•••••••••••••••••••••••••••••••••••••••••••••• 37

Visual Performance as a Function of Spectral Distribution of Sources at Used for General Outdoor

Alan L. Lewis

The need to provide outdoor lighting for orientation, The interest in the role of spectral power distribution identification, and safety at a cost that is reasonable in on visual performance, originally of concern only to the- terms of both dollars and has resulted in the use oreticians trying to understand visual physiology, became of light sources that differ significantly in spectral power of practical significance with the introduction of gaseous distribution from those used in interiors and for which discharge sources for general illumination. Because of almost all data on visual performance have been accu- the obvious advantages of using such energy efficient mulated. Furthermore, the illuminances used in and long-lived lamps for general illumination, studies outdoor lighting applications (0.01 cd/m2.....10 cd/rn") were performed to evaluate their performance relative are vastly lower than those under which the greatest to the more conventional incandescent and fluorescent amount of visual data is available. Under such condi- illuminants. Mercury, (both low and high pres- tions, the is, at best, in a mesopic state for sure), and metal halide lamps have all been studied in which the standard definition of "light," the , is some detail at high illuminances with the result that almost certainly inappropriate. suprathreshold visual performance, measured in terms The lumen is defined as visually evaluated electro- of speed and accuracy on achromatic tasks, is relatively !.I magnetic . Its magnitude was initially derived independent of light source or type as long as con- using viewing conditions that limited the visual response trasts are equated.' Although the color characteristics of to one that was mediated exclusively by the cone recep- HID lamps have reduced their use in interior applica- tor systems of the human . Most critically, a high level tions, they are widely employed for outdoor lighting. of and a central 2 degree visual field In the last few years two separate, but related, groups was used, which effectively eliminated any significant of investigations have raised new questions about the contribution by the rod system which is largely absent effects of both spectral power distribution (SPD) and from that area of the . The relative spectral sensi- light source color on performance. Berman et al., at the tivity of the eye under those conditions is nominally Lawrence Berkeley Labs, have shown that both pupil size described by the 1924 erE Standard Observer for (VI).1 For larger fields, or when retinal illu- CONTRAST THRESHOLD minances are insufficient to adequately stimulate the All Spatial cone systems, the use of the lumen to predict the human response to light is questionable. As a result, additional ~1~~:~ functions have been described for 13 §E larger fields (1964 erE Observer for 10 degree •... 1.015.1r======~-...... ::::::~ fields-VlO),)2 and for very low retinal illuminances (1951 ~ -~ ~ 1 , ~~----==~ eIE Scotopic Observer=-V'j )." o 0.95 Although spectral sensitivity functions are adequate to ~ 0.9 _------w I,: predict the amount of energy required to stimulate ~ 0.85 +.1---r-----,,-.,...... ,...... ,...,....,. -t-1..----..--...,...... ,...... ,..... -r+r- ..-i0 0 1 vision, they are very poor at describing the perceived (cd/sq m) effect of suprathreshold . For example, it is well Ii known that perceived differs significantly from photometric luminance under all but very restrict- -- INC -+- HPM -- HPS 1 Ii" 1 ---- LPS -- MH I: ed viewing conditions." The discrepancy between per- l" ceived effects and photometric quantities is especially Figure I-Relative contrast threshold as a function of adapted sensitive to conditions where lights are of different color luminance (mean luminance of sinusoidal gratings) for five illumi- or where viewing conditions are markedly different from nants (incandescent, low pressure sodium, high pressure sodium, those under which the light units were defined. high pressure mercury, and metal halide). Thresholds are plotted relative to those obtained with an incandescent source. Luminances Author's Affiliation: Michigan Collegeof Optometry, Ferris State were 0.1, 1.0, 3.0, and 10.0 c/m2. University, Big Rapids, Michigan.

