Heat Waves and Human Well-Being in Madrid (Spain)

: Heat waves pose additional risks to urban spaces because of the additional heat provided by urban heat islands (UHIs) as well as poorer air quality. Our study focuses on the analysis of UHIs, human thermal comfort, and air quality for the city of Madrid, Spain during heat waves. Heat wave periods are deﬁned using the long-term records from the urban station Madrid-Retiro. Two types of UHI were studied: the canopy layer UHI (CLUHI) was evaluated using air temperature time-series from ﬁve meteorological stations; the surface UHI (SUHI) was derived from land surface temperature (LST) images from MODIS (Moderate Resolution Imaging Spectroradiometer) products. To assess human thermal comfort, the Physiological Equivalent Temperature (PET) index was applied. Air quality was analyzed from the records of two air quality networks. More frequent and longer heat waves have been observed since 1980; the nocturnal CLUHI and both the diurnal and nocturnal SUHI experience an intensiﬁcation, which have led to an increasing number of tropical nights. Conversely, thermal stress is extreme by day in the city due to the lack of cooling by winds. Finally, air quality during heat waves deteriorates because of the higher than normal amount of particles arriving from Northern Africa.


Introduction
The warming of global temperatures during the last 150 years has been accompanied by an increase in the occurrence of heat waves [1,2], a trend that seems to be in close agreement with the most probable scenarios provided by climate models for the rest of the 21st century [3,4]. Heat wave impacts are diverse and widespread (crop failures, livestock losses, disruption of transportation, spreading of wildfires, power outages), but are particularly harmful in urban areas, due to the high population density and the urban heat island (UHI) effect, among other factors. UHI refers to a human-induced modification of the regional climate whereby the cities become warmer than their rural surroundings. Its magnitude is particularly intense at night, when an additional warming up to 7-10 • C have been recorded [5][6][7], although it can be reversed by day when the city becomes cooler than the surroundings [8]. Due to alterations in the natural landscape, urban environments absorb, preserve, and/or generate more heat, and gradually release heat compared with rural areas that cool off faster due to unobstructed outgoing radiation [9]. During heat waves, the prolongation of the diurnal thermal stress into the nocturnal hours deprives the population of relief, producing serious impacts on sensitive groups [10][11][12][13].
The Mediterranean area is widely recognized as one of the most vulnerable hot spots to climate change in the 21st century [14], particularly because of the predominance of the summer stable

