Outdoor Universal Thermal Comfort Index Climatology for Alaska

Data from 456 surface meteorological sites in Alaska, eastern Russia and northwest Canada for 1979-2017 were used to model hourly universal thermal comfort indices (UTCIs) under consideration of Alaska-appropriate clothing. The results served to determine a high-resolution climatology of thermal comfort levels for Alaska at various temporal and spatial scales as well as the frequency of thermal stress levels. On 1979-2017 average, various degrees of cold stress occurred with highest percentage on the Alaska West Coast and along the Arctic Ocean. In the continental and Inside Passage region, no thermal stress had the highest percentage of occurrence. In Interior Alaska, both strong heat and extreme cold stress occurred occasionally. At most sites and in all Alaska Köppen-Geiger bio-climate regions, the absolute range between monthly means of daily minimum and maximum UTCIs was larger than that of monthly means of daily minimum and maximum air temperatures. Major contributors to thermal discomfort (shortwave radiation, air temperature, moisture, wind speed) varied among bio-climate regions and in the diurnal and annual courses.

high levels of heat stress, extended exposure to extreme levels of cold stress can impact human health, reduce the efficiency of performed activities, lead to frost bite, hypothermia or even death when the exposure impairs thermoregulation [4] [5] [6] [7] [8]. Cold stress is a significant risk factor among patients, particularly women with cardiovascular diseases [9]. Unlike heat waves, during which mortality raises for several days, cold spells may increase levels of mortality for several weeks [5]. Thus, as more people visit, work and/or live in Alaska and its off-shore areas, adequate public health system and urban planning require assessment of the outdoor thermal environmental conditions in form of climatology and frequency of occurrence at high spatial resolution.
Typically, negative influences of thermal stress on human health and performance are expressed in terms of thermal stress indices. Various thermal stress indices have been developed. They differ with respect to their input data (airflow velocity, temperature, humidity, solar radiation, etc.), rational, concepts and location [10] [11] [12] [13]. For instance, most thermal stress indices assume a standard outfit (light shirt, slacks). Some indices are valid only for air temperatures above (e.g. heat index, humidex for T a > 20˚C) or below a threshold (e.g. T a ≤ 0˚C; wind chill index, wind chill equivalent temperature index). Analytical indices base on the principles of thermal heat exchange between the human body and its environment; empirical indices base on human response or comfort to various environmental factors; direct indices rely on measurements like the apparent (AT = T a + 0.33 × v a − 0.7 × v a − 4, where v a is wind speed), operative or the most commonly used Wet-Bulb Globe Temperature (WBGT = 0.567 × T a + 0.393 × v a + 3.94). The Standard Effective Temperature (SET) considers physiological parameters (skin temperature and wetness). The Effective Temperature (ET) [14] [15] and Physiological Equivalent Temperature (PET) were developed for indoor activity. The Mediterranean Outdoor Comfort Index (MOCI), for example, was optimized for the Cs climates [16]. The PET is based on energy balance considerations and is the output of the Munich Energy Balance Model for Individuals (MEMI). Some indices originally developed for indoor climate (e.g. SET) were adapted later to outdoor environments (OUTSET) by adding the impacts of shortwave radiation [17] [18].
In studies on weather impacts on health, the mean radiant temperature T mrt defined as the "uniform temperature of an imaginary enclosure in which the radiant heat transfer from the human body equals the radiant heat transfer in the actual non-uniform enclosure" [19] is a better measure than ambient air temperature [20]. T mrt depends, among other things, on shortwave and long-wave radiation fluxes from all sides, body posture (sitting, standing, etc.), the absorption of shortwave and long-wave radiation by the human body, and wind speed. Thus, it represents the combined effects of radiant and convective heat gains/ losses for the human energy balance in the outdoors. Studies in Freiburg, Germany, for instance, revealed that T mrt can differ up to 37˚C between shaded and sunlit areas, whereas air temperature, T a differs only 1˚C -2˚C between the two.
