Heat-related mortality in Moldova: the summer of 2007

Authors


ABSTRACT

Heat waves of 2007 in Chisinau (Moldova) were used to study the relationship between elevated temperatures and excess mortality caused by these events. As reference information, daily temperature and mortality data for an 8-year reference period (2000–2008 without 2007) were used. Mean (Tmean), maximum (Tmax) and minimum (Tmin) daily air temperatures, and corresponding apparent temperatures (ATs) in the warm season (April to September) were correlated with excess total mortality in 2007, taken as the difference of daily death counts or their 7-d moving averages with those of the reference period. Observed excess mortality was totalled about 190–200 deaths or 6.5–6.9% of the reference mortality. The average daily excess deaths above the threshold temperatures (TTs), in terms of a used ‘estimator’ (a temperature variable), were in the range of 2.0–4.4% per 1 °C. TTs were identified as the lowest 2 °C class intervals above which the excess mortality rates began a sharp increase from their zero reference value. For Tmean, Tmax and Tmin they were estimated as ∼25, 31 and 19 °C; TTs for ATs were somewhat lower. The heat waves were defined as a continuous period satisfying three conditions for daily Tmax: (1) it is above the 99th percentile of its reference distribution (2000–2008) for at least three consecutive days, (2) its average value is equal to at least this percentile for the entire period and (3) all daily values are above 90th percentile for the entire period. On the whole, eight heat events caused 146 excess deaths or about 73–77% of their total number in the warm period of 2007. Temperature–excess mortality relationships become stronger with an increasing time lag; maximal effects were revealed on the second–third days for Tmean and Tmax, and on the first–second days for Tmin. The total effect of mortality displacement was estimated as about 17–25% of ‘positive’ excess deaths.

1. Introduction

Mortality rates are ambient temperature dependent and have long been associated with the effects of both heat and cold. Research by epidemiologists and climatologists has grown rapidly following the European heat waves in 2003 (Schär et al., 2004; Carson et al., 2006; Laaidi et al., 2006; Confalonieri et al., 2007; Gosling et al., 2007, 2009a, 2009b; Matthies et al., 2008; Basu, 2009; Jendritzky and de Dear, 2009; Menne and Matthies, 2009; Tobías et al., 2010). That summer many western European countries experienced dramatic death tolls, and temperatures were considered as ‘a shape of things to come’ (Beniston, 2004). The analysis of isolated extreme events in addition to the mean climate is supported by empirical evidence that impacts of high temperatures are different from those associated with variability in the mean climate: they tend to be greater within a shorter time frame (Hanson et al., 2007). Moreover, such analysis provides an useful insight into the short-term response of populations to these events at the regional level.

Although the extremely hot summer of 2003 in Western Europe is well-known and well-studied (Parry et al., 2007; Euroheat, 2009; D'Ippoliti et al., 2010), another hot summer, in 2007, concentrated more significantly in southeastern Europe, has not received as much attention. This year was one of the warmest in the history of instrumental observations in Greece (Founda and Giannakopoulos, 2009), Romania (Busuioc et al., 2007) and Moldova; in the latter, practically all temperature records were broken in winter, spring and especially in summer, including the all-time record maximum temperature of 41.5°C, reported on 21st July (Bugaeva and Mironova, 2007). Similar to the perception of 2003 in western and central Europe (Beniston, 2004), the year 2007 also appears unusual given Moldova's historical climate (Corobov et al., 2010). This fact, along with general drying of Moldova's climate and a nearly complete lack of national biometeorological studies in recent times, has triggered this research.

