Impact of heat waves on mortality in Croatia

Ksenija Zaninović & Andreas Matzarakis

Received: 6 February 2013 /Accepted: 7 July 2013 /Published online: 1 September 2013# ISB 2013

Abstract The aim of this work was to determine the criteriafor heat loads associated with an increase in mortality indifferent climatic regions of Croatia. The relationship betweenheat stress and mortality was analysed for the period 1983–2008. The input series is excess mortality defined as thedeviations of mortality from expected values determined bymeans of a Gaussian filter of 183 days. The assessment of thethermal environment was performed by means of physiolog-ically equivalent temperature (PET). The curve depicting therelationship between mortality and temperature has a U shape,with increased mortality in both the cold and warm parts of thescale but more pronounced in the warm part. The thresholdtemperature for increased mortality was determined using ascatter plot and fitting data by means of moving average ofmortality; the latter is defined as the temperature at whichexcess mortality becomes significant. The values are higher inthe continental part of Croatia than at the coast due to therefreshing influence of the sea during the day. The sameanalysis on a monthly basis shows that at the beginning ofthe warm season increased mortality occurs at a lower tem-perature compared with later on in the summer, and thedifference is up to 15 °C between August and April. Theincrease in mortality is highest during the first 3–5 days andafter that it decreases and falls below the expected value.Long-lasting heat waves present an increased risk, but in verylong heat waves the increase in mortality is reduced due tomortality displacement.

Keywords Thermal load .Mortality . Physiologicallyequivalent temperature . Threshold temperature .

Thermo-physiological stress

Introduction

Living in extremely varied climate conditions, ranging frompolar cold to tropical heat, human beings are able to adaptgradually to a wide range of thermal conditions, but there areabsolute limits of heat load that a person can bear. Considerabledeviations from the conditions a human body is used to, i.e.warm or cold stress, can cause adverse effects and even long-term consequences. The heat regulatory mechanism of healthyadults is to some extent able to adapt to environmental condi-tions, but the ability to adapt is lower for risk groups (theelderly, children, or people with compromised health) (Frostet al. 1992; Pan et al. 1995; Díaz et al. 2002; Shao et al. 2008;Matzarakis et al. 2011; Kovats and Hajat 2008; Iñiguez et al.2010). Women are more vulnerable than men, probably due tothe overall higher number of elderly women and their differentthermo-physiology (Toulemon and Barbieri 2006). Particularlyvulnerable are socially sensitive people due to a reduced possi-bility of mitigation (Guest et al. 1999; Schwartz 2005; Laaidiet al. 2006;McGregor et al. 2007; Bassil et al. 2008; Nastos andMatzarakis 2008a; Farajzadeh and Darand 2009) and thoseexposed to high temperatures while at work (WHO 2003;Dunne et al. 2013). Heat waves are an environmental andoccupational hazard (Kovats and Hajat 2008), therefore earlywarning and the organisation of public health services canreduce the risk and protect the population.

A population becomes acclimatised to the local climate butthe adaptive capacity varies geographically and therefore con-ditions for heat wave occurrence are geographically variable innature (Yamanaka and Nakamura 1999; Keatinge et al. 2000;Saez et al. 2000; Laaidi et al. 2006; Gosling et al. 2007;Asanuma et al. 2008). The lowest mortality in warm climates

K. Zaninović (*)Meteorological and Hydrological Service, Grič 3,10000 Zagreb, Croatiae-mail: zaninovic@cirus.dhz.hr

A. MatzarakisMeteorology and Climatology, Albert-Ludwigs-University ofFreiburg, Werthmannstraße 10, 79085 Freiburg, Germany

Int J Biometeorol (2014) 58:1135–1145DOI 10.1007/s00484-013-0706-3

occurs at higher temperatures than in cold climates (Pan et al.1995; Ballester 1997; Keatinge et al. 2000; Laaidi et al. 2006;Iñiguez et al. 2010). Dependence on latitude is also manifestedin different temperature thresholds above which increased mor-tality occurs (Díaz et al. 2006; Blazejczyk and McGregor2008), as well as in the more significant increase in mortalitywith temperature at higher latitudes, despite similar temperaturethresholds (Kim et al. 2008).

Thus, there is no single definition of a heat wave, but eachcountry should determine the conditions for the occurrence ofheat waves and take into account that these rare events couldbe harmful for the population (Koppe 2005). Heat waves areoften defined as prolonged periods of high thermal load(Robinson 2001; Pascal et al. 2006; Gosling et al. 2009).

As early as the first half of the twentieth century, Gover(1938) observed that mortality rates can be four times higherthan expected after several consecutive days of extreme airtemperatures. Many investigations were instigated followingheat waves that had caused a number of unexpected deaths.Such was the case of heat waves that attacked the UnitedStates during the summers of 1966 and 1995 (Schuman1972; Kalkstein et al. 1996; Kunkel et al. 1996; Whitmanet al. 1997), or the well-known heat wave in 2003 that affecteda large area of Europe (WHO 2003; Johnson et al. 2005;Toulemon and Barbieri 2006; Robine et al. 2007; Mutherset al. 2010), causing more than 52,000 deaths (Larsen 2006).Extensive research after such incidents resulted in the imple-mentation of the Heat-Health-Warning-System (HHWS)(Pascal et al. 2006; Fouillet et al. 2006, 2008; Koppe andBecker 2009).

The motivation for this study was the necessity of intro-ducing the HHWS in Croatia. Therefore, the aim was todetermine the criteria for heat loads that are associated withan increase in mortality in different climate regions of Croatia.The method of investigation was chosen to include parametersthat can be accessed in the daily operational weather forecastof the national meteorological service and implemented easilyin an early warning system. This would allow public healthactivities aimed at preventing heat-related morbidity and mor-tality to be performed.

Data and methods

The relationship between mortality and the thermal environ-ment was investigated in different climatic regions of Croatia.For that reason four cities were chosen: Osijek as a representa-tive of the lowland continental climate, the capital—Zagreb—as a representative of the continental climate with a maritimeinfluence, and two representatives of maritime climates of thenorthern and southern Adriatic: Rijeka and Split, respectively(Fig. 1). The investigation was performed using mortality andmeteorological data in the 26-year period 1983–2008.

Graphical presentation of the results are shown only forZagreb in the article, while the results for other cities areprovided in tables and described.

