Children’s thermal comfort and adaptive behaviours; UK primary schools during non-heating and heating seasons

This paper aims to study children’s thermal comfort and related Adaptive Behaviours in UK primary schools. The study was carried out in 32 naturally-ventilated classrooms during Non-Heating (NH) and Heating (H) seasons. Alongside collecting environmental data, a self-reported questionnaire and an obser- vation form were employed to record children’s thermal comfort and adaptive behaviours. From eight primary schools, 805 children aged 9–11 were surveyed and 1390 questionnaires were collected. Children’s Thermal Sensation Votes (TSVs), Thermal Preference Votes (TPVs) and adaptive behaviours were compared against temperature offset from comfort temperature by EN 15 251 (T diff = T op -T C (CEN) ). Results sug- gest that children’s thermal comfort (T C (children) ) is 1.9 K and 2.8 K lower than that for adults (T C (CEN) ) during non-heating and heating seasons, respectively. Children have lower comfort temperature and higher sensitivity to temperature changes during heating seasons. This can be attributed to children’s lower practice of personal behaviours and more consistent indoor conditions during heating seasons. The propor- tion of children engaged with personal behaviours is one-third lower during heating seasons. As indoor temperature goes above children’s thermal comfort band, the proportion of children practising personal behaviours increases during non-heating seasons. Around 80% of window operation is carried out by teachers


Introduction
Due to climate change and rise in temperature, maintaining thermal comfort and reducing the risk of overheating in school buildings is becoming a major concern. Children are less resilient to adverse environmental conditions compared to adults, therefore, unacceptable environmental conditions affect them more significantly [1] . Reducing the risk of overheating and improving thermal environment in schools improve children's health, well-being, productivity, academic performance [1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16] and affects ener gy consumption [17][18][19] . It is shown that when classrooms' indoor temperature exceeds 23.9 °C students' respiration rate increases, which provides conditions for some other diseases [20] . High temperatures cause sluggishness, tiredness [21] , fatigue and reduced concentration [22,23] . It is shown that by reducing classroom temperature from 25 °C to 20 °C, task speed of 10-12 years old children increases by 2% per 1 °C reduction in temperature [1] . Similarly, by 1 °C reduction in temperature, academic performance in stan-dardized tests improves by 2-4% [5,24] . In another study, performance of 11-12 years old children who were exposed to temperatures of 20 °C and 30 °in the morning and afternoon was lower for higher temperatures and afternoon sessions [25] . Hence, concerns over thermal environment of primary school classrooms are growing [26] .
To improve thermal environment in primary schools, it is vital to estimate comfort temperatures. According to Nicol, Humphreys and Roaf, (2012), "Comfort temperature or the neutral temperature is the temperature at which the largest number of participants will be comfortable" [18] . Comfort temperature is also defined as "the operative temperature at which the average person will be comfortable" [27] . According to ANSI/ASHRAE, (2013), "Thermal comfort is the condition of mind that expresses satisfaction with the thermal environment and is assessed by subjective evaluation" [28] .
Comfort temperature varies in different studies under different climatic conditions around the world. In temperate climate of England, comfort temperature of 11-16 years old children is found 16.5 °C during winter [29] and 19.1 °C during summer [30] . Furthermore, comfort temperature of 7-11 years old children is found 20.5 °C during spring [31] . In temperate climate of Korea, comfort temperature is found 22.1 °C for 4-6 years old children durhttps://doi.org/10.1016/j.enbuild.2020.109857 0378-7788/© 2020 Elsevier B.V. All rights reserved.
ing spring [32] . In subtropical Australia, comfort temperature is found 24.2-24.5 °C during winter [33] and 22.5 °C during summer [17] for primary and secondary school children. In subtropical Taiwan, comfort temperature changes from 23 to 24 °C in [34] and from 22.4 to 29.2 °C in [35] for 11-17 years old students during Autumn. In subtropical China, comfort temperature is reported at 20.9 °C during summer [36] . In tropical locations, comfort temperature increases up to 26.8 °C in Hawaii, US [37] and up to 28.8 °C in Singapore [38] . In another study done in Iran with warm dry summers and cool winters, comfort temperature of 10-12 years old children is found 23.3 °C during summer [39] .
According to De Dear and Brager et al. (1998), differences in thermal comfort are related to occupants' physiological (acclimatization), psychological (expectations) and behavioural (clothing adjustments) adaptations [40] . Behavioural thermoregulation affects heat balance between human body and surrounding thermal environment [38,40] through change in clothing layers, posture, metabolic rate, location or use of buildings' controls [18] . According to Nicol et al., "If a change occurs such as to produce discomfort, people react in ways that tend to restore their comfort'' [18] . This reaction is either 'Personal Behaviour' with the occupants adapting to the building or 'Environmental behaviour' with the occupants adapting the building to suit their preferences. Adaptive Behaviours influence classrooms' environmental quality and school occupants' comfort significantly [41][42][43][44] . Therefore, adaptive behaviours should be facilitated in schools to achieve higher comfort levels for children [45] .
Change in occupant behaviour as one of the actions to mitigate the risk of overheating is proposed by the UK National Adaptation Programme (NAP), 2018 [46] . Therefore, a clear understanding of both environmental and personal behaviours in schools is required under various climatic conditions. This study aims to investigate children's perception of classrooms' thermal environment and estimate their comfort temperature in relation to the existing adaptive comfort models. It also examines children's personal and environmental adaptive behaviours as a response to thermal discomfort during non-heating and heating seasons.

