Abstract
Projected climate change will cause increasing air temperatures affecting human thermal comfort. In the highly populated areas of Western-Central Europe a large population will be exposed to these changes. In particular Luxembourg—with its dense population and the large cross-border commuter flows—is vulnerable to changing thermal stress. Based on climate change projections we assessed the impact of climate change on human thermal comfort over the next century using two common human-biometeorological indices, the Physiological Equivalent Temperature and the Universal Thermal Climate Index. To account for uncertainties, we used a multimodel ensemble of 12 transient simulations (1971–2098) with a spatial resolution of 25 km. In addition, the regional differences were analysed by a single regional climate model run with a spatial resolution of 1.3 km. For the future, trends in air temperature, vapour pressure, and both human-biometeorological indices could be determined. Cold stress levels will decrease significantly in the near future up to 2050, while the increase in heat stress turns statistically significant in the far future up to 2100. This results in a temporarily reduced overall thermal stress level but further increasing air temperatures will shift the thermal comfort towards heat stress.
1. Introduction
With very high confidence human health is influenced by climate change [1], as the thermoregulatory system of the human body interacts closely with the atmospheric system [2, 3]. The significant increase in air temperature is shown by different Global and Regional Climate Models (GCM and RCM) as presented, for example, in the framework of the ENSEMBLES [4] or EURO-CORDEX [5] projects. Most severe health effects could be associated with the projected increase of extreme events such as increasing frequency and intensity of heat waves [1, 6]. Nevertheless, adverse effects on human health could already be shown with moderate levels of heat stress, such as a characteristically temperature-mortality relationship even at air temperatures below 30°C [7].
Several studies on the impact of climate change on vegetation [8, 9], tourism [10], and human health effects [11] have been published for the region of Luxembourg. The latter focused on changes in extreme heat and cold periods. As Luxembourg is characterised by a temperate semioceanic climate with mild winters and moderate summers [8], the moderate heat and cold stress levels and their changes are dominant. A detailed analysis in space and time for human thermal stress in Luxembourg has not been conducted yet.
The thermal state of the environment is not sufficiently described by air temperature but comprises furthermore humidity, wind speed, thermal radiation (from the short and long wave domain), and facets of the physical environment [12]. To account for all parameters a large number of human thermal climate indices have been developed [12], whereof we chose two energy balance stress indices, the Physiological Equivalent Temperature (PET) [13] and the Universal Thermal Climate Index (UTCI) [14]. Both indices have been widely used and were shown to be suitable for human thermal stress assessments [3, 15–17].
The objective is to assess the impact of climate change on three different typical climatological regions of Luxembourg at a high spatial resolution. The assessment includes all classes of both human-biometeorological indices, evaluating the changes in time and space.
2. Materials and Methods
2.1. Climate Change Projections and Area of Investigation
For the climate projections we used a multimodel ensemble of 12 RCMs with a spatial resolution of 25 km per grid cell, in order to quantify the uncertainties related to these climate projections on a daily basis. Due to the calculation of a spatial mean from a 3 by 3 grid this results in only one time series for each ensemble member for Luxembourg [8]. In addition, we used a single RCM run with a spatial resolution of 1.3 km to perform a detailed analysis of different regions in Luxembourg on an hourly basis (Figure 1). Both approaches are based on the IPCC A1B emission scenario, which describes anthropogenic future emissions with rapid economic growth, a balanced use of energy resources, and an increasing global population until the middle of this century [18]. The measurements at Luxembourg Findel Airport from 2000 to 2010 lie within the spread of the multimodel ensemble [8] which accounts for the wide use of this GCM RCM combination in the area of investigation [10, 11, 19].
2.1.1. Global Climate Model (GCM)
To quantify the uncertainties related to the climate change projection, a multimodel ensemble of 12 transient simulations from 1971 to 2098 provided by the EU FP6 ENSEMBLES project has been used [4]. The GCM RCM combinations and the institutions running the models can be found in Table 1. The results including daily mean air temperature (at 10 m), relative humidity, wind speed, and global radiation were retrieved from the Danish Meteorological Institute (http://ensemblesrt3.dmi.dk/, November 30, 2013). The projections are all based on the SERS emissions scenario A1B [18]. Due to the coarse resolution of the ENSEMBLES members, a single data point located in the centre of Luxembourg at N49°36′E006°07′ at a height of 259 m AMSL was selected (Figure 1).
