Quantitative information about outdoor thermal comfort, on various temporal and spatial scales, is required to design better cities and mitigate heat problems not only in warm but also in temperate climates. The overall objective of this study is to explore the augmentation of global/regional climate changes by urban features such as geometry in a compact mid-rise high-latitude city (Gothenburg). The magnitude of spatial and temporal variations of intra-urban mean radiant temperatures (Tmrt) is quantified using the SOLWEIG (SOlar and LongWave Environmental Irradiance Geometry) model. Hourly time resolution, statistically downscaled meteorological data, based on the ECHAM5-GCM under the A1B emission scenario is used to simulate changes in Tmrt and physiologically equivalent temperature (PET) at the 2080–2099 time horizon.
Results show that urban geometry causes large intra-urban differences in Tmrt, on hourly, daytime and yearly time scales. In general, open areas are warmer than adjacent narrow street canyons in summer, but cooler in winter. According to the ECHAM5-based scenario, the daytime Tmrt will increase by 3.2 °C by the end of this century. This is 0.4° more than simulated increase in air temperature (2.8 °C) and is mainly a result of decreases in summer cloudiness. Occasions of strong/extreme heat stress are expected to triple. This equates to 20–100 h a year, depending on geometry. Conversely, the number of hours with strong/extreme cold stress decreases by 400–450 h. Furthermore, the number of hours with no thermal stress increases by 40–200 h a year.
The mean air temperature in Sweden is expected to rise by 2–5 °C by 2100 (IPCC, 2007). In temperate climates, a 2–3 °C increase in the average summer air temperature will double the frequency of periods characterized by extremely high air temperatures (WHO/WMO/UNEP, 1996). This means that heat waves will become more frequent, more intense and last longer (Meehl and Tebaldi, 2004). Extreme heat waves, such as the central European heat wave of 2003, have profound effects on people's health and well-being, with substantial economic consequences as a result (Pascal et al., 2006). Although European concern is focussed on the south, where summer air temperatures are already high, the predicted increase in heat in Scandinavia must also be taken seriously (Rocklöv and Forsberg, 2008). According to the Swedish governmental report ‘Sweden facing climate change—threats and opportunities’ (SOU, 2007), the number of heat-related deaths in the Stockholm area could rise by 5% in the summer if the mean air temperature rises by 4 °C during that season. The same report estimates that the increased costs of heat-related deaths due to climate change could be €50–70 billion in Sweden over 2010–2100. Although the number of cold-related deaths is expected to decrease due to milder winters, the risks associated with heat waves seem stronger, since northern populations have not yet adapted to heat as they have to the cold periods, over a long time (Rocklöv and Forsberg, 2008).
Urban centres are particularly vulnerable to heat waves because of the effect of the urban climate conditions and poor air quality. The urban heat island (UHI), which is mainly controlled by differences in geometry and thermal admittance between urban and rural environments (Oke et al., 1991), can reach up to 12 °C at night time in mega cities (Oke, 1981). The interaction between the extreme weather at a regional scale and the UHI at a local scale aggravated the heat stress of people, especially at night when the UHI is most pronounced (Jendritzky and Grätz, 1999; Pascal et al., 2006).
The fact that buildings retain heat at night if ventilation is inadequate implies that urban citizens can experience sustained heat stress both during day and night time, whereas people in rural areas may obtain relief from thermal stress at night (Jendritzky and Grätz, 1999). Although little is known about the effectiveness of certain urban planning measures for human health (WHO, 2003), it is well known that climate-responsive design, utilising urban geometry, vegetation, surface and building material, colour, etc., plays an important role in preventing heat stress at street level and also affects energy use (Ali-Toudert and Mayer, 2007; Silva et al., 2010).
In Sweden, heat is seldom a problem. Instead, buildings are designed to reduce the impact of cold and to admit sunshine in winter. The demand for sun gives Swedish cities an open-set, low- to midrise city structure. The exceptions are the city centres, which often have a more compact structure. Very few residential buildings in Sweden are airconditioned. In a warmer future climate, the demand for cooling in houses, health-care institutions, schools and work places is expected to increase during the summer (Holley et al., 1998). More airconditioners might be installed and used, which in turn will create additional demand for electricity and, additionally, release heat into the city itself in summer. In comparison, climate-responsive design is considered to be climate neutral.