Thermal environment

The balance between a human organism and the thermalenvironment is a complex system depending on air tempera-ture, wind, humidity, solar and terrestrial radiation, but also ongender, metabolism and clothing. Thermal impact on humanscan therefore be determined using a biometeorological indexderived using the equation of thermal balance between bodyand environment and taking into account all the componentsof this relationship. To assess the vulnerability of the popula-tion to the climatology of heat waves, the frequency of theiroccurrence, intensity and duration will be also analysed. Theassessment of the thermal environment is performed bymeans of the physiologically equivalent temperature PET(Mayer and Höppe 1987; Höppe 1999; Matzarakis et al.1999). This is based on the energy balance equation betweena human body and the environment derived from the MEMImodel (Munich energy-balance model for individuals;Mayer and Höppe 1987; Höppe 1999). PET is defined asequivalent to the air temperature at which a person (male,35 years; 1.75 m, 75 kg) sitting indoors (work activity 80 W,heat resistance of clothing Iclo=0.9, mean radiant tempera-ture equal to air temperature (Tmrt=T), velocity 0.1 m/s,water vapour pressure 12 hPa) would feel the same as inreal conditions.

The calculation of PET makes use of data on air tem-perature, relative humidity, wind speed and cloud cover, asthere are no radiation data (Matzarakis et al. 2007, 2010).Thermal perception based on the PET is determined as inTable 1.

Due to adaptation to the thermal environment, the effects ofheat waves on mortality at the beginning of the warm seasondiffer from those later on in the summer, with a strongerimpact of early heat waves compared with heat waves withthe same characteristics in late summer (Hajat et al. 2002;Kyselý 2004). In order to include short-term adaptation, in-tensifying earlier heat waves and reducing later ones, themethod introduced by Koppe and Jendritzky (2005) is used.The method modifies the thresholds of thermal perceptionexpressed by means of PET (Table 1) by introducing a vari-able part of the threshold (the relative threshold). Assuming a30-day short-term adaptation the latter authors used the one-fold Gaussian filter of 41 days (which corresponds to a half oftwo-fold 81-day filter) as this filter has 30 significant filterweights (Schönwiese 1992;Matzarakis and Nastos 2011). Theratio between absolute and relative values was chosen arbi-trarily as one-third and two-thirds; therefore, the changedthreshold that includes short-term adaptation (Tha) is calcu-lated as (Koppe 2005):

1136 Int J Biometeorol (2014) 58:1135–1145

Tha ¼ Thþ 1

3PETF41−Thð Þ ð1Þ

where Th is the absolute threshold and PETF41 is PETsmoothed with the Gaussian filter. In the case of a colderprevious period (PETF41<Th), the adapted threshold Tha islower than the absolute threshold (Th).

Mortality

The daily mortality data for the period 1983–2008 wereobtained from the Croatian Bureau of Statistics with noclassification according to age, gender or cause of death.Therefore, data are not age-, gender- or cause-standardizedand these effects are not fully controlled. Population data wereobtained from censuses carried out in 1981, 1991 and 2001,and the daily population was estimated by linear interpolation.The daily mortality data were standardized to 10,000 inhabi-tants, as most cities, with the exception of Zagreb, have fewerthan 200,000 inhabitants: at the end of the period analysed,Zagreb had 780,068; Split 180,459; Rijeka 128,741 andOsijek 103,993 inhabitants.

Heat-related mortality is defined as the incidence of deathsthat would not have occurred in the absence of heat stress(McMichael et al. 1996). Therefore, it is necessary to estimatethe “expected” mortality, or deaths that would have occurredwithout heat stress. In order to avoid long-term variability,trend or other events that can cause a greater excess mortality,the expected mortality is determined by using a low-pass filteras is recommended in such cases (Schönwiese 1992). Theadvantage of this method is that the expected value is basedon current data. Thus, a trend adjustment is not necessaryand the expected value adapts with flexibility to the currentreality (Koppe 2005). In this study, the expected mortality orthe baseline mortality is estimated by means of a two-sided

183-day filter. The advantage of applying a half-year filter liesin minimising the influence of extreme winter mortalityevents, as they are on the edge of the series for estimatingexpected mortality in summer, with the smallest impact onsmoothed values. In order to avoid shortening at the beginningand the end of the series, the series of mortality is extended by91 days at each end, with 5-year mean daily mortalities; i.e.,with the 1983–1987 daily means at the beginning, and the2004–2008 daily means at the end of the period. As it is a coldperiod, these data could not affect the results (Fig. 2).

The quality of the results depends on population size, andlow daily mortality is a limitation factor because the values arehighly influenced by noise (Pascal et al. 2006; Matzarakiset al. 2011). Accordingly, one could expect satisfactory resultsfor Zagreb with nearly 800,000 inhabitants, and to a lesserextent for other cities in Croatia.

According to the definition of heat-related mortality as thenumber of deaths excessive in comparison to the expectedmortality, the investigation was performed using series of devi-ations of mortality data from the expected values. Deviations

Table 1 Thermal sensitivity and the grade of physiological stress relatedto the physiologically equivalent temperature (PET; Matzarakis andMayer 1996)

PET (°C) Thermal sensitivity Grade of physiological stress

<4 Very cold Extreme cold stress (exCS)

4–8 Cold Strong cold stress (stCS)

8–13 Cool Moderate cold stress (moCS)

13–18 Slightly cool Slight cold stress (slCS)

18–23 Comfortable No thermal stress (noS)

23–29 Slightly warm Slight heat stress (slHS)

29–35 Warm Moderate heat stress (moHS)

35–41 Hot Strong heat stress (stHS)

>41 Very hot Extreme heat stress (exHS)

ZAGREBOSIJEK

RIJEKA

SPLIT

Hungary

Slovenia

Croatia

SerbiaBosnia and Herzegovina

Austria

Italy

ItalyAdriatic sea

Romania

Montenegro

Fig. 1 Location of the Osijek,Zagreb, Rijeka and Splitmeteorological stations

Int J Biometeorol (2014) 58:1135–1145 1137

are expressed as percentages; in the following text these valueswill be referred to as excess mortality and annotated by theabbreviation MR dev.

Relationship between thermal environment and mortality

The relationship between mortality and the thermal environ-ment was examined by means of scatter plots of meteorolog-ical parameters and excess mortality. The selected meteoro-logical parameters were PET during the warmest part of theday (at 2.00 p.m.), and the minimum and maximum air tem-perature. Besides the relationship between mortality and me-teorological parameters on the same day, the relationshipsbetween accumulated 3- and 5-day excess mortalities and 3-and 5-day mean PET values at 2.00 p.m. for the previous 3 or5 days were also analysed in order to clarify how prolongedheat stress affects mortality.