Several basic aspects shaped its content. The first issue for debate is how heat waves should be defined. In the absence of an adequate definition, it is impossible to assess either their change in the past or their possible consequences for the future. In the scientific literature, heat waves have various and in some cases overlapping definitions. According to the most recent (IPCC, 2012, p. 560) a heat wave is ‘A period of abnormally hot weather’. Reasoning by analogy, a heat day could be defined as ‘A day of abnormally hot weather’. Heat waves occur typically on synoptic scales, being often associated with the development of large stationary or quasi-stationary high pressure systems (Kyselý, 2007, 2010; Koffi and Koffi, 2008), and are usually characterized by a very large positive anomaly of air temperature of several days duration. Because heat waves differ spatially in their duration, intensity and time intervals between individual events, their studies are generally based on combinations of these three dimensions. In turn, heat intensity can either be based on the probability of occurrence of given percentiles or on temperature threshold exceedance with respect to a given reference period and given time frame (days, month and season). The choice of an appropriate reference period is tied to the task to be addressed.

Different probability-based (relative) and threshold-based (absolute) definitions of heat waves, which were used for their identification, can be found in numerous articles (Huynen et al., 2001; Vescovi et al., 2005; Beniston et al., 2007; Della-Marta et al., 2007; Medina-Ramón and Schwartz, 2007; Koffi and Koffi, 2008; Kyselý and Kříž, 2008; EuroHEAT, 2009; Kyselý and Kim, 2009; Menne and Matthies, 2009; Orlowsky and Seneviratne, 2012). At the same time, both these approaches have their advantages and limitations in their relations to impacts, which are summarized in the IPCC Special Report (IPCC, 2012) as follows:

  • Single absolute thresholds, which measure heat events of a fixed intensity, are space- and time-depended, being specific for every locations and time periods as well as to local population sensitivity to health.
  • Relative thresholds, which measure heat events of a fixed rarity from the extreme tails of probability distributions is not necessarily extreme in terms of impacts. Moreover, as Ballester et al. (2009) noted, the choice of percentiles can be a ‘trickier problem’.

These limitations hinder the formulation of a ‘standard’ heat wave definition and explain why it does not currently exist. From the view of this research, though heat waves are meteorological events, as the most suitable definitions appear those, which refer to human health impacts. One of such definition was proposed by Robinson (2001): ‘…an extended period of unusually high atmosphere related heat stress, which causes temporary modifications in lifestyle, and which may have adverse health consequences for the affected population’. Under such an approach, a critical moment in heat wave identification is to define a threshold above which it is believed that ambient temperatures may be hazardous to human well-being. The threshold value cannot be an arbitrarily chosen number; it has to be based on comprehensive studies that assess in terms of excess deaths the limits of human tolerance to hot temperatures.

The objective of this study was thus to estimate the effect of hot temperatures on mortality of Moldavian population, using the record drought and heat waves the country experienced in 2007 as a case study.

2. Materials and methods

We examined temperature exposure during the warm season, rather than limiting the research only to extremely high temperature periods or individual heat waves. The study period included 6 months (1 April to 30 September) for each of 9 years (2000–2008).

Daily mean (Tmean), maximum (Tmax) and minimum (Tmin) temperatures, as pertinent indicators of the observed and potential mortality impacts, were provided by the State Hydrometeorological Service of Moldova. In addition, apparent temperature (AT), as a measure of perceived exposure, was derived according to Steadman (1984):

display math

where Ta is air temperature and Td is dew point temperature. Because AT was one of the main exposure measures, humidity was not modelled separately.

Daily mortality data comprised all-cause deaths in the resident population of Chisinau. A total of 26 651 deaths were recorded in the city during the summer months of the 2000–2008 period, including 3106 who died in 2007. This information was retrieved from the death certificates archived at the National Center of Management in Health and represents both urban and rural population. As of 1 January 2009, of the 785 400 residents of Chisinau, 716 920 (91.3%) resided in the city itself, with the remainder in the suburban area. On the whole, the study encompasses about a quarter of Moldova's population, including about half of the urban one. To remove an inherent seasonality in daily mortality records, a common ‘observed−expected’ method (Dessai, 2002; Gosling et al., 2007) was used. An expected or reference mortality baseline was calculated as average death counts for each day of the study period where the warm period of 2007 was excluded. For greater reliability, both the expected and observed mortalities were taken as fixed daily death counts and their 7-d moving averages. Seven-day smoothing was selected as optimal in comparison with other alternatives (e.g. 3 or 31 d) because it is associated with small errors, yet still includes the entire weekly cycle of mortality. Subtraction of the reference mortality from daily mortality observed in 2007 resulted in mortality anomalies, or excess deaths.

