Color Temperature and Outdoor Lighting

Examining the Limits of CCT

Ian Ashdown, P. Eng., FIES

Chief Scientist, Lighting Analysts Inc.

[ Please send comments to allthingslighting@gmail.com ]

Related Posts

Botanical Light Pollution

Mobile Light Pollution

Light Pollution and Uplight Ratings

UPDATE: Sports field lighting analysis added 15/10/12.

[An edited version of this article was published as “STREET LIGHTS: Light pollution depends on the light source CCT” in the October 2015 issue of LEDs Magazine.]

Most of you will be familiar with the International Dark-Sky Association (IDA), which was founded in 1988 to call attention to the problems of light pollution. It reminds us that light pollution threatens professional and amateur astronomy, disrupts nocturnal ecosystems, affects circadian rhythms of both humans and animals, and wastes over two billion dollars of electrical energy per year in the United States alone.

The IDA’s Fixture Seal of Approval program “provides objective, third-party certification for luminaires that minimize glare, reduce light trespass, and don’t pollute the night sky.” Recent changes to this program have reduced the maximum allowable correlated color temperature (CCT) from 4100K (neutral white) to 3000K (warm white). Previously-approved luminaires with CCTs greater than 3000K will have one year to comply with the new standard.

There are several reasons for this revised CCT limit. One reason is that many people prefer low-CCT outdoor lighting, especially in residential areas. As noted by Jim Benya in his LD+A article “Nights in Davis” (Benya 2015), the City of Davis was obliged to replace newly-installed 4800K street lighting with 2700K luminaires at a cost of $350,000 following residents’ complaints. As was noted in the article, “2700K LEDs are now only 10 percent less efficacious than 4000K,” so there was presumably minimal impact on the projected energy savings.

But another, arguably more important, reason is that high-CCT luminaires contribute more to light pollution on a per-lumen basis than do low-CCT luminaires. This is perhaps best demonstrated by the widely-disseminated graph presented in Luginbuhl et al. (2014) and shown in FIG. 1:

Color Temperature -FIG. 1 FIG. 1 – Overhead scotopic sky brightness ratio (Source: Luginbuhl et al. 2014)

Luginbuhl et al. calculated this graph using a modified version of Garstang’s sky brightness model (Garstang 1986). What it shows is that the light pollution due to 5100K cool-white LED street lighting is approximately twice that of equivalent 2400K warm-white LED street lighting. According to the model, this relationship holds true regardless of the distance from the city to the remote astronomical observing site.

From the perspective of both professional and amateur astronomers as publicly represented by the International Dark-Sky Association, this graph is reason enough to require a maximum CCT of 3000K for the IDA’s Fixture Seal of Approval program.

There is however more to this story. While the graph shown in FIG. 1 may present clear evidence of the relationship between CCT and light pollution, we must remember that its data were calculated rather than measured. The question is whether it is reasonable to trust Garstang’s sky brightness model and its modification by Luginbuhl et al.

Garstang’s Model

Garstang’s sky brightness model is conceptually simple. Referring to FIG. 2, imagine a city C and a distant observer O. The sky glow as seen by the observer is due to light emitted by the city streetlights that is scattered by the air molecules and aerosols in the atmosphere along the path of the observer’s view direction. At any point along this path, the light will be scattered from the volume dV. The sky glow as seen by observer O is simply the sum of the scattered light for all such volumes along the path due to all of the luminaires within the city C.

Color Temperature -FIG. 2

FIG. 2 – Garstang’s sky brightness model

Understanding the mathematics of Garstang’s model requires a reasonably good understanding of atmospheric optics (e.g., Liou 2002). This topic will not be discussed here beyond presenting (without explanation) Garstang’s equation for sky brightness b:

Color Temperature - EQN. 1

What is important for this discussion is that Garstang’s model assumes that the street lighting is monochromatic. He assumed a wavelength of 550 nm as being representative for visual astronomy.

We can have confidence that Garstang’s sky brightness model is reasonably accurate, based on recently-published validation studies by, for example, Duriscoe et al. 2013. With cities ranging from Flagstaff, AZ to Las Vegas, NV however, it is simply not possible to measure the influence of correlated color temperature on light pollution.

Wavelength Dependencies

Light pollution is due to both Rayleigh scattering from air molecules and Mie scattering from aerosols such as dust, smoke, and haze. Rayleigh scattering is strongly wavelength-dependent, with a probability proportional to λ-4, where λ is the wavelength. The sky is blue because of Rayleigh scattering. Mie scattering is wavelength-independent, which is why the clear sky appears pale blue or white in heavily-polluted urban areas. (As an aside, the average distance a photon of blue light will travel through the atmosphere at sea level before undergoing Rayleigh scattering – its mean free path – is about 50 km. By comparison, the mean free path for a photon of red light is about 200 km.)

Luginbuhl et al. (2014) used these relationships to extend Garstang’s model for visible wavelengths between 400 nm and 700 nm in order to calculate FIG. 1:

Color Temperature - EQN. 1

While justifiable, this modification to Garstang’s model is somewhat ad hoc. In particular, the original model is a gross simplification of an exceedingly complex physical situation. While it has been validated in terms of sky brightness, this says nothing about whether Luginbuhl’s modifications result in similarly accurate solutions.

Radiative Flux Transfer

There have been more advanced light pollution models developed over the intervening thirty years, including Garstang 1991, Cinzano et al. 2000, Gillet et al. 2001, Aubé et al. 2005, Baddiley 2007, Kocifaj 2007, Luginbuhl et al. 2009, Kocifaj 2010, Kocifaj et al. 2010, Cizano and Falchi 2012, Kocifaj et al. 2014, Luginbuhl et al. 2014, and Aubé 2015.