Data were fitted by locally weighted scatter plot smoothing(lowess smoothing)—a robust statistical method where fittingsimple models to localised subsets of the data builds up afunction that provides a clearer picture of the overall shape ofthe relationship between variables (Cleveland 1979). Next, theaverage excess mortality was calculated at successive 3 °Ctemperature bands (PET, minimum, maximum and mean tem-perature) at increments of 0.5 °C to enable the determinationof the temperature interval with the lowest mortality and thethresholds at which a mortality increase occurs (Keatinge et al.2000; Kyselý 2004). The temperature thresholds were set ascentres of the first 3 °C temperature bands for which excessmortality was significantly higher than the mean excess mor-tality of the entire series (significance level P =0.05). Theregression lines between mortality and mean temperature inintervals with increased mortality were then determined. Insome cases, when the regression line deviated significantlyfrom the lowess smoothing, the peripheral bands with a smallnumber of cases that acted as outliers were rejected.

The displacement of mortality was analysed by means oflagged correlation with lags of 0 to 15 days between mortalityrates and PET at 2.00 p.m. higher than threshold temperatures[PET> heat cut point (HCP) in Table 2].

The following analysis will focus on the investigation ofmortality for different heat loads, taking into account theshort-term adaptation described in “Thermal Environment”.

Results

Both fitting methods, i.e. lowess smoothing and successive3 °C temperature bands, corresponded very well, with thegreatest discrepancies at the edges due to a small number ofcases, which is, as expected, more pronounced in smaller cities,highlighting the lower reliability of these results (Figs. 3, 4).

Threshold temperatures (HCPs) causing heat-related mor-tality vary geographically. The threshold temperatures forPET at 2.00 p.m. and the maximum temperature are thehighest in Osijek (38 °C PET and 33 °C maximum tempera-ture), somewhat lower in Zagreb (37 °C and 30 °C respec-tively), and lowest on the coast (36 °C and 31.5 °C in Split,36.5 °C and 30.5 °C in Rijeka) (Table 2). On the contrary,minimum temperature thresholds on the coast are higher thanon the mainland (17 °C), and are also higher at the southern(Split with a minimum temperature threshold of 24 °C) than atthe northern Adriatic coast (21.5 °C in Rijeka). These differ-ences appear to be due to the influence of the sea, which

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

1.1.83 1.1.88 1.1.93 1.1.98 1.1.03 1.1.08

Mor

talit

y

MRMR F183

Fig. 2 Time series of relative mortality (MR) and 183-day Gaussiansmooth for relative mortality (MR F183) in Zagreb, period 1983–2008

Table 2 Threshold temperatures [heat cut point (HCP) in °C] for PET at2 p.m., 3- (PET3) and 5-day mean PET at 2.00 p.m. (PET5), maximum(Tmax), minimum (Tmin) and average daily temperature (Tave), percen-tiles of HCP and increase in mortality for each °C rise above HCP(MR dev/°C)

PET PET3 PET5 Tmax Tmin Tave

Osijek

HCP 38.0 37.5 39.0 33.0 17.0 25.5

Perc 95.0 95.6 98.3 97.5 94.0 97.0

MR dev /°C 2.3 10.4 32.0 5.0 0.3 7.7

Zagreb

HCP 37.0 35.0 35.0 30.0 17.0 23.5

Perc 96.6 97.3 98.4 94.2 93.5 94.5

MR dev /°C 1.7 4.4 10.1 1.3 2.0 2.1

Rijeka

HCP 36.5 33.5 33.5 30.5 21.5 25.5

Perc 93.7 89.0 89.5 93.4 96.0 94.1

MR dev /°C 1.9 3.0 6.5 1.4 7.3 3.0

Split

HCP 36.0 35.5 35.5 31.5 24.0 27.5

Perc 92.0 91.8 92.3 92.5 94.1 93.0

MR dev /°C 2.7 9.7 20.8 5.4 5.4 6.5

1138 Int J Biometeorol (2014) 58:1135–1145

reduces the differences between extreme temperatures com-pared with continental ones. The threshold temperatures ofmean air temperature vary between 23.5 °C in Zagreb, 25.5 °C

in Osijek and Rijeka and 27.5 °C in Split. The thresholds forthe prolonged duration of high temperatures (PET3 and PET5for duration of 3 and 5 days) are lower than the previous

-20 -10 0 10 20 30 40 50

PET (°C)

-100

-50

0

50

100

150M

R d

ev (%

)

1 day

-20 -10 0 10 20 30 40 50

PET (°C)

-300

-200

-100

0

100

200

300

400

MR

dev

(%)

3 days

-20 -10 0 10 20 30 40 50

PET (°C)

-400

-300

-200

-100

0

100

200

300

400

500

600

MR

dev

(%)

5 days

Fig. 3 Scatter plots of excess mortality (MR dev in %) and physiolog-ically equivalent temperature (PET) at 2.00 p.m. (°C) for durations of 1–5 days in Zagreb. Circles Mean values of excess mortality set for 3 °Cbands, dashed line locally weighted scatter plot smoothing (lowesssmooth), full tick line linear regression betweenmortality and temperatureabove the threshold temperature

-20 -10 0 10 20 30

T (°C)

-100

-50

0

50

100

150

MR

dev

(%)

T min

-20 -10 0 10 20 30 40

T (°C)

-100

-50

0

50

100

150

MR

dev

(%)

T mean

-10 0 10 20 30 40

T (°C)

-100

-50

0

50

100

150

MR

dev

(%)

T max

Fig. 4 Scatter plots of MR dev (in %) and absolute maximum (T max),minimum (T min) and mean temperatures (T mean) (°C) in Zagreb.Circles Mean values of excess mortality set for 3 °C bands, dashed linelowess smooth, full tick line linear regression between mortality andtemperature above the threshold temperature

Int J Biometeorol (2014) 58:1135–1145 1139

thresholds (the exception is Osijek for 5 days, but this can beattributed to unreliability due to a small amount of data).Percentiles of threshold temperature for increased mortalityrange between 89 % and 98 %. The regression lines betweentemperature and excess mortality coincide closely with thelowess smoothing, with the exception of Osijek and minimumtemperature in Rijeka (Figs. 2, 3). In most cases, mortalityincreases by about 2% for each °C rise of PET. Slightly biggerchanges in Split (2.7 %) could be caused by a much largernumber of people staying in the central Adriatic area duringthe tourist season, especially if these are tourists coming fromdifferent climatic conditions who have not adapted to highertemperatures. A prolonged effect of high temperatures (PET3and PET5) can be a significant increase in mortality, and a riseof 1 °C is associated with an increase in mortality severaltimes larger than in the case of a 1-day impact of hightemperatures. In the case of a 3-day duration (PET3 exceedingthe determined threshold), a rise of 1 °C is associated with a 2-to 5-times higher mortality increase than in the case of a 1-dayimpact, and 3- to 15-times higher in case of a 5-day duration(Table 2).