Data and Methodology
The study area comprises the metropolitan area of Madrid, the capital of Spain. Madrid lies on the southern area of the Spanish Plateau, between the Jarama and Manzanares river basins, in the wider Tagus River catchment area. Orography is relatively rugged, since there are substantial differences in altitude, ranging from 846 m a.s.l. in the north to 543 m a.s.l. in the southeast. Due to its altitude and distance to the sea, the city experiences an extreme version of the Mediterranean climate, characterized by scarce precipitation, mostly in autumn and spring, and large annual and diurnal temperature changes. Summers are particularly warm (July average maximum temperature at Madrid-Retiro 32.1 • C) and dry (July 11 mm, August 9 mm), with most of the precipitation falling as convective showers.
Madrid is also the most populated urban area of Spain: 3.5 million people live in the municipality as of 2018, spread over 604.3 km 2 , but the population of the whole metropolitan area is approximately 6.5 million. The city experienced vigorous growth from the beginning of the century to 1970, increasing from approximately 500,000 to 3 million people. After that, urbanization focused on the surrounding locations ( Figure 1). Vulnerable population groups (e.g., the elderly and youngsters) represent 33.3% of the total, to which we must add people coming from other countries who can also be exposed to the heat, since Madrid has become a popular tourist destination [34]. There are 4,699,917 vehicles officially registered in the province of Madrid [35], 1,500,000 of which enter the core of the city every day [36]. Data used in this research originated from various sources. Meteorological data (daily maximum and minimum temperature; temperature, relative humidity, wind speed, and cloudiness at 00, 07, 13 and 18 UTC) from synoptic weather stations were obtained from AeMet, the Spanish meteorological office (Table 1 and Figure 2). Specifically, the long-term trends in heat waves were analyzed on the time series of Madrid-Retiro weather station retrieved from the SDATS (Spanish Daily Adjusted Temperature Series) database [37]. The SDATS timeseries were homogenized and adjusted to solve some temperature inhomogeneities derived from the "screen bias" (change from Montsouris to Stevenson screen), changes in placement and an artificial warming trend between 1894 and 1960, the period of the largest population increase.
Land surface temperature (LST) was analyzed using the product MYD11A1 (V6) -MODIS/Aqua Land Surface Temperature and Emissivity Daily L3 Global 1 km Grid SIN. Although the MODIS (Moderate Resolution Imaging Spectroradiometer) sensor is installed on two platforms, Aqua and Terra, only the first one was used because its approximate time of passage over the Iberian Peninsula (1:30 a.m. and 1:30 p.m.) is more suitable for the analysis of the surface UHI. Compared with other platforms (e.g., ASTER or LANDSAT), the lower spatial resolution (~1 km 2 ) is compensated for by a higher temporal resolution, which enables studying the temporal dimensions of the phenomenon. This product has already been used in research on surface UHI in cities of varying sizes and climates, for example in Europe [38,39] and Asia [40]. Previous studies confirmed the accuracy of the MODIS LST product, particularly by night under clear skies [41,42]. Additional technical details can be consulted on the NASA website (https://lpdaac.usgs.gov/). The original images were subjected to a basic quality control analysis using only those days with more than 90% valid pixels within the limits of the image (a total of 7,645 images). To relate the surface temperature to land uses, the MODIS images were cross-referenced with the Global Land Cover 2000 database [43]. This database covers the entire planet, providing 22 categories with a resolution of 1 km 2 , obtained from the VEGETATION Data used in this research originated from various sources. Meteorological data (daily maximum and minimum temperature; temperature, relative humidity, wind speed, and cloudiness at 00, 07, 13 and 18 UTC) from synoptic weather stations were obtained from AeMet, the Spanish meteorological office (Table 1 and Figure 2). Specifically, the long-term trends in heat waves were analyzed on the time series of Madrid-Retiro weather station retrieved from the SDATS (Spanish Daily Adjusted Temperature Series) database [37]. The SDATS timeseries were homogenized and adjusted to solve some temperature inhomogeneities derived from the "screen bias" (change from Montsouris to Stevenson screen), changes in placement and an artificial warming trend between 1894 and 1960, the period of the largest population increase.
Land surface temperature (LST) was analyzed using the product MYD11A1 (V6) -MODIS/Aqua Land Surface Temperature and Emissivity Daily L3 Global 1 km Grid SIN. Although the MODIS (Moderate Resolution Imaging Spectroradiometer) sensor is installed on two platforms, Aqua and Terra, only the first one was used because its approximate time of passage over the Iberian Peninsula (1:30 a.m. and 1:30 p.m.) is more suitable for the analysis of the surface UHI. Compared with other platforms (e.g., ASTER or LANDSAT), the lower spatial resolution (~1 km 2 ) is compensated for by a higher temporal resolution, which enables studying the temporal dimensions of the phenomenon. This product has already been used in research on surface UHI in cities of varying sizes and climates, for example in Europe [38,39] and Asia [40]. Previous studies confirmed the accuracy of the MODIS LST product, particularly by night under clear skies [41,42]. Additional technical details can be consulted on the NASA website (https://lpdaac.usgs.gov/). The original images were subjected to a basic quality control analysis using only those days with more than 90% valid pixels within the limits of the image (a total of 7,645 images). To relate the surface temperature to land uses, the MODIS images were cross-referenced with the Global Land Cover 2000 database [43]. This database covers the entire planet, providing 22 categories with a resolution of 1 km 2 , obtained from the VEGETATION sensor (SPOT 4 satellite). The types of land uses (Land Cover Classification System, LCCS) are inspired by the FAO (Food and Agriculture Organization).   