JOURNAL of the illuminating Engineering Society Winter 1999 38 ••••••••••••••••••••••••••••••••••••••••••••••

REACTION Indeed. Kinney showed a spatial dependent Realistic ask effect in her study. Such results are not inconsistent with

1200 the known properties of the dual visual system (known variously as the transient/sustained, XjY, or magnocel- ~ 1100 s lular/parvocellular systems) found in most vertebrate ..,§. 1000 animals, including humans . 900 E•• F aoo a General experimental parameters 700 a~: 600 Subjects 500 , 0.1 1 10 Subjects were five paid students each of whom met the LUMINANCE (cdl sq m) following criteria: (1) no current ocular disease or anatomical anomaly; (2) less than ± 0.50 -- INC -+- HPM ...... - HPS D in any meridian without correction; (3) normal color --=- LPS --- MH vision (Ishihara PIP); and (4) normal visual fields (Humphrey 30-2). Figure 2-Mean time to correctly identify the orientation of a grating There were three men and two women aged 20-23. as a function of adapted luminance for five illuminants (incandescent, They were informed of the nature of the experiments, low pressure sodium, high pressure sodium, high pressure mercury, and metal halide). Luminances were 0.1, 1.0, 3.0, and 10.0 c/m2. but not of the ultimate experimental questions. All testing was performed on the same five subjects and brightness are differentially affected by SPD;8.9i.e., who were carefully selected and highly trained in psy- environments that produce equal photopic luminances, chophysical . Training on threshold and but which differ in SPD, produce unequal responses-even reaction time determinations was conducted for each under nominally photopic conditions. They attribute the subject for several weeks until asymptotic performance differences to the fact that both pupil size and brightness was achieved. Training was conducted in the same appa- sensation are mediated, in part, by the rod system which ratus used in the experiment, but was accomplished has a different spectral sensitivity than does the cone sys- using different gratings (e.g., square instead of tem. The differences exist even when the sources are sine waves) and sources (e.g., fluorescent) from those metameric. employed in the experiments. The decision to use a few Kelly'? found up to a 40 percent increase in brightness highly trained subjects rather than a large number of sensation using filters as compared to an equilu- untrained persons was made to reduce the noise in the minous light at luminances between 7 and 40 data so that small performance differences would be cd/m2. She also attributed the effect to rod activity more apparent. Subjects served as their own controls. All because the effect was present only under conditions testing was done monocularly using the subject's right where rods are presumably active. eye; all subjects were right eye dominant. Data-gathering Kinney et al." showed an increase in visual perfor- sessions were of approximately 20 min duration after mance (defined as a decrease in reaction time) for some which a minimum of 15 rnin of rest was permitted. spatial frequencies when lights were viewed through yel- low filters. Kinney's group did not investigate the cause, Sources but speculated that it was due to reduced inhibition in Five sources were used for each condition of the the chromatic processing system. experiments, chosen because they are commonly used in While there is no doubt that SPD has important ram- outdoor lighting installations. They were incandescent, ifications for the perception of lights, it is not so clear high pressure mercury, high pressure sodium, low pres- that these effects translate into benefits that are of engi- sure sodium, and metal halide. neering significance. Indeed, if the effects are as large as Spectral power distributions were measured using an have been reported by Berman and Kelly, it is surprising Opticon Spectroradiometer that was calibrated with a to some that most controlled and properly analyzed stud- standard traceable to NIST. Relative spectral power dis- ies have shown little or no effect of SPD on visual per- tributions for each source are given in the Appendix. formance with achromatic tasks. However, since several Luminances were adjusted to account for the actual investigations have suggested that the effects are rod- SPDs rather than relying on filtered mediated, and since most of the performance tasks stud- (although a was used to monitor for possible ied have been dependent primarily on the resolution of changes in luminance during each experimental ses- high spatial frequencies (for which the rods are not par- sion). Luminances were calculated using the formula: ticularly sensitive), perhaps it isn't so strange after all.

