Introduction

The Universal Thermal Climate Index (UTCI) aims to assess outdoor thermal conditions in the major fields of human biometeorology in terms of a one-dimensional quantity summarising the interaction of environmental temperature, wind speed, and humidity, and of the long-wave and short-wave radiant heat fluxes. This assessment should be based on the physiological response of the human body, which in turn is simulated by a thermo-physiological model. For this purpose, based on an advanced multi-node model of human thermoregulation (Fiala et al. 1999, 2001, 2003, 2010) the ‘UTCI-Fiala’ model of thermo-physiological comfort was derived (Fiala et al. 2011) and coupled with a state-of-the-art clothing model. This model, which is described in detail in this special issue (Havenith et al. 2011), considers (1) the behavioural adaptation of clothing insulation observed from European field studies of the general urban population in relation to the prevailing environmental temperature, (2) the distribution of clothing over different body parts, and (3) the reduction of thermal and evaporative clothing resistances caused by wind and movement of the wearer, who is assumed walking at 4 km/h on the level.

UTCI was then developed following the concept of an equivalent temperature. This involved the definition of a reference environment with 50% relative humidity (but vapour pressure not exceeding 20 hPa), with still air and radiant temperature equalling air temperature, to which all other climatic conditions are compared. Equal physiological conditions are based on the equivalence of the dynamic physiological response predicted by the model for the actual and reference environments. As this dynamic response is multidimensional (body core temperature, sweat rate, skin wettedness, etc. at different exposure times), a strain index was calculated by principal component analysis (PCA) as a single dimensional representation of the model response. The UTCI equivalent temperature for a given combination of wind, radiation, humidity and air temperature is then defined as the air temperature of the reference environment that produces the same strain index value (Bröde et al. 2011).

Supporting urban planning by the assessment of the outdoor thermal environment constitutes one of the major application areas of human biometeorology (Vanos et al. 2010), as modifications in the physical attributes of urban space can promote improved thermal outdoor conditions and thus positively influence the use of open spaces. For instance, from observed or predicted effects on pedestrians, direct impacts of urban geometry on outdoor thermal comfort can be taken into account, e.g. to provide more shaded or sun-lit areas in urban spaces planned for pedestrian use. Furthermore, outdoor thermal conditions may have an indirect effect on indoor environments, interfering with indoor thermal comfort or enhancing/reducing energy loads for air conditioning. Especially in pedestrian areas, their overall thermal conditions will, at least partly, affect their attractiveness. In tropical and, to a certain extent, subtropical areas, pedestrian streets can gather substantial number of passers-by, including shoppers, office workers, etc., throughout the year. In this respect, once a link is established between urban planning aspects and the resulting outdoor thermal environment with the corresponding human reactions, such places can be improved accordingly.

Applying a complete heat budget model of the human thermal environment to assess the climate in street canyons dates back to a study by Jendritzky and Nübler (1981). However, more empirical thermal comfort data for outdoor spaces have been claimed, and a need for regional calibration of assessment procedures to different climate zones has been hypothesised (Spagnolo and de Dear 2003).

This contribution applies the operational procedure for UTCI (Bröde et al. 2011) to data from a survey on outdoor thermal comfort that was carried out in pedestrian areas in Curitiba, Brazil (Krüger et al. 2010). The main objectives of this study were to test the predictive capabilities of UTCI regarding thermal comfort assessment in the field of urban planning in a sub-tropical region, and to verify some of the assumptions of UTCI, including the definitions of the reference environment and the behavioural adaptation of clothing insulation to the environmental temperature.

Materials and methods

The study area

A survey was carried out in Curitiba (25°25′50″ S, 49°16′15″ W, 917 m elevation), a city of 2 million inhabitants in southern Brazil, with a long history of urban planning, starting in the 1930s. Later, with ever-increasing urban growth, the city plan underwent a series of revisions of the Master Plan, which led to an innovative mass transportation system capable of supplying current and future demands, as well as measures for preserving the city’s heritage and green areas. This planning process, which has attracted worldwide attention as a model of sustainable urban development, is based on the principle that land use in the city can be induced, restricted, and organised. Considering the long history of Curitiba’s urban planning, and the importance of sustainability considerations in this process, it is considered essential to develop innovative tools for urban climate analysis that could have an application in the city’s future growth.