Despite of the relevance of the phenomenon, no universal definition of a heat wave exists [48][49][50]. A heat wave can be considered as an extended period of hot days; from the statistical point of view, a hot day may be identified when air temperature rises beyond a threshold, usually calculated from the temperature distribution recorded at a meteorological station. Using this basic procedure, but considering the results of previous analyses of the impact of the high temperatures in Madrid [51][52][53], we identified those days in which the maximum temperature exceeds 36.5 °C at Madrid-Retiro as extreme hot days (EHD). This value is equivalent to the 95th percentile of the daily maximum summer (June, July, and August) temperature for the 1981-2010 period. Consequently, a heat wave was defined as a period of at least 3 consecutive EHDs, becoming heat wave days (HWDs). In case of  Despite of the relevance of the phenomenon, no universal definition of a heat wave exists [48][49][50]. A heat wave can be considered as an extended period of hot days; from the statistical point of view, a hot day may be identified when air temperature rises beyond a threshold, usually calculated from the temperature distribution recorded at a meteorological station. Using this basic procedure, but considering the results of previous analyses of the impact of the high temperatures in Madrid [51][52][53], we identified those days in which the maximum temperature exceeds 36.5 • C at Madrid-Retiro as extreme hot days (EHD). This value is equivalent to the 95th percentile of the daily maximum summer (June, July, and August) temperature for the 1981-2010 period. Consequently, a heat wave was defined as a period of at least 3 consecutive EHDs, becoming heat wave days (HWDs). In case of an intermediate day between two heat waves, recording a maximum temperature below such threshold, but above the 90th percentile, it was also considered an EHD and both heat waves were merged into a single heat wave.
The magnitude of the CLUHI was quantified as the air temperature difference (∆T) between the urban (Madrid-Retiro) and any other non-urban stations, calculated for both maximum (daytime) and minimum (nighttime). A positive value indicates that Madrid-Retiro is warmer, whereas a negative value indicates that is cooler. The Local Climate Zones (as defined by Stewart and Oke [47]) for each station are presented in Table 1. Madrid-Retiro's station is located in a park (LCZ B), approximately 100 m from a dense urban area (LCZ2). Barajas is the weather station of the busiest airport in Spain, and it is located over a Low Plant (LCZ D) area, close to the main runway, with edifications (terminals) at distances of less than 500 m (LCZ8). Cuatro Vientos, Getafe, and Torrejón de Ardoz are other meteorological stations located at smaller airports, nearby urban areas of different densities. Also, the number of summer tropical nights (Tmin > 20 • C) [54] was calculated for the long-term evaluation of the nighttime heat conditions.
To describe and quantify heat waves from a human-biometeorological point of view, the Physiological Equivalent Temperature (PET) index was calculated by the RayMan model [55,56]. PET describes the effect of the thermal environment as a temperature value ( • C), which facilitates interpretation of the results, and is based on the full human energy balance. PET is equivalent to the air temperature at which, in a typical indoor setting, the heat balance of the human body (work metabolism, 80 W of light activity, added to basic metabolism; heat resistance of clothing, 0.9 clo) is maintained with core and skin temperatures equal to those of a body in the assessed conditions [57]. The classification of PET and the related physiological strain are given in Table 2. To evaluate potential health risks for the population due to abnormal concentrations of air pollutants, daily averages, exceedances and ratios during normal days and HWD, plus a daily pollution index (PI), were calculated for selected pollutants (PM 10 , NO 2 and O 3 ). The PI [59] defines five distinct air quality classes and matches a reference concentration (C p ) of each individual pollutant (either daily PM 10 , maximum 8-hour O 3 , or maximum hourly NO 2 concentrations) to one of the five categories ( Table 3). The PI is expressed by: where BP hi and BP lo are the upper and lower concentration breakpoints of each PI category, respectively, to which C p corresponds; and PI hi and PI lo are the upper and lower PI breakpoints of the same category, respectively (Table 3). The chi-square test of independence was used to evaluate whether a significant association exists between the categories of two dichotomous variables, HWD and Saharan outbreaks (days). It is a non-parametric (distribution free) test designed to analyze group differences, being robust with respect to the data distribution. Specifically, it does not require equality of variances among the study groups nor homoscedasticity in the data.