The aggregate dose–response relationship between elevated temperatures and mortality was examined by grouping the daily excess death counts into non-overlapping 2 °C class intervals of temperature variables; the 0.01 °C increment of grouping ensured no loss of excess deaths. Because sample sizes (days) were small for very high temperatures, only the intervals with samples including at least 5 d were considered. Poisson regression analysis was used to assess excess mortality sensitivity to heat above threshold temperature (TT).

From different approaches to heat-wave identification we decided to use the one based on the combination of relative TT with their durations. A two-level criterion, similar to Meehl and Tebaldi (2004), was selected. According to this criterion, a heat wave is the longest continuous period satisfying the following three conditions:

  1. The daily maximum temperature must be above T1 for at least three consecutive days.
  2. The average daily maximum temperature must be at least T1 for the entire period.
  3. The daily maximum temperature must be above T2 for every day of the entire period.

T1 and T2 were identified as those percentiles of temperatures distribution in the current climate that best match the identified TTs.

Some details on the methods are described in the corresponding subsections of the article. All statistical analyses were performed with the StatGraphics Centurion Data Analysis and Statistical Software (Statgraphics Centurion XVI User Manual, 2009)

3. Results and discussion

3.1. Air temperature and mortality of 2007 compared with reference values

The entire warm period of 2007 was unusual by its temperatures, but the summer months were extremely hot. Some illustrations of their extreme character are given in Figures 1 and 2 as well as in Table 1 that show the summary statistics for each of the variable selected for the research. The climatological analysis was carried out for three subsamples: the baseline period (1961–1990), the reference period (2000–2008 without 2007) as well as 2007 itself.

Figure 1.

Mean monthly and summer air temperature in Chisinau in 2007 (vertical lines) relative to the temperatures of baseline 30 years (1961–1990), approximated by a normal distribution curve. T, observed temperatures; σ, standard deviation of the baseline period; δT/σ, normalized anomalies from baseline norms.

Figure 2.

Deviations of Chisinau daily air temperatures in summer 2007 from their baseline values (horizontal zero line).

Table 1. Descriptive statistics of the warm period (April to September) air temperature (°C) and daily mortality (deaths counts) in Chisinau for different periods of averaging
VariablePeriodStatistics
AverageSD (σ)MinMaxRange
 1961–199017.33.98.122.314.2
Tmean2000–200818.14.47.024.617.6
 200719.96.25.732.526.8
 1961–199022.83.913.728.414.7
Tmax2000–200823.54.611.230.319.1
 200725.57.06.739.432.7
Tmin1961–199012.33.74.017.213.2
 2000–200813.54.22.719.817.1
 200714.85.62.026.624.6
Mortality, case/day2000–200816.11.89.020.511.5
 200717.14.98.034.026.0

Figure 1 compares mean monthly and mean summer air temperatures in Chisinau with their distributions in the baseline period. The monthly temperatures in 2007 exceeded the baseline climate by 2.5–4.0σ, and the summer temperature by 5σ. Moreover, all three mean temperature variables for June through August 2007 exceeded the 95th percentile of the baseline climate (Table 2), exceeding the baseline normal by up to 12 °C. Such anomalies of monthly temperatures resulted from extremely hot days over the summer (Figure 2).

Table 2. Summer 2007 temperatures (°C) in Chisinau in comparison with 90 and 95% quantiles of their distribution in the baseline climate
 Monthly air temperature
MonthMeanMaximumMinimum
20071961–199020071961–199020071961–1990
90%95%90%95%90%95%
June23.220.721.028.926.326.717.715.615.9
July25.821.822.032.327.427.719.716.716.9
August23.922.022.429.327.828.219.116.817.2
Summer24.321.722.230.227.427.918.816.717.1

In the warm period of 2007, the average total daily mortality was higher than in the reference period by one death per day (Table 1). The connection between anomalous mortality and daily temperature is shown in Figure 3. The monthly analysis, carried out with a different approach, is shown in Table 3. Similar results are observed when using fixed daily mortality values or 7-d smoothed values.

