Effects of correlated color temperature of light on thermal comfort, thermophysiology and cognitive performance

Anecdotal evidence suggests that the correlated color temperature (CCT) of light can affect thermal comfort. Previous literature mostly investigated this effect over a short duration ( < 1 h) and often attributed it to the hue-heat hypothesis (color-temperature association)


Introduction
Heating, ventilation and air-conditioning (HVAC) systems consume nearly 20% of global energy [1,2].Improving their energy efficiency is paramount to meet our sustainability goals.HVAC systems aim to provide thermal comfort [3], mostly by regulating indoor temperature in a stringent temperature range.Expanding this temperature range reduces the cooling/heating load of HVAC considerably.Therefore, less strict temperature setpoints present one of the most effective solutions to improve HVAC energy efficiency [4,5].For example, a simulation study has shown that relaxing 1 • C of temperature setpoint can result in about 10% saving of the HVAC energy consumption [4].Moreover, repeated exposure to temperature variation (mild cold and warm) may also elicit important health benefits [6].Nevertheless, expanding the temperature range may compromise thermal comfort, particularly on an individual level, due to large interpersonal differences in thermal perceptions [7].
How to improve thermal comfort becomes an important question for practicing wider temperature setpoints.Light, as one of the environmental factors, has recently received increased attention because the energy efficiency of artificial lighting is increasing with the advances of LED technologies [8].Especially, the color appearance of the light, gauged by correlated color temperature (CCT), has been suggested to affect thermal perceptions [9][10][11][12][13][14][15][16][17][18][19][20][21][22].Adjusting the CCT of light requires nearly no extra energy in most color-tunable lighting systems [16], Abbreviations: HVAC, Heating ventilation and air-conditioning; CCT, correlated color temperature; pRGC, intrinsically photoreceptive retinal ganglion cells; BMI, body mass index; PMV, predicted mean vote; HSL, hue saturation luminance value; ICTAS, individual color-temperature association strength; EDI, equivalent daylight illuminance; SPD, spectral power distribution.therefore, thermal comfort may be improved with limited additional cost via CCT tuning.
Why CCT would affect thermal perception often is explained by the so-called hue-heat hypothesis.This hypothesis states that blue colors result in a cooler temperature sensation whereas red/yellow colors lead to a warmer perception of the thermal environment [23].White light with a high CCT appears more bluish than white light with a low CCT (which might be perceived as more yellowish).Therefore, exposure to a high CCT is expected to improve thermal comfort in thermally warm conditions compared to low CCT and vice versa.This cross-modal color-temperature association has been investigated using colored surfaces [24,25], goggles [26], virtual reality [27], and illumination [9][10][11][12][13][14][15][16][17][18][19][20][21][22], although luminous and non-luminous colors may have significant different impacts [28].Morgan et al. [29] suggest that color-temperature associations develop throughout childhood and adolescence, implying inter-individual differences in the color-temperature association.Furthermore, recent studies also observed an association between visual comfort (the overall comfort perception of the light) and thermal comfort, in both laboratory and field contexts [19,30,31].This indicates that visual comfort and thermal comfort may share the same underlying drive for comfort.
Besides the effects via visual experiences (e.g., hue-heat hypothesis, visual comfort), light also drives non-visual effects, i.e., effects beyond vision, which are largely driven via intrinsically photoreceptive retinal ganglion cells (ipRGC) with short-wavelength sensitive melanopsin photopigments [32][33][34][35].The non-visual mechanisms are known to affect hormone secretion (e.g.melatonin) [36], sleep [37], circadian rhythms [38], alertness [39] and arousal [40].Thus, the non-visual mechanisms of light may also affect thermophysiology (e.g., via changing hormone secretions), and hence, thermal sensation and comfort.For example, white light with a high CCT and monochromatic blue light at night have been suggested to increase vasoconstriction and core body temperature [41,42].The non-visual effects have been investigated widely at night, but effects during office hours (daytime) attract less attention [9,43,44].Furthermore, some non-visual responses may build up more gradually than the visual responses [10,45,46].As CCT manipulations imply changes in the spectrum, different CCTs may cause differences in thermophysiological responses.Interestingly, non-visual mechanisms would imply that a higher CCTi.e., visually cooler lightis related to higher arousal/alertness and core body temperature [39,40,42], which intuitively contrasts hue-heat hypothesized effects.
The overview in Table 1 also reveals that most studies had a relatively short exposure duration (less than 1 h).Previous studies showed that the exposure durations affect the magnitude of the non-visual effects of light [47][48][49].For example, at night, melatonin suppression was greater with an increase in light exposure duration [47,48].Therefore, with longer exposure duration, non-visual effects that elicit important thermophysiological changes may be more pronounced.It is also important to note that whereas most studies in Table 1 measured perceptive responses to the light (e.g.focusing on the hue-heat hypothesis), only a few of them measured thermophysiology and cognitive performance.Yet, how CCT affects work performance is also of interest to an office scenario for practicing wider ambient temperature ranges.
Although extensive studies have investigated CCT's effects on cognitive performance in the daytime, it remains an open question.Literature has reported positive [50][51][52][53][54][55][56][57], null [53][54][55][56][57][58][59][60][61][62] and even negative effects of high CCT lighting [59,63,64].One systematic review in 2018 reported that in 5 out of 15 cases, a positive alerting effect was found on subjective outcomes, and only one out of six studies reported a positive performance effect [65].A second one also reported extremely mixed results for daytime alerting effects induced by CCT or spectral manipulations [66].This inconsistency may be explained by many factors, including illuminance and spectrum distribution, exposure timing and duration, task characteristics (types and difficulties), participant characteristics (e.g., sex, age, motivation and mental fatigue), and their interactions.The underlying mechanism of light's effects on cognitive performance is often attributed to the lighting domain itself, via both visual and non-visual pathways [67].For example, cognitive Note: 'Result' indicates if the study finds significant relations between CCT and thermal perceptions: 'Yes' (significant effects in favor of the hue-heat hypothesis), 'No' (non-significant effects) or 'Partially' (significant effects in favor of the hue-heat hypothesis appeared only under a specific condition and/or for a specific outcome).No study showed significant effects that were against the direction of the hue-heat hypothesis.
W. Luo et al. performance may vary according to light's effects on workers' abilities to see the fine details in the tasks (visual performance) [68] and the non-visual mechanism of light may induce acute alerting effects, thereby affecting cognitive performance [46].However, CCT may also indirectly affect cognitive performance via moderating thermal comfort when wider ambient temperature ranges are applied, because thermal discomfort has been suggested to impair cognitive performance [3,69], though not definitively [70,71].For these reasons, the current study investigated the effects of CCT in an office-like scenario with relatively long exposure durations on 1) visual perceptions and eye-related symptoms, 2) thermal perceptions, 3) thermophysiology, 4) cognitive performance, 5) alertness, pleasure and arousal.The results are used to discuss the possible underlying mechanisms (both visual and non-visual mechanisms) regarding how CCT affects thermal perceptions.