Mortality increases by 1.3 % and 1.4 % for a rise of 1 °C ofmaximum temperature in Zagreb and Rijeka, and a little morefor minimum (2.0 % / 1 °C in Zagreb) and mean air temper-ature (2.1 % / 1 °C in Zagreb and 3.0 % / 1 °C in Rijeka). Thechanges in mortality with temperature are more pronounced inSplit than in other cities, varying between 2.7 % / 1 °C forPET, 5.4 % / 1 °C for minimum and maximum temperatureand as much as 6.5 % / 1 °C for mean temperature (Table 3).

Analysis of the relationship between mortality and PET at2.00 p.m. for eachmonth fromApril to October shows that thethresholds for the occurrence of increased mortality are notequal at the beginning of the warm season and in late summerFig. 5, Table 4). The values in Table 4 do not correspond to the

values in Table 2 because they present the thresholds of PETfor MR dev=0; i.e., the beginning of positive mortality, andnot threshold temperatures as in Table 2. In April, mortalitystarts to rise when PETexceeds 26.5 °C in Osijek and 29.5 °Cin Zagreb. In subsequent months these values increase,reaching the maximum mainly in August, when they varybetween 34 °C in Split and as much as 45.5 °C in Osijek.After that, thresholds fall again. Some deviations from thesemainly regular changes occur due to the small sample sizes.

Previous results show that, at all stations analysed inCroatia, increased mortality occurs when the PET at2.00 p.m. varies between 36 °C and 38 °C and exceedsthe threshold for strong thermal stress, according to thegrades of physiological stress shown in Table 1. According tothe distribution of mortality for different heat loads, therelationship between mortality and thermal conditionshas a U pattern, with increased mortality in the coldand warm part of the scale (Fig. 6). The lowest mortalityoccurs in the classes with no heat load or with slight heat

Table 3 Mean mortality rates at different levels of thermal loadaccording to PET at 2.00 p.m. ex Extreme, st strong, mo moderate, slslight, HS heat stress, CS cold stress, noS no thermal stress

Heat load Osijek Zagreb Rijeka Split

All −0.2±1.2 −0.1±0.5 0.0±1.0 0.1±1.1

exCS 2.1±2.2 1.7±1.0* 2.0±2.8 3.3±3.1

stCS 0.7±4.2 0.7±1.7 3.4±3.4 1.7±3.4

moCS −2.2±3.7 0.1±1.4 1.5±2.7 0.9±2.7

slCS −3.3±3.6 −0.8±1.4 −3.6±2.8* −0.7±2.8

noS −1.0±3.5 −2.8±1.3* −1.8±2.9 −3.5±3.0*

slHS −1.1±3.7 −3.0±1.3* −3.2±2.9* −5.3±2.9*

moHS −4.1±3.7* −1.5±1.4 −1.7±3.0 −4.8±3.0*

stHS −3.3±4.1 1.3±1.7 0.9±3.5 2.8±3.6

exHS 10.4±5.4* 6.0±2.3* 9.1±5.0* 17.7±5.3*

*Significantly different from the means of the complete series at the 5 %level

-20 -10 0 10 20 30 40 50

PET (°C)

-80

-60

-40

-20

0

20

40

60

80

MR

dev

(%

)

APRIL

-20 -10 0 10 20 30 40 50

PET (°C)

-80

-60

-40

-20

0

20

40

60

80

MR

dev

(%

)

AUGUST

Fig. 5 Scatter plot of excess mortality (MR dev in%) and PET at 2.00p.m. (°C) in Zagreb for April and August. Circles Mean values of excessmortality set for 3 °C bands, red curve lowess smooth, arrows border ofincreasing mortality rate (MR dev=0)

1140 Int J Biometeorol (2014) 58:1135–1145

stress. Mortality is also below the expected value for moderateheat stress between −1.5 % in Zagreb and −4.8 % in Split(Table 3). Therefore, moderate heat stress has no impact onincreased risk in Croatia.

The lagged correlation with lags of 0 to 15 days betweenmortality rates and PET at 2.00 p.m. higher than thresholdtemperatures (PET>HCP in Table 2) (Fig. 7), as well as thecourse of excess mortality during strong and extreme heatstress (Fig. 8), shows that, during heat stress, the highestmortality usually occurs in the first 3 days after exposure,followed by a decline to expected values or a fall below suchvalues 4 to 5 days after the stress (“mortality displacement”).In Zagreb and Osijek, the correlation between PET>HCP andmortality is positive 5–6 days after heat stress and then be-comes negative. According to the results of correlation inRijeka and Split, it seems that mortality increases 9–10 daysafter heat stress, and then declines below the expected value.Shortly thereafter, mortality in Split increases again, but thisincrease is probably not associated with previous heat stress.The quantification of the effect of mortality displacement isdetermined bymean cumulative excess mortality during 15 and30 days after PET exceeds the threshold temperature. In thecontinental part of Croatia (Zagreb and Osijek), the period withpositive excess mortality after the beginning of heat waves lasts

on average 4 days, resulting in mean cumulative excess mor-tality of around 20 % (Table 5). At the coast, where meanperiods with positive excess mortality last longer (from 7 daysin Rijeka to 11 days in Split), cumulative excess mortalityranges between 49 % and 92 %. On the other hand, the declineof excess mortality after that period is about 34 % in thecontinental part up to 15 days, and more than twice that up to30 days after the beginning of a heat wave. Due to the longerperiod of excess mortality at the Adriatic coast, the decline inmortality up to 15 days after that ranges between 7% and 15%,but is 32–50 % for an additional 15 days.