Figure 3.

Dynamic of the excess mortality (columns) and temperatures in Chisinau in 2007.

Table 3. Warm period excess mortality (death cases) in Chisinau in 2007, calculated using observed and 7-d smoothed data
Variants of comparisonMonthWarm period
AprilMayJuneJulyAugustSeptember
  1. a

    In this comparison, to have equal lengths of records, there were excluded three first (last) days of April (September) reference mortalities.

Observed mortality in 20075655384916085073973106
Reference mortality5365184814624754442916
Excess mortality29201014632−47190
7-d smoothed mortality in 20075045404876095083593007
7-d smoothed reference mortality4805164824644753932810
Excess mortality2424514533−34197
Fixed mortality of 2007a versus 7-d smoothed reference mortality2822914432−34201

In sum, in the warm season of 2007 the excess mortality in Chisinau, depending on the procedure of its calculation, reached between 190 and 200 deaths, or 6.5–6.9% of this period's reference mortality. Given the negative excess mortality in September, we can presuppose a certain mortality displacement that will be discussed below.

3.2. Temperature–mortality relationships above the ‘thresholds’

General temperature–mortality relationships in reference or ‘normal’ warm seasons are valid to certain ‘threshold’ temperatures, above which these relationships change. The TTs were identified as the centres of the lowest 2 °C class intervals of temperature variables, approximated by cubic splines, above which excess mortality rates began a sharp increase from their zero reference value (Figure 4). A comparison of the temperature variables' thresholds with their corresponding summer averages for the reference period (Table 4) shows that for Tmean and Tmax they exceed reference ‘norms’ by 3–4 °C, and for Tmin by 1–2 °C. This difference suggests a greater sensitivity of excess mortality to an increase in warm night temperatures.

Figure 4.

An example of the cubic spline of excess daily mortality grouped by 2 °C class intervals around the threshold value (mean air temperature).

Table 4. Poisson regression of excess mortality (EM) on temperatures above their thresholds' integer values
VariableStatisticsThe fitted model equation
RT (°C)TT (°C)R2P-valueError
ModelResidualMSEMAEME
  1. P-value for the model <0.05 indicates a statistically significant relationship between the variables at the 95.0% confidence level; P-value >0.05 for the residuals indicate that the model is not significantly worse than the best possible model for this data at the 95.0% or higher confidence level.

  2. RT, summer daily temperature of the reference period; TT, threshold temperature; MSE, mean standard; MAE, mean absolute; ME, mean errors.

Tmean21.625.9540.0030.7340.670.33−0.13EM = exp(−15.89 + 0.581 × Tmean)
Tmax27.131.8860.0070.6250.380.290.04EM = exp(−19.50 + 0.569 × Tmax)
Tmin16.819.9460.0000.6821.390.530.23EM = exp(−20.19 + 0.835 × Tmin)
ATmean21.424.9140.0010.9090.280.39−0.16EM = exp(−13.93 + 0.497 × ATmean)
ATmax26.330.9470.0000.8690.100.17−0.09EM = exp(−19.54 + 0.577 × ATmax)
ATmin16.117.9400.0050.9180.070.22−0.10EM = exp(−18.89 + 0.802 × ATmin)

The nonlinear Poisson regression analyses, performed on the data above TTs, produced suitable approximations of high temperature–excess mortality relationships (Table 4). Statistical significance and high levels of the explained variance (R2) of excess mortality allow estimating the possible excess deaths at air and ATs above their specific ‘tolerable-heat’ threshold values (Table 5). Here, the higher ends of heat exposure are close to the maximum temperatures observed in Chisinau in summer 2007. With mean daily summer mortality in the reference period equal to 16.0 deaths, mean excess mortality increases from 2.0 to 4.4% per 1 °C increase in temperature, depending on temperature variable chosen. These results are close to other European estimates (Kyselý and Huth, 2004; Kyselý and Kříž, 2008; Menne and Matthies, 2009).