Winter 1999 JOURNAL of the illuminating Engineering Society •••••••••••••••••••••••••••••••••••••••••••••• 39

REACTION TIME cpd) were tested at four levels ofluminance (0.1,1.0,3.0, Grating and 10.0 cd/m2). The visual field size of the gratings 1000 subtended approximately 13 degrees wide x 10 degrees ~ 900 high. Stimuli were controlled by an electronic shutter CD S 800 and were presented in a pre-determined random order '"E for a 750 msec duration every 3 see. To maintain adapta- F 700 l5 tion, a uniform field of the same -averaged lumi- Z 600 nance replaced the gratings between presentations. ••'" a: 500 Because there were no systematic source-dependent

400 effects due to spatial frequency-i.e., threshold differ- 0.1 1 10 LUMINANCE (cd! sq m) ences were similar across sources for all spatial frequencies except that the 10.0 cpd grating could not be resolved at the lowest (0.1 cd/rns) luminance-the data from all -- INC -- HPM ....•••....HPS

-€I- LPS -><- MH resolved frequencies were grouped. The results are presented in Table 1 and Figure 1. All thresholds were Figure 3-Mean time to correctly identify the facing direction of a normalized to incandescent so that differences among pedestrian located adjacent to a roadway (the scene was photo- sources would be more apparent. Data were analyzed by graphically displayed) as a function of adapted luminance for five illuminants (incandescent, low pressure sodium, high pressure a single factor analysis of variance combined with a sodium, high pressure mercury and metal halide). Luminances Scheffe post hoc comparison to identify means that are were 0.1, 1.0,3.0, and 10.0 c/m2. statistically significantly different (p < 0.05). 390 Results indicate that, for luminances at or near L = 683 L LeA.VA.!!.A. normal photopic levels, there are no significant differ- 728 ences in threshold among the five sources tested. However, at the lower two luminances, sources which where L, = spectral (W ster+ m-2) and V = rela- have more of their spectral power in the short - tive spectral efficiency of the eye (1924 CIE Standard lengths (e.g., high pressure mercury and metal halide) Observer). produce lower thresholds than do those which are richer in longer (e.g., sodium lamps). Experiment I: Visual performance as a function of spatial frequency Experiment ll: Reaction time as a measure of Contrast thresholds to sinusoidal contrast gratings performance were measured using a series of back-illuminated, photo- Reactions (the interval between onset of the graphically produced transparencies which varied in stimulus and the correct identification of the stimulus) contrast in steps of approximately 0.1 percent. A two- were measured for the 1.0 and 3.0 cpd gratings present- alternative forced choice (2AFC) procedure using the ed at high contrast (five times above each subject's own method of constant stimuli was employed. A 75 percent threshold contrast). Each subject was trained to asymp- probability of seeing was chosen to represent threshold. totic performance in the reaction time task using a high Thresholds were determined by varying the contrast in contrast 5 cpd grating prior to the experiment. The sub- small steps around a value determined for each subject jects' task was to correctly identify the orientation (hori- in preliminary testing. zontal or vertical) of a grating as soon as possible after its Stimuli were presented in Maxwellian view so that reti- onset Stimuli were self-presented by pushing a button nal would not be affected by fluctuations in (the space bar of a computer) which triggered an elec- pupil size. Four spatial frequencies (0.5, 1.0, 3.0, ans 10.0 tronic shutter which activated the replacement of a uniform field with a grating. When the grating was Table I-Relative contrast threshold. recognized, a second button was pushed which stopped Luminance INC HPM HPS LPS MH the timer and indicated the subject's determination of grating orientation. Only correct responses were utilized -0.01 1.09 1.10 1.18 0.89 in the data presented here. There was no statistically 1.0 1 1.06 1.09 1.12 0.93 significant difference in reaction time between the two 3.0 0.98 1.01 1.01 0.97 grating frequencies so the data were combined and are presented in Table 2 and Figure 2. 10.0 1 0.97 1.01 1.02 0.98 These results are consistent with the notion that reac- Cells in bold are significantly different from incandescent at that tion times decrease as adaptation levels increase (i.e., the luminance (p < 0.05). shortest reaction times are at the higher luminances), at