Method
The medical-ethical committee of Maastricht University approved this study.This study was performed in accordance with the Declaration of Helsinki and was registered at the Netherlands Trial Register (NL9387).Measurements took place between June 2021 and November 2021.

Participants
Sixteen participants (Caucasian, 8 males and 8 females) gave written informed consent and were enrolled.Only females who took hormonal contraception were recruited to minimize hormone influence.All participants met the following inclusion criteria: 1) healthy, no color blindness, non-smoker, 2) age 18-40 years old, 3) a BMI of 18-27.5 kg/ m 2 , 4) normal chronotype, 5) living in the Netherlands (or area near the Netherlands) for at least 2 months.Table 2 presents the participant characteristics.

Experimental protocol 2.2.1. Study design
This study is part of a within-subject, randomized, cross-over experiment consisting of four scenarios tested on separate days that has two investigation arms: the effects of CCT (2700 K and 5700 K) and the effects of personal control of light CCT.In the current paper, only the two CCT scenarios (constant-at-2700 K scenario and constant-at-5700 K scenario) were analyzed to answer the research question regarding the effects of CCT.The order of the scenarios was balanced by a 4 × 4 Latin square design.Each scenario consists of three periods (Fig. 1): Period 0baseline, Period 1 -habituation to high or low CCT, and Period 2testing (for comparisons with the other two personal control scenarios).
The designed CCTs are commonly found in office settings and are similar to the CCTs tested in previous studies [10,15,16,[18][19][20][21].Since the effect of CCT may depend on the illuminance [45], the illuminance remained constant at 500 lux at the eye in all scenarios.The scenarios were tested under constant thermal conditions, where the ambient air temperature was 17.2 ± 0.3 • C, the mean radiant temperature was 17.2 ± 0.2 • C, the relative humidity was 40-48% and the air speed was 0.15-0.26m/s.The thermal condition was intended to create mild thermal discomfort while avoiding obvious shivering and extreme discomfort [72].
The scenarios were performed on four separate days.To avoid possible short-term cold acclimation induced by the study intervention [73], sessions were scheduled at least two days apart.To avoid possible seasonal effects [74], participants completed all four scenarios within 28 days.The participants attended all four scenarios at the same time of the day, either in the morning 9:00-12:00 or in the afternoon 13:00-16:00 to reduce differences in circadian phase between sessions within subjects.Time of the day was balanced across participants.

Lighting system
To eliminate the lighting variation induced by daylight, a windowless room was used.The different light settings were provided using Hue Aurelle Rectangular Lights (Signify) that measure 120 by 30 cm.One Hue Aurelle was placed on the desk facing the participant and additionally one Hue Aurelle was integrated in the ceiling (see Fig. S1).Both lights were used to create the desired light conditions.To mimic daily office activity, a laptop screen was placed in the middle of the desk light.The luminance of the laptop screen was set at a low value.To control for possible influences of light directionality [45], the illuminance contribution at the eye level from the ceiling light was kept at ~80 lux, and the contribution from the luminaires on the desk (desk light and laptop screen) was kept at ~420 lux.Overall, the 2700 K scenario was realized at 2701 ± 39 K and 501 ± 5 lux, and the 5700 K scenario at 5733 ± 39 K and 502 ± 5 lux when different screen activities were performed (Table 3).The detailed mean light characteristics at eye level are shown in Fig. 2 and Table 3.

Preparation session and 24 h standardization before test days
Before the experiment, participants attended a preparation session to familiarize them with the test procedure and study setup.In addition, they practiced the cognitive tasks for 1.5 h to reduce learning effects later during the actual tests.They were encouraged to try their best for the cognitive tasks on the following test days, but they received no performance-dependent monetary incentive or further encouragement on the actual test days.The purpose was to make participants selfmotivated, which applies to most offices.
A day before each scenario, the participants followed a 24-h lifestyle standardization.During the standardization sessions, they were asked to 1) avoid strenuous exercise, alcohol, and coffee 2) consume similar food, 3) refrain from food intake after 22:00 h and after 9:00 h the next day for morning and afternoon test participants, respectively, 4) follow their usual bedtime and sleep duration.The purpose of standardization sessions was to reduce variations in participants' physiological and psychological states on the testing days.