Table 5. Excess mortality change with 1 °C temperature increase above threshold values. In bold there are shown excess deaths at temperature thresholds
 Excess death counts
T scaleTemperatureApparent temperature
MeanMaxMinMeanMaxMin
180.01
190.010.03
200.030.06
210.070.13
220.160.29
230.370.64
240.861.43
250.261.980.223.19
260.464.570.367.11
270.8210.50.615.9
281.460.99
292.611.62
304.662.660.11
318.340.164.380.19
3214.90.277.20.34
3326.70.4911.830.61
340.861.08
351.511.93
362.683.43
374.736.1
388.3510.9
3914.819.4
4026.2
Average related excess mortality (%)
4.23.823.34.42.9

3.3. Heat-episode analysis

3.3.1. Threshold temperatures versus percentiles

Relative thresholds, determined by percentiles of the selected variable's distribution, are usually used to determine potential thresholds for heat–health watch–warning systems. In Table 6, the absolute thresholds are compared with four percentiles of baseline and reference summer (June to August) climate. The 90th percentile is the indicator of an extreme event according to IPCC definition (IPCC, 2007), the 95th percentile is the delimiter of extreme events of low and high intensity in the EuroHeat classification (EuroHEAT, 2009), the 99th percentile characterizes the event of special rarity and severity (Schwartz (2005)), and the 97.5th percentile was selected as some interim percentile. As evidence of a warming climate, the thresholds identified for Chisinau were above the 99th percentile in the historical climate record. Assessing more recent years, the Tmean, Tmax and Tmin thresholds are close, respectively, to the 95–97.5th, 97.5th and 90th percentiles of their daily distribution in the reference period of 2000–2008 (excluding 2007). Thresholds for the ATs are somewhat lower.

Table 6. Threshold daily summer temperatures against different percentiles of their baseline (1961–1990) and reference (2000–2008) distributions
PercentilePeriodAir temperature (°C)Apparent temperature (°C)
MeanMaxMinMeanMaxMin
90.01961–199021.727.416.7
 2000–200823.929.719.024.329.419.0
95.01961–199022.227.917.1
 2000–200824.530.419.625.130.319.9
97.51961–199022.528.317.4
 2000–200825.131.120.225.931.020.6
99.01961–199022.928.717.8
 2000–200825.731.820.826.731.921.5
Threshold temperatures24.830.918.524.030.417.0

3.3.2. Heat episodes of 2007 and related excess mortality

We identified T1 and T2 in the heat wave two-level criterion (Section 2) in such a way as to cover the 90–99th percentile tails of Tmax distribution in the more recent (2000–2008) climate. On the one hand, T1 (the 99th percentile, ∼32 °C), being slightly above the Tmax threshold (Table 6), identifies heat days of exceptional intensity. On the other hand, T2 (the 90th percentile of Tmax, 29.7 °C), though being somewhat below the threshold value, allows the inclusion of days with temperatures that might be nevertheless associated with excess mortality. Heat days represent individual days that exceeded these thresholds, but for which the duration criteria for a heat wave was not met.

The heat waves during summer 2007 in Chisinau, identified by these criteria, are shown in Figure 5 and their characteristics in Table 7. On the whole, in May to September six heat waves and two individual heat days, with a total duration of 47 d, were observed. The longest and most intense heat wave, which lasted over the second-half of July, resulted in 116 excess deaths. This evidence supports the statement of the EuroHEAT Technical summary (Menne and Matthies, 2009) that heat waves with a long duration and high intensity have the highest impact on mortality.

Figure 5.

Heat-wave sequence in Chisinau in 2007.