JOURNAL of the illuminating Engineering Society Winter 1999 40 ••••••••••••••••••••••••••••••••••••••••••••••

Table 2-Reaction Time (msec) to Gratings. in experiment IT except for the substitution of the "real Luminance INC HPM HPS LPS MH life" stimuli for the gratings. The contrast of the complex 0.1 593 625 840 944 577 scene is not simply characterized, but was well above 1.0 510 531 568 638 470 threshold under steady-state viewing conditions at all luminances. Because no spatial frequency-dependent 3.0 457 459 452 444 449 effects were found in the other experiments, no attempt 10.0 452 449 457 446 454 to characterize the task according to spatial frequency . Cells in bold indicate significant differences from other sources at was made. Results are presented in Table 3 and Figure 3. that luminance (p < 0.05). Pupil size least for the lower luminances. At the two photopic levels Pupil size was monitored during testing at each lumi- (3.0 and 10 cd/m2), differences in reaction times are not nance level using a modified telepupillometer significant. Furthermore, at the two lower luminances, (Polymetric). The pupillometer remotely measured there are significant differences among the sources, with pupil size (diameter) by using a CCD camera that those that have more power at short wavelengths pro- imaged the pupil by means of a beamsplitter between the ducing shorter reaction times than do the sources that Maxwellian and the eye. The pupillometer was cali- are richer in long wavelengths. brated with a template placed in the same location as the eye. Average pupil sizes are presented in Table 4. Experiment ill: Reaction time to a "realistic" task Pupil diameters were difficult to measure precisely Differences that may be significant in highly visual because of the iris fluctuations that continuously occur, tasks may be less so when the task includes a larger per- especially during cognitive tasks. The data above represent centage of non- for completion. For the average horizontal diameters taken during blank field example, the detection of a spot of light on a uniform presentations. They are rounded to the nearest 0.1 mm. background requires almost no non-visual process- Earlier mention was made of Berman et al.'s on ing-it is either seen or not seen-and the amount of differences in pupil size that were found for equilumi- time to perform the task is limited primarily by the metameric sources." It should be noted that pupil amount of the time necessary to process the visual infor- size was not a factor in the experiments reported here mation, usually well under 1 sec. On the other hand, the because the Maxwellian view apparatus concentrates the task of tying a shoelace includes only a small amount of light entering the eye at the center of the entrance pupil; time for locating the lace; most of the time is spent retinal illuminance is therefore not affected by changes manipulating the lace. Only 1 percent of the time may be in pupillary area. Nonetheless, we did find pupillary area used to locate the lace while 99 percent would be spend changes of the type reported by Berman, although of in the act of tying. Consequently, a 10 msec difference in lesser magnitude. This reduced effect may be due to the visual processing would probably be lost in the large dif- smaller visual field used here (-13 degrees in this study ferences in time it takes people to tie laces. vs. Berman's full field) which probably stimulated a Experiment ill was designed to test whether the dif- smaller rod population or to differences in the spectra of ferences found in experiment IT, which was almost our sources. entirely a visual detection task, would also exist when the task included more non-visual cognitive processing. Discussion Furthermore, a task was chosen which more closely The results of this study strongly suggest that there are resembled the type of problem that might confront a real and significant differences in performance under driver who was operating a vehicle under conditions of sources with different spectral power distributions, but outdoor lighting than does the grating task. The task that those differences occur primarily at levels of adapta- stimuli were transparencies which depicted a woman tion where the spectral sensitivity of the visual system is standing at the right side of a roadway in the presence of well known to vary from that under which "light" is com- trees and a wooden fence (these stimuli have previously monly defined. The luminances used in this work were shown to be effective as a discrimination task). In one defined according to the photopic sensitivity function -a transparency, the woman was facing the roadway as if to definition that is suitable only for moderate to high levels walk onto the road; in the other transparency, the of adaptation. As the adaptation level decreases, the sen- woman was in the same place in the scene but was facing sitivity of the eye becomes progressively greater at shorter away from the road. The subject's task was to correctly wavelengths until, at very low luminances « 0.2 cd/m·2), identify which way the woman was facing. the peak sensitivity has shifted downward about 45 nm. _ The experimental conditions were identical to those It is, therefore, not unexpected that, at low lumi-