Procedure during test days
The test procedures were identical for all scenarios.A detailed schedule is provided in Fig. 1.The participants arrived in the morning at 9:00 h or in the afternoon at 13:00 h.The last meal before the measurement was standardized and provided by the researchers.Upon arrival, self-reports of last night's bedtimes and sleep duration were collected.Then, wearable sensors were attached to monitor heart rate, skin temperature and physical activity.Participants sat on an office chair and were dressed in standardized clothing (underwear, a long-sleeved shirt, sweatpants, socks and shoes).In total, the insulation was 0.8 clo including the office chair.Energy expenditure, skin temperature, heart rate, heart rate variability and physical activity were continuously measured.
As indicated above the test procedure included three periods (Fig. 1).Period 0 was intended to stabilize participants' states and perform baseline measurements.The participants resided in a respiration chamber for 30 min in a thermally and visually neutral environment (3700 K; 23 • C, PMV = 0, measured mean thermal sensation vote = 0.48 ± 0.73).During this period, participants performed cognitive tasks and completed the experience questionnaire (subjective perceptions of temperature and lighting) and the task questionnaire (subjective perceptions of tasks), followed by blood pressure assessments.
After completion of Period 0, the participants walked into another chamber and were exposed to cold thermal conditions (17 • C, PMV = − 1.65, measured mean thermal sensation vote = − 1.19 ± 0.87) and the experimental CCT setting (either 5700 K or 2700 K).Period 1 lasted for 70 min.The participants began with some video games like Candy Crush for the first 10 min, followed by skin blood flow measurements and the questionnaire probing thermal and visual perceptions ('experience questionnaire').In addition, the skin blood flow measurement and the experience questionnaire, combined with cognitive tasks and a task questionnaire, were scheduled every 20 min at timepoints t = 20, 40, and 60 min (Fig. 1).Blood pressure was measured once at the end of Period 1 (t = 60 min).
The environmental conditions of Period 2 were identical to that of Period 1. Period 2 also started with 10-min video games for relaxation.Skin blood flow was measured and the experience questionnaire was filled in at t = 80, 90, 110 and 130 min, while the cognitive task and task questionnaire were scheduled at timepoints t = 90, 110, and 130 min.Blood pressure and the local thermal perception were measured at the end of Period 2 (t = 130 min).At the end of the last testing day, visual analogue scales were administered to probe participants' colortemperature association strength.

Environmental measurements
Three temperature dataloggers (iButton, DS1922L, Maxim Integrated) were placed near the participants at 0.1 m, 0.6 m and 1.1 m from the ground to measure ambient temperature.The illuminance and CCT of the light were measured in the vertical plane at eye level facing the laptop using a spectrometer (MK350D, UPRtek).The air speed, mean radiant temperature and relative humidity were measured at 0.6 m from the ground before the actual measurements (Almemo 2890-9, Ahlborn),    during which the environmental condition was the same as during the actual measurements with the participants.

Participants' characteristics
Participants' chronotype was assessed by the Munich Chronotype Questionnaire [75].Normal chronotype was defined as a midsleep time between 3:30 h and 5:00 h in this study (average of workdays and work-free days).Body fat content was measured by an air displacement plethysmograph (Bodpod, Cosmed).The body surface area was calculated using the Du Bois formula [76].
Because the strength of color-temperature association may vary strongly among individuals, a warmth semantic differentials question was developed using visual analogue scales (Fig. S2).This question asked participants to rate how (cold:0 -hot:100) they typically perceive blue (HSL: 27, 100, 50) and yellowish-orange (HSL: 200, 100, 50).These two colors were displayed on the laptop screen with white backgrounds and only differ in hue, but with identical saturation and lightness.The individual color-temperature association strength (ICTAS) was defined as the difference between the scores for blue and yellowish-orange colors.

Experience questionnaire
The experience questionnaire consisted of four parts: thermal perceptions, visual perceptions, eye-related symptoms, and the fourth part that includes subjective alertness, pleasure and arousal.
Whole-body thermal perceptions were measured based on the ISO standard 10,551 [77].The thermal acceptance measurements used a binary scale ('acceptable' or 'unacceptable').Whole-body thermal sensation, preference, comfort, pleasantness and self-perceived shivering were assessed using a visual analogue scale (Fig. S3).In addition, local thermal sensation and comfort were obtained, including nine local body parts (head, neck, torso, upper arm, lower arm, hand, thigh, calf and feet, Fig. S3a).The thermal comfort rate was defined as the percentage of the votes equal or higher than 'just comfortable'.
Visual perception scales were adopted from previous literature [10,19] (Fig. S4).The sensation and preference were measured to probe the experienced and preferred light color and intensity.Visual acceptance, visual comfort and pleasantness were assessed without distinguishing light color and intensity because the two sub-dimensions regarding color and intensity have high internal consistency [78].In addition, eye-related symptoms (eye strain, eye discomfort and eye fatigue) were measured using 4-point Likert scales from Viola et al. [51] (Fig. S5a).
The Karolinska sleepiness scale was adopted to measure subjective alertness [79] (Fig. S5b).Moreover, self-assessment manikin scales were used to assess the pleasure and arousal components of affect [80] (Figs.S5c-d).

Physiological parameters
To measure skin temperature at a sampling rate of 1 min, fourteen skin sites were measured using iButtons (DS1922L, Maxim Integrated) according to ISO 9886 standard [81].Moreover, two additional iButtons were placed on the lower arm and middle finger to calculate the underarm-finger gradient, which indicates vasoconstriction [82].
The respiration chamber (Omnical, Maastricht Instruments) continuously gauged participants' oxygen consumption and carbon dioxide production.Based on participants' oxygen consumption and carbon dioxide production, energy expenditure was obtained according to Weir's equation [83], and carbohydrate and lipid oxidation were calculated by Péronnet's and Massicotte's formulas [84].A three-way acceleration meter was placed at the participants' thigh to measure physical activity at a sample rate of 1 min (MOX, Maastricht Instruments).
Heart rate and successive r-r intervals were measured with a chest belt (H10, Polar).The pNN50 index, the percentage of successive heartbeat intervals exceeding 50 ms, was used to indicate heart rate variability.Blood pressure was measured using an automatic blood pressure monitor (HEM-7322U-E, OMRON).A Laser Doppler Flowmetry (PF5000, Perimed AB) was employed to measure hand skin blood flow at the dorsal side of the left hand.Since the laser doppler flowmetry measures a relative value of the blood flow, the measured value of blood flow (Period 1 and 2) was normalized as the measured value divided by the baseline value (Period 0).
For skin temperature and heart rate measures, the mean values during the 10-min interval before the submission of the questionnaire were used as the representative values.For energy expenditure, the data were averaged per 20 min.Because the skin blood flow measurements are sensitive to movement, participants were asked to keep still for 3 min during each skin blood flow measurement.The average value of skin blood flow during the 3-min sitting still was used.