2.1.2. Regional Climate Model (RCM)
For the analysis of the regional differences in Luxembourg simulations with the spatial resolution of 1.3 km have been conducted with the regional climate model COSMO-CLM. Hourly values for air temperature, relative humidity wind speed at 2 m height, and global radiation at ground level were extracted from the model output. The nonhydrostatic limited area atmospheric prediction model of the consortium for small-scale modelling COSMO [22], version 4.8_clm11, is used as the underlying meteorological model. A three-step nesting approach (18, 4.5, 1.3 km) as described by Junk et al. [11] was used to generate the data for Luxembourg with a final resolution of 1.3 km. Three ten-year periods from 1991 to 2000 (reference period), 2041 to 2050 (near future), and 2090 to 2100 (far future) were calculated for three pseudo stations (Figure 1).
For the detailed spatial analysis of the CLM datasets, we implemented three pseudo stations in different characteristic landscapes of Luxembourg. The stations are (1) Reuler, at 492 m AMSL located in the north of Luxembourg in the “Oesling”, (2) Esch-sur-Alzette, at 291 m AMSL located in the south of Luxembourg in the “Gutland”, and (3) Wasserbillig at 190 m AMSL located in the Moselle valley (Figure 1).
RCM outputs suffer the problem of a bias compared to observational data [23]. Accordingly, bias-corrected data should be preferred over uncorrected ones. In the case of human-biometeorological indices the use of bias-corrected data is not possible, as there are no suitable bias correction schemes available for global radiation and especially for wind speed. Therefore, we used uncorrected data to calculate all human-biometeorological indices for this study, and we will focus our analyses on changes between the reference period and the future time spans.
2.2. Human-Biometeorological Indices
To evaluate the impact of the projected climate change on human health two common human-biometeorological indices were chosen for our study. The analyses are based on the PET [13] and the UTCI [14]. Both indices are based on air temperature, wind speed, relative humidity, and mean radiant temperature. The mean radiant temperature was approached with the RayMan model using the geographical location and temporal data together with the meteorological input (especially global radiation). These variables were extracted from the ground based 3D output fields of the RCM simulation. Both human-biometeorological indices are not directly driven by relative humidity, but in combination with the provided temperature data, the water vapour pressure can be calculated to be used for the calculations of the indices. In addition to these parameters, physiological aspects of the human body, such as activity, clothing, sex, and age, are taken into account [24].
Notwithstanding PET and UTCI use the same input variables, they express differences in the number of classes they have. PET divides the thermal stress in 9 classes ranging from “extreme cold stress” to “extreme heat stress” [20] (Figure 2), whereas the UTCI has 10 classes “extreme cold stress” to “extreme heat stress” [21] (Figure 3). As the classes are developed in different ways, we do not intend to compare the indices directly but decided to use them independently for the analysis.
We used the freely available RayMan Pro model Ver. 2.1 [24] to calculate both indices on an hourly basis for the single model runs at the three pseudo stations. For the calculations we used the standard physiological configuration for the indices. For PET this is a 35-year-old male with a weight of 75 kg, a height of 1.75 m, a clothing value of 0.9, and an internal heat production of 80 W. For UTCI we used the UTCI specific standard subject.
As the calculated human-biometeorological indices are based on non-bias-corrected data, a reference period must be defined in order to evaluate the human-biometeorological changes in the future projections. Instead of the standard reference period of 30 years from 1961 to 1990, defined by the World Meteorological Organisation [25], we used a 10-year reference period from 1991 to 2000. We chose the shorter period due to the fact that the computational resources limited the simulations at the high spatial resolution of 1.3 km to three time spans of ten years. Various studies show that the use of a shorter reference period also leads to reliable results [11, 26, 27].
The decision not to focus on extreme events is based on the research of Hajat and Kosatky [28] and Kovats and Hajat [7], who demonstrated that adverse effects on human health are already detectable with only moderate levels of heat stress, such as a characteristically temperature-mortality relationship even at air temperatures below 30°C.
Due to the absence of a suitable bias correction for the input parameters, it is not recommended to work with the absolute results in the context of index changes. One possibility to deal with that fact is the use of relative changes compared to a reference period. As some of the more extreme index classes result in zero for the reference period, but not for the future periods, we decided not to use a relative approach as this would have caused the division by zero. To keep the information of which classes are more and which are less represented in the region, we refrain from using relative anomalies using the absolute values instead.
