Observations and reconstructions of global temperature evolution indicate a pronounced warming during the last 150 years, with an increase in the occurrence of so-called heat waves, which are extended periods of anomalously high summertime temperatures (Schär et al., 2004). In the last decade, numerous heat waves occurred in Western and Central Europe and in the Mediterranean regions (2003, 2006 and 2007). Climate models for the 21st century suggest that the year-to-year variability of summer temperatures might experience a pronounced increase in response to greenhouse gases, affecting the frequency, intensity and duration of heat waves (Meelh and Tebaldi, 2004; Tebaldi et al., 2006). By the end of the century, if greenhouse gas emissions are not reduced, the fraction of days with temperatures above 30 °C in France might equal that currently experienced further south in Spain and in Sicily (Beniston et al., 2007).
Heat waves are especially deadly in cities due to population density and urban surface characteristics. Urbanization drastically modifies the partition of the heat fluxes compared to suburban or rural surroundings, generating a differential temperature effect called an urban heat island. The main contributing factors are (1) changes in the physical characteristics of the surface such as albedo, emissivity or thermal conductivity owing to the replacement of vegetation by asphalt and concrete, and consequently changes in the radiative fluxes; (2) a decrease in surface moisture available for evapotranspiration; (3) changes in the near-surface flow, owing to the complicated geometry of streets and tall buildings; and (4) anthropogenic heat. At night, urban areas gradually release the heat accumulated in structures during the day, whereas rural areas cool off by unobstructed outgoing radiation. As is demonstrated below, it is the lack of relief at night, rather than high daytime temperatures, that threatens people's health. The elderly, infants, young children, and people with chronic health problems such as asthma and cardiac diseases are more vulnerable. Air pollution, enhanced by high temperatures, further exacerbates the adverse health effects by stressing the respiratory and circulatory systems (Basu and Samet, 2002).
In August 2003, a persistent anticyclone over Western Europe blocked rain-bearing depressions from the Atlantic Ocean and advected dry air from Northern Africa. The average summer temperatures exceeded the 1961–1990 mean by ∼3 °C, an increase of up to 5 standard deviations (Schär and Jendritzky, 2004). These anomalous conditions resulted in a heat wave of exceptional strength and duration and a death toll estimated to have exceeded 70 000 (Robine et al., 2008). In France, this extreme heat wave event was preceded by an unusually warm and dry spring. Rainfall deficit and high net radiation enhanced by clear skies contributed to early green-up and drying of the soil by evapotranspiration (Zaitchik et al., 2006; Fischer et al., 2007). From March to August, temperatures exceeded the 1971–2000 long-term mean by 4.7 °C in June, 1.8 °C in July and 4.4 °C in August (Bessemoulin et al., 2004).
From 4 to 13 August, the Paris metropolitan area experienced 9 consecutive days with maximum air temperatures above 35 °C, reaching 39 °C at the peak of the heat wave on 12 August. Minimum temperatures steadily increased from 20 to 25.7 °C on 11 and 12 August. The atmosphere was very stable, the wind speed fluctuated between 1 and 4 m/s and the relative humidity and potential evapotranspiration were, respectively, below and above those of normal summers. From the start of the heat wave to its peak, the relative humidity decreased from 38% to 18% during daytime, and from 78% to 58% during night-time. Relative humidity regulates the rate of latent heat transfer from the human body, through perspiration, and high humidity augments the likelihood of heat stress and disorders. Clear skies intensified radiative forcing and strong sunlight triggered the photochemical production of ozone and its accumulation (Tressol et al., 2008). However, in Paris the levels of humidity or ozone had little influence on the excess mortality, which was mainly attributable to temperature (Filleul et al., 2006). The excess mortality reached +141% in the Paris region, resulting in an estimated 4867 heat-related deaths (Hémon and Jougla, 2003). Figure 1 shows the maximum and minimum air surface temperatures and associated mortality rate in summer 2003 in the Paris region. It indicates a short time lag between temperature increase and death, particularly during 11–13 August at the peak of the heat wave.