Methodology
This paper focuses on the relationship between thermal comfort and related adaptive behaviours when thermal environment is not within acceptable limits. The four main steps in this methodology are 1. Selecting samples, 2. Recording personal and environmental behaviours in relation to indoor environmental conditions, 3. Calculating comfort temperature and 4. Overviewing recorded data.

Sample selection
To investigate adaptive behaviours without any bias, samples were selected with specific attention to the a) climate in which buildings were located, b) buildings and their neighbourhood, c) controls within the buildings and d) children's age range.

Climate
To reduce the biased impact of extreme climates on children's behaviour, schools need to be selected from a mild climate. Therefore, Coventry as the second-largest city in the West Midland with a mild climate according to Koppen classification [47] was selected. The study was carried out from mid-July 2017 until the end of May 2018 to include a wide range of weather conditions. Table 1 shows the range of environmental variables during heating and non-heating seasons. During school's occupancy (9:00-15:30), outdoor air temperature ranged from 0.7 °C to 25.10 °C, relative humidity changed from 43% (RH) to 94% (RH) and air speed changed from 0.05 m/s to 9.6 m/s, Table 1 . Outdoor variables were taken from local weather stations that were maximum 3 miles away from each field study site [48] .

Buildings
To increase occupants' window operation, naturally-ventilated schools were selected in this study. Window opening can be restricted in naturally-ventilated schools that are located in neighbourhoods with a high background noise level [49,50] . To allow window operation without impairing acoustic comfort, schools were selected in quiet areas with a considerable distance to the main road. The regional Road Noise, LAeq 16 h, is less than 55 dB in all selected schools according to England Noise Map Viewer [51] . This is the maximum acceptable external noise level that allows natural ventilation [52] . Furthermore, to not restrict window opening due to outdoor pollution, all schools were selected in areas with low Daily Air Quality Index (DAQI) according to Air pollution Forecast provided by the Met Office [53] . In total, 32 naturallyventilated classrooms in 8 primary schools were selected and studied during non-heating (NH) and heating (H) seasons, Table 2 .

Windows
To categorize occupants' interaction with windows, schools with various window characteristics were selected after on-site visits and visual observations. Based on a comprehensive literature review on factors affecting window operation, selected classrooms were classified to ones with high and low opportunities for window operation. Review suggests that windows' ease of use [45,[54][55][56] and access and proximity to windows [45,[57][58][59][60][61] facilitate windows' operation. Windows at low heights that are manuallyoperated and accessible by children can provide more opportunities for children's window operation [45] . Windows at different levels (high and low-level openings) and sizes (small and large) can provide thermal comfort and different kinds of ventilation [45,50,[62][63][64][65] , therefore, they are operated more frequently to address different aspects of comfort.
Therefore, schools that provide high opportunities for window operation (Schools 1, 2 and 5) have many numbers of windows (8) in two different sizes and levels, have a low windowsill ( ≤1 m), are manually operated and are located within the length of the classroom. In this study, 18 classrooms provide low opportunities for window operation and 14 classrooms provide high opportunities for window operation, Table 2 .
Figs. 1 and 2 show classrooms with high and low opportunities for window operation. Fig. 1 shows a classroom with openings at two different sizes and levels that can be operated manually alongside the length of the classrooms. Fig. 2 shows a classroom with 5 small openings at high windowsill (1.7 m) located at the end of the classroom. Due to the potential impact of blinds on resisting airflows [50,66] , the study considers the impact of blinds on obstructing window open area. This study obtained its ethic approval before the start of the project and all ethical considerations were followed during field study, including getting consent from heads, teachers and children.

Occupants
To study adaptive behaviours of primary school children, it is important to select an age group that has a clear perception of environmental conditions. In this study, 9-11 years old children were targeted for two main reasons. 1) Primary school children in their late middle childhood (9-11 years old) compared to their peers in early middle childhood (6-8 years old) are more likely to operate controls because of their height. Children's heights were derived from UK-World Health Organisation (WHO) growth charts; average height of 9-11 years old children are reported to be 133, 138 and 144 cm, respectively [67] . Another study suggests that older children have more freedom to operate controls whereas the younger children are supervised more strictly inside the classrooms [68] .
2) Children in their late middle childhood (9-11 years old) compared to their peers can provide more valid responses to a structured questionnaire. They also have more developed language and literacy skills [69] , cognitive abilities [70] and attention span [71] . Children at this age compared to their peers think more productively and evaluate facts better [71] , which can increase data quality and consistency of findings [69] .