Results
In this section, we first review the main temporal characteristics of heat waves in the Madrid metropolitan area to continue with an analysis of how heat waves affect the UHI phenomenon. Finally, we show the impact of heat waves on the air quality conditions. Another important feature of Madrid's recent summer climate is the increase in the frequency of EHD after the 1980s, which has occurred in longer spells ( Figure 4a). Before 1980, occurrences of EHD was a rare phenomenon, most of the time as single-or two-day events. The most remarkable summer was 1949, which experienced both a four-day and a three-day event; however, 18 summers (60%) between the 1948 and 1979 recorded no heat wave. The situation reversed after 1979, as a clear increasing trend is detected; the slope of the linear fit equals 0.27 more HWD per year. We found that 2015 emerged as the absolute record year in the frequency of HWD, with 24 days, followed by 2017, with 20 days, and 1991 with 17 days.

Long-Term Trends on Heat Waves Occurrence
Derived from the increase in the number of HWD and the length of the episodes, the core of the summer, spanning from July 16 to August 15 (traditionally known as "canícula") has extended, particularly to June [60,61]. Most of the HWD until the 1980s were observed in July (74%) and August (22%); the earliest HWD occurred on June 29 (1950) and the latest on August 17 (1949). During the last four decades, summer is uniformly hotter than before, and the period prone to heat waves extends Regarding the possible causes of the warming experienced since the 1980s, Figure 4b shows the daily maximum temperature distribution function corresponding to almost equally long temporal periods, 1948-1979 and 1980-2017. Visually, both distributions are quite similar, but the mean and the median of the second period are displaced almost 1.5 • C to the right. This shift suggests the impact of the global trend to higher temperatures on the aforementioned warming. Another important feature of Madrid's recent summer climate is the increase in the frequency of EHD after the 1980s, which has occurred in longer spells (Figure 4a). Before 1980, occurrences of EHD was a rare phenomenon, most of the time as single-or two-day events. The most remarkable summer was 1949, which experienced both a four-day and a three-day event; however, 18 summers (60%) between the 1948 and 1979 recorded no heat wave. The situation reversed after 1979, as a clear increasing trend is detected; the slope of the linear fit equals 0.27 more HWD per year. We found that 2015 emerged as the absolute record year in the frequency of HWD, with 24 days, followed by 2017, with 20 days, and 1991 with 17 days.
Derived from the increase in the number of HWD and the length of the episodes, the core of the summer, spanning from July 16 to August 15 (traditionally known as "canícula") has extended, particularly to June [60,61]. Most of the HWD until the 1980s were observed in July (74%) and August (22%); the earliest HWD occurred on June 29 (1950) and the latest on August 17 (1949). During the last four decades, summer is uniformly hotter than before, and the period prone to heat waves extends to June (13%) and August (22%); the earliest have been recorded on June 11, 2017 and the latest on August 29, 2009. Regarding the possible causes of the warming experienced since the 1980s, Figure 4b shows the daily maximum temperature distribution function corresponding to almost equally long temporal periods, 1948-1979 and 1980-2017. Visually, both distributions are quite similar, but the mean and the median of the second period are displaced almost 1.5 °C to the right. This shift suggests the impact of the global trend to higher temperatures on the aforementioned warming.  However, a regional component of this summer warming may exist. Figure 5a displays the anomalous pattern of 500 hPa geopotential heights, composited from the 1980-2017 period minus the 1948-1979 period. The positive area over the Western Mediterranean Basin is consistent with the observed warming trend of maximum temperatures, since the large-scale pattern should induce a broad-scale subsidence and adiabatic compression upon the tropospheric air masses.  However, a regional component of this summer warming may exist. Figure 5a displays the anomalous pattern of 500 hPa geopotential heights, composited from the 1980-2017 period minus the 1948-1979 period. The positive area over the Western Mediterranean Basin is consistent with the observed warming trend of maximum temperatures, since the large-scale pattern should induce a broad-scale subsidence and adiabatic compression upon the tropospheric air masses. The pattern displayed in Figure 5a ensures an upper-level southerly flux with negative relative vorticity and a warm advection over the studied area, probably of Saharan origin, as portrayed by the 850 hPa temperature field (Figure 5b). The polar plots depicting the bivariate relationship between 13 UTC summer air temperature at Retiro and simultaneous surface wind and 850 hPa records from Barajas airport ( Figure 6) emphasize the southerly component of the wind, more evident at the surface due to the orientation of the Jarama valley, where the airport facilities are located. The pattern displayed in Figure 5a ensures an upper-level southerly flux with negative relative vorticity and a warm advection over the studied area, probably of Saharan origin, as portrayed by the 850 hPa temperature field (Figure 5b). The polar plots depicting the bivariate relationship between 13 UTC summer air temperature at Retiro and simultaneous surface wind and 850 hPa records from Barajas airport (Figure 6) emphasize the southerly component of the wind, more evident at the surface due to the orientation of the Jarama valley, where the airport facilities are located.

CLUHI
Due to the aforementioned increased frequency of heat waves after 1980, we relied upon the period 1980-2017 for the analysis of the impact of those events upon Madrid´s summer CLUHI, which is a frequent phenomenon by night, as a response of its large urban size and predominance of stable atmospheric conditions. Positive differences in minimum temperatures at Madrid-Retiro occur almost 9 of each 10 nights, with average differences up to 2.5 °C and maximums up to 5.5 °C ( Figure  7). Distance to the city center and topography explain the primacy of Torrejón de Ardoz (LCZ2/B vs. LCZ8/D difference). Barajas (also a LCZ2/B vs. LCZ8/D difference) experiences similar but less intense differences, probably because the station is surrounded by the facilities of the busiest airport in Spain. Cuatro Vientos and Getafe (LCZ2/B vs. LCZ5/D, and LCZ2/B vs. LCZ6/D differences, respectively), display weaker differences in response to their suburban and southward locations, and less favorable topography to nocturnal cooling. By day, urban-to-rural differences reverse for all