Cognitive performance
Four tasks from Cambridge Brain Sciences were selected to measure different cognitive processes involved in office activities: planning ability (Hampshire tree task), verbal ability (grammatical reasoning task), working memory (digital span task) and mental spatial manipulation (spatial rotation task).These four tasks were constructed based on classical psychological paradigms and have good test-retest reliabilities (readers are referred to reference [85] for the detailed introduction of these four tasks).To prevent possible ceiling and floor effects, the difficulties of the tasks are adjustable, depending on whether the answer is correct or not.A comprehensive performance score was defined as the mean of the standardized scores of the four tasks.

Task questionnaire
The task questionnaire assessed participants' perception of workload and how they interact with the cognitive tasks.The perceived workload was assessed using different aspects: physical demand, mental demand and temporal demand.The interaction between performers and the cognitive tasks included self-perception of performance, effort investments, frustration level and motivation.The physical demand, mental demand, temporal demand, perceived performance, effort, and frustration were measured by visual analogue scales (Fig. S6a), which are from a widely used NASA task load index questionnaire [86].The motivation was measured by a visual analogue scale adapted from Cui et al. [87] (Fig. S6b).

Statistical tests
For the cognitive tasks, scores that exceeded four times the standard deviation from the population mean were considered extreme outlier and therefore excluded from the analysis [87].
Paired t-test was used to test the parameters of the sleep questionnaire in section 3.1 and the warmth semantic differential question in section 3.2.
The effects of CCT were tested by mixed-effects models in section 3.3.The CCT, timepoints and their interactions were treated as fixed factors, and the participant was included as a random intercept.Since the actual ambient temperature (17.2 ± 0.3 • C) differed from the designed 17 • C, it was included in the model as a fixed factor.The baseline measurement in Period 0 was also added as a fixed factor to consider the possible baseline differences.These fixed factors were kept in the model regardless of their statistical significance.Covariates were added in the model, including age, sex, body surface area to mass ratio (heat loss ability [72]), fat-free mass to body surface ratio (heat production ability [72]), day timing and the order of the scenarios.Covariates were kept in the model only if they reached significance.A 'top-down' modelling strategy was adopted: the model started with the maximum model and proceeded with a backwards elimination procedure [88].After reaching the final model, conditional F-tests with degrees of Kenward-Roger's freedom correction were used to test the effects of the predictors [89].For the interest of effects of CCT at each timepoint, post-hoc comparisons were conducted and the p values were corrected using the false W. Luo et al. discovery rate method.The effects of CCT in Period 1 and Period 2 were tested using linear combinations of the coefficients from the final model.Cohen's d was calculated as the estimated effect of CCT divided by the square root of summed variance of the residuals and the random intercept [90].
To explore potential underlying mechanisms of how CCT affect thermal perceptions in section 3.4, relevant variables affected by the CCT in section 3.3 were added as fixed factors to correlate with thermal sensation and/or thermal comfort using mixed linear models.In addition, the participant was included as random intercept and the CCT condition was added as a fixed factor.Cohen's f 2 was reported to evaluate the effect sizes of the continuous variable based on Selya et al.'s method [91].
Statistical analyses were performed under open-source software R 4.2.0, including the LmerTest package [89] and Emmeans package [92].
The assumptions of the mixed linear models were checked.The cut-off value of significance used is 0.05.

Individual variations in the strength of color-temperature association
The warmth semantic differential question revealed that the yellow color was perceived as significantly warmer than the blue color (yellow vs. blue with mean ± standard error: 83.4 ± 2.5 vs. 14.6 ± 3.3, p < 0.001, d = 3.3, Fig. 3).However, visual inspection of the associational strength revealed large inter-individual variation in the strength of this color-temperature association (see Fig. 3).For example, one participant did connect yellow to hot and blue color to cold (strong colortemperature association) whereas another one rated yellow and blue color both rather close to neutral (Fig. 3).The ICTAS ranged from 29.7 to 97.7 (the score difference between the two colors, i.e. the slope in Fig. 3).

Effects of CCT on outcome parameters
The data in the following analyses are shown as 2700 K vs. 5700 K with estimated marginal means ± standard error and significance, for the mean responses over the whole exposure (main effects of CCT), unless stated otherwise.Cohen's d was reported for the effect sizes of CCT.
Moreover, there were clear interindividual differences in visual perceptions.The color sensations ranged from − 0.4 (between 'a bit cool' and 'neutral') to 2.0 ('warm') in 2700 K and from − 3.0 ('very cool') to 0.1 (close to 'neutral') in 5700 K. Similar variations also appeared in brightness sensation, visual preference and visual comfort.Visual perceptions showed no clear relation with exposure duration in the two CCT scenarios (main effects of timepoints: all p > 0.05, Fig. 4a-e).

Skin temperature.
No significant effects of CCT on mean skin temperature and underarm-finger gradient were found (Fig. 6a and b).On the other hand, mean skin temperature decreased over time (main effect of timepoints: p < 0.001, Fig. 6a) whereas the underarm-finger gradient showed a rising trend in Period 1 and remained stable in Period 2 (main effects of timepoints: p < 0.001, Fig. 6b).

