrow primaries can mix a white at any chromaticity, sky blues included, which is why “16 million color” smart panels are everywhere. The trap is that a white which looks right is not light that renders right. A display gets away with three primaries because you view its emitted light directly; a room is seen almost entirely by reflection, and surfaces reflect across the whole spectrum. Everything between the three peaks is simply absent (select “RGB display LED” in the chart above), so reds, wood tones, and skin land wherever the nearest peak happens to be. CRI routinely falls into the 20s to 60s, with deep-red R9 often negative.

Adding a white channel (RGBW) restores a phosphor continuum and lifts rendering into respectable territory at some settings, but panels built on display-style RGBW chip sets still fall short of daylight replication in four recurring ways:

• Their design brief is decoration. RGB and RGBW panels come from entertainment and mood lighting. Consumer tiles glow at tens of lumens, and even the troffer-format smart panels that do reach several thousand lumens carry documentation from that same world: typically a lone “CRI ≥80” describing the white channel alone at its native CCT, with no IES files, no R9, no spectral data, and no melanopic figures. The gap has causes. These products sell as consumer electronics through retail rather than through lighting channels, so nobody commissions LM-79 photometric testing; with millions of settings there is no single operating point to certify, no nominal CCT, no ANSI bin to claim; and the mixed-mode numbers, were they measured, would not flatter. The missing data is a symptom of intent: nothing in the category is engineered to carry a room’s daylight.
• Output collapses exactly where daylight needs it. Green and red LEDs convert less efficiently than phosphor whites (the “green gap”), and the color chips in a tile carry small current limits compared with the white channel. Daylight spectra above 6,500 K draw most of their power from the blue and green channels, so a panel that looks bright at warm white is dimmest at sky CCTs, where a daylight system needs its output most, and delivered lumens per watt land far below phosphor-white fixtures under a 0.6 W/ft² lighting power budget.
• Calibration to 1 SDCM does not happen, and 8-bit control cannot get there. No RGBW panel, commercial or consumer, is factory-calibrated to 1 SDCM at the fixture, and the parts make it structurally hard. The white chip is binned to the 3-5 SDCM commercial norm at best, and the color dies are not SDCM-binned at all: they are sorted into dominant-wavelength bins roughly 2.5 nm wide, and our modeling puts one bin of difference in a single primary at 3-5 MacAdam steps of shift in the mixed white at identical drive values, so two “identical” panels can disagree visibly out of the box. Nor can firmware correct it with the usual 8 bits per channel: near a white point, a single least-significant step on a dimmed channel moves the mixed color by a MacAdam step or more. Add temperature and aging drift, red being the worst offender, and arrays that matched on day one grow further apart.
• The melanopic gap stays. The cyan band between the blue and green primaries, 480-500 nm, right at the melanopic peak, remains empty unless the chip set adds a cyan channel, so even a convincing sky-blue white can underdeliver circadian stimulus for its appearance.

Since the mixed-mode data is unpublished, we modeled it with the CIE 13.3 test-color method, using typical tile primaries (625 / 525 / 465 nm) and a genuine 80-CRI 6500 K white channel, granting every advantage: each channel aimed exactly at the target color, the white channel carrying as much of the load as possible. The table answers one question: as the panel sweeps a daylight cycle, what is the best rendering it can deliver at each point? Each column is the panel dialed to that white setting. CRI runs to 100 (real daylight), the checklist target is 90, and below about 80 skin and wood tones visibly shift.

R9 grades the deep-red sample that CRI’s 8-sample average leaves out, and it goes negative when reds are distorted badly enough to fall through the bottom of the scale; zero already means “badly wrong.” The pattern to notice: the white chip’s CRI is the panel’s ceiling, and it is only available at the white’s own setting. At 6500 K the mix is 96% white channel and scores the chip’s 81; dialed warmer, the phosphor hands off to the narrow primaries and rendering collapses, with deep reds negative through the entire warm half of the cycle. A warm-white variant models better across the sweep on paper, because its phosphor supplies the red the primaries lack, but it leans on the current-limited color chips for its sky-CCT output, and neither choice changes anything about consistency, calibration, or the melanopic gap. These are ceilings under ideal aim; panels without per-unit calibration sit below them.

There is also a safety dimension: many RGB and RGBW “sky tiles” are sold online as uncertified consumer electronics, with no UL/ETL listing for recessed installation in a ceiling plenum (see safety certification). The approach that works keeps the multi-channel idea and chooses its channels to fill the daylight spectrum, leaving display gamut aside: four or more LED types selected to fill the daylight spectrum, including cyan, blended by a controller against per-fixture factory calibration. That is what delivers high CRI at every CCT, competitive efficacy, and matched color across an array at the same time.

In 2003 researchers confirmed a third class of photoreceptor in the human retina: intrinsically photosensitive retinal ganglion cells (ipRGCs). They contribute almost nothing to vision. Instead they report light, with peak sensitivity in the cyan band around 490 nm, to the brain region that sets your internal clock. Sleep, alertness, mood, cortisol, and melatonin all key off this signal. Ordinary lux ignores it, so a second measure exists: melanopic EDI (equivalent daylight illuminance), measured vertically at the eye where the light actually enters.

The scale of the outdoors-to-indoors gap surprises almost everyone, and it is the strongest argument that daylight replication is about far more than appearance:

Ten feet from a window on a cloudy winter day you receive around 19 m-EDI, less than a tenth of the recommended daytime dose, and a windowless interior receives essentially zero. This is the gap a virtual skylight has to close, and color temperature turns out to be the lever that makes it feasible: because the melanopic ratio climbs with CCT, a sky-blue source delivers several times the circadian stimulus of a warm one at the same visual brightness.

This is why a standard 3500 K office troffer at full output often fails to reach 250 m-EDI at the eye, while the same fixture geometry running a 40,000 K sky spectrum clears it comfortably. It also works in reverse: the same hardware, turned warm and dimmed, can drop below the evening and night limits so the room stays usable without disturbing sleep. That full day-night swing is the requirement; bright blue at noon is only half of it.

Getting the light right is half the problem. A virtual window also has to survive inspection by a visual system with nine independent ways of judging depth. Two of them do most of the damage to flat imitations. Binocular vision compares the slightly different images your two eyes receive; motion parallax compares how the scene shifts as you move. A photograph of sky, however bright and well printed, fails both instantly: each eye sees the identical image, and nothing changes when you walk. Two more cues fail the same way: accommodation (your lens refocusing by distance) and vergence (how far your eyes angle inward) both lock onto the ceiling plane itself. What a photograph does supply are pictorial cues, perspective, haze, relative size, the same cues in any poster, which is why printed sky panels photograph beautifully and disappoint in person.

Perceived depth is not cosmetic. Research in environmental psychology links views with visual depth to lower stress and restored attention, which is much of why real windows feel so different from a bright ceiling. James Turrell’s Ganzfeld installations demonstrate the same machinery in reverse: a perfectly even, featureless field of light removes every depth cue and visitors lose their sense of where the walls are.

An artificial skylight recesses into a ceiling or wall plenum, an enclosed cavity where heat accumulates, insulation may sit against the housing, and any fault is hidden from view. Independent safety listing therefore belongs in the specification itself: UL (or ETL) listing as a luminaire in North America and CE marking in Europe verify that the complete fixture, including its drivers and wiring, has been tested to confirm it will not create a fire or electrical hazard in exactly this kind of installation. Uncertified panels sourced as “LED sky tiles” frequently skip this testing entirely. Ask whether the listing number covers the complete assembly or only the power supply, and see the certification spec-help page for what the marks cover.