During the past decade, attention has increasingly turned to the potential effect of anthropogenic climate change on urban heat waves. A variety of approaches have been used to overcome the limited resolution and biases of general circulation models (GCMs) and to derive information about urban climate. One approach for investigating heat waves in GCM data is simply to use a model's temperature output, an approach used by Meehl and Tebaldi (2004). Although it is not appropriate to use the extremes of actual temperature values, percentile-based thresholds, which approximate the definition of heat waves in the real world, can be used. An alternative to using percentile thresholds is to first scale GCM's daily temperature outputs to match an observed temperature distribution (Hayhoe et al., 2004). An important limitation of this rescaling approach, however, is that the inter-variable correlations in the downscaled data may not realistically represent the observed climate. It is also much more difficult to rescale sub-daily data.
One way to create realistic, multi-variable time series is to use the relationships between large-scale atmospheric patterns and local climate. This is what is usually thought of as ‘statistical downscaling’ (Benestad et al., 2008). The advantage of this approach is that climate scenarios can be derived with much higher temporal and spatial resolution than is possible with numerical models (Wilby et al., 2009). This approach is therefore suitable for studying urban climate extremes. Meehl and Tebaldi (2004) showed that heat waves are linked with specific atmospheric circulation patterns. Wilby (2008) used such links directly to investigate potential changes in the nocturnal UHI using a hybrid statistical downscaling method.
A different approach for overcoming GCM's limited resolution is to use higher resolution numerical modelling (Christensen and Christensen, 2007). Although high-resolution models reduce model bias, the outputs are still not sufficiently close to observed data that they can be used directly as inputs in further simulations (Moberg and Jones, 2004; Rivington, 2008; Rivington et al., 2008). As a result, studies of urban heat-related phenomena that use regional model outputs use similar downscaling techniques to those based directly on GCMs. For example, Koffi and Koffi (2008) investigated heat wave indices using percentile-based threshold extremes. Belcher et al. (2005) rescaled outputs from regional models, using a ‘morphing’ algorithm in which changes in daily outputs were applied to hourly historical weather. Rivington et al. (2008) outlined a method to rescale precipitation, daily minimum and maximum temperature and solar radiation to better match the observed climate. However, it appears that very little research has been conducted into methods to rescale sub-daily regional model outputs.
Climate change is often discussed in terms of changes in air temperature, cloud, wind, etc., i.e. as trends in either averages or extremes. However, in order to evaluate its impact on people's thermal perception, health and well-being, it is necessary to analyse their combined effect. In summer, the mean radiant temperature (Tmrt) is one of the most important meteorological parameters governing human energy balance and thermal comfort (Mayer and Höppe, 1987). This is the sum of all short- and long-wave radiation fluxes to which a human body is exposed and is thus a critical issue in assessing the human comfort outdoors.
Quantitative information about outdoor thermal comfort, on various temporal and spatial scales, is required to design better cities and mitigate the problems of heat and cold stress (Mayer et al., 2008). The overall objective of this study is to explore the augmentation of global/regional climate changes by urban features such as geometry in a compact, mid-rise high-latitude city. The specific objectives of this study are as follows
(1)Quantify the effect of urban geometry on the temporal and spatial variations of intra-urban daytime (Tmrt) in the central parts of Gothenburg, Sweden. Our study uses maps, point values and areal means, and different temporal scales, i.e. hourly, daytime, seasonal and annual
(2)Simulate potential changes in Tmrt and thermo-physiological stress between 1980–1999 and 2080–2099 in different urban places in Gothenburg using hourly statistically downscaled meteorological data based on the ECHAM5/MPI-OM model under the SRES A1B emission scenario.
Our study demonstrates the magnitude of spatial and temporal variations of intra-urban Tmrt caused by urban geometry and highlights the potential for using urban geometry to mitigate heat or cold stress in temperate climates. Furthermore, it simulates how potential future changes in air temperature, wind, humidity and radiation would be perceived by a group of people, in terms of thermal comfort and thermo-physiological stress at street level. This knowledge could be used to develop design guidelines for mitigating extreme thermal stress, decrease energy use and improve the health and well-being of people living in urban areas.
2.1. Study area
The city of Gothenburg is located at 57.7°N latitude on the Swedish west coast. The city was founded in 1623 and is today the second largest city in Sweden, with about 510 000 inhabitants. The city centre has a classical European design, characterized by a compact mid-rise structure with little vegetation, as shown in Figure 1.
