Dr. C. Giannakopoulos Institute of Environmental Research & Sustainable Development, National Observatory of Athens, Greece. WP6.2 Progress and Plans Linking.

Dr. C. GiannakopoulosDr. C. Giannakopoulos

Institute of Environmental Research & Sustainable Development,

National Observatory of Athens, Greece.

WP6.2 Progress and PlansWP6.2 Progress and Plans

Linking impact models to probabilistic scenarios of climate

ENSEMBLES General AssemblyPrague, 12 – 16 November 2007.

Forest fire impact studyForest fire impact study

Completion of journal paper:  P. Good, Moriondo M., Giannakopoulos C. and Bindi M., The meteorological conditions associated with extreme fιre risk in Italy and Greece: relevance to climate model studies, Int. J. Wildland Fire, in press, 2007.

Collaborative study between National Observatory of Athens and University of Florence

Conference proceedings following ENSEMBLES Deliverable D6.91 Giannakopoulos, C., P.LeSager, E. Kostopoulou, A. Vajda, and A.

Venlainen, Intercomparison study of modelled forest fire risk in the Mediterranean for present day conditions, 6th International Workshop on Advances in Remote Sensing and GIS Applications in Forest Fire Management, Thessaloniki, Greece, 26-29 September 2007.

Collaborative study between National Observatory of Athens and Finnish Meteorological Insitute. Plan to convert to journal paper and involve also University of Florence.

Dr. C. GiannakopoulosDr. C. Giannakopoulos

Dr. B. E. PsiloglouDr. B. E. Psiloglou

Heat Stress and MortalityHeat Stress and Mortalityin Athens:in Athens:

Impact Model ConstructionImpact Model Constructionand Validationand Validation

ENSEMBLES General AssemblyPrague, 12 – 16 November 2007.

IntroductionIntroduction

Global climate change will have direct impacts on human health, including increased mortality due to heat stress and heat waves.

An empirical-statistical model for heat stress is constructed for the city of Athens, using the JUNE-AUGUST or JUNE-SEPTEMBER months of the observational period 1992-2006.

The ultimate aim will be to use ENSEMBLES multiple regional climate model output to estimate daily mortality under a climate change world.

Data AvailabilityData Availability

All-cause daily mortality data for the district of Athens, Greece, covering the period 1992-2006, were acquired.

Source: Greek National Institute of Statistics, Greece.

Daily climate data (Air Temperature & Rel. Humidity, Maximum & Minimum Air Temperature, Wind, Solar rad. Intensity),

for the same period 1992-2006 were provided

Source: National Observatory of Athens, Greece.

All-cause daily mortality data for Athens, Greece ,All-cause daily mortality data for Athens, Greece ,

and for the period 1992-2006.and for the period 1992-2006.

Light-blue line is a smoothed 30-days running mean.

Mean number of all-cause deaths per month, Mean number of all-cause deaths per month, for ATHENSfor ATHENS

MONTH 1992-2006 1992-1999 2000-2006

1 94.3 90.9 98.2

2 94.3 95.1 93.4

3 91.7 90.9 92.5

4 84.3 81.1 87.9

5 80.6 77.8 83.7

6 78.9 75.3 83.1

7 81.2 78.2 84.6

8 79.1 77.5 81.0

9 70.6 67.9 73.7

10 75.6 72.4 79.3

11 81.1 80.3 82.0

12 87.9 85.8 90.2

Calculation of EXCESS DEATHS for summer months,Calculation of EXCESS DEATHS for summer months,

for ATHENS and for the period 1992-2006for ATHENS and for the period 1992-2006

Two approaches were taken inorder to calculate EXCESS DEATHS, i.e. deaths beyond those expected for a specific period in that population:

A. Use of a FIXED MEAN of daily mortality for each summer month, for the period 1992-2006 (78.9 deaths for June, 81.2 in July and 79.1 in August).

B. Use of a 30-days running mean, which smoothes the fluctuations in the death data, but selecting only the summer months.

In each case, daily excess deaths were calculated by subtracting the expected form the observed daily death values.

Example: for the fixed mean approach, that meant that subtracting 78.9 from every observed daily death for June, 81.2 from every observed daily death of July, and 79.1 from every observed daily death of August.

In the 30-days running-mean method, from each observed daily death its corresponding expected value was subtracted.

Example for EXCESS Deaths calculationExample for EXCESS Deaths calculation

DATEMAXIM.