Environmental variables and adaptive behaviours
Personal and environmental adaptive behaviours and simultaneous environmental measurements were conducted in selected classrooms.

Environmental measurements
Environmental variables affecting thermal environment and adaptive behaviours were recorded at 5-min intervals by multifunctional SWEMA equipment [72] and standalone data loggers [73] . Table 3 shows details of the environmental equipment with their range, resolution and accuracy. SWEMA equipment, designed to comply with ISO 7726 [74] and ISO 7730 [72,75] standards, collects data from three sensors: 'air velocity and temperature', 'humidity and temperature' and 'radiant temperature' (globe thermometer Ø 150 mm). The location of the sensors varies in each classroom with regards to children's health and safety and the setup criteria. A measurement station was located at a height of 1.1 m as recommended by ISO 7726 [74] , away from heat sources (e.g. projectors), main airflows (e.g. windows) and sun patches. Equipment was placed within vicinity of children's desks without impairing their safety, seating arrangement or visual comfort. For instruments' acclimatization to the classrooms' thermal environment [18] , they were usually set up before children's arrival in the morning. To record state of windows and doors, time-lapse cameras were installed inside the classrooms alongside visual observations by the lead author.

Thermal perception and adaptive behaviours
To record thermal perception and related adaptive behaviours, a reliable and valid method which was validated by the authors [76] was employed. In this method, children were surveyed on personal adaptive behaviours including fanning, drinking and clothing through a self-reported questionnaire, Table 4 . Children and teachers' interactions with windows and doors were recorded using the observation form, Table 4 . Children were surveyed on their thermal sensation and preference by 5-point rating scales as (Cold, Cool, OK, Warm, Hot) and (Warmer, A little warmer, As it is, A little cooler and Cooler).
To record all adaptive behaviours during sessions, children were asked to fill out the paper-based questionnaire at the end of sessions. Children maintained a stable activity level at least 30 min before filling out questionnaire, as suggested by Goto et al., 2002 [77] . In total, 805 children were observed, and 1390 questionnaires were collected during field studies.
Schools in UK require pupils to wear uniforms which can restrict available clothing choices [78] , therefore, children have a specific range of school uniform options [31] . However, children in this study could wear a seasonal variant of the uniform if they wished to. Children's clothing uniform was surveyed ( Table 4 ), however, Top part of clothing uniform is not questioned as 'shortsleeve shirt/blouse' and 'light-weight long-sleeve shirt/blouse' have similar Clo values [75] . Clothing values in Table 5 were estimated according to ISO 7730 [75] by considering children's fixed layers (i.e. worn for the whole day) and adjustable layers (i.e. Jumper/cardigan) [76] . All combinations include underwear, and when jumper/cardigan is worn, 0.25 is added to Clo value [31] . Table 5 shows uniform combinations in studied classrooms with a total of eight different Clo values.
EN 15251 [79] adopts exponentially weighted running mean temperature (T rm ) that considers the significance of temperatures based on their distance in the past from Eq. (1) : where constant α is 0.8, T od−1 is the daily mean outdoor temperature for the previous day; To d-1 is the daily mean outdoor temperature for the day before that and so on [18] . Comfort temperature according to main studies on adaptive models [18,83] EN 15251 considers different building categories; Category I with high expectations for sensitive and vulnerable occupants, Category II for normal expectations in new or renovated buildings, Category III for moderate levels of expectation in existing buildings [79] . Eqs. (4) - (6) show the calculation of comfort temperatures in Building Categories I, II and III.

Overview of the recorded data
Outdoor (T out ) and indoor operative temperature (T op ) at the time of filling out questionnaires, day's running mean temperature (T rm ), comfort temperature predicted by EN 15251 [79] (T C (CEN) ), temperature offset form comfort temperature 'T diff = T op -T C (CEN) ', mean Thermal Sensation votes (TSVs) and mean clothing values (Clo) are presented in Table 6 to characterize classrooms' thermal environment and children's thermal perception. Fig. 3 shows the percent of children in each category of TSVs and thermal preference votes (TPVs). Around 15% of the children during non-heating seasons and 14% during heating seasons are overheated (i.e. proportion of children who feel warm or hot and prefer a cooler classroom).