Summer warming trends and increased weather variability in Western Europe have important implications for human health, particularly in large cities (Kalkstein and Green, 1997). Health risks increase with heat intensity, relative humidity, time exposure and high minimum night-time temperatures. Following the 2003 heat wave, the French Institut de Veille Sanitaire (InVS), which is a public health surveillance agency, in cooperation with Météo-France implemented a heat–health warning system based on biometeorological indicators, operating from 1 June to 31 August (Pascal et al., 2006). This alert system allows people to enact management and prevention measures when the thresholds are reached. However, the available information does not take into account the intraurban temperature variations, which would enable one to take not only emergency measures but also long-term ones concerning urban planning. For synoptic purposes, most weather stations are situated in parks and airports, away from the effects of built environments. In Paris, the reference weather station is located in the Montsouris Park and its measurements may underestimate the temperatures experienced by people living downtown, or working in suburban industrial areas by several degrees. Furthermore, the weather station network is too sparse to record smaller scale variations in temperatures that may pose an increased risk to human health and which are best resolved using satellite thermal infrared sensing.
Applications of remote sensing in urban climatology have demonstrated the relationship between sprawling conurbations and complex patterns of surface temperature and heat islands (Dousset and Gourmelon, 2003; Streutker, 2003; Kato and Yamaguchi, 2005; Nichol, 2005; Hung et al., 2006). Satellite monitoring of extreme heat events in urban areas (Dousset et al., 2007; Cheval et al., 2009) and estimations of associated public health impact (Johnson et al., 2009) are recent developments. The objectives of this research were first, to monitor and analyse the spatial gradient of radiometric surface temperature over the diurnal cycle during the 2003 heat wave; second, to implement indicators of heat exposure for the elderly to better represent the risks of heat stress and mortality and third, to demonstrate the use of satellites for monitoring summer heat waves. This article is divided into four parts. Data acquisition and processing are described in part two. The LST spatial variability over the diurnal cycle, the comparison with the summer of 1998, the relation to land use/cover, and new indicators of heat exposure are presented in part three. The results are discussed in part four.
2.1. The Paris metropolitan area
Paris (2°20′E, 48°50′N) is located in a sedimentary basin on the Seine River. The regional climate is moderated by the oceanic influence of the mid-latitude Westerlies. The Paris Basin is relatively flat with low relief hills, which may produce significant low-level atmospheric motions, although orography has little impact on the thermal parameters (Troude et al., 2002). The metropolitan area is characterized by compact urbanization, with a population of nearly 12 million and a high density of ∼20 000 inhabitants/km2 in the city.
2.2. Remotely sensed data
2.2.1. NOAA-AVHRRacquisition and processing
Two high-resolution thermal images were available during the 4–13 August 2003 heat wave, one from the Landsat-5 Thematic Mapper (TM) (9 August) and another from Terra-ASTER (10 August). However, monitoring requires repetitive imaging to resolve the diurnal cycle and high resolution to observe spatial variability. Geostationary satellites have frequent sampling, but their coarser spatial resolution is insufficient for urban observations. These are best resolved by 1.1-km resolution images from sun-synchronous polar orbiting satellites, such as NOAA, Terra and Acqua, that pass twice daily. In summer 2003, the simultaneous operation of satellites NOAA 12, 16 and 17 provided good coverage with a 4–6 h revisit period.
Eighty-four images, which were sensed from 21 July to 21 August by the advanced very high-resolution radiometer (AVHRR) on board NOAA satellites, were acquired by the high-resolution picture transmission (HRPT) receiving station at the Istituto Nazionale di Oceanografica e di Geofisica Sperimentale, in Trieste. Seventy-six scenes were selected according to clear sky, image quality and small satellite-zenith viewing angle to ensure ground resolution close to 1.1 km and minimize both atmospheric attenuation and anisotropic effects. The images were geometrically corrected for earth rotation and curvature, orthorectified and interactively referenced to a Lambert projection. The AVHRR scans in six spectral channels centred at 0.62 µm (ch. 1), 0.91 µm (ch. 2), 1.61 µm (ch. 3A) and 3.74 µm (ch. 3B), 10.8 µm (ch. 4) and 12 µm (ch. 5). Land albedo and daytime cloudiness were derived from channel 2, and night-time cloudiness from the difference between channels 3 and 4. Cloudy pixels were flagged according to a threshold derived from the histograms of cloud-free images. Normalized difference vegetation indices (NDVI) were computed from the visible and near-infrared channels (ch. 1 and ch. 2). Calibrated brightness temperature was inferred using the internal black-body references of the satellites. The main constraints in retrieving LST from infrared satellite images are (1) the partial absorption of black-body radiation by water vapour and other gases in the atmosphere; (2) the surface emissivities being less than 1 and spatially and spectrally variable (Becker and Li, 1995; Dash et al., 2002; Gustafson et al., 2006); (3) sub-pixel variations of surface temperature and hot spots averaged nonlinearly through Planck's law (Dozier, 1981, Dousset et al., 1993); (4) urban canyons trapping radiant and incident energy, increasing the pixel-averaged emissivity and (5) directional and anisotropic effects due to satellite viewing angles and urban structures (Voogt and Oke, 1998; Lagouarde et al., 2004).