Perhaps the most comprehensive light pollution model developed to date is Illumina, an open source program that was described in Aubé et al. 2005, and which is still under development. Unlike Garstang’s model (which was designed to execute on a 1980s-era Apple II computer), Illumina is a voxel-based radiative flux transfer program that can require weeks of computer time on a supercomputer with several thousand CPUs and terabytes of RAM (Aubé 2015).

The situation is similar to weather prediction models, where a simple model will give you a rough idea of what is going to happen, but it requires a supercomputer to perform massive amounts of data processing in order to have full confidence in the predictions. Simply put, Illumina models light pollution in a manner that would have been inconceivable thirty years ago.

Color Temperature -FIG. 3

FIG. 3 – Illumina calculation model (Source: Illumina Web site)

Unlike Garstang’s model, Illumina explicitly considers diffuse reflections from the ground and in-scattering of scattered light from volumes m into the volumes n visible to the observer. Garstang’s model includes an entirely ad hoc term for double scattering, but it is impossible to determine whether it correctly models the atmospheric optics.

The details of the program, however, are not as important for the purposes of this article as are the results recently reported by its author (Aubé 2015).

Modeling Sky Glow

What Aubé found with Illumina is that the combination of Rayleigh and Mie scattering results in a wavelength dependency described by λ, where α varies from 3.6 to 2.7 as the distance from the city center increases (FIG. 4).

Color Temperature -FIG. 4

FIG. 4 – Wavelength-dependent scattering exponent (Source: Aubé 2015, Fig. 20.)

What FIG. 4 shows is that near the city center, Rayleigh scattering dominates. This is to be expected, as Rayleigh scattering is not directional – the light is scattered equally in all directions, including back down towards the observer.

FIG. 4 also shows that away from the city center, Mie scattering begins to dominate. This is also to be expected, as Mie scattering is directional – the light is preferentially scattered in the forward direction. It is therefore more likely to be scattered to a remote observer as it travels horizontally through the atmosphere.

Sky Glow versus CCT

To apply Aubé’s results to the question of the influence of CCT on sky glow, we first need some representative white light LED spectral power distributions. The following normalized SPDs were digitized from Philips Lumileds’ Luxeon Rebel product catalog (FIG. 5):

Color Temperature -FIG. 5

FIG. 5 – Typical LED spectral power distributions (normalized)

To provide a fair comparison, these SPDs need to be scaled such that the LEDs generate the same luminous intensity. To do this, we multiply the SPDs by the photopic luminous efficiency function at 5 nm intervals (FIG. 6):

Color Temperature -FIG. 6

FIG. 6 – Photopic luminous efficiency function V(l)

and then sum the results to obtain the relative photopic intensities:

CCT Relative Luminous Intensity
2700K 0.88
3000K 1.00
3500K 1.12
4000K 1.17
5000K 0.94

Table 1

Dividing the normalized SPDs by these values gives:

Color Temperature -FIG. 7

FIG. 7 – Typical LED spectral power distributions (constant luminous flux)

What FIG. 7 shows are the different spectral power distributions of the street lighting at city center for each CCT, assuming the same luminous flux output.

Now, using Aubé’s results and assuming an observing site 80 km (50 miles) from the city center, we multiply each 5 nm interval by (λ/550 nm)-2.7 to represent the wavelength dependency (FIG. 8):

Color Temperature -FIG. 8

FIG. 8 – Overhead sky glow spectral power distribution at 80 km

This is precisely what we might expect – blue light is preferentially scattered, bolstering our assumption that high-CCT lighting results in increased sky glow. (These SPDs represent the relative spectral radiance distribution at zenith from the observing site, which is perhaps the most useful definition of sky glow.)

If we assume scotopic (i.e., dark-adapted) visual observing conditions, we need to multiply these SPDs by the scotopic luminous efficiency function at 5 nm intervals (FIG. 9):

Color Temperature -FIG. 9

FIG. 9 – Scotopic luminous efficiency function V’(l)

and sum the results to obtain the relative scotopic zenith luminance. The results are shown in Table 2:

CCT Relative Scotopic Luminance
2700K 0.96
3000K 1.00
3500K 1.04
4000K 1.12
5000K 1.42

Table 2 – Relative sky glow luminance at 80 km

This however is for a remote astronomical observing site, such as a dark-sky preserve. To understand what happens within the city center, we repeat the above procedure with α = 3.6 as per FIG. 4. Rayleigh scattering predominates, as shown by FIG. 10 with its greatly exaggerated blue peaks.

Color Temperature -FIG. 10

FIG. 10 – Overhead sky glow spectral power distribution at city center

When we calculate the relative scotopic luminance of sky glow, however, we find almost identical results (Table 3).

CCT Relative Scotopic Luminance
2700K 0.96
3000K 1.00
3500K 1.05
4000K 1.14
5000K 1.45

Table 3 – Relative sky glow luminance at city center

This assumes, however, that the observer is completely dark-adapted. In an urban setting, the surrounding street lighting will most likely result in only partial dark adaptation, and so mesopic vision will apply. This means a blending of the photopic and scotopic luminous efficiency functions (FIG. 6 and FIG. 9). With the photopic function being much less sensitive to 450 nm blue light, the differences in relative sky glow luminance at city center will be (depending on the visual adaptation field of the observer) somewhere between that of Table 3 and Table 4, which assumes photopic adaptation:

CCT Relative Photopic Luminance
2700K 0.99
3000K 1.00
3500K 1.01
4000K 1.02
5000K 1.06

Table 4 – Relative sky glow luminance at city center (photopic adaptation)

Of course, with full photopic adaptation, the observer will not be able to see anything but the brightest stars and planets in the night sky, so it is best to rely on Table 3 for comparison purposes.