CLUHI
Due to the aforementioned increased frequency of heat waves after 1980, we relied upon the period 1980-2017 for the analysis of the impact of those events upon Madrid´s summer CLUHI, which is a frequent phenomenon by night, as a response of its large urban size and predominance of stable atmospheric conditions. Positive differences in minimum temperatures at Madrid-Retiro occur almost 9 of each 10 nights, with average differences up to 2.5 • C and maximums up to 5.5 • C (Figure 7). Distance to the city center and topography explain the primacy of Torrejón de Ardoz (LCZ2/B vs. Atmosphere 2019, 10, 288 9 of 21 LCZ8/D difference). Barajas (also a LCZ2/B vs. LCZ8/D difference) experiences similar but less intense differences, probably because the station is surrounded by the facilities of the busiest airport in Spain. Cuatro Vientos and Getafe (LCZ2/B vs. LCZ5/D, and LCZ2/B vs. LCZ6/D differences, respectively), display weaker differences in response to their suburban and southward locations, and less favorable topography to nocturnal cooling. By day, urban-to-rural differences reverse for all stations.
Heat waves enhance the nocturnal temperature contrast by night. The average minimum temperature difference between Madrid-Retiro and Torrejón de Ardoz rises from 2.9 • C during normal conditions to 3.4 • C during heat waves; for Barajas, the corresponding values are 1.9 • C and 2.6 • C. Conversely, no differences are observed at Cuatro Vientos and Getafe. By day, no substantial differences are observed during heat waves.
Due to the aforementioned increased frequency of heat waves after 1980, we relied upon the period 1980-2017 for the analysis of the impact of those events upon Madrid´s summer CLUHI, which is a frequent phenomenon by night, as a response of its large urban size and predominance of stable atmospheric conditions. Positive differences in minimum temperatures at Madrid-Retiro occur almost 9 of each 10 nights, with average differences up to 2.5 °C and maximums up to 5.5 °C ( Figure  7). Distance to the city center and topography explain the primacy of Torrejón de Ardoz (LCZ2/B vs. LCZ8/D difference). Barajas (also a LCZ2/B vs. LCZ8/D difference) experiences similar but less intense differences, probably because the station is surrounded by the facilities of the busiest airport in Spain. Cuatro Vientos and Getafe (LCZ2/B vs. LCZ5/D, and LCZ2/B vs. LCZ6/D differences, respectively), display weaker differences in response to their suburban and southward locations, and less favorable topography to nocturnal cooling. By day, urban-to-rural differences reverse for all stations.
Heat waves enhance the nocturnal temperature contrast by night. The average minimum temperature difference between Madrid-Retiro and Torrejón de Ardoz rises from 2.9 °C during normal conditions to 3.4 °C during heat waves; for Barajas, the corresponding values are 1.9 °C and 2.6 °C. Conversely, no differences are observed at Cuatro Vientos and Getafe. By day, no substantial differences are observed during heat waves. As a direct consequence of the generalized warming, the number of summer tropical nights has increased since 1961 across the region, but the trend is faster in Madrid-Retiro (0.56 days per year) than in Torrejón de Ardoz (0.26 days per year) because of the intensification of nocturnal urban warming ( Figure 8). Both trends are statistically significant at a probability level of 0.001 according to a t-test. As a direct consequence of the generalized warming, the number of summer tropical nights has increased since 1961 across the region, but the trend is faster in Madrid-Retiro (0.56 days per year) than in Torrejón de Ardoz (0.26 days per year) because of the intensification of nocturnal urban warming ( Figure 8). Both trends are statistically significant at a probability level of 0.001 according to a t-test.

SUHI
The response of the earth surface to the atmospheric conditions created by heat waves follows the same direction observed in the case of air temperatures. Figure 9 displays the spatial pattern of

SUHI
The response of the earth surface to the atmospheric conditions created by heat waves follows the same direction observed in the case of air temperatures. Figure 9 displays the spatial pattern of average LST corresponding to the nocturnal and diurnal images during HWD, and their pixel-to-pixel differences with respect to normal summer conditions, obtained for the 2002-2017 period.
As a whole, the maps indicate that surface temperature anomalies coincide, to a great extent, with the built environment. The more irregular diurnal spatial pattern during daytime contrasts with the concentric disposition of the surface temperature field by night, when the center of Madrid is surrounded by an archipelago of smaller heat islands. By day, shadows and the greater heat capacity of urban materials explain the slower warming of the city, forming a surface urban "cool" island (SUCI). The lowest temperatures are visible at the western border of the city, in connection with the valley of the Manzanares river and the urban forest of Casa de Campo (about 10 • C less than the warmest spots). At the southeast, the warmest anomalies appear due to the presence of bare soils under gypsum rocks, accompanied by some isolated cool spots that coincide with wet surfaces (Jarama river). At night, the slow release of the heat accumulated in the city keeps the city warmer (about 24 • C) than the surroundings. Several lines, following the transportation system, join Madrid with the smaller suburban cities, like Móstoles and Getafe, whose islands are weaker (about 22 • C). The Barajas airport complex, in the north, becomes one the coolest spots, as well as the valleys of the Manzanares (NW) and Jarama (SE) rivers, highlighting the role of the topography in channeling cool air masses from the surrounding mountains and plateaus.
The map of nighttime anomalies confirms that, as with air temperature, SUHI intensifies during heat waves. The area corresponding to the highest anomalies is displaced to the more northerly neighborhoods of Madrid, which are also characterized by a high build-up density and narrow streets, but smaller urban nuclei also show positive anomalies of LST. Negative anomalies show a longitudinal N-S arrangement, which roughly corresponds to the background of the valleys of the Manzanares (to the west) and Jarama (to the east) rivers. By daytime, heat waves intensify the SUCI, by increasing LST at two particular locations: the NW and the SE.
To properly quantify the impact of the different land uses on surface temperature due to heat waves, we calculated the average LST values pixel-by-pixel, and then we grouped them together according to several categories ( Figure 10). According to the Global Land Cover 2000 database, urban uses represent approximately 14% of each image, woodlands 9.1%, rain fed crops 67.1%, and scrub/grassland 9.8% (for sake of simplicity, the two latter were grouped together into a class named herbaceous).
The thermal diversity is higher by day as a result of the multiple reflections of solar radiation associated with the sun's path over the horizon, the larger albedo values, and the shadows generated by buildings or the presence of vegetation within urban areas, in contrast to a constant emission of infrared radiation by night. Urban spaces are confirmed to be areas with lower thermal amplitude, cooler by day (46.9 • C) and warmer by night (23.1 • C). The type of vegetation also controls the values of LST by daytime (forests being the coolest), although the differences diminish by night.
Atmosphere 2019, 10, x FOR PEER REVIEW 11 of 22 Manzanares (to the west) and Jarama (to the east) rivers. By daytime, heat waves intensify the SUCI, by increasing LST at two particular locations: the NW and the SE. To properly quantify the impact of the different land uses on surface temperature due to heat waves, we calculated the average LST values pixel-by-pixel, and then we grouped them together according to several categories ( Figure 10). According to the Global Land Cover 2000 database, urban uses represent approximately 14% of each image, woodlands 9.1%, rain fed crops 67.1%, and scrub/grassland 9.8% (for sake of simplicity, the two latter were grouped together into a class named herbaceous).
The thermal diversity is higher by day as a result of the multiple reflections of solar radiation associated with the sun's path over the horizon, the larger albedo values, and the shadows generated by buildings or the presence of vegetation within urban areas, in contrast to a constant emission of infrared radiation by night. Urban spaces are confirmed to be areas with lower thermal amplitude, cooler by day (46.9 °C) and warmer by night (23.1 °C). The type of vegetation also controls the values of LST by daytime (forests being the coolest), although the differences diminish by night. The arrival of hot air masses during heat waves produces a general increase in LST at any time. Comparing the temperature of urban surfaces with the other land uses, we found that the enhancement of the diurnal "cool" island is slightly larger (−1.5 • C normal days, −2.1 • C heat wave days) than the nocturnal heat island (2 • C versus 2.3 • C, respectively) because of the greater temperature increase of forest and herbaceous surfaces.
The arrival of hot air masses during heat waves produces a general increase in LST at any time. Comparing the temperature of urban surfaces with the other land uses, we found that the enhancement of the diurnal "cool" island is slightly larger (−1.5 °C normal days, −2.1 °C heat wave days) than the nocturnal heat island (2 °C versus 2.3 °C, respectively) because of the greater temperature increase of forest and herbaceous surfaces.