Concerning the climate of Curitiba, average temperatures in summer range between 17 and 20°C and in winter between 12 and 14°C (Krüger and Suga 2009). Annual average temperature is about 16°C. Daily amplitudes vary between 0.5 and 25.7°C, and the average swing is 10.5°C. Absolute humidity ranges from about 4 to 18 g kg-1, with an average of about 11 g kg-1. Annual precipitation is around 1,600 mm. This study focusses on an area around a pedestrian street in downtown Curitiba named “XV de Novembro” (“15th of November”, Fig. 1), which was the first pedestrian street established in Brazil in the 1970s. Adjacent streets nearby have also been considered in outdoor comfort field campaigns.

Fig. 1
figure 1

Weather station located in 2 of 13 monitoring points on the pedestrian street “XV de Novembro” in Curitiba, Brazil. a With solar irradiation, b shaded area

Full size image

Microclimatic measurements

Measurements were taken between January and August 2009, over 14 days of sampling, covering a wide range of air temperatures, wind conditions and solar angles and height. Days of testing excluded rain and presented usually clear-sky conditions.

Climatic variables were monitored fulfilling the recommendations on accuracy according to ISO 7726 (1998). Measurements were carried out at two points each day, spanning up to 5 h (typically from 10.00 am to 3.00 pm local time). Two HOBO® weather stations (Onset Computer, Pocasset, MA) were used (Fig. 1), equipped with a three-cup anemometer at approximately 2.1 m height, air temperature and relative humidity sensors at 1.1 m, two Copper gray-colored globe thermometers (for measuring globe temperatures at 0.6 m and at 1.1 m), and a silicon pyranometer measuring the intensity of solar radiation at 1.6 m. Data from all sensors were recorded every 5 s, and averaged over 1 min. The exact times when surveys were conducted were used subsequently to match the inquired responses to the corresponding microclimatic data.

The mean radiant temperature (Tmrt in °C) was obtained from measurements with globe thermometers 2 inches (5.1 cm) in diameter and painted with RAL-7001 (gray) (Thorsson et al. 2007). From globe temperature measurements (Tg in °C), and taking into account wind speed (Va in m/s), air temperature (Ta in °C), and the globe’s emissivity (εg) and diameter (D in m), respectively, Tmrt was calculated according to ISO 7726 (1998) for forced convection (Eq. 1). For calculation of Tmrt, the value of Tg measured at 1.1 m was used corresponding to the measurement height of the other quantities.

$$ Tmrt = {\left\{ {{{\left( {Tg + 273} \right)}^4} + \left[ {\frac{{\left( {1.1 \times 1{0^8} \times V{a^{0.6}}} \right)}}{{\left( {{\varepsilon_g} \times {D^{0.4}}} \right)}}} \right] \times \left( {Tg - Ta} \right)} \right\}^{\frac{1}{4}}} - 273 $$
(1)

Acquisition of personal information

Personal data were based on answers to a thermal comfort questionnaire, designed according to the recommendations of ISO 10551 (1995). The first part of the questionnaire consists of items related to gender, age, height, weight, clothing [the various clothing ensembles were composed according to tables provided in ISO 9920 (2007), values were given clothing insulation in ‘clo’ units (1 clo = 0.155 m2 K W−1)], time of residency in Curitiba (to account for acclimatisation) and time spent outdoors before completing the survey. From weight in kilograms and height in centimetres, body mass index (BMI) was calculated as BMI = weight/(0.01 × height)2, and body surface area (AD) was computed as AD = weight0.425 × height0.725 × 0.007184. The second part of the questionnaire deals with items related to thermal perception and preference. The scales described in ISO 10551 encompass five kinds of judgment: perceptual, affective evaluation, thermal preference, personal acceptability and personal tolerance. In the present paper, we analyse results concerning thermal perception (thermal sensation vote) “How do you feel at this exact moment?”. We used the symmetrical seven-point two-pole scale ranging from −3=“cold” over 0=“neutral” to +3=“hot”.