Given the above, the answer to our question is yes, it is reasonable to trust Garstang’s sky brightness model and its modification by Luginbuhl et al. Aubé’s results, based on the much more comprehensive radiative flux transfer model used by Illumina, basically confirms the relationship between CCT and sky brightness as calculated by Luginbuhl et al. (2014).

Astronomical Considerations

According to Table 2, the increase in scotopic sky brightness for 4000K LEDs compared to 3000K LEDs is only 12 percent. Our perception of brightness, following Steven’s Power Law for extended light sources, means that we would see an increase in perceived sky brightness of only four percent! Surely this is not a reasonable justification for the IDA reducing the maximum allowable CCT from 4100K to 3000K for its Fixture Seal of Approval program?

Professional and amateur astrophotographers would vehemently disagree. Richard Wainscoat, Principal Investigator of the NASA-funded Pan-STARRS search for Near Earth Objects at the University of Hawaii, aptly called spectral power distributions of high-CCT LEDs such as that shown in FIG. 8 the “nightmare spectrum” (Betz 2015). Unfortunately, the peak 450 nm emission is right in the spectral region where natural airglow is low and there are important astronomical hydrogen and oxygen emission lines. Unlike the basically monochromatic emissions of low-pressure sodium lamps, it is impossible to filter out the blue LED emissions with band rejection filters. Limiting the CCT to 3000K reduces the contribution to light pollution in the blue region of the spectrum by a factor of two to three.

Allowing 4100K LEDs may be acceptable for casual stargazing, but not for astronomical research or astrophotography.

Ecological Considerations

According to the Fixture Seal of Approval requirements on the IDA Web site:

The case against blue light is well founded with regard to discomfort glare, circadian rhythm disruption, light scattering, sky glow, and biological system disruption in wildlife.

Outdoor lighting with high blue light content is more likely to contribute to light pollution because it has a significantly larger geographic reach than lighting with less blue light. In natural settings, blue light at night has been shown to adversely affect wildlife behavior and reproduction. This is true even in cities, which are often stopover points for migratory species.

The comment about cities is particularly germane in view of FIG. 10, where the light pollution in the blue region of the spectrum from 5000K LEDs is nearly three times that from 3000K LEDs. (To be fair however, this applies to clear skies only. For cloudy skies, Mie scattering from the water droplets dominates, and so the spectral power distribution of the reflected street lighting is essentially that of the lighting itself. On the other hand, much more light is reflected back towards the ground, greatly increasing light pollution.)

Summary

The purpose of this article was to examine the International Dark-Sky Association’s requirement of LEDs with CCTs of 3000K or less for their Fixture Seal of Approval program. Using recent research results based on a comprehensive light pollution model (Aubé 2015), it was found that the concerns over high-CCT LEDS are well-founded. While 4000K LEDs may be acceptable for casual star-gazing, they are anathema for astronomers and wildlife.

In short, requiring LED street lighting with CCTs of 3000K or less is completely justifiable.

UPDATE 2015/10/12

The analysis presented above assumes a gray world with spectrally neutral reflectance. This is a reasonable assumption in that most roadway surfaces – concrete and asphalt – are not strongly colored. In other words, the light reflected from the ground will have approximately the same spectral power distribution as the incident light.

Suppose, however, that we have an outdoor sports arena with a grass field. The spectral reflectance distribution for Kentucky bluegrass (Poa pratensis) is shown in FIG. 11. The pronounced green peak is expected, given the grass-green color. What is more interesting, however, is the relatively low reflectance in the blue region of the visible spectrum.

Color Temperature -FIG. 11

FIG. 11 – Kentucky bluegrass (Poa pratensis) spectral reflectance distribution

If we multiply the typical LED spectral power distributions shown in FIG. 7 with the grass spectral reflectance distribution on a per-wavelength basis, the overhead sky glow spectral power distribution at 50 km from the city center becomes that shown in FIG. 12. The blue peaks are still present, but they have been reduced by a factor of four relative to the remainder of the spectral power distribution.

Color Temperature -FIG. 12

FIG. 12 – Overhead sky glow spectral power distribution at 50 km due to grass field

The number of outdoor sports arenas may be relatively small, but they generate a surprising amount of light when they are illuminated at night. Using data from the U.S. Department of Energy’s 2010 U.S. Lighting Market Characterization report (DOE 2012), it can be estimated (with reasonable assumptions for typical lamp lumens) that the distribution of outdoor lighting in the United States is:

Outdoor Lighting Percent Lumens
Roadway 48.2
Parking 34.0
Building Exterior 10.2
Stadium 6.0
Billboard 0.8
Traffic Signals 0.7
Airfield 0.1
Railway 0.0

Table 5 – Light pollution sources (approximate)

This is, of course, a global view – light pollution next to a large outdoor sports arena can be a significant concern for residential neighborhoods. The best that can be done is shield the luminaires appropriately, and to turn on the sports field lighting only when it is needed.

In terms of correlated color temperature, the Fédération Internationale de Football Association (FIFA) specifies a minimum CCT of 4000K for football stadiums (FIFA 2007), while the National Football League (NFL) requires a CCT of 5600K (Lewis and Brill 2013). These are arguably acceptable in that green grass fields greatly alleviate the “nightmare spectrum” problem.

(Thanks to Brad Schlesselman of Musco Lighting for providing the grass spectral reflectance distribution and thereby inspiring this analysis.)

References

Aubé, M., L. Franchomme-Fossé, P. Robert-Staehler, and V. Houle. 2005. “Light Pollution Modelling and Detection in a Heterogeneous Environment: Toward a Night Time Aerosol Optical Depth Retrieval Method,” Proc. SPIE Volume 5890.