Children's comfort temperature (T C(children) ) vs en comfort temperature (T C(CEN) )
To investigate children's adaptive behaviour as an action to reach thermal comfort, there is a need to discover children's thermal comfort (T C (children) ) . The Equations by EN 15251 [79] for optimum comfort temperature were developed based on data collected from office workers in the SCATs project [18] . Therefore, predicted comfort temperature estimates adults' comfort temperature (T C (CEN) ) more reliably than that for adults. Evidence shows that outdoor climatic conditions affect thermal adaptation to indoor conditions significantly [79] . Therefore, the distance between indoor operative temperature (T op ) and the day's comfort temperature by EN 15251 (T C (CEN) ) [79] (T diff = T op -T C (CEN) ) is considered as the criteria for suggesting children's comfort temperature (T C (children) ) . Applying 'T diff ' for estimating comfort temperature is supported in similar studies exploring children's comfort temperature at schools [44,86] . Children's mean TSVs and TPVs for each survey were compared with 'T diff ' to provide a more detailed presentation of results. The method to calculate comfort temperatures is presented in the following three steps: Step 1) The difference between T op and the day's comfort temperature predicted by EN 15251 adaptive model [79] (T diff = T op -T C (CEN) ) was calculated. T diff values greater than 0 account for temperatures higher than comfort temperature predicted by EN 15251 and T diff values lower than 0 account for temperatures lower than comfort temperature by EN 15251.
Step 2) The proportion of children with Warm Sensation (i.e. 0 < TSV, the one who voted Warm or Hot), Cool Sensation (i.e. TSV < 0, the ones who voted Cool or Cold) and Neutral sensation (i.e. TSV = 0, the ones who voted OK) was calculated for each classroom and plotted against corresponding T diff , Figs. 5 and 6 . The intersection point of 'Warm sensation' and 'Cool sensation' graphs is the point at which proportion of children feeling warm and feeling cool is similar. Indeed, it introduces the point at which equilibrium is reached. To suggest this point as T C(CHILDREN) , the proportion of children feeling 'OK' should approximately be maximum at this point.
Step 3) . Similarly, the proportion of children with Warmer preference (i.e. 0 < TPV, the ones who preferred a bit warmer or warmer classroom), Cooler preference (i.e. 0 > TPV, the ones who preferred a bit cooler or cooler classroom) and 'As it is' preference was calculated and plotted against the related T diff , Figs. 5 and 6 . Similarly, the intersection point of 'Warmer preference' and 'Cooler preference' graphs suggests the 'preferred temperature'. At this point, the proportion of children preferring the classroom 'as it is' should approximately be maximum. This approach is supported in similar stud- Table 4 Questions on thermal perception and adaptive behaviours from questionnaire and observation Forms [76] .

Variables
Questions Scales and coding Thermal Comfort and Personal Behaviours (Method: Questionnaire) How do you feel now?
Hot ( + 2) How would you like the classroom to be now?     ies [17,34,86] that show intersection point of 'Want warmer' and 'Want cooler' probit models as the preferred temperature. Non-heating seasons: As it can be seen in Fig. 5 , the intersection point of warm sensation and cool sensation curves is at T diff = −1.9 during non-heating seasons, 1.9 K cooler than T C (CEN) . It represents the point at which 30% of the children have Cool sen-sation and 30% have Warm sensation. The proportion of children with neutral sensation (the rest 40%) is at its peak at this point, Fig. 5 . Similarly, the intersection point of warmer and cooler preference curves is at T diff = −0.8 where 34% of children prefer cooler, 34% prefer warmer and 32% prefer 'As it is', Fig. 5 . As shown in Fig. 5 , there is a 1.1 K difference between children's comfort and preferred temperature. However, this difference is still within 4 K distance between upper and lower margin of comfort band by EN 15251 for Category I buildings (T C (CEN) = 0.33T rm + 18.8 °C ± 2). T C (children) suggested in this study which happens at T diff = −1.9 is close to the lower margin of the comfort band predicted by EN 15251 (T diff = −2). Therefore, comfort temperature by EN 15251 (T C (CEN) ) overestimates children's comfort temperature (T C (children) ) by 1.9 K during non-heating seasons. At comfort temperature by EN 15251 (T diff = 0), the percentage of children who feel warm increases to 40% and the percentage of children who feel OK starts to decline ( Fig. 5 ). At upper limit of comfort band predicted by EN 15251 for Category I buildings (T diff =+ 2), more than 50% of children feel warm or hot and prefer a cooler classroom, Fig. 5 .
Heating seasons: As can be seen in Fig. 6 , the intersection point of warm and cool sensation curves is at T diff = −2.8 during heating seasons, 2.8 K cooler than T C (CEN) . It represents the point at which 30% of the children have Cool sensation, 30% have Warm sensation and the rest 40% have neutral sensation. At this point, the proportion of children having neutral sensation is approximately at its maximum. The intersection point of warmer and cooler preference curves is at T diff = −2.4 where 34% of children prefer cooler, 34% prefer warmer and 32% prefer 'As it is', Fig. 6 . At comfort temperature predicted by EN 15251 (T diff = 0), the proportion of children who feel 'warm or hot' increases to 47% and the proportion of children who feel 'OK' declines to 35%. The results confirm that comfort temperature predicted by EN 15251 (T C (CEN) ) overestimates children's comfort temperature (T C (children) ) by 2.8 K during heating seasons.
When T op equals to T C (CEN) (T diff = 0), the proportion of children who have warm sensation is higher during heating seasons (47%) than during non-heating seasons (40%). At T diff = 0, children are 12% and 20% more likely to prefer a cooler classroom than a warmer classroom during non-heating and heating seasons, respectively, Figs. 5 and 6 .     [17] .
Sensitivity: The regression slope is a measure of sensitivity to temperature changes [82] . The gradient of regression equation for linear models is inversely proportional to the adaptability of the building occupants [17] . A shallow gradient shows that subjects adapt more effectively to room temperature and accordingly their votes do not change quickly [17,39] . Figs. 8 and 9 suggest that children's adaptability to temperature changes is higher during nonheating seasons because slope of linear model is shallower during non-heating seasons ( b = 0.09) than heating seasons ( b = 0.14).  8 and 9 show that a temperature change of 11.1 °C is required to shift one score on thermal sensation scale during non-heating seasons, however, this change is 7.7 °C during heating seasons.