As a whole, from 14 monitoring campaigns, beginning on 9 January through 12 August, 2,024 comfort votes from pedestrians were obtained, of which 1,654 were included in the analysis. For inclusion, the respondents had to be permanent residents (i.e. living for more than 6 months in the city) and the responses on thermal perception and thermal preference had to be consistent, i.e. they had not to be the same values, either positive or negative: e.g. “feeling warm” (+2) combined with “I would prefer to be much warmer” (+3). In addition, respondents who reported having some kind of illness (cold, cardiovascular problems, physical disabilities, etc.) at the time of the interview were not considered. Following the recommendations of ASHRAE Standard 55 (ASHRAE 2004), only data from respondents who had spent at least 15 min in the outdoor space entered the analysis.

UTCI calculations

Values of the UTCI equivalent temperature were obtained by first running the UTCI-Fiala model (Fiala et al. 2011) coupled with the UTCI clothing model for the climatic conditions specified by the 1,654 measurements, and by subsequently computing the strain index and comparing it to reference conditions. This may be regarded as the most accurate method of UTCI calculation compared to simplified alternatives offered by the operational procedure (Bröde et al. 2011) to calculate UTCI without the need to run the complex UTCI-Fiala physiological model. These included look-up tables of pre-calculated UTCI values for all relevant combinations of climate parameters and a polynomial regression equation approximating the index values over the same relevant climate combinations. For the sample under study, the regression equation approximated the UTCI values derived directly from the UTCI-Fiala model quite well with averaged deviations (root-mean squared deviations) of −0.08°C (0.39°C). The table-lookup approach was even more accurate, with 0.01°C (0.12°C), and all the pairwise correlation coefficients between the values from the three different computational procedures were greater than 0.99.

As UTCI requires the input of wind speed at 10 m above the ground, the wind speed values measured at 2.1 m were scaled-up by a factor 1.29 according to the formula LOG(10/0.01)/LOG(2.1/0.01) as proposed by the operational procedure (Bröde et al. 2011). Whole body clothing insulation as applied by the UTCI simulation model was stored for later comparison with the clothing thermal insulation values obtained in the survey.

Allowing comparisons to the recorded thermal sensation votes, the UTCI-Fiala model also predicts thermal sensation (Fiala et al. 2003), termed dynamic thermal sensation (DTSFiala). These predictions were stored at times simulating short-term (0.5 h) and (quasi-) steady state (2 h) exposure conditions, respectively, as well as averaged over 0.5, 1, 1.5 and 2 h. Alternatively, predicted thermal sensations were also obtained for the reference environment (DTSRef) as simulated during the development of UTCI (Bröde et al. 2011), as well as from additional simulations with the individually determined clothing thermal insulation added to the input of the UTCI-Fiala model (DTSclo).

Data analysis and statistics

Besides gender, and following the recommendations of the World Health Organisation (WHO 1995; WHO expert consultation 2004), subgroups were considered as aged below 25 years (young), between 25 and 64 (adult) and above 64 (elderly). BMI groups were defined based on the 5th (P5), 85th (P85) and 95th (P95) percentile, and categorised as underweight for BMI < P5, as normal for P5 ≤ BMI < P85, as overweight for P85 ≤ BMI ≤ P95 and as obese for BMI > P95, with P5 = 18.5, P85 = 25 and P95 = 30 kg/m2, respectively, applied to persons above 20 years of age (WHO 1995). BMI values for younger persons (up to 20 years) were categorised according to percentiles obtained from CDC (Centers for Disease Control and Prevention) growth charts (Kuczmarski et al. 2002). To account for seasonal effects, subgroups were built for the warmer months (January–May) and compared to the colder months (June–August) during the observation period.

To consider the potentially non-linear relationships with environmental temperature and UTCI, respectively, general additive models with locally estimated smoothing functions (LOESS) were fitted to describe the average course of clothing thermal insulation and thermal sensation with 95% confidence bands computed for the whole sample as well as separately for subgroups defined by gender or season (Zuur et al. 2009).

For analysing the goodness-of-fit, the errors in predicting the thermal sensation votes by the diverse approaches (DTSRef, DTSFiala, DTSclo at different exposure times) for the whole sample and for the different subgroups were assessed by the averaged error (bias), with the error calculated as the difference of the predicted minus the observed values, by the root mean squared error (rmse) and by the Pearson correlation coefficient (rp) between observed and predicted values.