When the temperature (PET, minimum or maximum tem-perature) exceeds the threshold temperature, the longer theduration of unfavourable conditions, the more mortality in-creases (Fig. 9). In addition, when both minimum and maxi-mum temperatures exceed threshold temperatures, mortalityincreases with duration faster than with a minimum or maxi-mum temperature on its own. The increase in mortality withthe duration of unfavourable thermal conditions is more pro-nounced in the continental part of Croatia (Zagreb and Osijek)than on the coast, particularly in Split. After a 10-day durationof heat stress in the continental part, mortality increases on

-0.15

-0.10

-0.05

0.00

0.05

0.10

0.15

0.20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Time lag (days)

Cor

rela

tion

Fig. 7 Lagged correlations between PET>HCP (Table 5) and mortalityrates with lags of 0–15 days for Zagreb. Bold lines Statistical significancelimits of 5 % level

-10

-5

0

5

10

15

20

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Day after exposition

MR

dev

(%

)

extreme heat stressstrong heat stress

Fig. 8 Mean mortality rate per grade of thermo-physiological stress at2.00 p.m. for 0- to 15-day time lags for Zagreb

Table 4 Thresholds of PET at 2.00 p.m. for the start of increase inmortality (MR dev=0) by month

April May June July August September October

Osijek 26.5 35.5 36.0 36.5 45.5 – –

Zagreb 29.5 27.5 29.5 32.5 37.0 34.0 29.0

Rijeka – 35.0 35.0 35.5 40.5 – –

Split 29.0 34.0 35.5 35.0 34.0 37.5 –

-10

-5

0

5

10

15

20

MR

dev

(%

)

exCS stCS moCS slCS noS slHS moHS stHS exHS

Fig. 6 Mean mortality for different levels of thermal stress for PET at2.00 p.m. with the inclusion of short-term adaptation, Zagreb, period 1983–2008. ex Extreme, st strong, mo moderate, sl slight, HS heat stress, CScold stress, noS no thermal stress

Int J Biometeorol (2014) 58:1135–1145 1141

average between 10 % (Osijek) and 17 % (Zagreb) but only3 % in Split and 8 % in Rijeka, while during long-lasting heatwaves (10 days) cumulative mortality can rise to more than100 % (Fig. 10). Unlike other cities, where heat waves did notlast longer than 9 (Osijek) to 13 days (Rijeka), a heat wave in1994 in Split lasted for 25 days. At the end of prolonged heatwaves, as was the case in Split, mortality decreases due tomortality displacement (Tobias et al. 2010). If only heat waveslasting up to 13 days in Split are taken into account, theincrease in mortality is 17 % per 10 days. Similarly, the lowincrease in mortality with duration of maximum temperaturein Zagreb is a consequence of mortality displacement in thecase of several long-lasting heat waves as well as a 7-day heatwave with negative mortality in late August 2003, which isactually a continuation of the previous known long-lastingheat wave and can be also treated as mortality displacement.This is confirmed by the analysis of the relationship betweenthe frequency of heat waves in the season and excess mortalityduring the heat waves. This analysis shows that a greaternumber of heat waves does not cause greater cumulativeexcess mortality (Fig. 11). Moreover, it seems contradictorythat in 2003, when there were as many as 18 heat waves,including one lasting 7 days and one lasting 8 days, as well as

a total of 49 days with PET exceeding threshold values,cumulative excess mortality during the heat waves belongsto the lowest in the whole series and is even negative. On thecontrary, cumulative excess mortality was highest in 2005when there were 8 days with PET exceeding threshold valuesduring five heat waves, four of them lasting only 1 day andone lasting 4 days.

Discussion

The results of the relationship between temperature andmortality rate in Croatia show differences between variousgeographical regions and confirm the hypothesis about theadaptation of inhabitants to the local climate (Keatinge et al.2000).

The mean percentile of temperature threshold for increasedmortality for the cities analysed amounts to 94 %. Assuming asimilar relationship, that result can be used to estimate thresh-old temperatures for other areas in Croatia. The thresholds ofPET at 2.00 p.m. for increased mortality generally exceed35 °C. This is the lower limit for a class of strong heat stress,and the result confirms the appropriateness of the classifica-tion. The thresholds of PET causing heat-related mortality are

Table 5 Cumulative excess mortality after the beginning of the periodwith PETexceeding threshold temperature.MRdev cum Mean cumulativeexcess mortality during the period with MRdev>0; MRdev cum 15 ,MRdev cum 30 cumulative mortalities during the period of 15 and 30 daysafter the beginning of heat wave, respectively, dur MRdev>0 meanduration of period with consecutive days with MRdev>0

Osijek Zagreb Rijeka Split

dur MRdev>0 4 4 8 12

MRdev cum 21.1 18.4 49.3 91.8

MRdev cum 15 −33.6 −34.4 −14.9 −6.5

MRdev cum 30 −86.1 −70.8 −50.2 −31.6

-20

-10

0

10

20

30

40

50

60

1 3 5 7 9 11 13 15 17 19 21 23 25

Days

MR

dev

(%)

Tmax Tmin Tmax & Tmin

Fig. 9 Relationship between mortality and duration of periods with PET>HCP (above) and with minimum and maximum temperature exceedingthreshold temperatures, and simultaneously increased minimum andmaximum temperature (below) for Zagreb

0

20

40

60

80

100

120

1 3 5 7 9 11 13 15 17 19 21 23 25 27 30

Days

Cum

ulat

ive

MR

dev

(%) extreme heat stress

strong heat stress

Fig. 10 Cumulative mortality rates for strong and extreme heat stress at2.00 p.m. lasting 1–30 days for Zagreb

0

10

20

30

40

50

60

1983

1985

1987

1989

1991

1993

1995

1997

1999

2001

2003

2005

2007

Day

s

-60

-40

-20

0

20

40

60

80

100

MR

dev cum (%

)

Days

MR dev cum

Fig. 11 Number of days with PET exceeding threshold values andcumulative excess mortality during heat waves for Zagreb

1142 Int J Biometeorol (2014) 58:1135–1145

approximately equal to body temperature; therefore, when theenvironmental temperature exceeds this threshold, cooling ofthe body is difficult, especially if the temperature exceeds40 °C, which can cause a heat stroke and even mortality(Flynn et al. 2005; Kovats and Hajat 2008). The thresholdtemperatures for maximum temperature (30–33 °C) are higherthan those found by Kyselý and Huth (2004) for the CzechRepublic (26 °C) and Michelozzi et al. (2000) for Rome(28.0 °C). Differences between threshold temperatures forRome and Croatia may partly be a result of different methodsof investigation. On the contrary, due to the similarity ofmethods applied in the Czech Republic and in our study, thedifferences found are probably realistic and caused by climateconditions. Tobias et al. (2010) found threshold temperaturesof maximum temperature for Barcelona to be similar to thoseat the Adriatic coast, while Nastos and Matzarakis (2008a)have shown 33 °C for Athens. The highest threshold temper-ature of 41 °C was found by Díaz et al. (2002) for Seville. Formean temperatures, threshold temperatures range between23.5 °C and 25.5 °C and are similar to the values establishedin Budapest (Paldý et al. 2005; Ishigami et al. 2008), Rome(Michelozzi et al. 2000), Milan (Ishigami et al. 2008) andBarcelona (Saez et al. 2000), but higher than in London(Ishigami et al. 2008) and the Czech Republic (Kyselý andHuth 2004). The increase in the risk of death beyond thetemperature threshold (1–2 % for each degree of increase intemperature) is consistent with the results for different parts ofEurope (Ballester 1997; Iñiguez et al. 2010; Ishigami et al.2008).