Aubé, M. 2015. “Physical Behaviour of Anthropogenic Light Propagation into the Nocturnal Environment,” Philosophical Transactions of the Royal Society B 370(1667):20140117.

Baddiley, C. 2007. “A Model to Show the Differences in Skyglow from Types of Luminaires Designs,” Starlight 2007. La Palma, Canary Islands.

Benya, J. R. 2015. “Nights in Davis,” LD+A 45(6):32-34.

Betz, E. 2015. “A New Fight for the Night,” Astronomy 43(6):46-51.

Cinzano, P., and F. J. D. Castro. 2000. “The Artificial Sky Luminance and the Emission Angles of the Upward Light Flux,” Journal of the Italian Astronomical Society 71(1):251.

Cinzano, P., and F. Falchi. 2012. “The Propagation of Light Pollution in the Atmosphere,” Monthly Notices of the Royal Astronomical Society 427(4):3337-3357.

DOE. 2012. 2010 U.S. Lighting Market Characterization. U.S. Department of Energy Building Technologies Program.

Duriscoe, D. M., C. B. Luginbuhl, and C. D. Elvidge. 2013. “The Relation of Outdoor Lighting Characteristics to Sky Glow from Distant Cities,” Lighting Research and Technology 46(1):35-49.

FIFA. 2007. Football Stadiums: Technical Recommendations and Requirements, 4th Edition. Zurich, Switzerland: Fédération Internationale de Football Association.

Garstang, R. H. 1986. “Model for Artificial Night-Sky Illumination,” Publications of the Astronomical Society of the Pacific 98:364-375.

Garstang, R. H. 1991. “Dust and Light Pollution”, Publications of the Astronomical Society of the Pacific 103:1109-1116.

Gillet, M., and P. Rombauts. 2001. “Precise Evaluation of Upward Flux from Outdoor Lighting Installations (Applied in the Case of Roadway Lighting),” Proc. International Conference on Light Pollution. Serena, Chile.

Kocifaj, M. 2007. “Light-Pollution Model for Cloudy and Cloudless Night Skies with Ground-Based Light Sources,” Applied Optics 46(15):3013-3022.

Kocifaj, M. 2010. “Modelling the Spectral Behaviour of Night Skylight Close to Artificial Light Sources,” Monthly Notices of the Royal Astronomical Society 403:2105-2110.

Kocifaj, M., M. Aubé, and I. Kohút. 2010. “The Effect of Spatial and Spectral Heterogeneity of Ground-Based Light Sources on Night-Sky Radiances,” Monthly Notices of the Royal Astronomical Society 409:1203-1212.

Kocifaj, M., and S. Lamphar. 2014. “Skyglow: A Retrieval of the Approximate Radiant Intensity Function of Ground-Based Light Sources,” Monthly Notices of the Royal Astronomical Society 443:3405-3413.

Lewis, D., and S. Brill. 2013. Broadcast Lighting: NFL Stadium Lighting. The Design Lighting Group Inc.

Liou, K. N. 2002. An Introduction to Atmospheric Radiation, Second Edition. New York, NY: Academic Press.

Luginguhl, C. B., D. M. Duriscoe, C. W. Moore, A. Richman, G. W. Lockwood, and D. R. Davis. 2009. “From the Ground Up II: Sky Glow and Near-Ground Artificial Light Propagation in Flagstaff, Arizona.” Publications of the Astronomical Society of the Pacific 121 (876):204-212.

Luginbuhl, C. B., P. A. Boley, and D. R. Davis. 2014. “The Impact of Light Source Spectral Power Distribution on Sky Glow,” Journal of Quantitative Spectroscopy & Radiative Transfer 139:21-26.

The Kruithof Curve

A Pleasing Solution

Ian Ashdown, FIES

Chief Scientist, Lighting Analysts Inc.

[ Please send comments to allthingslighting@gmail.com ]

Related Posts

Kruithof Reconsidered

UPDATE 16/04/09 – This metastudy:

Fotois, S. 2106. “A Revised Kruithof Graph Based on Empirical Data,” Leukos. (Published online 08 April 2016, DOI 10.1080/15502724.2016.1159137.)

critically examined 29 studies in which the Kruithof curve was investigated. The author concluded that “… these [studies] do not support Kruithof. For pleasant conditions, these data suggest only avoiding low illuminances and do not favor any CCT.”

After 75 years of misconception and misuse, may this finally mark the end of the Kruithof Curve. If a lighting designer has a personal preference for warm or cool colors, fine – but please, do not try to justify it with scientific mumbo-jumbo.


[UPDATE 15/01/20 – Added Bartleson (1960) reference.]

Lighting designers today will surely recognize the Kruithof curve, in which the color temperature of the light source is related to a range of illuminances that we find “pleasing.” In its modern form, the Kruithof curve has become supposedly irrefutable evidence that the correlated color temperature (CCT) of LED-based lighting should not exceed 4000K for indoor applications.Kruithof - FIG 1FIG. 1 – Kruithof curve, modern version (source: Wikipedia).

In the process, luminaire manufacturers are being lambasted for promoting products with CCTs of 5000K and higher. Worse, some government agencies and non-profit organizations are adopting CCT limits that are presumably based on the Kruithof curve. The DesignLights Consortium, for example, stipulates that luminaires on its Qualified Products List must have CCTs of 5000K or less for most indoor applications.

Unfortunately, the modern version of the Kruithof curve is different from what A. A. Kruithof published 75 years ago. The upper and lower curves are approximately the same, but their interpretation is different from what Kruithof intended. In fact, the Kruithof curve appears to have been basically misinterpreted for the past three-quarters of a century.