Figure 2 shows a digital terrain model (DTM) which covers the central parts of Gothenburg that have been selected for this case study. The DTM in Figure 2 is based on a digital elevation model (DEM) that was derived from local governmental digital data (topography and urban geometry) according to Lindberg (2005). The DEM has a resolution of 1 m pixel.
To quantify the influence of urban geometry on Tmrt, four urban places within the case study area with different urban geometries—a large square, a small courtyard, a north–south orientated street canyon and an east–west orientated street canyon—have been selected for more detailed analyses (Figures 2 and 3). A detailed description of the places, including surface material, sky view factor (SVF) and height to width ratio (H/W), is found in Table I.
Table I. Surface material, SVF and H/W of the four urban places
Surface material (dominate)
Buildings: white/yellow plaster and bricks
Buildings: brown tile
Buildings: brown tile
Buildings: white/yellow plaster
2.2. Climate of Gothenburg
The area has a marine west-coast climate, dominated by alternating low- and high-pressure systems embedded in westerly flows. More rarely, there are long periods of stable weather conditions caused by blocking highs over Scandinavia. The area has relatively mild winters and cool summers for its latitude, with a mean air temperature of − 0.4 °C in winter (December–February) and 16.3 °C in summer (June–August). These air temperatures are based on measurements from 1960 to 1990 (SMHI, 2009). The length of the nights varies greatly throughout the year, from approximately 6 h in June to 16 h in December.
2.3. Meteorological data
For the simulations of Tmrt in SOLWEIG 1.0 (SOlar and LongWave Environmental Irradiance Geometry model), two meteorological datasets are used:
Dataset I consists of hourly meteorological data (temperature, wind, radiation and humidity) for 1977 from the Swedish Meteorological and Hydrological Institute (SMHI) weather station in Säve (at the city airport), just outside the city of Gothenburg. The climate of 1977 is considered to represent the typical climate of Gothenburg for the period 1965–1984, exhibiting all the normalities and extremes (Skanska Software, 1990). This dataset is used to analyse the interactions between urban geometry and the outdoor thermal environment and to examine spatial variations in Tmrt within the central parts of Gothenburg at different temporal scales, i.e. hourly, daytime, seasonal and annual.
Dataset II includes hourly statistically downscaled data, based on the ECHAM5/MPI-OM model under the SRES A1B emission scenario. The A1 scenario family describes rapid, global economic development; and the A1B scenario is considered a ‘mid-range’ emission scenario, in terms of 21st century global warming. The A1 family was thought to be the type of scenario ‘best represented’ in academic literature when the SRES scenarios were compiled (Nakicenovic and Swart, 2000). Subsequent rapid economic growth and increases in CO2 emissions in the emerging world (Le Quéré et al., 2009) continue to justify the use of the A1 scenario. Although short-term trends in emissions clearly cannot be extrapolated 90 years into the future, the emergence of coal as the world's largest source of energy (Le Quéré et al., 2009) potentially ‘locks energy production in CO2-intensive infrastructure for decades’ and suggests that a transition to the environmentally and sustainability-focused SRES B storyline branch is not imminent. Dataset II is used to simulate potential changes in outdoor thermal comfort due to climate change for the city of Gothenburg for the 2080–2099 time period.
The data were downloaded from the WCRP CMIP3 multi-model database, and climate change scenarios were derived using a downscaling algorithm based on historical re-sampling. The downscaling technique involves iterating over the daily GCM scenario data, and for each day, finding a day in the historical records which matches the large-scale atmospheric situation closely. The hourly weather time series for a future day is then constructed by taking the historical weather from the matched historical day. In this way, progressively warmer days from the historical sequence are selected to represent the climate of the future. For the downscaling procedure, the large-scale atmospheric situation is classified according to Chen (2000). This algorithm assigns daily circulatory regimes to one of 27 classes, based on daily-average sea-level pressure data from the NCEP/NCAR reanalysis project (Kalnay et al., 1996). Each day is also assigned a ‘regional temperature’. This is the daily-average temperature at a height of 2 m from the nearest model grid cell. NCEP/NCAR reanalysis data are used for the historical regional air temperatures, and the GCM temperature is used for the scenario days. Because the nearest pixel to Gothenburg corresponds to a land pixel in the NCEP reanalysis dataset and a ‘lake’ pixel in the ECHAM5 GCM, the latter are normalized to match the reanalysis air temperatures over the 1983–2000 period. A historical day is considered an acceptable analogue for a future day if both days (1) have the same circulation classes (or adjacent, i.e. North-West or North-East flow classes are adjacent to the North flow class), (2) have regional air temperatures within 5 °C, (3) are from the same month and (4) are continuous with the previous scenario day (defined as temperature and dew-point change at midnight < 2 °C). With these criteria, there exist an analogue day for around 65% of scenario days. If no acceptable analogue is found for a day, then restriction (1) is removed, allowing analogues to be found for over 95% of scenario days. Further relaxation of criteria allows historical analogues to be found for all scenario days. If more than one historical day is an acceptable analogue, then the day for which the sum of the regional temperature and the temperature and dew-point changes at midnight is a minimum is selected. Basic statistics from the downscaled emission scenario are shown in Table II.