However, due to short-term adaptation to previous thermalconditions, threshold temperatures vary during the warm sea-son by as much as 10°. Such differences were also found inother countries (Kyselý and Huth 2004; Saez et al. 2000; Hajatet al. 2002; Koppe and Jendritzky 2005; Laaidi et al. 2006),although the differences are not comparable due to differenttemperature indicators used by different authors.

Heat stress affects the increased mortality rate for the first3–5 days, but then the effect decreases sharply and mortalityfalls below average (“mortality displacement” or the“harvesting effect”) (Rooney et al. 1998; Smoyer et al. 2000;Kyselý and Huth 2004; Koppe 2005). This implies that theincreased mortality rate immediately after the heat wave is theresult of the mortality displacement of gravely ill patients whowould have probably lived a few days longer if there were noheat stress (Kunst et al. 1993). Smoyer et al. (2000) think thatmortality levels decrease below the baseline after the heatwave and suggest some pre-shifted mortality during the heatwave, while Kyselý (2004) believes that this short-term mor-tality displacement accounts for a substantial part of the mor-tality increase during heat waves. In the continental partof Croatia the decrease in cumulative mortality (up to 15 days)is greater than the cumulative increase of mortality in the firstdays of a heat wave. Prolonged heat stress causes significantly

higher mortality than 1-day heat stress because prolongedexposure of the human body to adverse conditions increasesthe risk, especially among vulnerable groups such as chronicpatients or the elderly (Díaz et al. 2002; Shao et al. 2008;Matzarakis et al. 2011; Kovats and Hajat 2008; Iñiguez et al.2010; Tobias et al. 2010). Extended duration of increased(both) minimum and maximum temperature is particularlydangerous due to the absence of night freshness that mayreduce the risk to the body after exposure to heat during theday (Rooney et al. 1998; Díaz et al. 2002; Nastos andMatzarakis 2008b). During long-lasting heat waves mortalityincreases mostly during the first 3–5 days; after that theincrease is weaker, but in the case of very long heat waves,mortality decreases at the end of the period due to mortalitydisplacement (Tobias et al. 2010). This happens in the south-ern Adriatic where particularly long-lasting heat waves occuras a result of the Azores anticyclone, which supports longperiods of fine and stable weather. In the mainland and in thenorthern Adriatic, cold air on the edge of cold fronts passing inthe north briefly interrupt periods of warmweather (Zaninovićet al. 2008).

The introduction of the HHWS, in addition to educating thepopulation about the possibilities of behavioural adaptation,may contribute to reducing the risk of heat stress (McGeehinand Mirabelli 2001; Tan et al. 2007; Fouillet et al. 2008;Koppe and Becker 2009). The risk of heat waves is particu-larly prominent in large cities due to the impact of the urbanheat island (UHI), which can be even more pronounced be-cause of atmospheric pollution (Smoyer et al. 2000; Fouilletet al. 2006). The UHI can be mitigated by planting greenareas, applying better insulation and the orientation of build-ings towards providing ventilation (Alberdi et al. 1998;Matzarakis and Nastos 2011). In addition to warning thepopulation, successful suppression of heat wave impactsrequires good organisation of public health services foreffective prevention and for provision of help during theheat waves (Kovats and Ebi 2006; Tan et al. 2007; Kovatsand Hajat 2008).

Conclusion

To summarise, it can be said that humans adapt to regionalclimate. The more variable conditions humans live in, the morethey adapt. The thresholds of PETat 2.00 p.m. generally exceed35 °C and can be taken as a good and relevant threshold. This isclose to core body temperature; therefore, when the environ-mental temperature exceeds this threshold, cooling of the bodyis difficult, especially if temperature exceeds 40 °C, which cancause heat stroke and even mortality.

Heat-related increased mortality can be observed after thefirst 3–5 days, but then the effect decreases sharply andmortality falls below average. During long-lasting heat waves

Int J Biometeorol (2014) 58:1135–1145 1143

mortality increases mostly during the first 3–5 days; after thatthe increase is weaker, but in the case of very long heat waves,mortality decreases at the end of the period due to mortalitydisplacement. This can be seen principally in the southernAdriatic where particularly long-lasting heat waves occur asa result of the Azores anticyclone with long-lasting stableconditions during summer.

Establishment of the HHWS, in addition to timely educa-tion of the population and the authorities may reduce theeffects of heat stress. The results presented here enabled theintroduction of the HHWS in Croatia in the summer of 2012.

References

Alberdi J, Díaz J, Montero JC, Mirón I (1998) Daily mortality in Madridcommunity 1986–1992: relationship with meteorological variables.Eur J Epidemiol 14:571–578

Asanuma S, Sato K, Kawahara I, Liang C, Ando M (2008) Impact of heatstress on urban population health in Japan and China. 18th InternationalCongress of Biometeorology, Harmony within Nature / Iriki, Masami(ur.). 22–26 September 2008, Tokyo, Japan, CD Hum1-O13

Ballester F (1997) Mortality as a function of temperature: a study inValencia, Spain, 1991–1993. Int J Epidemiol 26:551–561

Bassil KL, Cole DC, Tomic KE (2008) The effectiveness of public healthinterventions to reduce morbidity and mortality durig heat episodes,18th International Congress of Biometeorology, Harmony withinNature / Iriki, Masami (ur.). 22–26 September 2008, Tokyo, Japan,CD Hum3-O02

Blazejczyk K,McGregor G (2008) Biothermal thresholds for mortality inEurope. 18th International Congress of Biometeorology, Harmonywithin Nature / Iriki, Masami (ur.). 22–26 September 2008, Tokyo,Japan, CD Hum3-COST-O04

Cleveland WS (1979) Robust locally weighted regression and smoothingscatterplots. J Am Stat Assoc 74:829–836

Díaz J, García R, de Castro VF, Hernández E, López C, Otero A (2002)Effects of extremely hot days on people older than 65 years inSeville (Spain) from 1986 to 1997. Int J Biometeorol 46:145–149