Metabolism and physical activity.
Energy expenditure was not significantly affected by CCT in Period 1 (p = 0.371, Fig. 6c).However, in Period 2, 5700 K resulted in 0.112 kJ/min higher energy expenditure compared to 2700 K (5.46 ± 0.15 vs. 5.58 ± 0.15 kJ/min, p = 0.034, d = 0.19, Fig. 6c).Participants' physical activity intensities were constant at 0 for both scenarios, suggesting participants' physical movements were minimized and undetectable.In addition, no statistically significant difference in carbohydrate oxidation and lipid oxidation was found between the two scenarios (Figs.S8a and b).

Cardiovascular responses.
CCT had no significant main effect on hand skin blood flow, heart rate, heart rate variability and systolic blood pressure (Fig. 6d, e and Figs.S8c and d).On the other hand, diastolic blood pressure was 1.9 mmHg higher in the 2700 K scenario (72.8 ± 0.9 vs. 70.9± 0.9, p = 0.028, d = 0.46, Fig. 6f).The heart rate presented a declining trend over time, along with an increase in heart rate variability (main effects of timepoint p < 0.001, Figs.S8c and d).Consistent with the underarm-finger gradient, hand skin blood flow decreased in Period 1 and then remained stable in Period 2 (main effects of timepoints: p < 0.001, Fig. 6d).

Local body perceptions and skin temperature
Local body thermal sensation and skin temperature at t = 130 did not significantly differ between the two scenarios, regardless of local body parts (all p > 0.05, Figs.S9a and c).The local thermal comfort was also not affected by CCT in most of the distal body areas (arms, hands, feet, all p > 0.05, Fig. 7).In contrast, in the core body area (head, neck, torso), thigh and calf, 5700 K resulted in higher local thermal comfort than 2700 K (all p < 0.05, Fig. 7).

Cognitive performance
Performance on the rotation task was 21.2% higher in 5700 K  8d).There were no differences in performance on the digital span, grammatical reasoning and spatial planning tasks between the two scenarios (Fig. 8a, b, c).Overall though, the comprehensive task score (sum of the standardized scores of the four tasks) was 7.2% higher in 5700 K (15.3 ± 0.3 vs. 16.4 ± 0.3, percentage reference: 2700 K, p < 0.001, d = 0.61, Fig. 8d).

Subjective perception of workload and interaction between performers and cognitive tasks
CCT had no significant main effect on mental demand and physical demand (all p > 0.05, Fig. 9a and Fig. S10a).For temporal demand, participants felt more time pressure under 5700 K compared to 2700 K (51.6 ± 1.9 vs. 55.6 ± 1.9, p = 0.010, d = 0.33, Fig. 9b).This difference of temporal demand only occurred in Period 2 (p = 0.008, Fig. 9b), but not in Period 1 (p = 0.296, Fig. 9b).Furthermore, mental and physical demand and did not significantly differ over time (main effects of timepoints: all p > 0.05, Fig. 9a and Fig. S10a) whereas temporal demand increased over time (main effects of timepoints: p < 0.05, Fig. 9b).
The average vote of all measurements in 5700 K generally indicated more effort investment (58.2± 2.4 vs. 61.3± 2.4, p = 0.039, d = 0.23, Fig. 9c) and motivation (− 0.1 ± 0.1 vs. 0.2 ± 0.2, p = 0.014, d = 0.30, Fig. 9d) than in 2700 K.No significant differences were found in perceived performance and frustration levels between the two scenarios (Figs.S10b and c).Moreover, no clear patterns related to time were found in motivation, perceived performance and frustration levels (main effects of timepoints: all p > 0.05, Fig. 9d and Figs.S10b and c).On the other hand, participants' effort investments showed growth in Period 1 and levelled off in Period 2 (main effects of timepoints: p < 0.05, Fig. 9c).

Analysis of potential underlying mechanisms of CCT effects on thermal perceptions
CCT may affect thermal sensation and thermal comfort via different mechanisms, such as hue-heat hypothesis, visual comfort and non-visual effects inducing physiological changes.For hue-heat effects, we showed that CCT significantly affected light color sensation (p < 0.001, Fig. 4a), hence the light color sensation was added to the models for thermal sensation and thermal comfort.For visual comfort, studies [19,30,31] have indicated an association between visual comfort and thermal comfort.Therefore, visual comfort was included when thermal comfort is the outcome variable, although we showed that visual comfort votes were not significantly different between the two CCT scenarios (p = 0.121, Fig. 4e).Moreover, we showed that the perceived shivering was affected by the CCT (Fig. 5d).The (perceived) physiological strain may influence thermal comfort [93,94].Therefore, the perceived shivering was incorporated in the thermal comfort model.The participant was included as random intercept and the CCT condition was also added as a fixed factor.
The model revealed that light color sensation did not show a statistically significant relation with thermal sensation (p = 0.292, Table 4) and thermal comfort (p = 0.590, Table 4).Whereas one unit decrease of perceived shivering was associated with an increase in thermal comfort by 0.717 units (p < 0.001, Cohen's f 2 = 0.20, Table 4).Visual comfort was positively related to thermal comfort (p < 0.001, Cohen's f 2 = 0.05, Table 4).After adding light color sensation, perceived shivering and visual comfort in the thermal comfort model, the direct effects of CCT on thermal comfort became insignificant (p = 0.946, Table 4).

Discussion
The current study investigated the effects of CCT with a relatively long exposure (more than 2 h) in an office-like scenario at 17 • C (2700 K vs. 5700 K).By strict control of possible confounders (such as thermal conditions, illuminance, daylight, daytime and light direction), we observed that CCT did not significantly affect thermal sensation.But unexpectedly, the blue light improved thermal comfort in a cold condition, along with decreased perceived shivering, lower preferred warmth and higher thermal pleasantness.Most of the employed physiological parameters did not respond to CCT.Meanwhile, alertness and arousal were higher in 5700 K than 2700 K. Visual comfort was not significantly affected by the CCT, but in 5700 K participants experienced less eye-related symptoms.Effects of CCT on cognitive performance depended on the task types.On average, the blue-enriched light improved the comprehensive cognitive score, which was accompanied by increases in effort investments and motivation.After incorporating different possible mechanisms of CCT effects on thermal sensation and comfort in the mixed linear model, we found light color sensation was neither associated with thermal sensation, nor thermal comfort.However, thermal comfort showed a correlation with perceived shivering and visual comfort.