AIR TEMP

OBSERVED NUMBER OF

DEATHS

EXCESS DEATHS (FIXED MEAN

FOR JULY=81.2)

30-DAYS RUNNING

MEAN

EXCESS DEATHS USING 30-DAYS RUNNING MEAN

01/07/01 34.9 77 -4.2 80.9 -3.9

02/07/01 33.3 88 6.8 81.1 6.9

03/07/01 29.1 88 6.8 81.3 6.7

04/07/01 32.5 74 -7.2 82.5 -8.5

05/07/01 34.0 68 -13.2 83.8 -15.8

06/07/01 35.5 78 -3.2 84.2 -6.2

07/07/01 34.5 65 -16.2 85.8 -20.8

08/07/01 36.2 81 -0.2 86.3 -5.3

09/07/01 38.4 82 0.8 86.6 -4.6

10/07/01 39.0 90 8.8 86.1 3.9

11/07/01 36.2 88 6.8 86.2 1.8

12/07/01 34.3 80 -1.2 85.4 -5.4

13/07/01 35.9 96 14.8 85.4 10.6

14/07/01 36.3 95 13.8 86.1 8.9

15/07/01 37.4 115 33.8 86.8 28.2

16/07/01 39.3 96 14.8 86.9 9.1

17/07/01 37.0 77 -4.2 86.5 -9.5

18/07/01 39.2 90 8.8 86.4 3.6

19/07/01 39.2 103 21.8 86.7 16.3

20/07/01 38.4 104 22.8 86.9 17.1

Calculated

EXCESS Deaths: Description of MethodEXCESS Deaths: Description of Method

1. Heat-related deaths were defined as the number of deaths occurring in excess of the number that would have been expected for that population in the absence of stressful weather (McMichael et al., WHO, 1996).

2. Each number of excess deaths was then grouped into the corresponding 1oC interval of maximum air temperature for simplification purposes (common in an ecological study design)

Example: if on the 16th of July the maximum temperature was 39.3oC and there were 10 excess deaths, 10 would be put in the 39-39.9oC interval.

3. All excess deaths in each 1oC interval for the entire period were added in order to find out where heat-related deaths were no longer detectable. In this way only temperatures over a certain threshold were regressed.

Example: if the maximum temperature on the interval 39-39.9oC was observed 5 times, and the calculated excess deaths were: +20, -15, +12, -7 and +10, then SUM=(+20)+(-15)+(+12)+(-7)+(+10)=20.

4. Finally, the sum of the excess deaths in each interval was divided by the frequency of occurrence of that temperature interval in the 1992-2006 period, to give the number of deaths per day for a particular temperature interval.

Example: if there were 681 deaths (the sum of all excess deaths negative & positive that occurred) in the 39oC interval (i.e. 39-39.9oC), and the number of times this temperature interval was observed in the period 1992-2006 is 27, then the number of excess deaths per day is equal to 681/27=25.2

EXCESS Deaths: Description of MethodEXCESS Deaths: Description of Method

Example for EXCESS Deaths per DAY calculationExample for EXCESS Deaths per DAY calculation

LOW LIMIT

UPPER LIMIT

SUM OF ALL EXCESS DEATHS (POS. or NEG.)

NUMBER OF TIMES

OBSERVED

EXCESS DEATHS PER DAY

FREQUENCY OF APPEARENCE

25 25.9 -67 6 -11.167 0.004762

26 26.9 -87.3 7 -12.471 0.0055556

27 27.9 -183.3 18 -10.183 0.0142857

28 28.9 -248.2 42 -5.91 0.0333334

29 29.9 -406.1 52 -7.81 0.0412698

30 30.9 -579.2 90 -6.436 0.0714286

31 31.9 -595.8 116 -5.136 0.0920635

32 32.9 -816.4 136 -6.003 0.1079365

33 33.9 -245.6 169 -1.453 0.1341270

34 34.9 -28.4 189 -0.15 0.1500001

35 35.9 173 138 1.254 0.1095238

36 36.9 696.1 112 6.215 0.0888889

37 37.9 869.6 81 10.736 0.0642857

38 38.9 450.6 47 9.587 0.0373016

39 39.9 681.7 27 25.248 0.0214286

40 40.9 238.2 13 18.323 0.0103175

41 41.9 73.8 2 36.9 0.0015870

42 42.9 75.1 5 15.02 0.0039683

43 43.9 33.6 2 16.8 0.0015873

44 44.9 33.4 1 33.4 0.0007936

Period: 1992-2006, Months: JUNE-JULY-AUGUST, using FIXED MEAN values

Total number of data=1260

Summer Deaths per Day for each temperature Summer Deaths per Day for each temperature interval for Athens, during the period 1992-2006interval for Athens, during the period 1992-2006

A V-shape relationship between mortality and temperature has been observed.

The observed/expected analysis, under both approaches, showed that hotter days were associated with greater mortality risk. It becomes clear that the 30-days running mean approach gives a more conservative estimate of excess deaths than the fixed-summer months mean. Both approaches were consistent in showing that heat-related deaths were not discernible below 34oC. Substantial heat-related deaths, occurred at very high temperatures. The temperature interval/frequency curve shows that very high maximum temperatures rarely occurred in the 1992-2006 period.

Summer Deaths per Day: Changing the month periodSummer Deaths per Day: Changing the month period

Climate-Mortality relationshipClimate-Mortality relationship

The non-linear regression of excess deaths per day above 34oC (in 1oC intervals) by its maximum temperature interval led to exponential equations.