Table II. Mean air temperature, wind speed, relative humidity and global radiation for the different periods (1980–1999 and 2080–2099) according to seasons, downscaled from ECHAM5/MPI-OM model under the SRES A1B emission scenario
Control period (1980–1999)
Scenario period (2080–2100)
Air temperature ( °C)
Wind speed (m s−1)
Relative humidity (%)
Global radiation (W m−2)
Our downscaling approach is somewhat similar to that proposed in Wilby (2008), in that weather patterns are used to drive a weather generator. Our weather generator is much simpler, however, being based on historical re-sampling. We chose this approach instead of rescaling outputs from a regional model because the SOLWEIG model requires realistic, multi-variable climate inputs at sub-daily time resolution. The downside of using a historical re-sampling algorithm, of course, is that our future time series will not contain more extreme weather than has been experienced in the past. Nevertheless, this is not a major problem for our study because we are only interested in the number of occurrences of extreme heat stress and not in how extreme those events are. Furthermore, our study concerns only one location, so we are not concerned with spatial coherence, a concern that is easier to address by scaling regional models.
2.4. Modelling approach of the mean radiant temperature
The SOLWEIG 1.0 model is used to calculate spatial variations of Tmrt. SOLWEIG is a 2.5-dimensional model, in the sense that it applies a raster-based DEM (i.e. x and y coordinates with height attributes). The model is based on the framework of raster analysis of urban form originally presented by Ratti and Richens (1999), where high-resolution DEMs are used to extract geometrical parameters such as surface shadow patterns and SVFs. The meteorological input parameters in the current version are global short-wave radiation, air temperature and relative humidity. In addition to meteorological parameters, geographical location and urban geometry information (urban DEM) are required. The design of the DEM consists of both ground and building heights, i.e. the model allows for varying ground topography. Emissivity is given for surface and buildings (0.95 and 0.90). One value of albedo (0.15) is given for the entire study. In the current version of the SOLWEIG 1.0 model, vegetation is not considered. SOLWEIG is suitable for use with extended meteorological datasets. The computational time for a DEM of 350 × 350 pixels and 20 years of hourly meteorological data is approximately 12 h on a regular PC. The model has been evaluated and reveals good agreement between modelled and measured values of Tmrt (Lindberg et al., 2008). Since SOLWEIG can calculate spatial variations in Tmrt, areal means for the four selected urban places are derived in this study. The extent of the areal means is shown in Figure 2.
The physiologically equivalent temperature (PET) index of Mayer and Höppe (1987) is used to quantify the combined effect of future changes in air temperature, air humidity, wind speed and radiation on the people's thermal perception and thermo-physiological stress. To determine PET, the meteorological variables—air temperature, vapour pressure, wind speed and mean radiant temperature—are needed. Characteristics of human beings are set as constants, i.e. the internal heat production is 80 W and the heat transfer resistance of clothing is 0.9 clo (Matzarakis and Mayer, 1996). The wind input data at a height of 10 m is recalculated to 1.1 m, which corresponds to the average height of the centre of gravity for adult (Mayer and Höppe, 1987). The waning of wind speed can be approximated with a power function:
where W is the wind speed at height H or h and a is the exponent of altitudinal swell determined by the surface roughness (Carruthers, 1943). Given the surface roughness in Gothenburg, the 1.1-m wind velocity is calculated to be 25% of the 10-m velocity for the square, and 10% for the courtyard and street canyons (Holmer, 1978).