Hue-heat hypothesis
According to the hue-heat hypothesis [23], it is expected that 5700 K would result in a cooler thermal sensation than 2700 K.However, our results showed that both whole-body and local thermal sensations were unaffected by CCT at 17 • C.There are several potential explanations.Firstly, the participants performed engaged screen activity in this study.The attention mechanism of our visual system suggests that the brain only processes a small amount of visual information consciously, and that the unattended objects are largely ignored when one object is attended [95].Therefore, some participants might not even have experienced or been aware of the color appearance created by the light CCT when they focused on the screen activity.
Secondly, our result showed the interpretation of color's meaning (i.e., the strength of color-temperature association) varied substantially among participants.Some participants in this study have a weak colortemperature association (Fig. 3).This finding supports the study [29] demonstrating that color-temperature association depends on individual experience.Morgan et al. [29] investigated the color-temperature associations in three age groups (6, 12, 18 years), and reported that the color-temperature association was strongest in the 18-year-old group, and modest in the 12-year-old group, but absent in the 6-year-old group.Their finding suggests that the color-temperature association develops throughout childhood and adolescence, and therefore, it may depend heavily on individual experience.Moreover, light color sensation in our study varied substantially between subjects as well (Fig. 4a), which is also reported in other studies [10,19].This interpersonal variation indicates that 2700 K and 5700 K were not by everyone perceived as 'warm' and 'cool' colors.

How does CCT affect thermal comfort? Potential co-effects of visual and non-visual mechanisms
Surprisingly, 5700 K improved thermal comfort at 17 • C compared to 2700 K (0.26 unite higher, Fig. 5b).This improvement was consistent with the increase of thermal pleasantness (Fig. S7b) and a lower preference for a higher temperature (Fig. 5c).Correlation analyses indicated that whole-body thermal comfort is negatively associated with perceived shivering (coefficient: 0.717, Cohen's f 2 = 0.20) and positively with visual comfort (coefficient: 0.161, Cohen's f 2 = 0.05) but not with light color sensation.The 5700 K resulted in 0.28 units lower perceived shivering (Fig. 5e) and 0.16 units higher visual comfort (Fig. 4e).Therefore, perceived shivering (potential indications of muscle tension) explained 77% of the improvements of thermal comfort (0.28 × 0.717 divided by 0.26) whereas visual comfort only explained 10% of the improvements (0.16 × 0.161 divided by 0.26).Thus, the improvements in thermal comfort were mainly related to reduced perceived shivering at 5700 K.
The reduced perceived shivering in 5700 K in this study was likely caused by non-visual mechanisms instead of visual mechanisms.For the visual mechanisms, the main manipulation of this study is the color appearance of the light (5700 K vs. 2700 K).Notably, the blue color is typically perceived as less exciting than the red/yellow color [96,97], and therefore, less arousal of the blue color (5700 K) might have decreased the perceived shivering.However, this explanation is contrary to our results: 5700 K resulted in higher subjective arousal and alertness than 2700 K. Thus, it is unlikely that the responses of perceived shivering, arousal and alertness in 5700 K are induced by visual effects.
On the other hand, non-visual mechanisms would imply that a higher CCT is related to higher arousal/alertness [39,40,42], which is consistent with our observations.Therefore, the reduced perceived shivering in 5700 K in this study was more likely due to greater non-visual effects than the effects from vision.Interestingly, the decreased perceived shivering in 5700 K in this study is contrary to that of Kompier [10] and te Kulve [19].Kompier [10] et al. tested two CCT (2700 K vs. 5900 K) in combination with two illuminances (100 lux and 1000 lux at eyes) at 18 • C for 45 min (PMV is around − 1.7).They found no statistically significant effects of CCT on perceived shivering.Furthermore, te Kulve et al. [19] even found an increased perceived shivering in 5800 K compared to 2700 K in a mild cold condition (PMV is around − 2.2).The apparent discrepancy may be explained by the differences in melanopic illuminances and exposure durations among these studies (see Table 4).A recent review study [98] suggests that melanopic illuminance (indicated by melanopic equivalent daylight illuminance (EDI)) is the best available predictor for acute non-visual effects and the relation follows a sigmoid function (Fig. 11).With the increase of melanopic illuminance, the non-visual response is indiscernible initially (non-response zone), then rises fast (response zone), and finally saturates (saturated zone, Fig. 11).The melanopic illuminance is affected by the manipulation of both CCT, illuminance, and their interaction, of which illuminance is the most influential factor.This indicates that the non-visual effects of CCT are dependent on the illuminance.When the illuminances are fixed at relatively high or low levels for different CCT scenarios, the melanopic EDI of different CCTs may all fall into the saturated zone or non-response zone, hence the non-visual effects of two CCT scenarios will be similar.For example, if the melanopic EDI range for the response zone is 100 lux-400 lux, the melanopic EDIs of two CCTs in te Kulve and Kompier's studies (Table 5) would be in saturated zone or non-response zone, therefore the non-visual effects of CCTs in their studies would be similar.
Moreover, some non-visual responses may evolve more slowly [10,45,46], suggesting a role of exposure duration.In this study, participants received 140-min light exposure and the effects of CCT on perceived shivering only emerged from t = 60 min onwards.Consistently, the effects of CCT on alertness, mood, arousal and energy expenditure were significant in Period 2 (t = 70-140 min) whereas they appeared to be non-significant in Period 1 (t = 0-70 min).These results indicate that, for some outcomes, the differences in non-visual effects may need some time to build up.On the other hand, the exposure durations were 45 min in Kompier's study [10] and 75 min in Kulve's study [19] (Table 5), which may have been insufficient to develop differences in non-visual effects between the two CCTs.
In summary, we argue that light may affect thermal perception via both visual and non-visual pathways.Given the multiple mechanisms and their characteristics discussed above and in section 4.1, it is not surprising that the literature presents a mixed result (Table 1).In the current study, the non-visual effects of CCT may have played the most influential role in thermal comfort.Therefore, the 5700 K improved thermal comfort compared to 2700 K in this study.