The Kruithof curve itself was thoroughly debunked a quarter-century ago with three exhaustive studies involving up to 400 participants (as opposed to two people in Kruithof’s study, including himself). The Kruithof curve was somewhat belatedly removed from the IES Lighting Handbook five years ago (IES 2010).

This article is however not so much about the validity of the Kruithof curve as it is about a careful re-examination of his 1941 paper. While Kruithof has been rightly criticized over the years for not providing experimental details, he wrote enough for us to infer how he arrived at his findings. When you realize that he was working with early prototypes of the first fluorescent lamps, it is in itself an interesting story.

In the Beginning

The year was 1941. Captain America made his first appearance in a comic book, Europe was being torn asunder by World War II … and Philips Research was quietly developing its own fluorescent lamp technology in Eindhoven, the Netherlands. As part of this effort, Philips Technical Review published a paper by A. A. Kruithof titled, “Tubular Luminescence Lamps for General Illumination” (Kruithof 1941).

Kruithof’s paper was primarily about fluorescent lamp technology, which had been commercially released by General Electric in 1938. His experimental meter-long T12 “luminescence lamp” (FIG .2) was designed have a luminous flux output of 1000 lumens.

Kruithof - FIG 2FIG. 2 – Philips “luminescence” lamp (from Kruithof 1941).

Halophosphate-based phosphors were not invented until the following year (McKeag and Ranby 1942), and so these “white” fluorescent lamps used a combination of cadmium borate, willemite (zinc orthosilicate), and magnesium tungstate, which respectively fluoresce red, green, and blue when excited by the ultraviolet radiation emitted by the mercury-argon gas fill.

By themselves, the phosphors resulted in maximum luminous efficacies of approximately 70 lumens per watt for willemite and 35 lumens per watt for cadmium borate and magnesium tungstate. By combining the phosphors in various proportions, it was possible to generate white light with CCTs ranging from 2650K to 10000K.

Kruithof fabricated fluorescent lamps with CCTs of approximately 4200K and 5800K, plus a third lamp type that was so far off the blackbody curve as to be considered colored rather than white. Comparing these to the extant incandescent lamp technology with its typical 15 lumens per watt luminous efficacy, Kruithof was well justified in writing, “These properties give reason to expect that luminescent lamps will be widely used in the future.”

Kruithof also performed an extensive analysis of the color rendering capabilities of his lamps by observing the color shifts of 313 color cards (as opposed to the eight colors used for the CIE General Colour Rendering Index Ra). Using the phosphor and visible mercury line spectra published in his paper, it is possible to estimate CRI values of his lamps as:

Lamp CCT Ra R9
4200K 36 -110
5800K 54 -60

As for the original form of the Kruithof curve (FIG. 3), the paper includes a description of what the author referred to decades later as a “pilot study” of lamp CCT versus illuminance level. To fully understand this study, it is necessary to quote Kruithof (1941) extensively, beginning with:

“In the first place at a given level of illumination it is found that the colour temperature must lie within certain limits if the effect of the illumination is to be pleasing. Roughly, it may be said that a low or a high colour temperature corresponds to a low or a high level of illumination, respectively. We have investigated this relation experimentally somewhat more closely by introducing in a room a variable number of electric lamps whose current (i.e., the temperature of the filaments) could be varied.”

Kruithof - FIG. 3FIG. 3 – Kruithof curve, original version (from Kruithof 1941).

With vague phrases like “pleasing” illumination, “a room,” and a “variable number of electric lights,” we can at best only infer the experimental conditions that were used to develop this curve.

Kruithof continues:

“Below the lowest curve the illumination is ‘dim’ (at low colour temperature) or ‘cold’ (at high colour temperature). Above the highest curve the unnatural colour reproduction was unpleasant.”

There are two items of immediate interest here:

  1. Kruithof used the adjectives “dim” and “cold” rather than the modern interpretation of “appears bluish.”
  2. Kruithof used the adjectives “unnatural” and “unpleasant” rather than “appears reddish.”

It could be argued that the modern interpretation is intuitively valid, but that is not the point. Kruithof measured a specific psychometric parameter that he termed “pleasing” illumination. Recasting the results to support a different hypothesis effectively invalidates the experiment.

In terms of the illumination referenced in FIG. 1 appearing reddish or blush, it must be noted here that no academic studies published to date support this hypothesis. The modern form of the Kruithof curve appears to be an interpretation with no scientific evidence to support it.

Continuing:

“These obviously vague limits within which the illumination is considered ‘pleasing’ could in our experiments be determined at least with an accuracy of 20 or 30 percent.”

This is a blazing red flag that something may be seriously wrong with this study – how can you measure something as subjective as “pleasing” with an accuracy of 20 to 30 percent in terms of illuminance? What Kruithof said – without providing any evidence – is that we can apparently tell the difference between say 100 lux and 125 lux of illumination. This is not a side-by-side comparison of two illuminated surfaces, but by simply walking into a room and deciding whether or not the colors are “unnatural.”

Much has been written in the following years that Kruithof did not provide any significant details of his experimental apparatus or protocols, and so it is difficult to accept the Kruithof curve as being valid. However, the rest of his paper is reasonably detailed and informative, indicating that Kruithof was a careful researcher. He must therefore have had some reason for claiming an accuracy of 20 to 30 percent.

Continuing with Kruithof, the extensive caption he wrote for FIG. 3 reads in part:

“The left-hand part of the limiting curves, up to a colour temperature of 2850 ºK, is recorded by allowing electric lamps with variable (decreased) current to burn in a room, and varying the number of lamps. The illumination intensity on a table 80 cm high was here measured. In the right-hand part the lowest level which does not give the impression of coldness was determined by experiments with daylight itself and with the daylight luminescence lamps to be described below.”