3.1. Intra-urban spatial variations in mean radiant temperature
One clear summer day, i.e. 21 June 1977, is selected to examine intra-urban spatial variations in Tmrt within the central parts of Gothenburg. Results are presented as hourly and daytime averages (Figures 4–6).
Figure 4 shows simulated hourly average Tmrt at 10 a.m. There are large spatial variations in Tmrt within short distances as a result of the complex urban geometry, e.g. street direction, street width and building height. The Tmrt is generally higher in open areas compared to narrow street canyons and small courtyards as a result of the higher amount of direct-beam solar radiation. The Tmrt of the square, the courtyard, the adjacent north–south orientated street canyon and the adjacent east–west orientated street canyon are 53.1 °C on average (Figures 2 and 3(a)), 41.6 °C (Figures 2 and 3(b)), 42.1 °C (Figures 2 and 3(c)) and 31.7 °C (Figures 2 and 3(d)), respectively. This gives Tmrt differences between the square, the courtyard and the two street canyons of 11.5, 11.0 and 21.4 °C, respectively. Note also the large spatial variations within the square (32.3 °C) and the increase in Tmrt of about 6.0 °C in front of south-facing walls that are lit by the sun.
Figure 5 shows the temporal variation of Tmrt at single points within the study area from sunrise to sunset. The square is generally warmer than the courtyard and the two street canyons, as a result of more incoming solar radiation. However, when the point within the courtyard and the street canyons is lit by the sun, the Tmrt is 1.5–8.6 °C higher than that of the point within the square.
Daytime (i.e. from sunrise to sunset) average Tmrt on a clear summer day is shown in Figure 6. The Tmrt of the square, the adjacent courtyard, the north–south orientated street canyon and the adjacent east–west orientated street canyon are 37.3 (on average), 25.5, 26.8 and 27.4 °C, respectively. This gives Tmrt differences between the square, the courtyard and the two street canyons of 11.8, 10.5 and 10.1 °C, respectively. The direction of the narrow street canyons (H/W ≈ 2) appears to be less important when looking at daytime average Tmrt. Furthermore, the spatial variation within the square is less pronounced on a daytime average (14.1 °C) when compared with an hourly average (32.3 °C).
The average of daytime Tmrt for 1977 (typical year) is shown in Figure 7. Despite weather and seasonal differences, large spatial variations in Tmrt still exist within short distances on a yearly basis; these are similar in pattern, but smaller in magnitude, than those simulated during a clear summer day (Figure 5). On a yearly average, the square is 5.2 and 5.7 °C warmer than the adjacent north–south and east–west orientated street canyons during daytime.
3.2. Seasonal variations of daytime mean radiant temperature in different built-up structures
Figure 8 shows box plots of simulated areal means of daytime hourly Tmrt of the square, the courtyard, the north–south and the east–west orientated street canyon, displayed according to season. As shown, there are large seasonal differences in the areal means of daytime hourly Tmrt, with the highest median value (21.3 to 28.8 °C) in summer (JJA) and the lowest median value (−0.2 to 1.9 °C) in winter (DJF).
In summer, the square is considerably warmer (6.2 to 7.2 °C) than the other three places. In addition, the square has the largest range of Tmrt. The square can, therefore, be both the coldest and the warmest place. The other three places have about the same magnitude and range of Tmrt, indicating that the orientation of a narrow street canyon is of less importance. The situation is roughly the same in spring, but in autumn there are almost no differences in daytime median Tmrt between the four places. In winter, however, the daytime median Tmrt of square is 0.7, which is between 0.6 and 1 °C lower than within the two street canyons and the courtyard, respectively (Figure 8(a)).
3.3. Potential change in thermal stress in Gothenburg due to climate change
According to the ECHAM5-based scenario, the average annual daytime Ta is expected to increase by 2.8 °C between 1980–1999 and 2080–2099. This is 0.4 °C less than the calculated increase in average annual daytime Tmrt (3.2 °C).
Table III shows simulated average annual numbers of daytime hours with different grades of thermo-physiological stress for the two periods (1980–1999 and 2080–2099) and the four urban places (point values). The square is both the coldest and the warmest place of them all. The relative difference between the four places is larger in summer than in winter, which corresponds well with the results presented in Figure 8.