In August 2003, the air was dry and the differential atmospheric attenuation from thermal channels 4 and 5 yielded negligible water vapour corrections equivalent to errors less than 1 °C at night and 1 °C to 1.7 °C during the day for brightness temperatures between 30 and 40 °C, respectively. Uncertainties related to radiosounding and modelling of atmospheric transmittance to account for atmospheric absorption would exceed actual errors, and this correction was not justified. The correction for time-independent emissivity was also neglected, because the objective was to study the relative temporal variations of surface temperatures. Most images with satellite-zenith angles higher than 35° were rejected; however, a few were kept and empirically corrected based on consecutive images with smaller satellite-zenith angles (Roujean et al., 1997; Lagouarde et al., 2004). Median LST images were constructed from 50 NOAA-AVHRR images over six time intervals of satellites passes (i.e. 1–3, 4–7, 9–12, 12–15, 15–18 and 20–23 UTC) during the 9-day heat wave episode. Figure 2 shows the diurnal distribution of the 50 images, and Figure 3 displays the six median LST images, each averaged over six to ten individual images.
2.2.2. SPOT HRV-4 acquisition and processing
Surface characteristics and properties that govern the surface energy balance were extracted from a SPOT-4HRV multi-spectral image acquired on 13 July 2003, three weeks before the extreme event. The processing of the SPOT image included detector radiometric equalization, geometric processing to remove the earth rotation and resampling across-track to a uniform 20-m pixel size. Figure 4(a) represents the unsupervised land cover classification derived from the four visible and near-infrared channels, which yielded six classes corresponding to water, densely built urban areas, suburban residential areas, light bare soils, forest and lawns and fields. The classification was further validated using the NDVI, calculated from the SPOT visible and near-infrared channels. A geographic information system (GIS) was used to generate land use/cover fractional images and merge 20-m SPOT and 1-km AVHRR pixels. Ancillary data on urban administrative delineations, parks, rivers and industrial areas were obtained from the Paris Department of Planning (APUR) and integrated in the GIS database.
2.3. In situ meteorological data
The in situ meteorological data were recorded at the Paris automated weather station in Montsouris Park. Hourly data include surface air temperature (Tair) at 10 cm, 50 cm and 1.50 m above the surface, ground temperature at − 10 cm, dew-point, wind speed and direction, relative humidity, water vapour and insolation and net radiation. Additional data at 00:12 UTC include ground temperatures at − 20 cm, − 50 cm and − 1 m depth. Colocated in situ and remotely sensed data from Montsouris Park were used to interpret LST results, taking into account the different nature of these measurements.
2.4. Public health data
Following the summer of 2003, the InVS conducted a case–control study to identify the mortality risk factors for elderly people living in the Paris region during the heat wave (Ledrans et al., 2005; Vandentorren et al., 2006). The study included 241 people aged 65 and over, who died between 8 and 13 August and remained at home for at least 24 h before death or hospitalization. These 241 cases were matched with 241 controls, who were people living in a district of Paris under similar socio-economic conditions, having the same sex and age class as the cases, but survived the heat wave (Ledrans et al., 2005). The surface temperature near the domicile of the considered cases and controls was one of the many risk factors related to death that were investigated. Other factors included age, sex, socio-economic conditions, family entourage, behaviour during the heat wave, mobility, health status, housing conditions and environment. In the previous work, LSTs on 200-m radius around the 482 addresses were derived from a single Landsat TM image at 10:17 UT on 9 August. Although the spatial resolution of this image is very high, this timing was not optimal to assess thermal patterns. Because the 16-day repeat cycle of the Landsat image precludes daily monitoring, thermal indices for this study were estimated from a time series of 61 NOAA-AVHRR images (1–13 August). The thermal index of an address corresponds to the LST of the pixel that contains the address. Different indicators of temperature exposure were constructed based on minimum and maximum LST, taking into account different lags between temperature and death, and daily temperatures as well as temperatures averaged over 3, 7 and 13 days, based on the knowledge of the lag between temperature extremum and health impact. These indicators were alternately integrated into the conditional logistic model of the previous case–control study, replacing the single indicator. This conditional logistic model (Breslow and Day, 1980) was adjusted for the same variables as in the previous study (age, sex, etc.) so that the relationship between thermal indicators and mortality could be tested.
