Spectrum & Rendering: SPD, CRI & the R9 Problem

The spectral power distribution is a light source's fingerprint: how much power it emits at every wavelength. Everything else on this site (chromaticity, CCT, gamut, CRI) is computed from the SPD; two sources can share the exact same white point and CCT while having wildly different spectra. Chromaticity tells you where the light lands for the eye; the SPD tells you what it actually contains. Color rendering lives entirely in that difference.

Sunlight is a ~5800 K black body filtered by atmosphere: power at every visible wavelength, no gaps. Halogen is a small black body at ~3200 K, warm-tilted but perfectly continuous, which is why tungsten held the reference-light role for a century (CRI 100 by construction). Xenon arc is the cinema projector classic: a high-pressure discharge whose visible output is a smooth, daylight-like continuum. All three light every pigment a face or a fabric can contain. Their differences are of balance, not of missing content; a white balance fixes balance, but nothing fixes missing content.

Almost every white LED is a blue pump (~450 nm) exciting a yellow phosphor. The signature is unmistakable in the 3D landscape: a sharp blue spike, a valley near 480 nm (the cyan gap), a broad phosphor hump, and (on cheap emitters) a cliff after ~620 nm where deep red should be. Narrow spectra are efficient (lumens per watt reward putting energy where the eye's sensitivity peaks), so economics pulls exactly opposite to rendering quality. High-CRI LEDs add red phosphors to rebuild the long-wavelength tail; you pay for those photons in efficacy, which is why honest R9 costs money.

What reaches the camera is a wavelength-by-wavelength product: reflected(λ) = SPD(λ) × reflectance(λ). An object can only return light the source delivered. Skin reflects strongly above 600 nm (that long red slope is what reads as life and warmth), but if the source's SPD is near zero there, the product is near zero, and the sensor records nothing. No matrix, no LUT, no grade can reconstruct spectral information that never arrived; correction can only redistribute what was captured, amplifying noise where the signal is starved.

CRI renders a set of standard test-color samples under the test source and under a reference (a black body or daylight of the same CCT), and scores the shift of each. The headline Ra averages only R1–R8: eight pastel, low-saturation samples. R9, the saturated red, is computed but excluded from the average. Because red sits exactly where blue-pump LEDs are weakest, a fixture can advertise CRI 90+ while its R9 is single-digit or negative. Two rules of thumb: never accept an LED spec without R9 (and ideally R13, the skin sample), and treat Ra ≥ 90 with R9 ≥ 50 as a floor for camera work, not a target.

Skin's reflectance is built by hemoglobin and melanin: the oxyhemoglobin W (twin dips near 542 and 577 nm) and a steep rise through the reds. Starve the reds (low R9) and every skin tone flattens toward gray-yellow wax, across all complexions, because the differentiating information lives in that red slope. Cameras make it worse: sensor spectral sensitivities are not the eye's color-matching functions, so a spiky SPD that happens to fool the eye (metameric white) can still land wrongly on the sensor: faces that looked fine on set go wrong in dailies, and no two camera models fail identically. Continuous spectra are the only lighting that is safe for both observers.

Practical consequences: light faces with full-spectrum fixtures (high-R9 LED, HMI, tungsten) and spend the cheap narrow sources on backgrounds and effects. Never mix metamerically-different matching whites on one face: they will match to the eye and split on camera. When sources must intercut, match them by spectrum, not by CCT: that is what SSI (Spectral Similarity Index) measures. And verify with the actual camera: a color chart under the actual fixture, on a scope, tells the truth a spec sheet won't.

CRI's successors fix its blind spots: TM-30 scores 99 real-world samples and reports both fidelity (Rf) and gamut (Rg) with hue-by-hue detail; TLCI evaluates rendering through a standardized camera instead of the eye (the broadcast-relevant question); SSI compares two SPDs directly for source matching. A serious fixture spec quotes several of these. The measurement side of this story (how spectroradiometers capture the SPDs you've been orbiting) is its own module. Book a calibration →