- :"'ter 1999 JOURNAL of the illuminating Engineering Society •••••••••••••••••••••••••••••••••••••••••••••• 41

Table ~Reaction time - realistic task (msec). tively more foveal processing than did our task will pro- Luminance INC HPM Hl'S LPS MH bably show fewer differences across sources; higher reso- 0.1 812 827 1129 1186 810 lution tasks can only be performed at higher levels of 1.0 688 701 767 852 658 luminance where the VA.sensitivityfunction is more suitable and where, consequently, the lumen is an appropriate unit 3.0 622 614 604 609 597 of light. 10.0 610 595 617 596 630 These data strongly suggest that the lumen is not an Cells in bold indicate significant differences (p < 0.05) appropriate metric to characterize the visual effects of lights for tasks which are performed at low luminances nances such as those encountered in outdoor lighting and which commonly occur in many lighted environ- environments, sources which have relatively more of ments such as roadways, parking lots, and pedestrian their spectral power at short wavelengths will be more walkways. effective at stimulating vision and, consequently, enhan- cing visual performance. It remains to be seen whether References the shift in sensitivity alone will account for the differ- 1. Commission Intemationale de l'Eclairage. 1926. ences found here or whether more complex explana- Compte Rendu, Sixth Session, Geneva, 1924. tions will be required to account for the data. It is also Cambridge, UK: Cambridge University Press. important to assess whether or not the relative increase 2. Commission Intemationale de l'Eclairage. 1951. in performance found for the short -rich Compte Rendu, Bureau Central. Paris, France. CIE. sources is sufficient to offset the energy efficiency advan- 3. Commission Intemationale de l'Eclairage. 1964. tage that is currently enjoyed by sodium sources because Compte Rendu, 1963 Session, Bureau Central. Paris, of the (incorrect) photopic assumption. France: CIE. The only statistically significant effects of light sources 4. Chapanis, A., Halsey, RM. 1955. Luminance of occur at the lower two of the four adaptation levels used equally bright .] opt Sac Am 45: 1-6. in these experiments. This suggests that the choice of 5. Boynton, RM., Kaiser, P.K 1968. Vision: the addi- source for outdoor lighting is dependent on the lumi- tivity law made to work for heterochromatic photometry nance provided by the lighting installation. For relatively with bipartite fields. 161: 366-368. high luminances, the choice of source is less important, 6. Kaiser, P.K, Wyszecki, G. 1978. Additivity failures in at least from the standpoint of visual performance. heterochromatic brightness matching. Color Res Appl 3 However, for luminances at or below 1.0 cd/m-2, there (no.4): 177-182. appears to be an advantage to using sources that are rel- 7. Hopkinson, RG., Rowlands, E., Waters, I., Loe, D.L. atively richer in power at short wavelengths. 1973. Visual performance in illumination of differing A great deal of additional work under actual condi- spectral . Report, Environmental Research Unit, tions is necessary to further quantify the benefit of such University College Environmental Research Group, a choice and the benefit may vary with the particular visu- March 1973. al task. The magnitude of the effect will also depend on 8. Berman, S.M., Jewett, D.L., Bingham, L.R., Nahass, the particular spectral power distribution of the source. R.M., Perry, F., Fein, G. 1987. Pupillary size differences Given the large effect, we would expect similar results under incandescent and high pressure sodium lamps.] within classes of sources (e.g., comparing metal halide to oftheIES. 16(no. 1): 3-20. high pressure sodium lamps). The degree of advantage 9. Berman, S.M., Jewett, D.L., Fein, G., Saika, G., and of one source over another will greatly depend on the Ashford, F. 1990. Photopic luminance does not always particular SPD of the selected lamps. predict perceived room brightness. Lighting Res Technol The degree of difference between lamp types will likely 22: 37-41. also depend on the choice of visual task. A task that 10. Kelly, SA. 1990. Effect of yellow-tinted on requires a high resolution ability and consequently rela- brightness.] opt Sac Amer (A7 7: 1905-1911. 11. Kinney, j., Schlichting, C.L., Neri, D., and Table 4-Pupil Diameter (mm) Average of five subjects. Kindness, S.W. 1983. Reaction time to spatial frequencies Luminance INC HPM Hl'S LPS MH using yellow and luminance-matched neutral goggles. 0.1 5.1 5.0 5.2 5.2 4.9 Am] Optom. Physiolog 60: 132-138. 12. Bichao, I.C., Yager, C.D. and Lewis, A.L. 1993 1.0 4.5 4.5 4.6 4.6 4.4 Computer simulation of roadway lighting evaluation: a 3.0 3.8 3.8 3.9 3.9 3.7 preliminary report. In: Proceedings of thefifth international 10.0 3.5 3.5 3.5 3.6 3.5 conferenceon vision in vehicles, Glasgow.