3.1. The LST diurnal cycle
Figure 3 shows six composite images of median LST, constructed from 50 images taken between 4 and 13 August. The LST spatial variability over the diurnal cycle is evident. The images reveal contrasting daytime and night-time heat island patterns, reflecting the different day and night rates of heating and cooling between urban and suburban areas, enhanced by the stable atmosphere that characterized the heat wave episode. The LST variability was analysed as a function of urban surface characteristics by merging data from the 20-m resolution classified SPOT image, and from APUR, with the 1-km resolution NOAA-AVHRR images (Dousset and Gourmelon, 2003).
At night-time, the images show strong urban heat islands of up to 8 °C over the 20–23 UTC and 1–3 UTC averaging periods (Figure 3(a) and (b)), and up to 6 °C over 4–7 UTC (Figure 3(c)). The LST distribution was well correlated with increasing built density from the suburbs to downtown. At the peak of the heat wave, low relative humidity of 27–40% in the evenings (20–23 UTC) in Montsouris Park suggests that vegetated areas in the residential suburbs and rural areas may have been conducive to evaporative and radiative cooling. In the rural areas north of Paris, LST decreased by ∼4 °C, and remained constant the rest of the night. In downtown Paris, temperatures decreased slowly as the heat stored in buildings and trapped in urban ‘canyons’ was progressively lost and anthropogenic heat was continually produced. Eventually, the LST decreased after sunrise, both in the suburbs (∼19 °C) and downtown (∼22 °C) (Figure3(c)).
During the daytime, multiple temperature anomalies were scattered in the densely built and industrial suburbs, conveying mostly variations of the surface heat balance between dry and comparatively moist surfaces. In Montsouris Park, relative humidity dropped from 38% to 18% at the peak of the heat wave, and the wind fluctuated between ∼0 and 2 m/s. From 9 to 12 UTC, the northeast region quickly warmed because of unobstructed field of view, low thermal inertia surfaces and anthropogenic heat produced by heavy traffic and industries. From 12 to 15 UTC (Figure 3(e)), heat islands up to 11 °C were observed between the industrial suburbs and the forests. The highest LSTs of 38–42 °C occurred in the suburbs: to the north in the industrial areas of St. Ouen, Aubervilliers, St. Denis and in the freight zones of Le Bourget, Garonord and Roissy airport; to the west in the industrial area of Argenteuil and to the south in the freight and industrial areas of Ivry-sur-Seine and the Orly airport.
Figure 4(b) displays the diurnal cycle of mean LST at three locations: downtown Paris, Garonord (industries near Roissy Airport) and the Boulogne park, shown in Figure 4(a). Urban canyons in downtown Paris consist of asphalt pavement, stone or concrete buildings of up to 7–10 floors and roofed with slate or zinc. The industrial area comprises large warehouses of 2–3 levels, with corrugated steel or fibre roofs, asphalt or concrete parking lots, and roads. The ∼8 km2 Boulogne Park is mainly composed of lawns, trees, bushes and a small lake. The mean diurnal cycles indicate a near-constant difference of ∼1.5–2.2 °C between downtown and the park, but differences of +3 °C at night, and − 3.5 °C at noon between downtown and the industrial area, and +4.5 °C at noon between the industrial area and the park (Figure 5). The industrial surfaces with lower thermal inertia and unobstructed sky view are consistently warmer at daytime and cooler at night-time than the downtown and park surfaces.
3.2. Colocated satellite and in situ data in Montsouris Park
Tair and LST measurements are intrinsically different. The former corresponds to the ambient temperature at 1.50 m above the surface, whereas the latter represents the surface radiometric temperature averaged over a pixel, mostly composed of horizontal surfaces within different levels of the canopy layer. These measurements are complementary, because the sensible heat flux is determined by the temperature difference between the surface and the air immediately above it. Their correlation arises from the surface heat balance.