The Global Outdoor Comfort Index (GOCI), for instance, uses empirical rela- tions depending on T mrt , T a , relative humidity, v a , latitude, mean annual temperature, mean temperatures of the hottest and coldest months [21]. Only few studies on thermal comfort in high latitudes exist. A study in Göteborg, Sweden (Dfb climate), for instance, showed that T mrt can be much lower than T a on extremely cold days with almost no shortwave radiation reaching the surface [22]. A study in Göteborg (57.70N, 11.94E Dfb), Luleå (65.54N, 22.11E Dfc) and Stockholm (59.35N, 18.06E Dfb) found that during clear sky conditions, the highest and lowest T mrt occurred near sunlit walls and in shaded areas, respectively; spatial variation decreased with increasing cloudiness [22]. Under cloudy conditions, highest T mrt occurred in open areas due to high shortwave diffuse radiation from the sky vault. While in winter, T mrt strongly differed with latitude, no such sensitivity occurred in summer [22]. A summer study on the thermal comfort in Apatity (67.567N, 33.4E) in the Murmansk region of Russia north of the Arctic Circle found significant spatial variation of PET during contrast weather conditions [23].
Various studies showed that cold stress caused different responses depending on the bio-climate region. Cold spells, for instance, can cause notable and often significant increases in mortality in elderly people (65 years or older) from cardiovascular causes, especially strokes, coronary heart disease events, and respiratory causes in subtropical Cfa [7], in mid-latitude Cfb [4], and high latitude Dfb climate [6]. In regions with warm winters, mortality from cardiovascular diseases (CVD) increases with a given temperature decrease; in Ireland (Cfb climate), the CVD mortality rate is 45% higher in December than August, while it amounts to 28% in the much colder winters of Norway [5] (~46% in Dfc, ~26% in Cfb, ~26% in Cfc, ~1% in ET, less than 1% in Cfc climate).
Alaska still lacks an assessment of human thermal comfort at a regional scale. Weather differs widely over the state depending on the location with respect to the semi-permanent high-pressure systems, major storm tracks, elevation above sea level, distance from the coast and the presence or absence of seasonal sea-ice [27]. Consequently, Alaska falls into several climate regions [24] [28] [29] [30] (e.g. Figure 1). Unfortunately, currently available thermal stress maps like, for instance, from global simulations of various heat-stress metrics within the Community Land Model (1˚ × 1˚) [12] or derived from ERA-Interim (T255 ~79 km) [31] are too coarse to capture Alaska's complex topographic and environmental and hence thermal comfort conditions. The goals of this study were to fill this gap with a climatology and to identify regions of huge thermal hazards as well as the frequency thereof. To achieve this goal, I used near-surface meteorological observations between 1979 and 2017 (both included) to assess the thermal comfort using the Universal Thermal Comfort Index (UTCI) model [32] [33].
The UTCI was chosen for the following reasons: The UTCI bases on the well evaluated [34] advanced multi-node human thermoregulation model [35]. The UTCI proved itself to be able to represent bioclimatic conditions in terms of human strain under a wide range of climatic outdoor conditions [10]. The  [24] based on datasets from the Climatic Research Unit (CRU) [25] and Global Precipitation Climatology Centre (GPCC) [26]. This resolution is between ~34.94 km ( [36]. This means the clothing model permits a realistic description of clothing insulation for Alaska. Furthermore, a comparison using one year of 6 am to 9 pm daily data revealed significant correlations (P < 0.0001) between UTCI and other heat indices (PET, PMV, SET, WBGT), and environmental parameters (e.g. dry temperature); UTCI correlated the strongest with PET (r = 0.96; r 2 = 92%) followed by WBGT (r = 0.88), SET (r = 0.87), and dry temperature (r = 0.90). Note that PET and SET are based on the body thermal equation, while the thermal perception of WBGT is more similar to that of the UTCI than that of other indices [13]. See [10] and [12] for comparisons of thermal indices.