Personal adaptive behaviours
Children in schools adapt themselves to the environment by a number of personal adaptive behaviours including changing clothing level [43,[87][88][89][90] , changing activity type and posture [18,91] , drinking and fanning [42,44] . In this study, clothing, drinking and fanning behaviours of children were investigated by applying a questionnaire that was validated by the authors [76] . ' Cooling personal adaptive behaviours' in this study refer to all personal actions that children adopt to reach a cooler sensation. Fig. 10 shows proportion of cooling personal adaptive behaviours such as fanning, drinking cold water or not wearing jumper/cardigan. Fig. 10 shows that proportion of children who practice two and three cooling personal behaviours (total of 45.7%) is higher during non-heating seasons and proportion of children who do not practice any cooling adaptive behaviour is higher during heating seasons (39.4%) and Fig. 11 shows the breakdown of cooling personal adaptive behaviours during non-heating and heating seasons. When children practice only one cooling personal adaptive behaviour, drinking cold water is the most frequent one, followed by taking off jumper/cardigan, Fig. 11 . This is mainly because children have cold drinks frequently during breaks, after or before PE and assembly. When two personal behaviours are practised, the combination of having cold drink and removing jumper/cardigan has the highest frequency.
Children's Clothing Behaviour: To investigate how children's sensitivity for adaptive behaviours change in relation to comfort temperatures, Spearman correlation tests were run between 'clothing values' and 'T diff '. Spearman Correlation which is a test to examine the relationship between an ordinal variable with skewed dependent variable [92,93] is used in this study.
Non-heating seasons: Children's clothing values and T diff are significantly correlated during non-heating seasons (Spearman Correlation coefficient = −0.3, P < 0.001). Fig. 12 shows that by increase in T diff , the proportion of children wearing lighter levels of clothing [Clo value = 0.3 and 0.39] increases significantly and the proportion of children wearing thicker layers of clothing decreases. At children's comfort temperature (T diff = −1.9 K and T C (children) = 22.9 °C), average Clo value is around 0.58, however, it decreases to 0.38 when T op is 6 K higher than T C (children) (T diff = 4 K and T op = 28 °C), Fig. 12 . Previous studies confirm that children's clothing level is correlated with running mean temperature, sequence of temperature, long term fluctuation in temperature [43,[87][88][89] and operative temperature [32,89] .
As can be seen in Table 7 , the proportion of children with a certain clothing value within the comfort band (T C (children) ±2 K) is more stable than that outside of the comfort band (T diff > T C (children) + 2 K). Standard Deviations (SDs) are significantly lower within comfort band than outside of it, Table 7 . This finding confirms the suggested comfort band for children in this study. At upper limit of comfort band (T diff = T C (children) + 2 K), there is a turning point in the proportion of children who follow a certain clothing behaviour, Fig.  12 . At this point, the proportion of children who follow category A (the lightest level of clothing) starts to increase, however, the proportion of children who follow categories B, C and D (the heavier levels of clothing) starts to decrease. According to Fig. 5 , the proportion of children with warm sensation and cooler preference at the upper limit of comfort band (T diff =+ 0.1 and T op ≈23 °C) is 42%, however, the proportion of children with the lightest clothing level is only 20% at this point. This suggests that higher proportion of children could potentially achieve thermal comfort at this point by adopting personal adaptive behaviours.   Heating Season: Children's clothing values and T diff are correlated during heating seasons (Spearman Correlation coefficient = −0.1, P < 0.01). However, the correlation is less significant than that during non-heating seasons because most of the children (76%) have the same clothing values (0.72 or 0.74) during heating seasons, Fig. 4 . Fig. 13 shows that by in-   heaviest clothing level) drops only by 8%. At this point, the proportion of children with warm sensation and cooler preference is 43%. When T op is 6 K higher than T C (children) [T op ≈28 °(NH) and T op ≈26.5 °C(H)], average Clo value decreases 0.2 and 0.03 during non-heating and heating seasons, suggesting that children make fewer changes to their clothing uniform during heating seasons.
Cooling Personal Behaviours: The probability of practising cooling personal behaviours differs at different tem perature intervals during non-heating seasons. Besides clothing adjustment, having cold drinks and fanning are also investigated as cooling per-sonal behaviours in Fig. 14 . When children feel in discomfort, the proportion of them having cold drink is the highest, followed by choosing lighter levels of clothing and then fanning, Fig. 14 . The proportion of children having cold drink is always high irrespective of temperature changes because having cold drink can be related to several other factors such as occupancy patterns, activity levels and thirst. The results show that T diff is a statistically significant predictor of fanning (Logistic Regression coefficient = −0.57, P < 0.001) during non-heating seasons, however, it is not a predictor of fanning ( P = 0.74) during heating seasons. Logistic regression