Results

The data from 710 females and 944 males aged 13 to 91 years with their characteristics as presented in Table 1 were available for further analyses. The time spent outdoors varied between 15 and 30 min, mean ± SD were 27.5 ± 5.1 min for females and 27.7 ± 5.0 min for males, respectively. Ambient temperatures measured during the 10 am to 3 pm intervals ranged from 6.4° to 30.7°C (20.1 ± 5.4°C), and relative humidity from 23% to 93% (56 ± 13%), while the solar radiation intensity varied within the range of 0.6–1,199 W m−2 (285 ± 305 W m−2).

Table 1 Distribution of the personal characteristics of the survey participants. SD Standard deviation
Full size table

Climatic parameters

The distribution and the smoothed values of the climatic parameters relating to air temperature (Fig. 2) showed that wind speed did not vary systematically with air temperature, but showed a rather constant average value of about 1 m/s (Fig. 2a). The smoothing function representing average relative humidity stayed between 50% and 60%, but at higher temperatures the course of the smoother was truncated and vapour pressure rarely exceeded 20 hPa (Fig. 2b). The intensity of thermal radiation expressed as the difference of mean radiant temperature to air temperature (ΔTmrt) appeared to increase at higher temperatures (Fig. 2c), but the smoothed function may not represent mean conditions here, as shown by the bi-modal distribution observed for solar radiation indicating that shaded conditions were more prevalent in the study sample (Fig. 2d). So the increased averaged thermal radiation at temperatures above 25°C may be explained by a higher proportion of conditions with solar irradiation (cf. Fig. 1).

Fig. 2
figure 2

Distribution of a monitored wind speed, b humidity and c thermal radiation, expressed as difference in mean radiant temperature to air temperature (ΔTmrt) in relation to air temperature with locally estimated scatter plot smoothers (LOESS) added as solid lines, as well as d a histogram showing the relative distribution of 1,654 measurements of solar radiation intensity

Full size image

Clothing thermal insulation

Clothing thermal insulation decreased with rising air temperature (Fig. 3a) and showed a large inter-individual variability, which was slightly reduced at the lower and upper bounds of the temperature range. The smoothed curve was in good agreement with the clothing model of UTCI, but showed a deviation of about 0.2 clo at 20°C and a steeper decrease along higher temperatures compared to UTCI. The latter could be related to the higher proportion of sunny conditions in this temperature range as suggested by the concomitant increase in thermal radiation intensity (Fig. 2c). However, analysing observations from warmer and colder periods separately (Fig. 3b) showed a close fit of the observed data to the UTCI model for ‘summer’ conditions and also for low temperatures, but a larger deviation when people wear ‘winter’ clothes at temperatures around 20°C.

Fig. 3
figure 3

Distribution of clothing thermal insulation in relation to air temperature. LOESS with 95% confidence bands (dashed lines) are added for the whole sample in a, and separately for data from warmer and colder periods in b; black solid lines indicate clothing insulation as assumed by the Universal Thermal Climate Index (UTCI) model

Full size image

These properties were well reflected by UTCI, as shown by plotting clothing insulation versus the calculated values of UTCI (Fig. 4a). The insulation postulated by the model was predominantly within the confidence band computed for the observed data. Significant deviations, which, however, were bounded above by 0.1 clo, occurred only in the range between 16 and 22°C. Comparing the clothing behaviour of males and females (Fig. 4b) revealed that females chose slightly less insulating clothing in warm conditions with UTCI above 24°C.

Fig. 4
figure 4

Observed clothing thermal insulation values related to UTCI with fitted LOESS smoothers and 95%-confidence bands for the whole sample (a) compared to the thermal insulation as applied by the UTCI model (solid black line), and with LOESS separately fitted for 710 females and 944 males, respectively (b). Note that the ordinates in a and b are scaled differently

Full size image

Thermal sensation

Figure 5 presents the thermal sensation votes, which showed considerable inter-individual variation, related to UTCI. The course of the LOESS smoother for the whole sample in Fig. 5a suggests, that on average, neutral sensations were observed for UTCI between 18°C and 23°C. The increase in thermal sensation votes above this UTCI range and the decrease with lower UTCI values was approximately linear and did not reveal any significant difference between males and females (Fig. 5b).