Data Sources and Data Processing
Near-surface meteorological data collected at land, well, ship and buoy sites be- Since not all data are from operational weather monitoring, data cover timeframes of varying lengths and at different times during the period. Also some operational sites were moved or closed and new sites were set up during the period. Some sites lacked winter precipitation data or winter data at all. Thus, the available data were considered as a sample within the period following [38] and [27].
Observational data were converted to SI units as needed. Daily accumulated precipitation was calculated from hourly or 6-hourly values when not reported as daily values. Furthermore, except for precipitation some data came at smaller time steps than an hour. These data were averaged to hourly values. Relative humidity served to determine dew-point temperatures and vice versa.
Unfortunately, in many of the datasets, the practice to just leave out hours and/or days without observations was used. However, creating a climatology requires a continuous dataset in time for 1) comparison of data from different sites at the same time, and 2) calculation of daily and monthly means as well as diurnal and annual cycles. Therefore, the data were examined for being in con- When also no cloud-cover observations existed, cloud cover was estimated from relative humidity as [39] where rh is relative humidity as fraction of 1 and rh crit = 0.7 is a critical threshold for cloudiness to occur. Multiplication of c with 100 provides cloud cover in percent.

Calculation of the UTCI
The UTCI combines thermo-physiology and heat exchange theory [32] [33] and These reference conditions assume a metabolism of 135 W•m −2 , light activity (walking 4 km•h −1 ), a mean radiant temperature being equal to the 2 m air temperature, a relative humidity of 50% for T a ≤ 29˚C and water vapor pressure e = 20 hPa for T a > 29˚C, and a 10 m wind speed, v a of 0.5 m•s −1 , which corresponds to about 0.3 m•s −1 at 1.1 m above ground level [33]. All other weather conditions are compared to this reference.
As aforementioned, the calculation of the UTCI bases on the well evaluated [34] advanced multi-node human thermoregulation model [35]. This numerical model simulates, among others, the environmental heat exchanges, the heat and mass transfer with the body, thermoregulatory reactions of the central nervous system as well as perceptual responses [33] ( Figure 2). The advanced multi-node human thermoregulation model is valid for −50˚C ≤ T a ≤ 50˚C, −30˚C ≤ T mrt -T a ≤ 70˚C, 0.5 m·s −1 ≤ v a ≤ 30.3 m·s −1 , 5% ≤ RH ≤ 100% and water-vapor pressure e < 50 hPa [33].
Thermo-regulation refers to the human's ability to keep the body temperature within certain limits despite of a very different ambient temperature. When body-core temperature and/or skin temperature change, effector responses occur; effector actions depend on weather conditions, clothing and activity [40] [35]. In moderate and warm environments, a person's physiology can maintain body temperature within acceptable limits via thermoregulation; sweating and vasodilation of skin vessels occur to cool the skin by evaporation of sweat [35] [41]. In cold environments, increased convection is the main path to body-heat loss. In response, the thermoregulatory system reduces the peripheral blood flow to decrease skin temperature and to increase the thermal insulation of the skin tissue. Shivering produces heat [40]. Adequate clothing, for instance, is a behavioral effector response to thermal conditions [36]. The advanced multi-node human thermoregulation model [35] runs in coupled mode with an adaptive clothing model [36] and predicts the dynamic thermal sensation as an equivalent temperature, the UTCI ( Table 1). The model (and hence UTCI) neglects influence of physiological adaptation or acclimation.
The clothing model [36] simulates people's adjusting their clothing to the outside conditions. Insulation changes depending on ambient temperature. At temperatures below −20˚C, the clothing model assumes special clothing. Clothing insulation values stem from data of actual behavior in the field [36]. A partial clothing approach is used. It accounts for the head and face being covered differently than the torso. It assumes the face to be exposed to the ambient air. Movement of the person and wind reduce thermal and evaporative clothing resistances. At ambient temperature, radiation, relative humidity and wind modulate the physiological response while the clothing remains the same. The incorporated comfort model uses physiological states (skin temperature, core temperature, sweat rate, skin wittedness, etc.) to predict thermal sensation responses to steady state and transient conditions.
The difference between the UTCI value and air temperature, T a depends on the actual air and mean radiant temperature, T mrt , wind speed, v a and humidity (water-vapor pressure, e or relative humidity, RH in %). The model converts the observed 10 m wind speed to body level assuming a logarithmic wind profile following [42].
The mean radiant temperature is determined following [43] ( ) ( ) where T g , v a , T a , ε and D are the globe temperature, wind speed, air temperature, emissivity (0.95) and diameter of the globe. See [11] for a comparison of different methods to determine the mean radiant temperature and their impacts on UTCI (and PET) outdoor comfort levels. The globe temperature was calculated following [44]. The ground surface was assumed as grass until onset of snow cover. Global albedo for shortwave and long-wave radiation was set to 0.2 for dry, snow-free conditions. Surface albedo for grass was set to 0.15 on days with at least one rain event.
A snow cover was assumed when snow depth and/or snowfall were reported.
When no snow-depth data existed, a closed snow cover was assumed at latitudes north of 60˚ between November and end of March in accord with [29]. Once a snow cover existed, surface albedo was set to 0.55 (fresh snow) and 0.3 (aging snow) for days with at least one and no snow event, respectively.