The effects of light CCT on visual perception and eye-related symptoms
We intended to create different color appearances using CCT of light.As expected, CCT affected mean light color sensation, indicating that participants noticed the CCT manipulation.However, 5700 K led also to a brighter sensation compared to 2700 K, although the actual illuminances of the two CCTs were the same.This finding has also been reported in a few other studies [10,20,53,99], though not consistently.In fact, a review of multiple studies concluded that CCT has "negligible effect on ratings of brightness" [100], and suggests that this inconsistency between studies may be attributed to differences in the spectrum that are not captured in the CCT metric.Which additional SPD-based metric might consistently explain findings between studies is, to our knowledge, unknown to date.Visual comfort did not significantly differ between the two CCT scenarios.This finding also appeared from other studies [10,100] observing a null effect of CCT on visual comfort.For example, Kompier et al. [10] demonstrated that the CCT, on average, did not significantly affect visual comfort after a 30-min adaptation to the light exposure.Notably, although the classic Kruithof curve indicates that the effects of CCT on visual pleasantness may depend on the illuminance [101], a thorough review study [100] suggests that, based on nine well-performed and well-reported studies, visual pleasantness is insensitive to CCT based on the empirical data, independent of illuminance.We should note that we of course did not manipulate light level and hence cannot test such interactions in our study.The absence of an effects of CCT on visual comfort may be caused by large individual differences, which is shown in both our paper and previous literature [10,19,102].Nevertheless, participants indicated less preferences for changes in 2700 K compared to 5700 K in terms of color and brightness.Furthermore, 5700 K resulted in less eye-related symptoms compared to 2700 K in this study, which is in line with the results of Viola [51].

The effects of light CCT on cognitive performance
In addition to participants' comfort, higher CCT improved the comprehensive task score by 7.2% in this study.One possible explanation is that higher CCT improved cognitive performance indirectly via   enhanced thermal comfort.However, this is unlikely, partly because the effect of CCT on thermal comfort is rather small in this study, but, more importantly, our previous study using the same cognitive tasks showed that thermal comfort does not affect performance on the employed cognitive tasks in cold conditions [103].Therefore, this improvement in cognitive performance is more likely a direct, so-called acute, non-visual effect of the light manipulation.Previous literature reported mixed results of CCT effects on cognitive performance, including positive [50][51][52][53][54][55][56][57], null [53][54][55][56][57][58][59][60][61][62] and even negative effects [59,63,64].It is difficult to directly compare the current study to previous literature because methodological differences may moderate the effects of CCT on cognitive performance, including light parameters (illuminance and spectrum), exposure timing and duration, tasks (types and difficulties), participant's characteristics, and their interactions.Furthermore, this inconsistency also appeared if we look into different sub-tasks.We found that the 5700 K condition only benefited mental spatial manipulation ability (21.2% improvement) whereas working memory, planning and verbal ability were not responsive to the CCT manipulation in the current study.This finding confirms that the effect of CCT depends on the task types [54,57].It should be noted that in principle the CCT may affect task performance via both visual and non-visual pathways [67].In the current study, the sub-tasks were completed within 12 min and the order of the sub-tasks was randomized.Considering that the employed sub-tasks have similar interfaces and we used a normal CCT range for office applications (2700 K and 5700 K), it is unlikely that CCT differentially affected visual performance or visual experiences during different sub-tasks.
It is also important to understand how the office workers perceive their workload and their interaction with the tasks.Our result showed that the CCT did not affect perceived mental demand, physical demand, perceived performance and frustration level.However, participants did feel more motivated and invested more effort in the task in 5700 K.It should be noted that motivation and effort investment are one of the most powerful determinants of complex cognitive performance [70,87].Therefore, it is highly likely that the increase in motivation and effort investments may have led to the improvements on the comprehensive task score.In addition, a high CCT also resulted in higher perceived temporal demand, indicating greater time pressure perceptions.There are two possible explanations for the increase in time pressure perception.Firstly, exposure to the 5700 K condition elevated subjective arousal level, which may affect time perception [104].Secondly, participants were more motivated in 5700 K, therefore, they might aim for better performance, which likely led to more time pressure perceptions due to the time restriction for the tasks.

Practical implications, limitations and future perspectives
The interaction between light and temperature provides great potential to decrease building energy consumption.The classical theory behind CCT effects on thermal perceptions is the hue-heat hypothesis that suggests using blue light in warm conditions and reddish/yellow light in cold conditions.However, we revealed a much more complex nature of how CCT affects thermal perceptions under mild cold temperature conditions.Both visual effects and non-visual effects also may play a role.
Our results do not support the hue-heat hypothesis in mild cold.However, we cannot exclude the possibility that the hue-heat effects might exist in some populations or contexts, because null effects of CCT on thermal sensation might be caused by the weak color-temperature associations among some of the participants.Future studies are needed to investigate the color-temperature association in various populations and contexts, in terms of its magnitude, direction and variability.Moreover, color-temperature association may be dynamic, therefore, an intervention itself using a specific pair of light color and temperature may change the strength of color-temperature association.For example, repeated exposure to reddish light and cool ambient temperature may attenuate the red-warm association.Thus, it requires more investigations to demonstrate the possible adaptation of colortemperature association in long-term exposures.
On the other hand, the non-visual effects of CCT on thermal perceptions may depend on the illuminance, durations and temperature.As discussed above, non-visual effects likely follow a sigmoid function [98].Therefore, there may be a range of illuminances where CCT can alter non-visual effects.Notably, the illuminance in the building also needs to meet the requirements of visual comfort, visual experience, visual performance and health (e.g.support circadian rhythm).In the current study, we only tested the CCT in mildly cold conditions.It is reasonable to assume that the non-visual effects of CCT on thermal perceptions in warm conditions are different.Thus, it is worthwhile to investigate the CCT in warm conditions with long exposure durations.
In summary, we cannot yet unequivocally recommend how to tune CCT to improve thermal comfort in practice, although we did observe that the 5700 K improved thermal comfort compared to 2700 K during mild cold exposure.Future studies are necessary in quantifying the magnitude of CCT effects on thermal comfort in various contexts (different lighting conditions, thermal environments and populations).Nevertheless, this study provides a possible theoretical framework to understand how CCT affect thermal comfort.