Table 8
Changes in proportion of children engaged with personal adaptive behaviours.

Changes in proportion of children engaged with personal adaptive behaviours Within Comfort Band
Outside of Comfort Band when is suitable for testing relationships between a categorical outcome variable and one or more categorical or continuous predictor variables [94] . The proportion of children engaged with cooling behaviours has a turning point at the upper limit of comfort band (T diff = T C (children) + 2 K, T op ≈23 °C). The speed of children's engagement with cooling behaviours within and outside of the comfort band is shown in Table 8 . The speed of engagement is higher outside of the comfort band than inside of the band. The speed of engagement with clothing behaviour is higher than that with fanning and drinking behaviours, especially outside of the comfort band.

Environmental adaptive behaviours
Window Operation: Window operation as one of the most important environmental behaviours [95] was recorded using nonparticipant observation method which was validated by authors [76] . Results show that teachers or teacher assistants undertake around 78% of windows' adjustments, Fig. 15 . Children carry out another 5% of adjustments which are requested by teachers. Around 16% of window operations are carried out directly by children and 2% of them are requested by children, Fig. 15 . In total, 82% of operations are carried out based on teachers' perception of thermal environment and 18% are done based on children's perception. Hence, teachers and teacher assistants are mainly in charge of operating windows, as supported in previous studies [31,44,96,97] . In only three of the studied classrooms (10%) children were encouraged on environmental adaptive behaviours. among window operations done by children, 87% of adjustments were done in Schools 1, 2 and 5 that have high potentials for window operation, Table 2 .
Window Opening Temperature (WOT): To investigate how window operation in classrooms is related to thermal discomfort, window opening temperature (WOT) is compared with T C (CEN) and T C (children) , Fig. 16 . Temperatures at which windows were opened upon teacher's arrival to the classroom were removed from the database. A total number of 35 window openings during nonheating seasons and 20 window openings during heating seasons are presented in Fig. 16 . Results show that among 97% of the cases during non-heating seasons and 80% during heating seasons, WOT is higher than T C (children) , Fig. 16 . However, among 63% of the cases during non-heating seasons and 20% during heating seasons, WOT is higher than adults' comfort temperature (WOT > T C (CEN) ). Table 9 shows that among 63% of the cases during non-heating seasons and 50% during heating seasons, the difference between WOT and T C (children) is more than 2 K (WOT-T C (children) > 2 K). Hence, more than half of the windows are opened at a temperature that is outside of the children's comfort band during non-heating seasons. However, almost all windows are opened within adults' comfort band during non-heating seasons (Only in 3% of the cases, WOT-T C (CEN) > 3 K). This indicates that WOT follows teachers' thermal perception rather than children's thermal perception.
Open Area vs Comfort Temperature: The probability of opening windows as a function of thermal discomfort is estimated via calculating percent of open areas at 10-min intervals against T diff ,   Table 9 Relation between WOT, T C (CEN) and T C (children) .

Discussion
This study investigated thermal comfort and adaptive behaviours of primary school children during heating and nonheating seasons. The main findings of the study are listed below:

Children's comfort temperature
This study suggests T C(children) of 20.9 °C during non-heating seasons and 20.2 °C during heating seasons which are 1.9 K and 2.8 K cooler than comfort temperature predicted by EN 15251 (T C (CEN) ). A similar study on 7-11 years old children in UK suggests comfort temperature of 20.5 °C during spring [31] . In a study in Australia during summer seasons, thermal comfort is found to be 1.5 K and 0.8 K cooler than comfort temperature predicted by ASHRAE in primary and secondary schools, respectively [86] . In another study in primary schools in the UK during summer, the proportion of children who feel comfortable and OK is the highest at T diff = −3 [44] . In a study in kindergartens in Korea from June to May, children's comfort temperature is 0.5 °C and 3.3 °C lower than that for adults during summer and winter, respectively [89] . In another study in elementary and high schools in Taiwan from September to January, comfort temperature is 1.7 °C lower than that recommended by ASHRAE [34] .
Children's thermal comfort in this study is lower than that for adults that is also supported in similar studies [17,31,34,39,44,86,102] . The discrepancy between children's and adults' comfort temperature can be explained by children's more limited adaptive behaviours [19,39] and their physical and physiological differences [31,32,34,[103][104][105][106][107][108][109][110]111] . The main physical difference between children and adults affecting thermoregulation is children's higher surface-area-to mass-ratio [103,109,110] , which results in a higher rate of heat absorption or loss [103] . The main physiological differences are children's higher metabolic rates per body weight [32,89] and children's lower sweating rate [103,111] . Therefore, children are more sensitive to higher temperatures [31,89] and they have a higher sensitivity to core temperature changes [110] .
Results of this study show that children's comfort temperature during heating seasons is lower than that during non-heating seasons, as supported in similar studies in educational buildings [89,107] . Having higher comfort and preferred temperatures during non-heating seasons can be related to children's more practice of personal adaptive behaviours and exposure to more variant environmental conditions during non-heating seasons. Results show that children's preferred temperatures are 1.1 K and 0.4 K cooler than their comfort temperatures during non-heating and heating seasons. This discrepancy indicates that comfort temperature does not necessarily represent the preferred temperature of occupants, as supported in [31,86] .

Adaptive behaviours
Children practice personal adaptive behaviours more than environmental behaviours in this study; around 90% of the children during non-heating seasons and 60% during heating seasons practice at least one cooling personal adaptive behaviour while only around 16% of window operations are done by children. A similar study in UK primary schools during non-heating season shows that 74% of children adopt personal behaviours and 19% adopt environmental behaviours [44] .
• Personal Behaviours: The proportion of children who adopt personal adaptive behaviours starts to increase when classroom temperature goes above children's comfort band during nonheating seasons. By 2 K increase from comfort temperature (at T op ≈23 °C (NH) and T op ≈22.3 °C (H)) more than one-third (42-43%) of the children feel 'warm or hot' with 'a bit cooler or cooler' preference. However, less than one-fifth of children have chosen lighter clothing levels at these temperatures. This suggests that children in discomfort could potentially be reduced by adopting effective personal behaviours. Around 40% of children during heating seasons and 12% during non-heating seasons practice no personal adaptive behaviours. These children need to be encouraged to adopt effective personal behaviours when feeling overheated, noting that 15% of children are overheated in this study. There are circumstances that restrict children's personal behaviours in schools, such as school dress codes, social background [39,44,97,112] or limitations in modifying activity levels during teaching periods [39] . According to Fig. 12 , children's personal behaviours start to increase significantly outside of the comfort band (T diff > T C (children) + 2 K), suggesting that children are uncomfortable outside the 4 K band. Therefore, children's comfort band should not exceed 4 K which is also recommended by EN 15251 for category I buildings that accommodate vulnerable occupants. Another study supports that children have relatively smaller ranges of thermal comfort compared to adults [89] . • Environmental Behaviours: This study shows that operation of windows is mainly carried out by teachers (up to 77%), also supported in [31,44,96,97] . Children are usually passive recipients of classroom conditions rather than active users [39] .
One of the reasons that teachers usually decide for the entire classroom is that practising environmental adaptive behaviours in shared spaces with many occupants can be challenging, as supported in [45,58,113,114] . Children might disagree over preferred environmental behaviour, especially if one's adaptive behaviour results in someone else's local discomfort. This problem can be solved to some extent by providing more local controls [45] . There is a direct link between children's perception of thermal environment and their related adaptive behaviours [87] ; more opportunities to control the environment make occupants more tolerant of uncomfortable conditions [82,90,115] . Therefore, lack of opportunities for controlling classroom environment results in students' increased level of dissatisfaction, especially at higher indoor temperatures [32,39] .
The study highlights that windows' operation (i.e. WOT and the proportion of open window) is based on teachers' thermal perception. The proportion of open area is higher within adults' comfort band than children's comfort band. Furthermore, the difference between WOT and T C (children) is more than 2 K in more than half of the cases. This difference can be explained by following reasons: First, classrooms are mainly controlled by teachers who have higher comfort temperatures than children [17,31,34,39,44,106,116] . Second, children's reaction to the rise of temperature is slower than that for adults as children are less sensitive to temperature changes [39,78,88,117] . This is because children have faster heat loss rates [109] and higher metabolic rates [105,108] . Third, opportunities for practising effective environmental adaptive behaviours are not sufficiently provided for school children [45] . Fourth, teachers do not encourage children to engage in environmental adaptive behaviours. Fifth, teachers are not fully aware of their differences with children in perceiving thermal environment.
The findings suggest when T diff > −2.5, the proportion of open area in relation to T diff is higher during heating seasons, Fig. 17 . This can be explained by the following reasons: 1) School occupants practise fewer personal adaptive behaviours during heating seasons, therefore, environmental behaviours are adopted first. The sequence of adaptive behaviours can potentially be more efficient by adopting personal adaptive behaviours as the first reaction to thermal discomfort instead of opening windows at low outdoor temperatures during heating seasons. There is evidence that the sequence of practising adaptive behaviours can change energy consumption of the buildings [118] . 2) Windows in this study are opened at lower temperatures during heating seasons to improve indoor air quality, as supported in [12,98,[119][120][121] . This can compromise thermal comfort by letting draughts in [59,119,122,123] and result in heat loss and waste of energy [9,124] .
3) The temperature at which heating systems are operated during heating seasons can result in occupants' thermal discomfort and accordingly window opening. Therefore, heating setpoints need to be revised to provide children's pleasant thermal environment, reduce the number of overheated children and save energy. A similar study suggests that if students' comfort temperature is used for classrooms' heating, 12% of heating energy can be saved [125] .
Children's climatic adaptation to coldness should be considered in running classrooms during heating season [125] .