3. Results
3.1. Multimodel Ensemble Mean, Trends, and Spread
The analysis of the multimodel ensemble has been conducted to (1) account for the spread in future climate change projections and (2) get an overview of the change in the input parameters for human-biometeorological indices.
Figure 4 shows that there is a significant trend in daily mean air temperature () and vapour pressure (VP), whereas wind speed (WV) and global radiation (GR) do not show clear trends over the period from 1971 to 2098 for annual values. The slopes (calculated by Sen’s method [29]) of the significant () trends calculated from annual mean values by the Mann-Kendall test have a slope of 0.029 (0.29°C/10a) for daily mean air temperature () and 0.017 (0.17 hPa/10a) for vapour pressure (VP).
(a)
(b)
(c)
(d)
In addition to the trends, Figure 4 shows that, for all four model-driving parameters (air temperature (Figure 4(a)), vapour pressure (Figure 4(c)), wind speed (Figure 4(b)), and global radiation (Figure 4(d))), the results from the high-resolution CLM run (10-year average plotted) are within the spread of the multimodel ensemble. Wind speed results from the CLM model are, for all pseudo stations, below the multimodel mean. This results probably from the fact that the ENSEMBLE multimodel output includes wind speed at 10 m only, whereas the CLM model output contains wind speed at 2 meters. In Figure 4(c), the CLM results for Esch-sur-Alzette and Wasserbillig resemble each other so that the lines overlay each other. The increasing air temperatures and vapour pressure are causing the human-biometeorological indices to increase [30].
3.2. CLM Model Results
A detailed analysis of the human-biometeorological index PET in Luxembourg can be found in Figures 5 and 6. The corresponding results for the UTCI are presented in Figures 7 and 8.
(a)
(b)
(c)
(a)
(b)
(c)
(a)
(b)
(c)
(a)
(b)
(c)
3.2.1. PET
Figure 5 shows boxplots of the PET index classes at the three pseudo stations Esch-sur-Alzette (Figure 5(a)), Reuler (Figure 5(b)), and Wasserbillig (Figure 5(c)) for the reference period 1991–2000, near future 2041–2050, and far future 2091–2100. Each boxplot represents the number of hours with the corresponding PET class per year in the respective time period. Groups of boxplots per index class that are not significantly different () are marked by the same letter underneath the plots. If a box shares the letters ab, it is not significantly different from a and b.
PET classes with the highest number of hours are the classes “strong cold stress”, “moderate cold stress”, and “slight cold stress” at all stations (Figure 5). The highest PET category “extreme heat stress” is not present at any station during the reference period and the near future. In the far future this category appears 10 and 9 times, respectively, in Esch-sur-Alzette and Wasserbillig (Figures 5(a) and 5(c)), but still not in Reuler (Figure 5(b)). Due to the low number of occurrences this change is not statistically significant at . The tendency of decreasing cold index classes and increasing hot index classes can be observed throughout all pseudo stations. Most significant changes occur in the transition between the near and far future. The only significant changes between the reference period and the near future can be found in the “extreme cold stress” index class at all pseudo stations. In Esch-sur-Alzette and Wasserbillig a significant change in the “slight cold stress” index class can only be observed in the transition from the reference period to the far future (Figures 5(a) and 5(c)); the amount of hours in this category in the near future is not different from the reference period and the far future. The most represented index class “moderate cold stress” does not show any significant changes at all.
While Figure 5 focuses on the changes in time, Figure 6 addresses the comparison between the three pseudo stations during the three time slices, reference period 1991–2000 (Figure 6(a)), near future 2041–2050 (Figure 6(b)), and far future 2091–2100 (Figure 6(c)).
In addition to the information obtained from Figure 5, Figure 6 highlights that Reuler shows lower human-biometeorological indices compared to the other stations. This is only significant in the nonextreme index classes “strong cold stress”, “slight cold stress”, and “no thermal stress” in the reference period (Figure 6(a)). In the “slight heat stress” and “moderate heat stress” index classes only the difference from Reuler to Esch-sur-Alzette is significant () in the reference period. In the near future, the differences between the stations decrease temporarily (Figure 6(b)). Here only in the index classes “strong cold stress”, “slight cold stress”, and “slight heat stress” differences between Reuler and other two stations appear. Most of the significant differences between Reuler and the other two regions of Luxembourg can be found in the far future (“strong cold stress”, “no thermal stress”, “slight heat stress”, and “moderate heat stress”) (Figure 6(c)). At least one significant difference can be found in that period for the “slight cold stress” and “strong heat stress” index classes. This is the first appearance of a significant regional difference in the “hot” index class. It becomes apparent that the number of hours in the higher index classes increases at all stations, but they are still the lowest at Reuler (Figure 6).