This is all frustratingly vague, but there is a key point: “… varying the number of lamps.”

Continuing:

“The shape of the upper curve has been extrapolated in this region with the help of the fact that in direct sunlight (colour temperature 5000 ºK) even with the highest illumination intensities occurring (104 or 105 lux) the colour rendering is never found ‘unnatural’.”

This is a crucial quote in that Kruithof decided that direct sunlight was not – and by definition could not – be “unnatural.” This brings us back to the question of the upper curve for color temperatures below 2850K (FIG. 3), where dimmed incandescent lamps were used.

Looking deeper into the question of “unnatural color reproduction,” Kruithof stated that he used a selection of color cards from the Ostwald Color Atlas, a contemporary color classification scheme of the Munsell color system we use today (FIG. 4). With respect to this, he wrote:

“While the colour rendering can be judged by comparison when luminescence is used in combination with other light and the designation of the colour impression obtained and the saturation of the colour obtained must agree, when only luminescence lamps are used no comparison is possible. In judging the colour rendering therefore in this case one must have recourse to ‘colour memory’ which is chiefly confined only to the designation of colours.”

Kruithof - FIG. 4FIG. 4 – Ostwald color atlas (source: Wikipedia).

This is even more puzzling in that our ability to recall colors is mediocre at best. In general, we tend to remember colors as being more saturated than they really were (Bartleson 1960). It therefore makes even less sense that an accuracy (or more properly repeatability) of 20 to 30 percent could be perceived.

It does make sense, however, if Kruithof switched lamps on and off to vary the illuminance while maintaining constant color temperature. This would be a form of flicker photometry. We are mostly insensitive to absolute illuminances, but we are highly sensitive to changes in illuminance. Switching illumination levels would reveal even subtle changes in the perceived chromaticities of the color cards.

… which brings us to the first of two color appearance effects, the Bezold-Brücke hue shift effect. This effect was first reported in the 1870s, but it was not studied extensively until the 1930s (Purdy 1931). Even then, the study was published in the American Journal of Psychology. A lamp research engineer like Kruithof could be excused for not being aware of the paper and its implications.

The Bezold-Brücke effect results in perceived color hues changing with changes in luminance. As shown in FIG. 5, the wavelength shifts required to maintain constant perceived hue for monochromatic colors can be huge, particularly for red and cyan, with a ten-fold increase in luminance. The effect on Kruithof’s color cards would have been generally less noticeable, but may have been still evident with, for example, saturated red colors.

Kruithof - FIG. 5FIG. 5 – Bezold-Brücke effect for monochromatic light (from Wyszecki & Stiles 1982, Fig. 2(5.9))

The second color appearance effect would not have been known to Kruithof because it was not reported until the 1950s. The Hunt effect (Hunt 1950, 1952, 1953) results in the perceived chroma (i.e., colorfulness) of illuminated objects increasing with increased illuminance. This effect is important enough to have been built into the CIECAM02 color appearance model that is widely used for color management for displays, printers, and other imaging devices.

For a given CCT, Kruithof presumably began with a low illumination level and then increased the illuminance until color appearance of the color cards more or less matched his color memory of them. The changes in perceived color would presumably be due to both the Bezold-Brücke and Hunt effects.

As however he continued increasing the illuminance, the continuation of these effects would result in the colors no longer matching his color memory. This would result in, to use Kruithof’s own words, “unnatural color reproduction.”

If Kruithof had reduced the daylight illuminance using, for example, neutral density filters, he would likely have observed the same behavior. However, even if he had done so, the time taken to move the filters into position would likely have masked the color differences. He also would have had the conundrum of having to call natural daylight “unnatural” and “unpleasant.”

In Kruithof’s defense, he may have been one of the first researchers to observe the Bezold-Brücke and Hunt effects using white light illumination with constant CCT. Certainly color shifts with changes in CCT were well-known at the time and modeled by the von Kries chromatic adaptation model. Color shifts with constant CCT were, however, a different matter.

Kruithof was certainly aware of the color shifts that his fluorescent lamps produced, as shown by FIG. 6 and FIG. 7, where the circled areas labelled ‘2’ represent “clearly appreciable” color shifts.

Kruithof - FIG. 6FIG.6 – Color shifts between daylight and 5800K fluorescent lamp (from Kruithof 1941).

Kruithof - FIG. 7.FIG. 7– Color shifts between electric light and 4200K fluorescent lamp (from Kruithof 1941).

Interestingly, Kruithof wrote:

“Most colours are somewhat less saturated in luminescence light than in daylight.”

which is exactly what you would expect from the Hunt effect. Nevertheless, Kruithof had little choice but to describe the 5800K lamp as generating “pleasing” illumination above some threshold.

It must be emphasized, however, that this is only a hypothesis – it would need a carefully designed large-scale experiment to determine whether in fact the Bezold-Brücke and/or Hunt effects can satisfactorily explain Kruithof’s results. Even then, it will be impossible to know with any certainty because Kruithof described his experiments so frugally. On the other hand, they at least offer a plausible explanation of his claim to 20 to 30 percent repeatability.

Time Marches On

If Kruithof considered his work to be a mere “pilot study,” the follow-up studies have been anything but. Bodman (1967) for example performed studies wherein he varied the illuminance in a conference room illuminated by fluorescent lamps. Remarkably, more than 400 subjects took part in these studies. Like Kruithof, he found that people had a preferred illuminance range over which 90 percent of the subjects found the lighting to be “good.” However, As can be seen from FIG. 8, the preferences appear to be influenced more by the lamp spectral power distributions than by their CCTs. At the time that these studies were conducted, “deluxe” fluorescent lamps had CRI values of 90 or so, but warm white fluorescent lamps with their halophosphate phosphors had CRI values as low as 50.