Table III. Average daytime annual numbers of hours with different grades of thermo-physiological stress according to Matzarakis and Mayer (1996) for the period 1980–2000 and 2080–2100 and the four urban places
PET ( °C)
Grade of thermo-physiological stress
1980–1999 yearly average number of hours
2080–2099 yearly average number of hours
Change in yearly average number of hours
(1) Square, (2) east–west oriented street canyon, (3) north–south oriented street canton and (4) courtyard.
The future climate is simulated using statistically downscaled data from ECHAM5/MPI-OM model under the SRES A1B emission scenario.
Extreme cold stress
Strong cold stress
Moderate cold stress
Slight cold stress
No thermal stress
Slight heat stress
Moderate heat stress
Strong heat stress
Extreme heat stress
The number of hours with strong and extreme heat stress will triple in Gothenburg between 1980–1999 and 2080–2099, according to the selected climate scenario. This equates to an increase of about 20–100 h a year depending on geometry. The more open built structure the larger increase in number of hours of strong/extreme heat stress. Conversely, the number of hours with strong and extreme cold stress will decrease in winter. The decrease is more or less the same for all four places, i.e. an average of 400–450 h a year or 20–25%. Furthermore, it is shown that the number of hours with no thermal stress will increase in all four places by about 40–200 h a year. The increase is especially pronounced in the two streets where the number of hours with no thermal stress will increase by about 200 h a year.
4.1. The influence of urban geometry on thermal comfort
Our results show that large spatial differences in Tmrt exist within short distances, on hourly, daytime and yearly average (Figures 4–7). This is mainly due to the urban geometry, e.g. street direction, spacing and width and building height, which control the amount of solar radiation that reaches the ground and building surfaces and the reflection of short- and long-wave radiation. On clear summer days, the hourly average Tmrt of the square can be more than 20 °C higher than that of an adjacent narrow street canyon (Figure 4). However, when a narrow street canyon is lit by the sun, the Tmrt there can be up to 8.6 °C higher than within the square, as a result of multiple reflection of long- and short-wave radiation from the surrounding walls (Figure 5). Narrow street canyons and courtyards also tend to be warmer than large open areas in late afternoons. This is due to diminished loss of long-wave radiation through multiple reflections at the canyon surfaces (Figure 5), which leads to less cooling around sunset (Holmer et al., 2007). The largest intra-urban differences in Tmrt occur in summer and spring (Figure 8). This is a result of more solar radiation and fewer clouds in summer and spring than in autumn and winter.
As shown in Figures 4–8, a dense built-up structure mitigates extreme swings in Tmrt improving outdoor comfort conditions both in summer and in winter. The study confirms the potential for using geometry to mitigate daytime thermal stress (Gulyas et al., 2006; Johansson, 2006; Emmanuel and Fernando, 2007; Ali-Toudert and Mayer, 2007; Mayer et al., 2008). Urban geometry has a direct impact on Tmrt (Figures 4–8). This is to be compared with Ta, which is characterized by rather small daytime spatial variations (Emmanuel and Fernando, 2007; Mayer et al., 2008). Since Tmrt is critical to outdoor thermal comfort, mitigation options ought to focus on Tmrt, rather than Ta alone (Emmanuel and Fernando, 2007).
4.2. Potential change in thermal stress in future warmer climates and its implication for health and well-being
Hourly statistically downscaled meteorological data based on the ECHAM5 model under the A1B emission scenario were used to calculate potential changes in Tmrt, and PET between 1980–1999 and 2080–2099 for the central parts of Gothenburg. The results show that the increase in average yearly daytime Tmrt (3.2 °C) is larger than the increase in average yearly daytime Ta (2.8 °C). This difference could be explained by changes in cloudiness. As shown in Table II, the cloudiness decreases in summer in our scenario (as seen from the global radiation), which leads to an increase in Tmrt. The fact that the increase in Tmrt is not linearly proportional to the increase in Ta highlights the importance of including information on Tmrt (or another thermal comfort measure) in future climate scenarios, in order to provide a more realistic measure of the combined effect of future climate changes on human health.
According to the ECHAM5-based scenario, the number of hours with strong and extreme heat stress is expected to triple in Gothenburg by the end of this century. This equates to 20–100 h a year depending on geometry. However, the number of hours with strong/extreme cold stress will decrease more, i.e. by 400–450 h a year. Furthermore, the number of no thermal stress hours will increase by about 40–200 h a year. This means that although the problems of excessive temperature will increase in summer, outdoor thermal comfort will improve significantly in winter, spring and autumn in a future warmer climate. A similar result has been reported from Germany (Matzarakis and Endler, 2010).