Physiologic Urban Heat Island
Urban comfort conditions were assessed at each meteorological observatory using PET values, also calculated for the 1980-2017 period. Figure 11 presents the frequency of occurrence of each PET classes during normal days and HWD, during both nighttime and daytime. A clear shift to higher PET classes is depicted during HWD, particularly during daytime. The figure also reveals that the distribution of PET classes on each observatory is in accordance with the distance to the city center. As expected, the population suffers very hot conditions during most of the HWD days at 13 UTC, particularly those living in the city center (almost 100% of the days); very hot conditions were also common in Cuatro Vientos and Getafe, as the former weather station also experiences a weaker UHI and the latter occupies a southward location. Night hours provide relief of the heat stress, since comfortable sensations are predominant; however, slightly warm conditions are present in urban observatories, whereas slightly cool conditions characterize nights in the outskirts of Madrid. Figure 10. Statistical distribution of LST by land uses for (a) nighttime and (b) daytime during normal summer days (NO) and heat wave days (HWD). Grey boxes account for the lower (25%) and upper (75%) quartiles, red rectangles represent the median, blue dots the average, and whiskers indicate the value of the 5% and 95% percentiles.

Physiologic Urban Heat Island
Urban comfort conditions were assessed at each meteorological observatory using PET values, also calculated for the 1980-2017 period. Figure 11 presents the frequency of occurrence of each PET classes during normal days and HWD, during both nighttime and daytime. A clear shift to higher PET classes is depicted during HWD, particularly during daytime. The figure also reveals that the distribution of PET classes on each observatory is in accordance with the distance to the city center. As expected, the population suffers very hot conditions during most of the HWD days at 13 UTC, particularly those living in the city center (almost 100% of the days); very hot conditions were also common in Cuatro Vientos and Getafe, as the former weather station also experiences a weaker UHI and the latter occupies a southward location. Night hours provide relief of the heat stress, since comfortable sensations are predominant; however, slightly warm conditions are present in urban observatories, whereas slightly cool conditions characterize nights in the outskirts of Madrid. Atmosphere 2019, 10, x FOR PEER REVIEW 13 of 22

Air Quality During Heat Waves.
The mean pollution levels during HWD and their ratios (% of variation) regarding normal days for selected pollutants are presented in Figure 12. The mean concentration of all pollutants increased during HWD, but the increase was significant in most stations only for PM10 levels according to a ttest. The increases in NO2 levels were less pronounced, and non-relevant for O3 levels. The higher increases of PM10 concentration (sometimes more than 60%) were observed at stations inside the city, rather than in suburban places (less than 25% on average), mainly at the NE, which might be a consequence of the displacement of the urban plume by the southerly winds during HWD. As stated previously, the moderate increase in NO2 concentrations showed no clear spatial pattern.