Table 7. Heat wave (HW) and heat day (HD) characteristics
HW, HDDuration (d)PeriodAbs Tmax (°C)Excess deaths
HW1822.05–29.0534.22
HW2514.06–18.0632.73
HD1126.0634.7−6
HW3302.07–04.0733.35
HW4309.07–11.0733.111
HW51616.07–31.0739.4116
HD2112.0832.51
HW61017.08–26.0839.214
Total47146

On the whole, enumerated heat events caused 146 excess deaths or about 77% of the total number in the warm period of 2007. This percent is higher than the EuroHEAT results – from 7.6 to 33.6% (Menne and Matthies, 2009; D'Ippoliti et al., 2010) and differs from suggestions that mortality attributable directly to heat events is relatively small, and the majority of deaths occur at times other than during identified heat wave periods (WHO, 2006). We explain our results by the greater number, intensity and total duration of heat events in the Chisinau study of one very extreme summer of 2007, whereas EuroHEAT project included nine European cities; that project also showed a high heterogeneity of the effect between cities and populations.

3.3.3. Sensitivity analysis: lag effects and mortality displacement

Different approaches are used to examine possible lag effects. Basu et al. (2008) examined several lag times through an exploratory analysis and found that a single-day effect of same-day AT exposure (lag 0) had the best model fit in comparison with moving averages over the same day and previous 3 d (lag 0–3) or the previous 3 d (lag 1–3). Gasparrini et al. (2010) also showed a very strong and immediate effect of heat, but suggested a more delayed effect for extremely hot temperatures. Gosling et al. (2007) studied the possible influence of lag effects, calculating excess deaths per day for up to 12 d afterwards. Taking into account the continuity of the heat waves observed in 2007 in Chisinau, we decided to test a likely lag effect through a cross-correlation analysis where in a simple linear analysis the excess mortality and temperature were, correspondingly, the dependent and independent variables. The best fitted lag was estimated and the results are presented in expanded form in Table 8. In our results, the same-day correlation between temperature and mortality is not statistically significant. Rather, these relationships become stronger as time lag increases. Maximal effects are revealed on the second and third days (except in April) for Tmean and Tmax, but on the first and second day for Tmin. This result is a further evidence of the greater sensitivity of mortality to night temperatures.

Table 8. Lag effect in the dependency of excess deaths on air temperature in 2007
MonthLag (d)
01234
rprprprprp
  1. Bold values indicate the best fit of models.

  2. r, Pearson correlation coefficient; p, statistical significance of correlation.

Mean temperature
April−0.2270.229−0.3330.078−0.1720.381
May−0.2310.210−0.2240.223−0.3190.080−0.2900.113
June0.0240.901−0.0060.973−0.3100.096−0.4990.005−0.0770.686
July0.2960.1060.6280.0000.6530.0000.6860.0000.5250.002
August0.2700.1420.1120.5500.3450.0580.2530.170
September−0.1370.4670.0670.7380.1850.3300.1720.347
Maximum temperature
April−0.1260.508−0.1800.351−0.1830.351−0.0980.627
May−0.1080.564−0.2270.218−0.3020.099−0.3380.063−0.1990.284
June−0.1300.4940.1360.473−0.3000.107−0.5930.0010.1670.379
July0.3000.1010.6510.0000.7030.0000.6370.0000.5310.002
August−0.0890.6380.1990.2840.2690.1430.2110.255
September−0.1740.3580.1150.5430.1370.4700.2810.1320.0190.920
Minimum temperature
April−0.3420.064−0.2940.121−0.0920.642
May−0.2430.188−0.2550.167−0.3120.088−0.2310.212
June0.1400.460−0.2210.241−0.2540.156−0.2540.175
July0.3740.0380.6630.0000.6060.0000.6420.000
August0.2970.1040.1840.3220.3720.039−0.0120.950
September−0.0390.838−0.0530.7790.1540.416−0.1020.591

Heat waves create additional impacts on the human population, beyond those of the general temperature–mortality relationship (Hajat et al., 2006). In the first part of Chisinau's ‘normal’ warm season an inverse relationship between temperature and daily mortality is observed, lasting usually up to its optimal value in late July (Corobov and Opopol, 2010). In 2007, this relationship is also evident from the negative sign of r in April to June that is manifested with different lags. In other words, when temperature is below its optimal value one cannot speak about ‘excess death’ because of so-called added positive effects of higher temperatures on negative temperature–mortality dependencies. These results differ from other studies that have shown an increased impact of early-season heat waves (Sheridan and Kalkstein, 2010); but in this case it may be due to the lack of substantial heat events early in the 2007 warm season. Further research on the seasonal trends in heat impacts on a longer data set is warranted. The combination of two effects – the seasonal rise in temperature, which accompanies a decrease in mortality rates, and the occurrence of heat waves that cause an increase in mortality – makes the temperature–mortality relationship difficult to quantify. However, in July we observe a sharp ubiquitous transition to high positive correlations between temperature and excess mortality.