Analysis
Bio-climate studies typically apply the Köppen-Geiger classification (KGC) [ For each site i with more than one year of observations, the monthly mean over the period for the j th month (j = 1, 12) reads Here m is the number of hourly values in the sample of site i in all j th months of the period and X stands for T a and UTCI, respectively. Analogously, the higher moments as well as monthly means of daily minimum

Results
In some continental locations, summer maximum and winter minimum temperatures differed more than 80 K during the period (not shown). Coastal areas without seasonal sea-ice varied the least between summer and winter temperatures.
The annual course of solar insolation at the TOA differs strongly between     Figure 4, Figure 5). Absolute differences  between actual air temperature and UTCI varied among sites and among bioclimate regions (see Section 3.1).
Obviously, a daily or monthly climatology would exclude hourly or daily extremes (cf. e.g. Figure 6). This means that with only a climatology of monthly or annual means hazardous situations may be underestimated. Therefore, the percentage of hours in each sample for each comfort level was determined as well (see Section 3.3).  In this section, the thermal comfort for 1979-2017 is discussed based on the data of these 370 sites. For most sites and all bio-climates, the absolute difference between monthly means of daily maxima of UTCI and air temperatures were typically much larger than those between monthly means of UTCI and air temperatures. (e.g. Figure 7) The same was true for monthly means of daily minima of UTCIs and air temperatures.

Marine West Coast Climate (Cfc, Cfb)
Along the coast of the Gulf of Alaska, frequent cyclones cause windy, wet conditions [29]. Thus, frigid thermal conditions are often paired with high humidity and/or high wind speeds, which can lead to wind chill. Under windy winter conditions blowing snow may occur as well.
Cfc bio-climate has the same annual precipitation pattern as Cfb bio-climate, but with lower temperature conditions (e.g. Figure 6(a), Figure 6(b)). Overall, in both Cfb and Cfc climates, thermal comfort levels encompassed a wider range than actual temperatures. Monthly means of daily minimum UTCI and maximum air temperatures were lower than those of daily minimum air temperature and maximum UTCI, respectively.
Spatially-averaged over all sites with Cfb bio-climate, the period-annual course of monthly mean UTCI spanned 21.8 K. Monthly mean UTCI was highest in July (16.3˚C ± 1.5˚C), lowest in January (−5.4˚C ± 3.2˚C), and below the freezing point from November to March (Figure 8(a)). Monthly mean UTCIs typically exceeded monthly mean temperatures from May to August (Figure 8(b)).  zone. Typically, slight cold stress occurred from April to June and in September.
During the period, the lowest minimum and highest maximum UTCI were −20.6˚C and 14.9˚C, respectively (Table 2).
In both Cfb and Cfc bio-climates, spatial-inter-annual variability in thermal comfort levels was higher for the cold than warm thermal range. In Cfb bio-climate, spatial-inter-annual variance in UTCI-values was largest in February followed by November and March, and smallest in July. Spatial-inter-annual variability in monthly mean minimum UTCI, for instance, was more than twice as high as that of the monthly mean maximum UTCI. In Cfc climate, spatial-  inter-annual variance was largest in January and smallest in September (cf. Figure 8(a), Figure 8(b)). The pattern in inter-annual variability can be explained as follows: High UTCI values are associated with low cloudiness, high-pressure situations. Under such synoptic situations, shortwave radiation, calm winds and topography govern the temperature and moisture conditions over a large area [37] [49]. However, low UTCI values occur under cyclonic conditions. Wind speeds, cloudiness, humidity and temperatures strongly differ before and after a frontal passage. The resulting strong gradients in meteorological parameters over the bio-climate region yield great variability in UTCI.