Conclusions
Beyond merely the hue-heat hypothesis, this paper has argued a more complex nature of the CCT effects on thermal perception in office scenarios.We investigated the effects of CCT in mild cold conditions (17 • C) with relatively long exposure durations (>2 h).The main conclusions are as follows: • A large interindividual variation exists in the color-temperature association.The results did not support the hue-heat hypothesis: thermal sensation was not significantly affected by CCT or light color sensation in mild cold.5700 K even improved thermal comfort and thermal pleasantness, and lowered warmth preference compared to 2700 K in a cold condition.In line, local thermal comforts of the core body area were greater in 5700 K than 2700 K. • The thermal comfort was significantly associated with perceived shivering and visual comfort but not light color sensation.• The CCT affects specific cognitive processes.Overall, the blueenriched white light (5700 K) improved the comprehensive cognitive score, and increased effort investments, motivation, pleasure, and time pressure perceptions.• The CCT significantly affected light color and brightness sensation and preference.Participants evaluated the light in the 2700 K condition as warmer and dimmer, and preferred the 2700 K over 5700 K in terms of color and brightness.However, 5700 K resulted in less eye-related symptoms than 2700 K. Visual comfort did not significantly differ between the two CCTs.• 5700 K increased subjective alertness and arousal, which likely indicates 5700 K elicits greater non-visual effects than 2700 K in this study.• Most physiological parameters did not significantly respond to the CCT manipulation in the current study.However, energy expenditure was higher after 1-h exposure in 5700 K than in 2700 K. On average, the mean diastolic blood pressure was lower in 5700 K.
Together, we argue that the CCT may affect thermal comfort via multiple mechanisms in a mild cold condition (17 • C), including both visual and non-visual pathways.Our results suggest a more important role of the non-image-forming process that slowly built up during the experiment.Likely, greater non-visual effects of 5700 K enhanced thermal comfort, alertness, arousal, pleasure, and cognitive performance in mild cold.On the other hand, the effect sizes of different mechanisms may differ in various contexts.Therefore, future studies in various contexts are necessary in quantifying the magnitude of CCT effects.

Fig. 1 .
Fig. 1.Experimental design and testing procedures.Two static CCT scenarios (high/low) included three periods: baseline, habituation and testing periods.The illuminances were kept at 500 lux.The ambient temperature was maintained at 23 • C during Periods 0 and at 17 • C during Periods 1 and 2.

Fig. 2 .Fig. 3 .
Fig. 2. Mean spectral power distribution of the light settings of two CCT conditions (2700 K and 5700 K).Fig. 3. Individual warmth association ratings of blue and yellow colors.The grey lines represent the votes of specific individuals.Dashed lines highlight two participants with the most extreme difference in strengths of color-temperature association.

Fig. 4 .
Fig. 4. Visual perceptions over time: a) light color sensation, b) light color preference, c) light illuminance sensation, d) light illuminance preference, e) visual comfort, f) eye strain, g) eye discomfort, h) eye fatigue.The symbol × indicates p < 0.05, ** indicates p < 0.01, *** indicates p < 0.001, x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the main effects of CCT.(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

Fig. 5 .
Fig. 5. Thermal perceptions over time: a) whole-body thermal sensation, b) whole-body thermal comfort, c) whole-body thermal preference, d) perceived shivering.The symbol × indicates p < 0.05, ** indicates p < 0.01, x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the effects of CCT.

Fig. 6 .
Fig. 6.Physiological responses over time: a) mean skin temperature, b) underarm finger skin temperature gradient, c) energy expenditure, d) normalized hand skin blood flow, e) systolic blood pressure, f) diastolic blood pressure.The symbol x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the effects of CCT.

Fig. 7 .
Fig. 7. Local body thermal comfort at t = 130 min.The symbol × indicates p < 0.05, x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.

Fig. 8 .
Fig. 8. Cognitive task performance over time: a) Grammatical reasoning task score, b) digital span task score, c) spatial planning task score, d) rotation task score, e) comprehensive task score.The symbol ** indicates p < 0.01, x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the effects of CCT.

Fig. 9 .
Fig. 9. Workload perceptions and interactions between participants and tasks over time: a) mental demand, b) temporal demand, c) effort investments, d) motivations.The symbol x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the effects of CCT.

Fig. 10 .
Fig. 10.Alertness and affective states over time: a) alertness, b) arousal, c) pleasure.The symbol x indicates potential outliers outside the whiskers of the box plot, and n. s.Represents non-significance.The p-values for Period 1 and Period 2 refer to the effects of CCT.

Table 1
Studies testing the effects of light on thermal perceptions.

Table 3
Measured mean light settings at the eye level.

Table 4
Analysis of potential underlying mechanisms of CCT effects on thermal perceptions.Significant effects are shown in bold.

Table 5
Comparisons of lighting conditions among studies.