Air Quality During Heat Waves
The mean pollution levels during HWD and their ratios (% of variation) regarding normal days for selected pollutants are presented in Figure 12. The mean concentration of all pollutants increased during HWD, but the increase was significant in most stations only for PM 10 levels according to a t-test. The increases in NO 2 levels were less pronounced, and non-relevant for O 3 levels. The higher increases of PM 10 concentration (sometimes more than 60%) were observed at stations inside the city, rather than in suburban places (less than 25% on average), mainly at the NE, which might be a consequence of the displacement of the urban plume by the southerly winds during HWD. As stated previously, the moderate increase in NO 2 concentrations showed no clear spatial pattern. As expected, the increase in the average concentration levels during HWD resulted in the air quality standards being exceeded. Figure 13 shows the probability of occurrence of a PM10 exceedance (more than 50 μg m −3 ) during HWDs with respect to normal days. This dimensionless value was calculated by dividing the probability of a PM10 exceedance occurrence during a HWD by the probability of an exceedance occurrence during a normal summer day (probability of occurrence of an exceedance calculated as number of days with exceedances by total number of days at each station). The aforementioned spatial pattern of concentration around the city center is also visible in Figure 13, where a PM10 exceedance is up to 12 times more probable under a heat wave than a normal day.   As expected, the increase in the average concentration levels during HWD resulted in the air quality standards being exceeded. Figure 13 shows the probability of occurrence of a PM 10 exceedance (more than 50 µg m −3 ) during HWDs with respect to normal days. This dimensionless value was calculated by dividing the probability of a PM 10 exceedance occurrence during a HWD by the probability of an exceedance occurrence during a normal summer day (probability of occurrence of an exceedance calculated as number of days with exceedances by total number of days at each station). The aforementioned spatial pattern of concentration around the city center is also visible in Figure 13, where a PM 10 exceedance is up to 12 times more probable under a heat wave than a normal day. As expected, the increase in the average concentration levels during HWD resulted in the air quality standards being exceeded. Figure 13 shows the probability of occurrence of a PM10 exceedance (more than 50 μg m −3 ) during HWDs with respect to normal days. This dimensionless value was calculated by dividing the probability of a PM10 exceedance occurrence during a HWD by the probability of an exceedance occurrence during a normal summer day (probability of occurrence of an exceedance calculated as number of days with exceedances by total number of days at each station). The aforementioned spatial pattern of concentration around the city center is also visible in Figure 13, where a PM10 exceedance is up to 12 times more probable under a heat wave than a normal day.     Figure 14 shows a clear shift to higher PI classes during HWD, which is more pronounced at the urban traffic station.
Atmosphere 2019, 10, x FOR PEER REVIEW 15 of 22 background). PI values were exclusively determined by PM10 concentrations. Figure 14 shows a clear shift to higher PI classes during HWD, which is more pronounced at the urban traffic station. As discussed above, we hypothesized the advection of warm air masses from Northern Africa as the most probable cause of heat waves in Madrid. Saharan dust plumes are one the main origins of elevated concentrations of PM10 in Madrid [62,63]. For that reason, we assessed the relationship between heat waves, Saharan dust outbreaks, and air quality conditions in Madrid (Table 4). More Saharan dust outbreaks are expected to occur simultaneously with heat waves (X 2 value of 47.81, significant at 0.001 level); outbreaks simultaneous with heat waves more probably produce a widespread occurrence of PM10 exceedances, as shown by the average ratio between stations with/without exceedances.

Discussion and Conclusions
We analyzed the evolution of heat waves in Madrid (Spain) and their impacts on human comfort. Those extreme events show an increase in frequency and duration after 1980, in close agreement with similar findings about the temporal evolution of heat waves in the Iberian Peninsula and the Mediterranean region [64][65][66][67][68][69][70]. Regarding the underlying cause of this trend, it might be primarily related to the increase in global temperatures. However, it is not possible to rule out a regional component in the form of a change in atmospheric circulation, which might favor a higher As discussed above, we hypothesized the advection of warm air masses from Northern Africa as the most probable cause of heat waves in Madrid. Saharan dust plumes are one the main origins of elevated concentrations of PM 10 in Madrid [62,63]. For that reason, we assessed the relationship between heat waves, Saharan dust outbreaks, and air quality conditions in Madrid (Table 4). More Saharan dust outbreaks are expected to occur simultaneously with heat waves (X 2 value of 47.81, significant at 0.001 level); outbreaks simultaneous with heat waves more probably produce a widespread occurrence of PM 10 exceedances, as shown by the average ratio between stations with/without exceedances.