Humid Continental Climate (Dfb, Dwb)
Dfb and Dwb climates are hemi-boreal [28]. Temperatures of the four warmest months are 10˚C or above, but less than 22˚C. These bio-climates mainly differ by the timing of precipitation (not shown).
In the annual course, period-spatial highest and lowest mean monthly UTCI  (Figure 8(c)). Lowest and highest UTCIs obtained at sites with Dfb climate were −28.5˚C and 9.9˚C (Table   2) indicating strong and no thermal stress for the extremes, respectively.
In Dwb climate, on period-spatial average, January and July had the lowest (−18.0˚C; strong thermal stress) and highest monthly mean UTCIs (20.6˚C; no thermal stress). This wider range in UTCI values in Dwb than in Dfb bio-climate results from the enhanced humidity due to winter precipitation and less humidity in summer. In Dwb bio-climate, the lowest and highest UTCI were −27.3˚C (strong thermal stress) and 24.3˚C (no thermal stress), respectively ( Table 2).
Monthly mean UTCI values indicated moderate or worse cold stress from October to March (Figure 8(e)).
In both Dfb and Dwb bio-climates, spatial-inter-annual variance in monthly mean UTCIs was lowest in September. In Dfc bio-climate, this variability was notably higher from November to March than in the other months, largest in December followed by November and February. In Dwb bio-climate, January saw the highest spatial-inter-annual variance followed by November and February. The strong decrease and increase of solar radiation with latitude ( Figure 3) in these months are major reasons. In summer, latitudinal differences in solar radiation are much smaller over these regions.

Continental Subarctic Climate (Dfc, Dwc)
Dfc and Dwc climates are colder than Dfb and Dwb climates. Again, Dfc and Dwc climate are the same with respect to their temperature thresholds, but differ by the timing of precipitation.
The vicinity to the Canadian High leads to mostly calm or no wind conditions in Interior Alaska. In Dfc bio-climate, highest and lowest period-spatial monthly means of UTCI differed 35.9 K. Typically, January and July had the lowest (−20.6˚C; strong cold stress) and highest monthly mean UTCIs (15.3˚C; no thermal stress), respectively. Averaged over the period and all Dfc sites, monthly mean UTCIs indicated various degrees of cold stress from October to April with at least moderate cold stress (Figure 8(d)). The lowest and highest UTCI obtained at sites with Dfc bio-climate were −40.7˚C (extreme cold stress) and 25.0˚C (no thermal stress), respectively ( Table 2).
Due to the high humidity associated with winter precipitation, highest and lowest period-spatial monthly mean UTCI differed 39.8 K in Dwc bio-climate.
On period-spatial average, the lowest minimum (−24.8˚C) and highest maximum monthly mean UTCIs (14.9˚C) occurred in January and July, respectively.
Monthly mean UTCIs were below 0˚C from October to April (Figure 8(f) Figure 6(c)) and Dwc (dry summer) bio-climate, August UTCIs showed the least variability in space and time. In Dfc bio-climate, highest spatial-inter-annual variance in thermal comfort conditions occurred in March followed by April (melting season), while in Dwc bio-climate, highest spatial-inter-annual variance occurred in April followed by March, May, and February.

Mediterranean-Influenced Subarctic Climate (Dsc)
At the few sites with Dsc bio-climate, highest and lowest period-spatial monthly mean UTCIs differed 39.9 K. On period-spatial average, the minimum (−23.3˚C) and maximum monthly mean UTCIs (16.6˚C) occurred in January and July, respectively. Monthly mean UTCIs stayed below freezing from October to April (Figure 8(g)). The humidity and wind from the ocean caused moderate to strong cold stress during these months. During 1979-2017, the lowest and highest UTCI in Dsc climate were −31.7˚C (very strong cold stress) and 19.5˚C (no thermal stress), respectively ( Table 2). The high heat capacity of the ocean delays warm-up in spring and may result in cool sea-breezes keeping summers comfortable. In fall, the high heat capacity of the ocean contributes to mild conditions. The winds and high humidity yield cold stress of various degrees in winter and spring. Spatial-inter-annual variance in UTCIs was highest in April caused by occasional late snowfall and the presence or absence of sea-ice close to the shores. During June to September spatial-inter-annual variance in thermal conditions was notably lower than during the cold season with September having the least variability. The strong inter-annual variability of thermal comfort in winter can be attributed to the inter-annual variability in storm tracks and extension/location of high-pressure systems. Storms from the south advect Mediterranean warm air masses, while a strong high-pressure system can advect Arctic cold air from the Interior and Yukon Territory.