Kruithof - FIG. 8FIG. 8 – Preferred illuminance range versus fluorescent lamp type (from Bodman 1961)

What are interesting are the terms that Bodman’s subjects used to describe the lighting (FIG. 9). The warm white fluorescent lighting (CCT < 3000K) was described as “excessive” and “artificial” (terms which may have influenced by the low CRI values) at high illuminances, but for white (CCT » 4000K) and daylight (CCT > 6000K), the terms used were “pleasant” and “lively.” This is consistent with what you might expect from an increase in colorfulness as provided by the Hunt effect.

Average Illuminance (lux) Color of light
Warm white White Daylight
< 700 Not unpleasant Dim Cool
700 – 3000 Pleasant Pleasant Neutral
> 3000 Excessive, artificial Pleasant, lively Pleasant

FIG. 9 – Subjective impressions of illumination levels (Bodman 1961).

Boyce and Cuttle (1990) performed similar experiments using fluorescent lamps with CCTs ranging from 2700K to 6300K, and with average illuminances ranging from 30 to 600 lux, to illuminate a small office space. All of the fluorescent lamps had CRI values ranging from 82 to 85.

Doing a detailed statistical analysis of 410 questionnaires completed by 15 subjects who spent 20 minutes becoming visually adapted to the room, the authors found that the lamp CCT had no statistically significant influence on the subjective assessments. Instead, the major factor in both color discrimination tests and subjective assessments was illuminance.

Interestingly, none of the subjects who were unfamiliar with lighting design used the terms “warm” or “cool” to describe the lighting.

Davis and Ginthner (1990) also performed similar experiments with 40 test subjects. Using 2750K and 5000K fluorescents lamps with CRI values of 90 and illuminance levels of 270, 600, and 1350 lux, they confirmed the findings of Boyce and Cuttle (1990) that the subjective ratings of preference were influenced by illuminance only. They also found that low light levels were rated as less colorful than high light levels for the same CCT, which again suggests the Hunt effect.

Finally, Viénot et al. (2008) performed experiments designed to investigate the validity of the Kruithof curve using LED modules rather than fluorescent lamps. To ensure high CRI with variable color temperature, they used LED clusters with independently-controlled blue, cyan, green, amber, orange, red, cool white, and warm white LEDs. These provided CRI values from 91 to 96 over an illuminance range of 150, 300, and 600 lux and a CCT range of 2700K, 4000K, and 6500K.

Unlike the previous studies, however,the  Viénot et al. (2008) experimental setup was not a room but a light booth measuring some 41 x 35 x 38 cm (16 x 14 x 15 inches) that had an 80-degree field of view with a dark surround. It is debatable whether the results of the 20 subjects can be applied to offices spaces, but the authors concluded that:

“In one sense, we have validated Kruithof’s statement that high CCT at low illuminance is unpleasant. Nevertheless, we cannot conclude that low CCT should be confined to low illuminance to arouse pleasant sensations.”

and:

“When the colour rendering index is very high and the light spectrum is under control, there is no indication that high colour temperature is judged more pleasant than low colour temperature at higher illuminance levels.”

Summary

There have been many more studies related to the Kruithof curve, including Cockram et al. (1970), Denk et al. (2014), Dikel et al. (2014), Fotois et al. (2013), Hu et al. (2006), Ishi and Kakitsuba (2003), Juslén (2006), Küller et al. (2006), Logadóttir and Christoffersen (2008), Mills et al. (2007), Naoyuki and Tomimatsu (2005), Navvab (2001), Park et al. (2010), Pinto et al. 2008), Weintraub (2000), and Zhai (2014).

None of these, however, are as focused or comprehensive as those of Bodman (1967), Boyce and Cuttle (1990), and Davis and Ginthner (1990). The common conclusion of these authors is that while dim lighting at any CCT is seen as unpleasant, there is no observational evidence in support of Kruithof’s upper curve (FIG. 3).

What this discussion has shown, however, is that Kruithof may not have meant “unpleasant” in the sense of poor lighting quality, but rather in the sense of optimal color reproduction.

It must remembered that Kruithof was working with incandescent lamps with CCTs varying from 1800K to 2850K, 4000K and 5800K fluorescent lamps that probably had CRI values of less than 60, and “natural” daylight with unknown CCT. He likely would have never seen chromaticity shifts with changes in illuminance at constant CCT prior to his experiments, and probably (and quite reasonably) saw them as unnatural and hence unpleasant. This despite the fact that he observed exactly this when comparing his 5800K lamp with daylight, and explicitly commented on the fact.

Conclusion

The conclusion is straightforward, and indeed was established more than a quarter-century ago with three major studies: the Kruithof curve is essentially meaningless. There is no upper boundary to “pleasant” illumination at any CCT, and the best that can be said about the lower boundary is the obvious: dim lighting can be unpleasant, regardless of the CCT.

At the same time, however, Kruithof deserves credit for having been the first to investigate the topic. His failure to describe his experiments in more detail is regrettable, but perhaps understandable. Done as a pilot study, the brief discussion in his paper is basically a progress report with preliminary findings.

It has basically been through our continued misunderstanding of his term “pleasing” that the Kruithof curve continues to persist in lighting design practice. If Kruithof were able to comment on this today, he would likely have only two words to say (in Dutch):

“Stop daarmee!” (English translation: “Stop that!”)