The number of hours with strong/extreme heat stress increases more in the square compared to the courtyard and the two streets, as shown in Table III. On the other hand, the positive effects of a warmer climate (decrease in strong/extreme cold stress and increase in no thermal stress) are less in the square than in the other three places. This means that a dense built structure will not only mitigate the swings in PET (thermal stress) but also mitigate the negative effects and benefit more from a warmer climate in terms of outdoor thermal comfort at street level, compared to a more open built structure.
Several studies have shown that both extreme cold and hot air temperatures increase the rate of mortality (Kalkstein and Greene, 1997; Keatinge et al., 2000; Pascal et al., 2006; Medina-Ramón and Schwartz, 2007). However, determining the overall potential health impacts of climate change is a complex issue. Although some studies suggest that increases in heat-related death will be so dramatic that they will not be compensated by decreases in cold-related deaths (Kalkstein and Greene, 1997; Medina-Ramón and Schwartz, 2007), others believe that human populations will adjust to warmer air temperatures and increases in heat-related deaths will be outweighed by larger declines in cold-related deaths (Keatinge et al., 2000). In Scandinavia, relative risks associated with heat and heat waves are expected to be stronger than the cold effects (Rocklöv and Forsberg, 2008). One explanation for this is that northern populations have not yet adapted to heat as they to the cold over a long period of time (Rocklöv and Forsberg, 2008). On the other hand, it has been shown that a moderate increase in air temperature and relatively warm summers have a positive impact on peoples' outdoor activities (Thorsson et al., 2004; Eliasson et al., 2007), mood (Eliasson et al., 2007) and mental health (Hartig et al., 2007) in Sweden.
Note that the hours with thermo-physiological stress presented in Table III are 20-year averages, which means that single years can have both less and more extreme heat and cold stress hours. Furthermore, the UHI effect is not taken into account in this study. According to Svensson et al. (2003) the central parts of Gothenburg can be up to 2–3 °C warmer at noon on clear and calm summer days than rural areas. Since the input climate data in the present study are from a rural weather station just outside Gothenburg, the number of hours with heat stress presented in Table III is most likely underestimated.
4.3. Performance and sensitivity of the statistically downscaled datasets
In this paper, we have used outputs from ECHAM5 model. We note that the model simulates the distribution of weather classes in Sweden realistically, and thus gives a realistic description of climate in the region.
The mid-range SRES A1B emissions scenario was selected to simulate the potential change in outdoor thermal comfort between 1980–1999 and 2080–2099. If future emissions are higher or lower than in this scenario, we would expect that changes in Tmrt would be similarly higher or lower. But the scaling may not be linear, because of the dependence on changes in solar radiation, which are closely related to changes in cloudiness and hence atmospheric circulation. This could be investigated by considering two substantially different emission scenarios, i.e. a low- and a high-end version.
The downscaling method has a number of limitations. Most obviously, the analogue technique cannot generate days with more extreme climate than those observed historically. We expect that days with higher maximum air temperatures than those that occurred in the historical record would occur in the future under the A1B scenario, but our downscaled data cannot show this. Instead, the downscaled scenario shows an increased number of days with air temperatures that are high (but not unprecedented) by current standards. Thus, we are confident that the number of hours categorized as extreme heat stress (highest category in Table III) is realistically simulated, even though the magnitude of the extreme heat stress is most likely underestimated. Secondly, the downscaling algorithm assumes that the relationships between the regional atmospheric situation and the station record continue to hold in the future. This assumption is present in some form in all statistical downscaling studies and is unavoidable when scenarios that realistically represent local weather time series are required.