Fig. 5
figure 5

Observed thermal sensation votes related to UTCI with fitted LOESS smoothers and 95%-confidence bands for the whole sample (a) compared to the dynamic thermal sensation predicted by UTCI for the reference conditions (DTSRef) averaged over 0.5, 1, 1.5 and 2 h (black solid line), and with LOESS fitted separately for 710 females and 944 males, respectively (b)

Full size image

Though values of UTCI are based on the physiological strain in response to the thermal conditions (Bröde et al. 2011), thermal sensation votes were well predicted by the DTS calculated for the UTCI reference environment (DTSRef) averaged over 2 h of exposure time (Table 2, Fig. 5a). There was a small bias underestimating the observed votes by only 0.13 points on the seven-point scale, with an rmse below 1 point and a correlation between observed and predicted values of 0.62. Averaging the predicted votes over the 2-h exposure time improved the agreement compared to sole steady state (2 h) assessments, which overestimated thermal sensation, and were more accurate than taking the short-term predictions at 0.5 h (Table 2), although the latter matched more closely the reported 28 min spent outdoors before completing the survey.

Table 2 Averaged errors (bias), root mean squared errors (rmse) and Pearson correlation coefficients (rp) of the observed 1,654 thermal sensation votes compared to the dynamic thermal sensation (DTS) predicted by the UTCI reference conditions (DTSRef), by running the UTCI-Fiala-model for the specified climatic conditions (DTSFiala), and by providing the actual clothing insulation values as additional input to the simulation model (DTSclo). Results were obtained from short term (0.5 h) and steady state predictions (2 h), as well as from predictions averaged over simulated exposure times of 0.5, 1, 1.5, 2 h
Full size table

Predictions of the UTCI-Fiala model for the monitored climatic conditions (DTSFiala) were not superior to predictions calculated by the UTCI-Fiala model for UTCI reference conditions (DTSRef). Using the recorded individual clothing insulation values as input into the model (DTSclo) gave a poorer performance (Table 2).

Table 3 summarises the results obtained for the different subgroups of the most simple but accurate DTS predicted for the reference conditions (DTSRef) averaged over 2 h of exposure time. Gender-related differences in sensation votes were negligible. Youngsters showed a somewhat greater bias, but also higher correlations compared to adults and elderly. Concerning the influence of body composition, there was an increase in bias and in rmse accompanied by a decrease in correlation from the ‘underweight’ and ‘normal’ to the ‘overweight’ and ‘obese’ groups, respectively. A similarly increased bias was observed for persons interviewed during the colder period. However, overall differences between the subgroups were small, with rmse around 1 point of the 7-point scale in any case.

Table 3 Averaged errors (bias), root mean squared errors (rmse) and Pearson correlation coefficients (rp) of the observed thermal sensation votes compared to the DTS predicted by the UTCI reference conditions averaged over simulated exposure times of 0.5, 1, 1.5, 2 h (cf. DTSRef in Table 2) for different subgroups defined by gender, age, body mass index (BMI) and seasonal period (warmer: January–May, colder: June–August)
Full size table

Discussion

This survey on urban thermal comfort had its particular strength in monitoring the relevant climatic and clothing parameters onsite during interviews. Focussing on wind speed data, the simultaneous measurements of two non-stationary weather stations, located at two different spots on the pedestrian area, allowed for the assessment of air quality and pollutant dispersal at several spots in the urban canyons (Rasia and Krüger 2010). The calculation of BMI values from self-reported height and weight might raise methodological concerns over their validity, as a large-scale validation study (Kuczmarski et al. 2001) showed that, on average, younger women underreported weight by approximately 1.5 kg and older persons overrated their height by 2–4 cm. However, classifying overweight status using BMI calculated from self-reported values was performed with very high sensitivity and specificity ranging between 84% and 99% in all age and gender groups. In that study, sensitivity dropped to 73% only for persons older than 79 years. As less than 1% of our sample was of such an old age, we considered this potential effect to be of minor importance for the purposes of our study.