While the heat wave analysis provides a useful insight into the short-term human response to a heat event, it can overestimate the temperature effect due to mortality displacement, or ‘harvesting’. O'Neill and Ebi (2009) describe this phenomenon as an idea that some deaths associated with exposure to extreme temperature are those already expected among a frail subset of the population who would have died shortly thereafter anyway, regardless of the weather. Such deaths are only advanced by a few days, and hence do not represent a substantial shortening of a person's life. But although mortality displacement is often observed during heat waves, the real estimates of its contribution are not as common in the literature. At the same time, this concept is critically important for estimating the overall population-burden associated with exposure to extreme temperatures; the explicit evaluations whether the temperature effects represent ‘harvesting’ are also valuable for predicting the future impacts.

Mortality displacement is usually calculated as the relative percentage comparison of the mortality deficit (a number of ‘negative excess deaths’ after a heat wave, i.e. their quantity below the expected values) with that during heat waves (Le Tertre et al., 2006; Gosling et al., 2007; Kyselý and Kříž, 2008). Sometimes, such estimates are made for defined periods that include ‘before’, ‘during’ and ‘after’ the heat wave terms typically lasting less than 2 months (Gosling et al., 2009a). However, any simple algorithms are good only when dealing with an isolated heat wave. When we have a high-frequency series of heat events, as in Figure 5, the interference of their effects complicates this task. Here, the excess mortality was observed each month of the May to August period. Therefore, as a total ‘harvesting’ effect we decided to consider the ‘negative excess mortality’ in September (about 35–45 deaths) that, based on different methods of estimations (Table 2), amounts to about 17–25% of the total ‘positive’ excess deaths. For comparison, in the Czech Republic this effect was found to account on average for as much as 90% of the excess mortality during heat waves (Kyselý and Huth, 2004), though in one French study it varied from 1 to 30% (Le Tertre et al., 2006).

4. Conclusion

Although the human impacts of excessive heat has been well established in the literature, this study contributes to our knowledge by assessing a region – the northern Black Sea region – and an event – the summer of 2007 – that has been relatively understudied. In Moldova, we uncovered a significant correlation between excess human mortality and temperature variables, with similar results appearing across several temperature and AT variables. In 2007 at least, a shift in the temperature–mortality relationship was observed in mid-summer, coinciding with the onset of the most severe heat waves. Our results support the use of lagged data in heat-episode studies, as a lag in heat wave mortality effect depends on the peculiarity of the selected temperature variable, being minimal (1–2 d) for increases in minimum temperatures and more delayed (2–3 d) for increases in mean and maximum temperatures. This specificity, as well as the lower temperature threshold for minimum temperature, supports a higher sensitivity of excess human mortality to night temperatures.

Given the warming of Moldova's climate, it was shown that, defining the TTs for heat wave identification, one should take into account the observed trends and changes in regional climate, as the extreme temperature percentiles have increased. Otherwise, the parameters of possible heat waves can be underestimated. Further, such a city-specific analysis is important because of evident regional differences in the human acclimatization to heat and imperative for the proper development of a heat health watch warning system. Under the projected future increases in heat wave frequency, severity and duration, today's population tolerance to heat may become insufficient to cope with heat wave consequences for human health.

Acknowledgements

The research described in this publication was made possible in part by Award No. MOB-2928-CS-08 of the U.S. Civilian Research & Development Foundation for the Independent States of the Former Soviet Union (CRDF). Any opinions, finding and conclusions or recommendations expressed in this material are those of the authors and do not necessary reflect those of the CRDF.

Ancillary