Tundra and Ice-Cap Climate (ET, EF)
As can be seen in Figure 3, solar insolation gradually decreases northward over the tundra leading to no insolation from late November to end of January in the northernmost areas. During that time, long-wave radiation loss governs the energy budget and thermal comfort. Once the Sun comes back, the high albedo of snow reflects huge amounts of shortwave radiation that reaches the surface.
In ET bio-climate, monthly mean UTCI were above freezing only in July meaning still slight cold stress on average. Very strong cold stress with monthly mean UTCIs below −30˚C occurred from December to March.
In EF climate, the annual course of period-spatial monthly mean UTCI was asymmetric with only negative values (not shown). Various degrees of cold stress existed year round reaching very strong cold stress in January to March (<−35˚C). Monthly mean UTCIs showed a steep increase in April and peaked in June. Thereafter, monthly means of UTCI steadily decreased to −28.6˚C (strong cold stress) in December. In ET bio-climate, spatial-inter-annual variance in monthly mean UTCI was highest in March and lowest in August (Figure 8(h)). Skewness and kurtosis were highest in February when the Sun comes again above the horizon. Both skewness and kurtosis had a secondary maximum in November when the solar insolation reached zero (Figure 8(h), Figure 3).
In EF climate, spatial-inter-annual variance in monthly UTCIs was least in November to January, largest in June and marginally less in May. In the former months, long-wave radiation dominates, while in June shortwave radiation at the TOA is the largest (Figure 3). The high inter-annual variability results from variability in cloudiness and hence shortwave radiation reaching the surface. Skewness ranged between −0.3 (September) and 0.5 (February). Kurtosis was negative in all months with higher absolute values in summer than winter.

Climatology of Alaska Thermal Comfort
In this analysis, for each of the 456 sites, its total available data were taken as 100%. Figure 9 shows for each of the sites the most often occurring thermal comfort level and the percentage of hours it occurred. No thermal stress existed between 20% and 30% of the time in the Interior of Alaska as well as 25% to 40% of the time in the Inside Passage. Along the Gulf of Alaska, slight and moderate cold stress dominated the thermal comfort levels. In the Aleutians, moderate and strong cold stress were the most frequent conditions governing the thermal comfort levels 15% to 35% of the time.
Crews on oil and gas platforms in the Chukchi and Beaufort Seas were exposed to very strong or strong cold stress 45% to 50% of the time (Figure 9, Figure 10). Slight cold stress existed in the Chukchi and Beaufort Seas up to 20%  of the time. In the eastern Beaufort Sea, moderate cold stress occurred up to 35% of the time (Figure 9). Note that ships cruise in these seas from June to late October [2] [50].
The continental regions of Alaska (and the Yukon Territory) had slight heat stress up to 10%. In some places, strong heat stress occurred on occasions as well. The continental regions of Alaska experienced a higher fraction of time with very strong cold stress than those along the Gulf of Alaska or in the Aleutian ( Figure 9). Along the Alaska West Coast, the fraction of time with very strong cold stress increased with latitude reaching up to 25% on the North Slope. This means the low or lack of insolation played a key role for deep winter thermal comfort. Note that pockets with even more hours of very strong cold stress occurred. Köppen-Geiger bio-climates. The 1979-2017 thermal climatology of Alaska reveals the following. At most sites and in all bio-climates in Alaska, thermal comfort levels encompassed a wider range than actual air temperatures. In general, differences between actual air temperature and UTCI varied with the time of the day, day and month of the year (i.e. shortwave radiation reaching the ground, long-wave radiation loss, wind speed, humidity). The various bio-climates differed in the range of thermal comfort levels as well as timing of when monthly means of air temperatures exceeded those of UTCI and vice versa. However, at most sites, monthly mean UTCI values exceeded monthly mean air temperatures in summer, while the opposite was true in winter. Typically, monthly means of daily minimum UTCI were lower than monthly means of daily minimum air temperature; the opposite was true for the maxima.