Díaz J, García-Herrera R, Trigo RM, Linares C, ValenteMA,Miguel JMD,Hernández E (2006) The impact of the summer 2003 heat wave inIberia: how should we measure it? Int J Biometeorol 50:159–166

Dunne JP, Stouffer RJ, John JG (2013) Reductions in labour capacityfrom heat stress under climate warming. Nat Climate Change.doi:10.1038/nclimate1827

Farajzadeh M, Darand M (2009) Analyzing the influence of air temper-ature on the cardiovascular, respiratory and stroke mortality inTehran. J Environ Health Sci Eng 6:261–270

Flynn A,McGreevy C,Mulkerrin EC (2005)Why do older patients die ina heatwave? Q J Med 98:227–229

Fouillet A, Rey G, Laurent F, Pavillon G, Bellec S, Ghihenneuc-JouyauxC, Clavel J, Jougla E, Hémon D (2006) Excess mortality related tothe August 2003 heat wave in France. Int Arch Occup EnvironHealth 80(1):16–24

Fouillet A, Rey G, Wagner V, Laaidi K, Empereur-Bissonnet P, Le TertreA, Frayssinet P, Bessemoulin P, De Crouy-Chanel P, Jougla E,HémonD (2008)Has the impact of heat waves onmortality changedin France since the European heat wave of summer 2003? A study ofthe 2006 heat wave. Int J Epidemiol 37:309–317

Frost DB, Auliciems A, de Freitas C (1992) Myocardial infarct death andtemperature in Auckland, New Zealand. Int J Biometeorol 36:14–17

Gosling SN, McGregor GR, Páldy A (2007) Climate change and heat-related mortality in six cities part 1: model construction and valida-tion. Int J Biometeorol 51:525–540

Gosling SN, Lowe JA, McGregor GR, Pelling M, Malamud BD (2009)Associations between elevated atmospheric temperature and humanmortality: a critical review of the literature. Clim Chang 92:299–341. doi:10.1007/s10584-008-9441-x

Gover M (1938) Mortality during periods of excessive temperature.Public Health Rep (1896–1970) 53(27):1122–1143

Guest CS,Willson K,Woodward AJ, Hennessy K, Kalkstein LS, SkinnerC, McMichael AJ (1999) Climate and mortality in Australia: retro-spective study, 1979–1990, and predicted impacts in five majorscities in 2030. Clim Res 13:1–15

Hajat S, Kovats RS, Atkinson RW, Haines A (2002) Impact of hottemperatures on death in London: a tume serie approach. JEpidemiol Community Health 56:367–372

Höppe P (1999) The physiological equivalent temperature—a universalindex for the biometeorological assessment of the thermal environ-ment. Int J Biometeorol 43:71–75

Iñiguez C, Ballester F, Ferrandiz J, Pérez-Hoyos S, Sáez M, López A(2010) Relation between temperature and mortality in thirteenSpanish cities. Int J Environ Res Public Health 7:3196–3210.doi:10.3390/ijerph7083196

Ishigami A, Hajat S, Kovats RS, Bisanti L, RognoniM, RussoA, Paldy A(2008) An ecological time-series study of heat-related mortality inthree European cities. Environ Health 7:5. doi:10.1186/1476-069X-7-5

Johnson H, Kovats RS, McGregor G, Stedman J, Gibbs M, Walton H(2005) The impact of the 2003heat wave on daily mortality inEngland and Wales and the use of rapid weekly mortality estimates.Eurosuveillance 10(7) (http://www.eurosurveillance.org)

Kalkstein L, Jamason P, Greene J, Libby J, Robinson L (1996) ThePhiladelphia hot weather-health watch/warning system: develop-ment and application, summer 1995. Bull Am Meteorol Soc77:1519–1528

Keatinge WR, Donaldson GC, Cordiolo E, Martinelli M, Kunst AE,Mackenback JP, Nayha S, Vuori I (2000) Heat-related mortality inwarm and cold regions of Europe: observational study. Br Med J321:670–673

Kim J, Lee DG, Kyselý J, Choi BC (2008) Regional acclimation effectson heat-related mortality in Korea, 18th International Congress ofBiometeorology, Harmony within Nature / Iriki, Masami (ur.). 22–26 September 2008, Tokyo, Japan, CD Hum1-P02

Koppe C (2005) Gesundheitsrelevante Bewertung von thermischerBelastung unter Berücksichtigung der kurzfristigen Anpassung derBevölkerung an die lokalen Witterungsverhältnisse (Health-relatedassessment of thermal stress considering the short-term adaptationof the population to local weather conditions), Albert-Ludwigs-Universität Freiburg, http://www.freidok.uni-freiburg.de/volltexte/1802/, p 208

Koppe C, Becker P (2009) Comparision of operational heat healthwarning systems in Europe. In: Matthies F, Menne B (eds)Preparedness and response to heat-waves in Europe, from evidenceto action. Public health response to extreme weather events. WHORegional Office for Europe, Copenhagen

Koppe C, Jendritzky G (2005) Inclusion of short-term adaptation tothermal stresses in a heat load warning procedure. Meteorol Z14:271–278

Kovats RS, Ebi KL (2006) Heatwaves and public health in Europe. Eur JPublic Health 16:592–599

Kovats RS, Hajat S (2008) Heat stress and public health: a critical review.Annu Rev Public Health 29:9.1–9.15

Kunkel KE, Changnon SA, Reinke BC, Arritt RW (1996) The July 1995heat wave in theMidwest: a climatic perspective and critical weatherfactors. Bull Am Meteorol Soc 77:1507–1518

1144 Int J Biometeorol (2014) 58:1135–1145

Kunst AE, Looman CW, Mackenbach JP (1993) Outdoor air temperatureand mortality in the Netherlands: a time-series analysis. Am JEpidemiol 137:331–341

Kyselý J (2004) Mortality and displaced mortality during heat waves inthe Czech Republic. Int J Biometeorol 49:91–97

Kyselý J, Huth R (2004) Heat-related mortality in the Czech Republicexamined through synoptic and ‘traditional’ approaches. Clim Res25:265–274

LaaidiM, Laaidi K, Besancenot J (2006) Temperature-related mortality inFrance, a comparison between regions with different climates fromthe perspective of global warming. Int J Biometeorol 51:145–153

Larsen J (2006) Setting the record straight: more than 52,000 Europeansdied from heat in summer 2003. Earth Policy Institute, http://www.earthpolicy.org/plan_b_updates/2003/update29

Matzarakis A, Mayer H (1996) Another kind of environmental stress:thermal stress. WHO Newsletter 18:7–10