Discussion and Conclusions
We analyzed the evolution of heat waves in Madrid (Spain) and their impacts on human comfort. Those extreme events show an increase in frequency and duration after 1980, in close agreement with similar findings about the temporal evolution of heat waves in the Iberian Peninsula and the Mediterranean region [64][65][66][67][68][69][70]. Regarding the underlying cause of this trend, it might be primarily related to the increase in global temperatures. However, it is not possible to rule out a regional component in the form of a change in atmospheric circulation, which might favor a higher frequency of hot air masses advections from southern latitudes. Confirmation of these hypotheses pends future work.
An intensification of the nocturnal CLUHI due to heat waves was identified from the analysis of summer daily temperatures in Madrid, but there is no general agreement so far on the directions of the UHI response to warmer conditions. Some studies report an intensification of the nocturnal UHI during extreme hot days [71][72][73][74]; others report a slight intensification of both daytime and nighttime UHI [75], or even a simple modification of the diurnal spatial pattern of the UHI [39]. Some studies found no intensification of the nocturnal UHI during heat waves [21,76]. Additional discussions have arisen about the physical mechanisms that might enhance the nocturnal UHI during heat waves. Whereas lack of surface moisture and low winds are usually considered candidates that boost the impact of heat waves on UHI intensity, recent research argues the existence of a cascade of synergetic reactions, with diverse responses depending on the regional climate, location, mesoscale mechanisms, and urban structure [77][78][79]. The possible enhancement of the nocturnal CLUHI under warmer conditions has profound implications, but simulations of the magnitude of the CLUHI under the global warming also produced divergent results [80][81][82].
Worldwide, the SUHI varies considerably depending on urban density, surface properties, extent of vegetation, etc. [83][84][85]. Madrid, as the majority of cities, experiences a well-developed nocturnal SUHI, whereas its diurnal SUCI is not as frequent [86,87]. The structure of the nocturnal pattern displays higher values in the city center and a sharp gradient in the western part of the city, which occurs due to the predominance of forests and a concave topography. By day, the rural surfaces, mostly occupied by dry farming fields or scrublands under sandy or gypsum soils, experience extreme warming that reverses the urban-to-rural thermal contrast, generating a SUCI, whereas aquatic surfaces become the most significant cooling sources [88]. Under heat wave conditions, both the diurnal SUCI and the nocturnal SUHI intensify, but the magnitudes are not symmetrical, since the average value of the former is about 1 • C and the latter 2 • C.
Regarding the impact of urban spaces on human comfort, a diurnal intensification of the heat stress in the city during the daytime, using the Physiological Equivalent Temperature, was found. The nocturnal physiological heat island is less intense than its nocturnal air temperature counterpart. Two explanations are possible: (1) The 00 UTC synoptic observation does not represent the maximum intensity of the summer urban heat island, since sunset occurred just a few hours prior and the differential cooling of the countryside is beginning.
(2) During daytime, high air temperatures, dry conditions, and strong solar radiation-the basic elements of the extreme diurnal heat stress in Madrid-are approximately the same in the city (Madrid-Retiro) and the suburban areas (Barajas, Cuatro Vientos, Getafe, and Torrejón de Ardoz). However, there is a substantial contrast regarding wind speeds. Convective movement derived from the strong surface warming should enhance the turbulent mixing in the lower levels of the troposphere, leading to higher wind speeds in the suburban areas (3.6 m·s-1 on average), whereas buildings slow down the wind (2.1 m·s −1 on average) in the urban weather station. Consequently, a slight refreshing effect is experienced in the suburban areas. During nighttime, the stable layer in the suburban areas reduces the wind speed, so differences in PET between urban and non-urban observatories are weaker.
Finally, we highlighted the coincidence between heat waves, strong nocturnal urban heat islands, and worsening of the air quality in the Madrid city center due to rising levels of PM 10 . This is not surprising due to the common African origin of some, but not all, air masses. Some HWD are accompanied by an autochthonous high dust atmospheric load, when, under large-scale atmospheric stagnation conditions, the poor vegetal cover, a dry environment (low precipitation, high radiation intakes), and the warming of unprotected soils, a strong low tropospheric convection develops, favoring the resuspension of crustal or soil particles [89].
Excessive nocturnal heat negatively influences not only human health -including increasing mortality rates due to heat stress [90,91] and more frequent insomnia [92]-but also impacts labor productivity [93]. Many analyses have highlighted the vulnerability of Madrid's population to extreme temperatures; likewise, both mortality and morbidity are closely related to the worsening of air quality in the city [94][95][96]. Besides, research has shown a distinctive spatial pattern of mortality: the city center districts are more vulnerable because of the predominance of aged people, and the antiquity and lower quality of the buildings [97]. Our research agrees with this pattern of vulnerability, since those districts experience an enhanced nocturnal UHI and the worst air quality during heat waves. Thus, as recently developed at the national scale [98], the implementation of prevention plans designed to minimize the impact of high temperatures on health should include the impact of the higher nocturnal urban temperatures, resulting from urban heat islands, to reduce the effect of heat waves on population health. This action should be accompanied by studies analyzing the spatial and temporal variability of urban heat islands at finer scales.
Author Contributions: D.R. and F.F. conceived and designed the study; D.R., F.A., A.M., and F.F. conducted the research, analyzed the data and wrote the paper.
Funding: This research was funded by the research project number CGL2016-80154-R "Análisis y modelización de eventos climáticos extremos en Madrid: olas de calor e inversiones térmicas" funded by Convocatoria 2016 de Proyectos de I+D+I, correspondientes al programa estatal de Investigación, Desarrollo e Innovación Orientada a los Retos de la Sociedad, from the Spanish Ministry of Education.