Using Summer Fixed Means

Y=0.0005236 exp(0.2559 X)

R2=0.615

Using 30-days running mean

Y=0.0000072 exp(0.3569 X)

R2=0.725

Climate-Mortality relationship: Model ValidationClimate-Mortality relationship: Model Validation

The testing of predictive models is the most critical stage of an impact assessment (Parry & Carter 1998) :

1. In order to validate the model, available all-causes mortality data for Athens (1992-2006) were split into two samples, hereafter referred to as 1990s (1992-1999) and 2000s (2000-2006).

2. For both periods, associations between mortality and climate were established for maximum air temperatures over 34oC (the observed threshold for both periods), using the same observed/expected analysis and the regression methods described/used above. This test was performed both for summer fixed-mean values and 30-days running means.

(see the following diagrams)

3. Finally, each function was applied to the daily maximum temperature series of the other time period, and observed and modeled data were compared by using the correlation between predicted vs. observed values and residual analysis approach.

Climate-Mortality EXP-relationships for 1990s and 2000sClimate-Mortality EXP-relationships for 1990s and 2000s

1990s

2000s

Using Summer Fixed Means

Y=0.0008745 exp(0.2472 X)

R2=0.722

Using 30-days run. Mean

Y=0.0000647 exp(0.3075 X)

R2=0.769

Using Summer Fixed Means

Y=0.00244 exp(0.2176 X)

R2=0.619

Using 30-days run. Mean

Y=0.0001309 exp(0.2873 X)

R2=0.862

Climate-Mortality relationship: Summer FIXED meansClimate-Mortality relationship: Summer FIXED means

Model Validation: Residual analysisModel Validation: Residual analysis

Climate-Mortality relationship: 30-days running meansClimate-Mortality relationship: 30-days running means

Model Validation: Residual analysisModel Validation: Residual analysis

Climate-Mortality relationship: Summer FIXED meansClimate-Mortality relationship: Summer FIXED means

Model Validation: Correlation between obs. – pred.Model Validation: Correlation between obs. – pred.

Corr. coef.=0.8555

p>99%

N=10

Corr. coef.=0.6321

p>95%

N=8

Climate-Mortality relationship: 30-days running meansClimate-Mortality relationship: 30-days running means

Model Validation: Correlation between obs. – pred.Model Validation: Correlation between obs. – pred.

Corr. coef.=0.8256

p>99%

N=10

Corr. coef.=0.9120

p>99%

N=8

Calculation of EXCESS DEATHS for summer months,Calculation of EXCESS DEATHS for summer months,

for ATHENS and for the period 1992-2006,for ATHENS and for the period 1992-2006,

using the HEAT INDEX valueusing the HEAT INDEX value

The same methodology was used in order to calculate EXCESS DEATHS based on the values of HEAT INDEX.

Two approaches for the Heat-Index estimation:

A. MAX-Heat-Index: was calculated using the Maximum air temperature and the corresponding Relative Humidity between 13:00 – 17:00 LST time period.

B. MIN-Heat-Index: was calculated using the Minimum air temperature and the corresponding Relative Humidity between 5:00 – 7:00 LST time period.

Heat-Index equationHeat-Index equation (C.SCHOEN, J.Appl.Meteor., 2005):

HEAT-INDEX = T - 1.0799 EXP(0.03755 T)

(1.0 - EXP(0.0801 (Td-14.0)))

where T is the air temperature and Td is the dew point temperature:

Td = (237.3 * AB)/(17.27 - AB), where

AB = ((17.27 * T)/(237.3 + T)) + LOG(RH/100.0), RH=rel.humidity

Heat Index MAX and MINHeat Index MAX and MIN

Summer Deaths per Day for each Summer Deaths per Day for each Heat IndexHeat Index interval interval for Athens, during the period 1992-2006for Athens, during the period 1992-2006

Heat-Index-Mortality relationshipHeat-Index-Mortality relationshipThe non-linear regression of excess deaths per day above 35oC (in 1oC intervals) by its MAX-Heat-Index interval, and above 24oC (in 1oC intervals) by its MIN-Heat-Index interval led to exponential equations.

ConclusionsConclusions

Our study has so far shown that there have

been considerable heat-related deaths in the city of Athens,

both from moderate and extreme heat, during the

summer months of 1992–2006. The empirical-statistical

model constructed is shown to reproduce well the

observed heat-related deaths. This makes the model

more reliable for the quantification of the potential impacts

of climate change on health to be studied using ENSEMBLES regional model output.

Acknowledgement :Acknowledgement :

E-mail:E-mail: [email protected] [email protected]

[email protected] [email protected]

This work was supported by EU project This work was supported by EU project ENSEMBLESENSEMBLES

under contract  number GOCE-CT-2003-505539under contract  number GOCE-CT-2003-505539

THANK YOU FOR YOUR ATTENTIONTHANK YOU FOR YOUR ATTENTION