4.4. Urban sprawl or compact cities?
For future urban planning it would appear necessary to find a solution between urban sprawl and compact cities. It has been shown that compact cities are more vulnerable to impacts of climate change and hot air temperatures than sparsely populated locations, especially for those citizens residing on the top floor of apartment buildings (McGeehin and Mirabelli, 2001) and in the urban core (Buechley et al., 1972). During extreme heat waves, the interaction between the weather at a regional scale and the UHI at a local scale aggravate the heat stress of inhabitants of compact cities (Jendritzky and Grätz, 1999; Pascal et al., 2006). However, on an intra-urban scale, a densely built structure has shown to mitigate daytime heat stress in warm climates (Ali-Toudert et al., 2005; Ali-Toudert and Mayer, 2006; Johansson, 2006; Emmanuel and Fernando, 2007). As shown in Figures 4–8, this is also valid for cities in temperate climates. But although a densely built structure mitigates daytime heat stress, it has been shown to cool more slowly than open-set built structure around sunset (Holmer et al., 2007), as a result of diminished loss of long-wave radiation through multiple reflections at the canyon surfaces. In urban areas, vegetation has been shown to effectively reduce air temperature and solar radiation at street level during periods of excessive temperatures (Picot, 2004; Ali-Toudert and Mayer, 2007; Bowler et al., 2010; Shashua-Bar et al., 2010) and could therefore be used to reduce heat stress in urban areas. The cooling effect depends on the vegetation type and coverage, as well as the characteristics of the surroundings, such as urban geometry, surface and building material, colour, etc. (Shashua-Bar et al., 2010).
For Scandinavian cities, with their cool summers and few air-conditioned buildings, a future warmer climate is most likely to increase the demand for cooling by airconditioners in residential buildings, health-care institutions, work places, etc. By designing with climate in mind and by providing a good network of outdoor public spaces, some of the problems caused by climate change will be mitigated, both outdoor and indoor, decreasing cooling needs and energy use.
The question of urban sprawl versus compact cities, in terms of thermal comfort, is complex. As pointed out by Mayer et al. (2008), high spatio-temporal resolution information about human thermal comfort is required for application in city planning. Over the years, several models have been developed to simulate outdoor thermal comfort, such as Rayman (Matzarakis, 2000; Matzarakis et al., 2010), ENVI-met (Bruse, 1999, 2006), TownScope (Teller and Azar, 2001) and SOLWEIG (Lindberg et al., 2008). SOLWEIG is a two-dimensional model, which makes it suitable for analysing the complex interaction of urban geometry and thermal environment at a neighbourhood or city block scale, or when extended meteorological datasets are used. In the current version of the SOLWEIG model (version 1.0), vegetation is not considered. However, inclusion of a vegetation and land-use scheme in the SOLWEIG model is under development and will be incorporated in future model versions.
The overall objective of this study was to explore the augmentation of global/regional climate changes by urban features such as geometry in a compact mid-rise high-latitude city. The magnitude of spatial and temporal variations of intra-urban Tmrt were quantified and future changes in Tmrt and thermal stress due to anthropogenic climate changes at the 2080–2099 time horizons were simulated for the central parts of Gothenburg, Sweden.
The study shows that large intra-urban variations of Tmrt exist on hourly, daytime and yearly average, as a result of urban geometry. The study also shows that a densely built structure mitigates extreme swings in Tmrt, improving outdoor comfort conditions both in summer and in winter. The fact that urban geometry has a direct impact on Tmrt confirms the potential for using geometry to mitigate daytime thermal stress in cities.
The ECHAM5-based climate scenario shows that the daytime Tmrt will increase by 3.2 °C by the end of this century. This is 0.4° more than the simulated increase in daytime mean Ta (2.8 °C), which is mainly a result of decreases in summer cloudiness. The fact that the increase in Tmrt is not linearly proportional to the increase in Ta highlights the importance of including information on either Tmrt or thermal comfort in climate scenarios, in order to describe the combined effects of changes in multiple climate variables and to more realistically measure the impact on humans.
The number of hours with strong and extreme heat stress is expected to increase by 20–100 h in Gothenburg by the end of this century according to the selected climate scenario. However, the number of hours with strong/extreme cold stress will decrease more, i.e. by 400–450 h a year. Furthermore, the number of hours with no thermal stress will increase by about 40–200 h a year. This means that although the problems of excessive temperature will increase in summer, outdoor climate will improve significantly in winter, spring and autumn in a future warmer climate.
The project is financially supported by Formas, the Swedish Research Council for Environment, Agriculture Sciences and Spatial Planning, within the European Commission programme Urban-Net. We gratefully acknowledge the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, for providing the NCEP Reanalysis data, and the modelling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) and the WCRP's Working Group on Coupled Modelling (WGCM), for their roles in making the WCRP CMIP3 multi-model dataset available. Support for this dataset was provided by the Office of Science, U.S. Department of Energy.