Figure 6(a) shows an hourly time series of Tair at 1.50 m and LST at the 4–6 h satellite sampling interval. Figure 6(b) and (c) shows the hourly time series of relative humidity and wind speed. Figure 6(d) and (e) shows the scatter plots of satellite LST versusTair at night (01–06 UTC) and by day (12–15 UTC) for 1–13 August. Slopes and offsets of the linear regressions are shown in the respective figures. The linear correlation coefficient was 0.92 at night, but only 0.68 during the day, presumably reflecting stronger sub-pixel variations of surface cover and heat balance regimes. Minima of both Tair and LST occurred before sunrise at ∼05:00 UTC, just before insolation resumed and net radiation ceased to be negative. At night, Tair was generally 1–2 °C higher than LST, except on 12 August at the peak of the heat wave, when the near-surface gradient reached 5 °C presumably because of lack of mixing. Maximum LSTs occurred at the time of maximum solar irradiance, while maximum Tair lagged typically by ∼3 h. At ∼13:00 UTC, Tair-LST varied from − 2 to − 4 °C, implying a significant near-surface temperature gradient due to intense radiative heating. Some residual difference may also be attributed to the effects of the satellite viewing geometry (Lagouarde et al., 2004).
3.3. LST comparison of the August 2003 heat wave with August 1998
The time series of images of August 2003 were compared with those of August 1998, previously analysed by Dousset and Gourmelon (2003). The comparison indicates a large difference in the vegetation indices and high minimum temperatures. The metropolitan area is surrounded by fields and forests, and the city comprises a dozen small parks (0.1–0.6 km2), and two large ones (8–10 km2) at the west and east edges. In spring 2003, strong incident radiation and large precipitation deficit forced an early spring green-up and progressively reduced the soil moisture. Consequently, the vegetation index of August 2003 is lower than normal summers and evidences a significantly increased drought and reduced primary productivity (Ciais et al., 2005; Zaitchik et al., 2006). Figure 7 represents the bivariate histogram of the composite afternoon LST (12–15 UTC) versus the mean NDVI for 4–13 August 2003 (a) and its distribution (b). The figure indicates a significant negative correlation with a slope of ∼− 0.2 °C/%NDVI that illustrates the importance of vegetation in the partitioning between latent and sensible surface heat fluxes. Analysis and manual sampling of individual images indicate that the LST is 2–3 °C lower in small parks than in their built surroundings and 4–5 °C lower in large ones.
Figure 8 shows median LST diurnal cycles over the Paris metropolitan area, one from 5 to 11 August 1998, and the other from 1 to 13 August 2003. In 1998, the period from 5–11 August was relatively warm, with area-mean maximum LSTs between 28 and 36 °C, and LSTs for 4 consecutive days were over 35 °C. However, the highest area-mean minimum LSTs never exceeded 15 °C, thus averting the occurrence of a heat wave. In 2003, the diurnal amplitude was 5 °C lower (16 °C vs 21 °C) than that in 1998, and the mean minimum was 8 °C higher which confirms the impact of night-time minimum temperature on the heat wave. For two consecutive nights, on 10–11 and 11–12 August, stable meteorological conditions and low winds prevented convective mixing. At 05:00 UTC (07:00 local time), the minimum Tair was 25.7 °C for both nights. On 11 August, night-time LSTs were still 21.5 °C in Montsouris Park and 25–26 °C in downtown Paris. These high minimum night-time temperatures contributed to a lack of night rest for vulnerable people and to 500 excess deaths in the Paris region, on these two particularly deadly days during the heat wave (Figure 1). On the 13th, both temperatures and mortality excess began to decrease, and returned to normal on the 19th.
3.4. LST and mortality
Multilayer maps were generated over the six composite thermal images to identify risk factor areas. Standard layers, which were selected first, included surface albedo, percentage of built and vegetated surfaces, hydrology, population density, industrial areas and transportation.
Figure 9(a) represents the LST from 01 to 03 UTC and Figure 9(b) that of 12 to 15 UTC, with some of the land use overlays. Although the 1-km spatial resolution of the infrared images may be coarse for urban climate analysis, the overlays demonstrate the high sensitivity of the signal, such as a 1.5 °C decrease or increase attributable to the Jardin des Plantes Park and to the Palais du Louvre, respectively. At night, in the 20–22 UTC and 0–3 UTC composite images, the spatial distribution of the highest LSTs of 24 °C to 26 °C in the districts south and northeast of the Seine River matched the spatial distribution of the highest mortality ratios described in Cadot and Spira (2006), which were not related to the highest daytime LSTs. These maps would be useful for taking preventive measures to reduce strenuous activities in industrial areas north of Paris, where LST reached 40 °C in the afternoon, and assist elderly people living in the southern districts, where LST remained 25 °C at night.
Figure 10 displays the spatial distribution of the 482 cases and control addresses over Paris and the department of Val de Marne, in the suburbs of Paris. A total of ∼29 000 individual LST measurements were extracted at these addresses, using 61 satellite images recorded between 1 and 13 August.