3.2.2. UTCI
The analyses for the UTCI show similarities to the PET results for the changes in time (Figure 7) and on a regional level (Figure 8). In most mid index classes (UTCI class “moderate cold stress” to “moderate heat stress”), the results for Reuler (Figure 8) differ significantly from the other regions. The cold stress decreases through the three time slices at all stations, while the heat stress increases in all regions. A significant increase in the “strong heat stress” class can be found at all stations in the far future (Figure 7). The significant change in the amount of hours in the “strong cold stress” class already takes place in the transition from the reference period to the near future.
3.2.3. Stress Periods
Merging the index classes into three categories “cold stress”, “no thermal stress”, and “heat stress”, it can be seen for PET that the number of hours causing any form of cold stress decreases through all stations (−5.6 to −8.0% for the far future compared to the reference period). In contrast, the number of hours in the comfortable and heat stress class increases. The increase in heat stress for the far future ranges from 264.6% (Reuler) to 447.6% (Esch-sur-Alzette). In total these changes result in fewer hours with thermal stress in the future. For the far future this reduction ranges from 3205 h to 3394 h over ten years. Similar results can be obtained when analysing the UTCI. The decrease in cold stress ranges from −17.4 to −23.4% for the far future, while the increase in heat stress for Reuler is 358.7% and for Esch-sur-Alzette 596.9%. The total reduction of heat stress ranges from 5745 h (Reuler) to 7407 h (Esch-sur-Alzette) over ten years regarding the UTCI.
To account for the increasing stress level at prolonged periods of thermal stress exposure, we calculated the duration of “cold stress”, “no thermal stress”, and “heat stress” (Figures 2 and 3) for both indices at all pseudo stations during all three time slices. To attribute one of the three categories to each day, we identified the most frequent category per day and assigned it as the day’s heat stress level. The generated daily data was used to calculate the duration of days in the same category. From the reference period to the near future, the duration of cold stress is reduced by 27.7 to 53.9% (PET), respectively, 0.5 to 15.4% (UTCI). The lowest reduction can always be observed at Reuler, which had the longest periods of consecutive “cold stress” days in the reference period. For the far future, the reduction of consecutive “cold stress” days continues, resulting in a reduction up to 78.3% for PET, respectively, 35.0% for the UTCI. Regarding heat stress, the results for near and far future differ. In the near future, the length of “heat stress” periods is reduced by 4.0 to 26.7% (PET), respectively, 5.0 to 10.0% (UTCI). In the far future, the length of these periods, however, increases by 29.4 to 34.8% (PET), respectively, 20.0 to 57.1% (UTCI). Only Reuler represents an exception for the PET as the length of “heat stress” periods is reduced by 8.3%.
4. Discussion
Both human-biometeorological indices are capable of reflecting the climate change towards higher air temperatures as a positive trend in air temperature (Figure 4) goes along with a trend towards higher human-biometeorological indices (Figures 5 to 8). These findings correspond well with the results of Molitor et al. [31] who showed that the frost risk in the region of Luxembourg will almost disappear until the end of the century and the finding of Matzarakis et al. [10] who found a strong increase in heat stress based on the REMO (Regional Model of the Max Planck Institute for Meteorology, Hamburg) data sets. In addition, Junk et al. [11] concluded that there is not only an increase in the average air temperature or human-biometeorological index, but also an increased duration of hot periods in the future. This indicates a possible underestimation of heat stress by only analysing the index classes, as prolonged heat waves cause more severe health outcomes due to the higher thermal stress [32]. Our analysis of the duration of thermal stress showed that the overall stress level in the near future is reduced, while in the far future the duration of heat stress increases. The more frequent occurrence of heat stress, as well as its longer duration, shows the potential health risk of the far future heat stress. In our study we analysed all hours of the day including night time hours as heat stress during night times has been identified as significantly dangerous to human health [33, 34]. This results in less pronounced heat stress, as temperatures over night are usually lower than at daytime. The fact that the projected CLM wind speed results are given at 2-meter height adds to the underestimation of heat stress.