JOURNAL of the Illuminating Engineering Society Winter 1999 42 ••••••••••••••••••••••••••••••••••••••••••••••

Appendix Relative Spectral Power Distributions of Illuminants 1. Incandescent 2. Low pressure sodium 3. High Pressure Sodium 4. High Pressure Mercury 5. Metal Halide

Relative Spectral Power Distribution Relative Spectral Power Distribution Incandescent High Pressure Sodium

0.9 0.9 0.8 0.8 0.7 (I; 0.7 ~ 0.6 IV 0- ~ 0.6 0.5 0- ~ 0.5 s 0.4 ~ '0 1ii a: '0 0.4 0.3 a: 0.3 0.2 0.2 0.1 0.1 ~·~~~~;;~~~~~W.5~~m=~~~~~~·~~~~=m~6~OO~ ~ 430 470 510 590 630' 670 710 4~ 460 500 540 seo 620 660 700 410 450 490 530 570 610 650 690 730 Wavelength (nm) Wavelength (nm) Relative Spectral Power Distribution Relative Spectral Power Distribution Low Pressure Sodium High Pressure Mercury (OX)

0.9 0.9

0.8 0.8 0.7 0.7 a; .(1; 0.6 0.6 0~- ~ 0.5 0- ~ ~CIl 0.5 ~ 0.4 'iii 0 '0 0.4 a: a: 0.3 0.3

0.2 [ 0.2 0.1 0.1 '" ~·~"~~~"~~~"!5~~~~~"~~~~~~~"~6!00~ ~I 430' 470 510 550 590 630 570 710 4~ 460 500 540 500 6~ 660 700 410 450 490 530 570 610 650 690 730 Wavelength (nm) Wavelength (nm) Relative Spectral Power Distribution Relative Spectral Power Distribution Metal Halide Metal Halide

0.9 0.9 0.8 0.8 a; 0.7 0.7 0.6 0.6 0-~ 0~.. Il> 0.5 II> 0.5 > > ~ 1ii 0.4 a; 0.4 '0 a: a: 0.3 0.3


0.1 0.1

470 510 550 590 630 670 710 ~~~w=430~==~4~70~~T=~~~~~630~m=~67~O~~7~10~ 450 490 530 570 610 650 690 410 450 490 650 690 Wavelenglh (nm)

Winter 1999 JOURNAL of the Illuminating Engineering Society