Table I lists the adjusted odds ratio (OR) of mortality for the different thermal indicators (OR adjusted on age, sex, socio-economic conditions, family entourage, behaviour during the heat wave, mobility, health status, housing conditions, and so on). The OR is a relative measure of risk, which indicates the probability for someone exposed to a given factor to develop the outcome, as compared to someone with less or no exposure. If exposure is not linked to health impact, the OR is close to 1. If exposure is positively linked to health impact, OR > 1. If exposure is negatively linked to health impact (protective effect), then OR < 1. OR is given with a confidence interval (CI): if the value 1 is included in the CI, the result is not statistically significant. Table I shows that thermal indicators taking the average or maximum daytime temperatures into account were not significantly linked to deaths. However, those taking minimum nocturnal temperatures into account were significantly linked to death. For example, for the mean minimum nocturnal temperature in the 7-day interval until the date of death, the OR associated with a 0.5 °C LST difference between cases and controls was 2.2, indicating a death risk more than twice as high. This value of 0.5 °C corresponds to the 90th percentile of the distribution of the differences between the cases and controls for this indicator.
Table 1. Results of the logistic model showing adjusted odds ratios of mortality for different thermal indicators. The rows italicized indicate significant results. The odds ratios were adjusted for age, sex, socio-economic conditions, family entourage, behaviour during heat wave, mobility, health status and housing conditions
Land surface temperature indicators from NOAA-AVHRR images
OR (Cl 95%)
OR, odds ratio; Cl 95%, 95% confidence interval.
Mean of the minimal temperatures for 7 days (day of death and 6 preceding days)
Mean of the maximal temperatures for 7 days (day of death and 6 preceding days)
Mean of the minimal temperatures from 1 to 13 August
Mean of the maximal temperatures from 1 to 13 August
Mean of the average temperatures from 1 to 13 August
4. Discussion and conclusion
This study documented the satellite monitoring of a 9-day heat wave over a metropolitan area and the associated epidemiological risk and time lag of death for elderly persons at given locations. The thermal images showed contrasting night-time and daytime heat island patterns, which were related to surface characteristics and land uses. The results confirm the influence of nocturnal temperatures on heat wave intensity and excess mortality, and show the contribution of urban heat islands in intensifying the heat wave by absorbing heat during the day and progressively raising minimum nocturnal temperatures linked with heat stress and mortality. Urban parks had a significant impact on surface temperature with a 0.2 °C LST decrease per %NDVI, despite a prior warm and dry spring that lowered moisture availability.
The effect of urban heat islands on mortality is well known and the variability of death risk within cities has been investigated (Canouï-Poitrine et al., 2006). Some studies (Vandentorren et al., 2006; Johnson et al., 2009) have integrated societal factors with LST derived from a single Landsat TM image. This study, consisting of the joint analysis of a large time series of thermal images and of colocated validated cases of elderly mortality, allowed us to infer the relative impact of heat exposure on elderly population at given locations and assess the time lag between heat exposure and death. It demonstrated the predominance of mean minimum nocturnal temperature, where a 0.5 °C increase doubles the risk of death, in the temperature range of the heat wave period.
There are implications regarding temperature monitoring and public health surveillance that are key issues in Europe given the summer warming trend and the resulting increase in heat wave incidence, duration and intensity, and also the aging of the population, which brings greater physiological and social vulnerability. Heat-related deaths are more frequent in large cities. In France, in 2003, the excess mortality was 40% in small towns, but 80% in Lyon and 141% in Paris (Vandentorren et al., 2004). An analogous heat wave in USA would result in a mortality rate of +150% for New York City (Kalkstein et al., 2008).
Satellite monitoring of summer heat waves provides surface temperature values that can be merged with urban databases to identify risk areas at the scale of a district or a neighbourhood, and supply criteria for adaptation strategies. Satellite monitoring can provide data to substantially improve understanding of urban surface processes and societal vulnerabilities to near-term climatic changes. Although Paris was chosen as a case study, having recently experienced an extreme heat wave with well-documented health consequences, the methods used in this study are general and should be applicable to other cities as well.
This work was funded by the MAIF Foundation. B. Dousset was partially supported by a National Research Council scholarship at the National Oceanic and Atmospheric Administration, Air Resources Laboratory (Silver Spring). Land-cover layers were supplied by the Atelier Parisien d'Urbanisme. Discussions with Julian Wang and Dian Seidel are gratefully acknowledged. Nancy Hulbirt prepared the figures. Contribution number 7995 of the School of Ocean and Earth Sciences and Technology, University of Hawaii.