Matzarakis A, Nastos P (2011) Human-biometeorological assessmentof heat waves in Athens. Theor Appl Climatol 105:99–106.doi:10.1007/s00704-010-0379-3

Matzarakis A, Mayer H, Iziomon MG (1999) Applications of a universalthermal index: physiological equivalent temperature. Int JBiometeorol 43:76–84

Matzarakis A, Rutz F, Mayer H (2007) Modelling radiation fluxes insimple and complex environments—application of the RayManmodel. Int J Biometeorol 51:323–334

Matzarakis A, Rutz F, Mayer H (2010) Modelling radiation fluxes insimple and complex environments—basics of the RayMan model.Int J Biometeorol 54:131–139

Matzarakis A, Muthers S, Koch E (2011) Human biometeorologicalevaluation of heat-related mortality in Vienna. Theor ApplClimatol 105:1–10. doi:10.1007/s00704-010-0372-x

Mayer H, Höppe P (1987) Thermal comfort of man in different urbanenvironments. Theor Appl Climatol 38:43–49

McGeehin MA, Mirabelli M (2001) The potential impacts of climatevariability and change on temperature-related morbidity and mortalityin the United States. Environ Health Perspect 109(Suppl 2):185–189

McGregor GR, PellingM,Wolf T, Gosling SN (2007) The social impactsof heat waves. Environment Agency, Bristol

McMichael AJ, Haines A, Slooff R, Kovats RS (eds) (1996) Climate changeand human health: an assessment by a task group on behalf of theWorldHealth Organization, the World Meteorological Organization and theUnited Nations Environment Programme. WHO, Geneva

Michelozzi P, Fano V, Forastiere F, Barca A, Kalkstein LS, Perucci CA(2000) Weather conditions and elderly mortality in Rome duringsummer. WMO Bull 49:348–355

Muthers S, Matzarakis A, Koch E (2010) Summer climate and mortalityin Vienna—a human-biometeorological approach of heat-relatedmortality during the heat waves in 2003. Wien Klin Wochenschr122:525–531. doi:10.1007/s00508-010-1424-z

Nastos P, Matzarakis A (2008a) The effect of air temperature andthe thermal index PET on mortality in Athens, Greece. 18thInternational Congress of Biometeorology, Harmony within Nature /Iriki, Masami (ur.). 22–26 September 2008, Tokyo, CD, Hum1-P09

Nastos P, Matzarakis A (2008b) Human-biometeorological effects onsleep disturbances in Athens, Greece: a preliminary evaluation.Indoor Built Environ 17:535–542

Paldý A, Bobvos J, VámKovats RS, Hajat S (2005) The effect of temper-ature and heat waves on daily mortality in Budapest, Hungary, 1970–2000. In: Kirch W, Menne B, Bertollini R (eds) Extreme weather

events and public health responses.WHORegional Office for Europe,Copenhagen, pp 99–108

Pan W, Li L, Tsai M (1995) Temperature extremes and mortality fromcoronary heart disease and cerebral infarctions in elderly Chinese.Lancet 345:353–355

Pascal M, Laaidi K, Ledrans M, Baffert E, Caserio-Schönemann C, LeTertre A, Manach J, Medina S, Rudant J, Empereeur-Bissonet P(2006) France’s heat health watch warning system. Int JBiometeorol 50:144–153

Robine JM, Cheung SL, Le Roy S, Van Oyen H, Herrmann FR (2007)Report on excess mortality in Europe during summer 2003. EUCommunity Action Programme for Public Health, http://ec.europa.eu/health/ph_projects/2005/action1/docs/action1_2005_a2_15_en.pdf

Robinson PJ (2001) On the definition of a heat wave. J Appl Meteorol40:762–775

Rooney C, McMichael AJ, Kovats RS, Coleman MP (1998) Excessmortality in England and Wales and Greater London during the1995 heatwave. Epidemiol Community Health 52:482–486

Saez M, Sunyer J, Tobias A, Ballester F, Antó JM (2000) Ishaemic heartdisease mortality and weather temperature in Barcelona, Spain. EurJ Public Health 10:58–63

Schönwiese CD (1992) Praktische Statistik für Meteorologen undGeowissenschaftler, 2nd edn. Gebrüder Borntraeger, Berlin, Stuttgart

Schuman S (1972) Patterns of urban heat-wave deaths and implicationsfor prevention: data from New York and St. Louis during July 1966.Environ Res 5:59–75

Schwartz J (2005) Who is sensitive to extremes of temperature?Epidemiology 16:67–72

Shao L, Luo M, Walker R, Liu X, Hwang S, Robert C (2008)Impact of hot weather conditions on respiratory and cardiovascularhospital admissions in New York city, USA. Epidemiology19:S302–S303

Smoyer KE, Rainham DGC, Hewko JN (2000) Heat-stress-related mor-tality in five cities in Southern Ontario: 1980–1996. Int JBiometeorol 44:190–197

Tan J, ZhengY, TangX,GuoC, Li L, SongG, ZhenX,Yuan D, KalksteinA, Chen H (2007) Heat wave impacts on mortality in Shanghai,1998 and 2003. Int J Biometeorol 51:193–200

Tobias A, de Ollalla PG, Linares C, Bleda MJ, Caylà DJ (2010) Short-term effects of extreme hot summer temperatures on total dailymortality in Barcelona, Spain. Int J Biometeorol 54:115–117.doi:10.1007/s00484-009-0266-8

Toulemon L, Barbieri M (2006) The mortality impact of the August 2003heat wave in France, population of America Association Meeting,30 March–1 April, Los Angeles

Whitman S, GoodG, Donoghue ER, Nanette BenbowN, ShouW,Mou S(1997) Mortality in Chicago attributed to the July 1995 heat wave.Am J Public Health 87:1515–1518

WHO (2003) The health impacts of 2003 summer heat-waves. Briefingnote for the delegations of the fifty-third session of the WHORegional Committee for Europe, http://www.irisknet.cn/riskdb/globalrisk/juzai/10.6.pdf

Yamanaka S, Nakamura Y (1999) Relationship between CVD mortalityand environmental temperature, selected papers from the ConferenceICB-ICUC’99, Sydney, 8–12 November 1999, WCASP-50, WMO/TD No. 1026: 157–162

Zaninović K, Gajić-Čapka M, Perčec Tadić M et al (2008) Klimatskiatlas Hrvatske climate atlas of Croatia 1961–1990. Državnihidrometeorološki zavod, Zagreb

Int J Biometeorol (2014) 58:1135–1145 1145