References

Bartleson, C. J. 1960. “Memory Colors of Familiar Objects,” Journal of the Optical Society of America 50(1):73-77.

Bodman, H. W. 1967. “Quality of Interior Lighting Based on Luminance,” Transactions of the Illuminating Engineering Society of Great Britain 32(1):22.

Boyce, P. R., and C. Cuttle. 1990. “Effect of Correlated Colour Temperature on the Perception of Interiors and Colour Discrimination,” Lighting Research and Technology 22(1):19-36.

Cockram, A. H., J. B. Collins, and F. J. Langdon. 1970. “A Study of User Preferences for Fluorescent Lamp Colours for Daytime and Night-Time Lighting,” Lighting Research & Technology 2(4):249-256.

Davis, R. G., and D. N. Ginthner. 1990. “Correlated Color Temperature, Illuminance Level, and the Kruithof Curve,” Journal of the Illuminating Engineering Society 19(1):27-38.

Denk, E., P. Jimenez, and B. Schulz. 2014. “The Impact of Light Source Technology and Colour Temperature on the Well-being, Mental State and Concentration of Shop Assistants,” Lighting Research & Technology (in press).

Dikel, E. E., G. J. Burns, J. A. Veitch, S. Mancini, and G. R. Newsham. 2014. “Preferred Chromaticity of Color-Tunable LED Lighting,” Leukos 10(2):101­-115.

Fotois, S., S. Atli, C. Cheal, K. Houser, and A. Logadóttir. 2013. “Lamp Spectrum and Spatial Brightness at Photopic Levels: A Basis for Developing a Metric,” Lighting Research and Technology 0:1-23.

Hu, X., K. W. Houser, and D. K. Tiller. 2006. “Higher Colour Temperature Lamps May Not Appear Brighter,” Leukos 3(1):69-81.

Hunt, R. W. G. 1950. “The Effects of Daylight and Tungsten Light-Adaptation on Color Perception,” Journal of the Optical Society of America 40(6):362-371.

Hunt, R. W. G. 1952. “Light and Dark Adaptation and the Perception of Color,” Journal of the Optical Society of America 42(3):190-199.

Hunt, R. W. G. 1953. “The Perception of Color in 1º Fields for Different States of Adaptation,” Journal of the Optical Society of America 43(6):479-484.

IES. 2010. IES Lighting Handbook, Tenth Edition. New York, NY: Illuminating Engineering Society.

Ishi, M., and N. Kakitsuba. 2003. “Preferred Color Temperatures at 200 lx during Exposure to Cool or Warm Environments for Middle-Aged Female Subjects,” Journal of the Human-Environmental System 6(2):93-100.

Juslén, H. 2006. Influence of the Colour Temperature of Preferred Lighting Level in an Industrial Work Area Devoid of Daylight,” Ingineria Illuminatului 18(8):25-36.

Kruithof, A. A. 1941. “Tubular Luminescence Lamps for General Illumination,” Philips Technical Review Vol. VI, No. 3, pp. 65-73.

Küller, R., S. Ballal, T. Laike, B. Mikellidesa, and G. Tonello. 2006. “The Impact of Light and Colour on Psychological Mood: A Cross-Cultural Study of Indoor Work Environments,” Ergonomics 49(14):1496-1507.

Logadóttir, A., and J. Christoffersen. 2008. “Individual Dynamic Lighting Control in a Daylit Space,” Proc. Eleventh International Conference on Indoor Air Quality and Climate (Indoor Air 2008). Technical University of Denmark.

McKeag, A. H., and P. W. Ranby. 1942. Great Britain Patent 578,192. Improvements in Luminescent Materials.

Mills, P. R., S. C. Tomkins, and L. J. M. Schlangen. 2007. “The Effect of High Correlated Colour Temperature Office Lighting on Employee Wellbeing and Work Performance,” Journal of Circadian Rhythms 5(2):1-9.

Naoyuki, O., and N. Tomimatsu. 2005. “Preference on Illuminance and Colour Temperature of Interior for Various Behavior Settings,” Proc. 2005 Annual Conference of the Illuminating Engineering Institute of Japan (in Japanese).

Navvab, M. 2001. “A Comparison of Visual Performance under High and Low Colour Temperature Fluorescent Lamps,” Journal of the Illuminating Engineering Society 30(2):170-175.

Park, B.-C., J.H. Chang, Y.-S. Kim, J.-W. Jeong, and A.-S. Choi. 2010. “A Study of the Subjective Response for Corrected Colour Temperature Conditions of a Specific Space,” Indoor and Built Environment 19:623-637.

Pinto, D. P., J. M. M. Linhares, and S. M. C. Nascimento. 2008. “Correlated Color Temperature Preferred by Observers for Illumination of Artistic Paintings,” Journal of the Optical Society of America 25(3):623-630.

Purdy, D. M. 1931. “Spectral Hue as a Function of Intensity,” American Journal of Psychology 43:541-559.

Wyszecki, G., and W. Stiles. 1982. Color Science: Concepts and Methods, Quantitative Data and Formulae, Second Edition. New York, NY: John Wiley & Sons.

Viénot, F., M.-L. Durand, and E. Mahler. 2009. “Kruithof’s Rule Revisited Using LED Illumination,” Journal of Modern Optics 56(13):1433-1446.

Weintraub, S. 2000. “The Color of White: Is there a ‘Preferred’ Color Temperature for the Exhibition of Works of Art?”, Western Association for Art Conservation Newsletter 21(3).

Zhai, Q.-Y., M.-R. Luo, and X.-Y. Liu. 2014. “The Impact of Illuminance and Colour Temperature on Viewing Fine Art Paintings under LED Lighting,” Lighting Research & Technology 2014 (in press).