Discussion, Conclusions and Outlook
The thermal conditions with the highest percentage of occurrence ranged from slight cold stress to very strong cold stress except for Interior Alaska and along the Inside Passage. In Interior Alaska and along the Inside Passage, no thermal stress had the highest percentage of occurrence. Nevertheless, cold stress of different degrees dominated overall in these two regions. These two regions and the very continental Dwc bio-climate had the widest range of thermal conditions. However, in the former two regions, thermal comfort levels reached from extreme cold to very strong heat stress, while they reached from very strong cold stress to strong heat stress in Dwc bio-climate. However, the frequency of occurrence was marginal for these extremes in all three regions.
Along the coasts of West and North Alaska and the northern Gulf of Alaska some degrees of cold stress had the largest percentage of occurrence. Along the northern Gulf of Alaska, wind and enhanced humidity from the ocean were notable contributors to low UTCI levels. Along the Bering Sea coast north of the Arctic Circle and along the coast of the Arctic Ocean, the low or lack of solar insolation in combination with long-wave radiation loss in winter and windy, humid conditions in summer led to on average uncomfortable thermal conditions year round.
Except for Dwc climate, spatial-inter-annual variability in thermal comfort levels was higher for the cold than warm thermal range, but differed in timing of minima and maxima among bio-climates. In Dwc bio-climate, spatial-inter-annual variability in UTCI was larger in the first than second half of the year.
In ET climate, the skewness and kurtosis of UTCI increased strongly in February and November due to the incremental northward increase and decrease, respectively, of shortwave radiation reaching the surface. In all other bio-climates, skewness and kurtosis showed only marginal differences among months.
While the UTCI was developed to be applicable universally to different climates, a human behavior is the thermal adaptation to their own bio-climate [51].
Their past experiences and culture may influence their expectations and percep- showed that those living in Melbourne wore more clothing (0.1 clo) than those in Hong Kong [52]. This means that the Alaska thermal comfort climatology might underestimate the thermal stress of new residents and tourists.
In the regions of increased exploration, oil and gas production, strong to very strong cold stress dominated. Thermal comfort rarely exceeded slight cold stress.
Similar was true for the regions of increased shipping. Based on these findings one may conclude that here the capacities for treatment of cold stress related symptoms have to be extended as the population, ship and air traffic increases on the North Slope and its coastal waters. The increasing tourism for aurora watching, dog-mushing, hiking, kayaking, fishing, boating etc. in Interior Alaska will require expansion of capacity to treat patients with cold and heat thermal stress.
The more than 23% growth in Alaska's population mainly increased the outskirts of the three major cities (Anchorage, Fairbanks, Juneau). Recent studies showed that increasing urbanization and improved urban planning, among other things, altered the thermal stress in European cities of different background climates over time [53]. Given the few urban sites in Alaska, of which many have only short time series, the assessment of small-scale differences in thermal comfort and planning for its improvement remain a challenge. Thus, the development of urban thermal comfort models is an urgent need as well.
Major challenges in determining a thermal comfort climatology for Alaska were the quantity and quality of available data. While at a few sites, over 100 years of data exist [29], the number of available sites and the spatial resolution of sites were too coarse for a meaning-full spatial climatology starting at that time.
Furthermore, some of these early time series only reported daily maxima, minima and means, i.e. they had a coarse temporal resolution. Thus, using the data from 1979 to 2017 was a compromise between sufficient spatial resolution, number of sites and length of the time series.
However, when looking at a limited timeframe, extreme events with low probability of occurrence may be missed. For instance, the highest and lowest recorded temperature since onset of recording were 37.8˚C (Ft. Yukon, June 27 1915) and −62.2˚C (Prospector Creek, January 23, 1971), respectively [29]. This means that the 1979-2017 sample missed these and probably other local extremes.
The investigations of thermal comfort at different temporal scales revealed that a climatology based on coarse temporal resolution data would exclude hourly or daily extremes. This means that in assessment of future thermal comfort conditions, thermal indices will have to be determined within the simulation of the climate projection itself. Using, for instance, output data of climate-model projections with 3 h or 6 h or even longer time increments may underestimate the severity and frequency of hazardous thermal situations.