Concerning the recently developed operational procedure of UTCI (Bröde et al. 2011), the survey provided useful data for checking some of the assumptions underlying the development of UTCI. With respect to the definition of the reference environment required for computing UTCI equivalent temperatures, the bi-modal distribution of solar radiation intensity confirmed the choice of a shaded condition as reference. The high prevalence of shaded conditions corresponded to the obstructed urban settings evaluated with averaged height-to-width ratio (H/W) of 1.4 and a sky view factor ranging from 0.2 to 0.55 (Emmanuel et al. 2007; Matzarakis et al. 2010; Lin et al. 2010). Also, the reference humidity profile as assumed by UTCI, with 50% relative humidity with a maximum of 20 hPa water vapour pressure, was supported by the climate data of this survey.

The dependence on ambient temperature found for clothing insulation in this study was in good agreement with the behavioural adaptive clothing model of UTCI (Havenith et al. 2011). Although the average profile showed slightly increased insulation values compared to the UTCI model at 20–22°C, which might be explained by the observation that survey participants used ‘winter’ clothing at temperatures that might as well represent ‘summer’ conditions, and although there were still smaller, but significant, differences between the insulation worn by males and females in warm conditions, the overall pattern of clothing choice was properly reflected by UTCI.

The observed thermal sensation votes were predicted well by the DTS simulated by the UTCI-Fiala model for reference conditions. The predictions averaged over the simulated 2 h exposure time gave better approximations than the steady-state values after 2 h, which might have been expected from presumed differences in the assessment of indoor and outdoor thermal comfort with a higher relevance of the dynamic aspects of thermal comfort for the outdoor environment (Höppe 2002). Surprisingly, the predictions for 0.5 h, which matched closely the average time spent outdoors (28 min) before completing the survey, were still worse than the 2 h values. One explanation could be related to expectations or experience of the outdoor climate prior to the period actually considered (Nikolopoulou and Steemers 2003; Knez et al. 2009), which could also be associated with local building characteristics, usually with low thermal insulation, not air-conditioned and, in many cases, similar to enclosed semi-outdoor spaces resembling outdoor conditions (Spagnolo and de Dear 2003).

The simple procedure to predict thermal sensation from DTS values for the reference environmental conditions, which was already available from UTCI development (Bröde et al. 2011), provided the same degree of accuracy as the more complex approach of running the UTCI-Fiala model for the climatic parameters actually recorded. This suggests that UTCI allows for adequate comparisons of climatic conditions and again confirms the simplified approach taken by UTCI to develop an assessment scale based on responses to reference conditions. Interestingly, the inclusion of individually recorded clothing insulation as additional input to the UTCI-Fiala model even impaired the predictive capacity, although individualised models performed better in other contexts (van Marken Lichtenbelt et al. 2004; Havenith and Middendorp 1990). One reason might be that differences in the choice of clothing insulation under identical climate conditions, which are evident by the large inter-individual variation observed in this study sample, could partly reflect differences in individual thermoregulatory systems. Thus, considering individual variations in chosen clothing insulation without at the same time taking into account individual differences in thermoregulation and potentially interacting personal and psychological characteristics (Knez et al. 2009) might be counterproductive. However, information on individual thermoregulation besides simple anthropometric characteristics such as height and weight is difficult to obtain in a survey.

The analyses of subgroups revealed a negligible influence of gender and age on prediction error, whereas there was only a limited increase in the error for obese persons and in the colder season. Within the conditions studied here, UTCI appears to be applicable to males and females, to young and adult people as well as to the elderly, although it should be noted that the study population was restricted to healthy and permanent, i.e. acclimatised, residents. The small rmse of less than 1 point on the 7-point thermal sensation scale observed in all subgroups suggests a reasonable goodness-of-fit and does not advocate the need for calibration of the UTCI procedure, which was developed mostly from European data, to the sub-tropical region under study.

As the UTCI may be coupled easily to measurements from meteorological monitoring stations (Spagnolo and de Dear 2003) or to computational fluid dynamics software (CFD) simulating the physical parameters of the thermal environment (Cropper et al. 2010), our results suggest that the UTCI may serve as a suitable planning tool for urban thermal comfort in sub-tropical regions.