Sensitivity and adaptive behaviours
This study confirms that children are more tolerant of temperature changes during non-heating seasons than heating seasons. It is found that TSV shifts one score by a temperature change of 11.1 °C during non-heating seasons and 7.7 °C during heating seasons. In another study in primary and secondary schools in Australia during summer, children's mean TSV shifts one point on the sevenpoint rating scale by the temperature change of 8 °C [17] . Temperature change in this study is higher than that in [17] which can be attributed to the five-point rating scale used in this study and practice of personal adaptive behaviours. In similar studies, university students' TSV shifts one score by temperature change of 4.16 K in [126] and 6.39 K in [125] . This study suggests that children are less sensitive to temperature changes than adults, as supported in [39,78,117] .
Children in this study have a higher comfort temperature and less sensitivity to temperature changes during non-heating seasons compared to heating seasons. Two reasons can be discussed for this finding: 1) The proportion of children engaged with personal behaviours of clothing, fanning and drinking is significantly higher during non-heating seasons. Reactions for adopting personal behaviours is slower during heating seasons; at T diff = 4 K only 26% of the children change their ' clothing level at comfort temperature', however, this number is 60% during non-heating seasons. Previous studies support that adaptive behaviours increase occupants' tolerance of high temperatures, uncomfortable conditions [82,90,115,127] , occupants' forgiveness and satisfaction [44,90,[127][128][129][130][131] and decease their reported discomfort [132] . 2) Evidence shows that thermal sensitivity can be affected by indoor and outdoor temperature variations [39] and by the difference between mean T op on survey day and T op at the time of filling out questionnaire [133] . More diverse thermal exposures in classrooms can possibly account for greater degrees of thermal sensitivity [17] . In this study, SDs for T op and T out are higher during non-heating seasons (SD Top = 2.1 and SD Tout = 3.7) than heating seasons (SD Top = 1.7 and SD Tout = 2.8). Therefore, higher diversity of indoor and outdoor conditions during non-heating seasons can potentially contribute to children's higher adaptability, less sensitivity and also their acceptance of higher temperatures during non-heating seasons.

Conclusion
The results of this study are significant in improving the resilience of the UK primary schools in the light of climate change by understanding adaptive behaviours of school occupants.
Children's comfort temperature is found to be lower than that for adults. During heating seasons, children have a lower comfort temperature and they feel overheated quicker which can be attributed to fewer personal adaptive behaviours and more consistent environmental conditions during heating seasons. Around 15% of children are overheated in both seasons, however, practice of personal and environmental behaviours is different in each season. During heating seasons, 40% of children practice no personal behaviours, however, the ones who adopt personal behaviours engage more slowly compared to the ones during non-heating seasons. Teachers are mainly in charge of environmental adaptive behaviours and classrooms are controlled based on their perception of thermal environment rather than children's perception. To deliver effective learning environments, providing opportunities for adaptive behaviours should be considered as a part of design process for both newly-built and refurbished schools. The study suggests that: • Schools designers should consider design strategies that can facilitate the efficient engagement of both teachers and children with controls. • School protocols should encourage school occupants (teachers and children) to practise personal and environmental behaviours in an efficient sequence to reach comfort and save energy. • Teachers should be informed about the gap between adults and children's thermal comfort. • Teachers should encourage children to adopt effective personal and environmental behaviours when feeling in discomfort. • Children should be informed about the impact of their adaptive behaviours on thermal sensations and energy consumption so that they consciously adopt adaptive behaviours. • Children should be encouraged to communicate with their teachers about their thermal perception and their preference over controls.

Declaration of Competing Interest
None.