The analysis of the CLM data set in terms of regional differences within Luxembourg showed that the lowest human-biometeorological indices are found in the north of Luxembourg at the pseudo station Reuler throughout all time periods, whilst the highest occurrence of hot index classes in both human-biometeorological indices can be found in Esch-sur-Alzette. These findings are in line with the results from the REMO model analysed by Matzarakis et al. [10]. Regional differences that are already present in the reference period remain throughout the following time slices in most cases. There are two exceptions. The first one is the “slight cold stress” PET class, where the difference between Reuler and Esch-sur-Alzette becomes no longer significant in the far future. The second one is the “strong cold stress” UTCI class where the difference between Reuler and the two other stations disappears in the near future and becomes a significant difference between Reuler and Wasserbillig in the far future.
In general, the index classes above “no thermal stress” show regional differences in the future periods. The only exception here is the “warm” PET class where an existing regional difference temporarily disappears in the near future, to come back in the far future.
In terms of the periods, where changes in the classes occur, we can distinguish again between the index classes below and above “no thermal stress.” The changes in the classes below “no thermal stress” mostly occur in the transition from the reference period to the near future or the far future, while changes in the classes above “no thermal stress” become significant only for the far future.
Our analysis shows that the stress level in total decreases in the future. This is caused by the fact that the absolute number of hours in the cold stress range decreases, while the number of days in the heat stress range does not increase by the same quantity. The global climate change also leads, for the area of Luxembourg, to a higher amount of hours in the “no thermal stress” range for the projected future. As this process continues, hours that are now present in the comfort range might then shift towards heat stress levels. These findings might lead to the conclusion that the climate change including temporarily reduced stress levels, especially reduced cold stress, leads to reduced morbidity and mortality during winter time. According to Donaldson and Keatinge [35, 36], low temperature increases the risk of mortality. In contrast, a recent study by Staddon et al. [37] shows that decreasing cold stress no longer leads to decreasing mortality rates in temperate countries. This is due to the better housing and the fact that people prevent themselves well from being exposed to the winter climate. Staddon et al. [37] identified influenza as the only significant parameter influencing the number of excess winter deaths from 1971 on.
On the other hand, the increasing heat stress is likely to cause an increase in heat related mortality and morbidity [32, 38, 39]. Together with the findings of the increasing duration of heat waves in Luxemburg by Junk et al. [11] it can be assumed that the influences of heat stress tend to be even more severe than assumed by just the number of hours in the heat stress classes.
5. Conclusion
Our study confirms the general decrease in cold stress and the general increase in heat stress for the region of Luxembourg by analysing the human-biometeorological indices PET and UTCI in detail. The analysis revealed that the change in stress levels is caused by significant trends in air temperature and vapour pressure.
It could be shown that there are significant differences in stress levels amongst the different regions of Luxembourg. The coldest climate conditions are found in the “Oesling”, a region in the north of the country, resulting in the highest levels of cold stress and the lowest levels of heat stress. The lowest level of cold stress can be found in Esch-sur-Alzette in the “Gutland” region in south Luxembourg, which also exhibits the highest levels of heat stress. The Moselle valley exhibits slightly lower levels of heat stress than found in the “Gutland” region.
Regarding the changes in the index classes, we can distinguish between cold and heat stress. The changes in heat stress tend to already appear in the near future (2041–2050), whereas the heat stress levels changes become statistically significant in the far future (2091–2100).
In total the number of hours in index classes that are considered to be stressful for the human body decreases in the future. This is caused by the fact that the number of hours with cold stress decreases, but the number of hours causing heat stress does not increase by the same amount. To evaluate if this also causes a decrease in thermal stress related mortality and morbidity, further studies for the region are foreseen.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
Acknowledgments
The authors gratefully acknowledge the financial support of the National Research Fund in Luxembourg (4965163-FRESHAIR). Parts of the work have been done in the framework of the “Small Particles-environmental behaviour and toxicity of nanomaterials and particulate matter” (SMALL) project. Dr. Lucien Hoffmann and Vanessa Peardon are thanked for final proofreading. COSMO-CLM data were taken from the CLIMPACT Project funded by the National Research Fund of Luxembourg through Grant FNR C09/SR/16.
Supplementary Materials
The supplementary material visualizes the average length of “cold stress”, “no thermal stress” and “heat stress” periods at the three pseudo stations compared to the reference period in percent explained in 3.2.3.