3.1 Time series of the JJA maximum air temperature at three coastal sites along the Croatian Adriatic coast in the period 1961–2018
Current analysis shows that the daily maximum temperatures during the summer months (JJA; 1961–2018) exhibit significant differences in the maximum and air temperature range at the three locations (Fig. 2). Thresholds for the 95th and 99th percentile criteria are indicated in the figure.
Figure 2 indicates that there are different amplitude ranges in the daily maximum temperature among the stations but similar variability during the period. The maximum temperatures are well spatially correlated among these stations with the correlation coefficients between 0.77 and 0.88 (figure not shown). Greater maximum values appear to be in the second part of the period at all stations. An increasing trend of the daily maximum temperature can be visually inferred and will be tested in latter discussion.
Time series of BSSDMAT show that the southern location has lower air temperature values and smaller amplitude of variation compared to the other two sites, which could be due to cooling effects of the deep southern Adriatic water and blocking of inland effects on temperature by coastal mountains (Fig. 1).
3.2 Basic statistics of the BSSDMATs
Basic statistics of the BSSDMATs at these three weather stations (Table 2) generally follow the results from the climatologically average temperatures as mentioned in the previous text.
The results show that Rijeka and Split belong to a similar region of the maximum temperature regime with the mean of the maximum temperature greater in July than in August for both locations, while for Dubrovnik the maximum in July is lower than in August. The maximum temperature distributions approximately follow a normal distribution evidenced by small differences between the mean and median values (figures not shown). Although at a lower latitude, maximum temperatures at Dubrovnik are lower than in Rijeka and Split. Consequently, the standard deviation is also smallest for Dubrovnik. The mean maximum temperature in August for Dubrovnik is lower than the mean temperatures for June in Rijeka and Split. Although an isolated maximum of 40 °C is in Rijeka, the greatest mean of JJA maximum temperatures is in Split.
Table 2. Statistics of the daily maximum temperature (MxT) (°C) in June, July, and August, for the whole period 1961–2018 and separately for 1961–1989 and 1990–2018 for Rijeka, Split, and Dubrovnik.
MxT 1961–2018
|
Rijeka
|
Split
|
Dubrovnik
|
Mean MxT June
|
25.51
|
27.26
|
22.55
|
Mean MxT July
|
28.52
|
30.38
|
25.13
|
Mean MxT August
|
28.47
|
30.02
|
25.20
|
Mean MxT 1961–2018
|
27.52
|
29.24
|
24.31
|
Standard deviation
|
3.93
|
3.50
|
2.78
|
Abs. max MxT
|
40.0
|
38.5
|
33.4
|
Abs. min MxT
|
14.4
|
15.9
|
12.7
|
Med. MxT 1961–2018
|
27.6
|
29.5
|
24.5
|
MxT 61_89 vs.91_18
|
|
|
|
Mean MxT 1961–1989
|
26.37
|
28.53
|
23.57
|
Mean MxT 1990–2018
|
28.67
|
29.95
|
25.05
|
Diff. MxT (91_18-61_89)
|
2.30
|
1.42
|
1.48
|
Note that the average maximum temperature increased by 1.42-2.30 °C in the second part of the period (1990–2018) compared to the first period (1961–1989). It appears that the location in the northern Adriatic with shallow bathymetry has the most pronounced increase in the later period. The contrast in the maximum temperature distribution can be clearly seen when comparing data for the first (1961–1989) and the last (1990–2018) period (Fig. 3). Every calendar day in the first half of the period is compared with the same calendar day after 29 years – for example, 1 June 1961 is plotted against 1 June 1990, 2 June 1961 against 2 June 1990, etc.
An increase in the maximum temperature for all stations is clearly present in the second part of the period. The number of cases when the temperatures are greater in the second period (1990–2018) compared to the first period (1961–1989) is 2.1 times larger for Rijeka and Dubrovnik and 1.7 times larger for Split. Considering the first decade (1961–1970) compared to the last one (2009–2018), the ratios are even more significant: 2.8, 2.1, and 2.8 for Rijeka, Split, and Dubrovnik, respectively.
3.3 Very hot days (VHDs)
The analysis also includes a determination of Very Hot Days (VHDs) when a daily maximum temperature equals or exceeds 35 °C (Hoy et al. 2016). A distribution of VHDs by years is shown in Fig. 4.
There is a definite increase in frequency and peak temperatures of VHDs in the second part of the period after 1990 compared to the first part (113 vs. 2 for Rijeka and 145 vs. 42 for Split). This is significant evidence of broader regional warming and generally of global warming. Maximum daily temperatures in Dubrovnik never reached the threshold of 35 °C (Table 2).
Hoy et al. (2016) mentioned that there are only a few VHDs annually over the central Europe, but they are more frequent in southeast Europe. For example, they reported 18 VHDs in Vienna in 2015 which corresponds well with 17 and 15 VHDs in Rijeka and Split, respectively.
In addition, high temperatures and associated heat stress cause degradation of health and eventual increase in mortality rates. One of the parameters for probability estimation of increased mortality rate is the Physiologically Equivalent Temperature (PET) (Mayer and Höppe 1987; Zaninović and Matzarakis 2014). Zaninović and Matzarakis (2014) estimated that the PET temperature thresholds for Rijeka and Split are 36.5 and 36 °C, respectively. There were 71 such days (more than one per year on average) in Split and 32 days in Rijeka during 1961–2018.
3.4 Trends of mean annual summer temperatures 1961–2018
Since BSSDMAT trends can be visually inferred (Fig. 2), it is important to examine time series of average summer JJA maximum temperatures for actual trends. The time series is discontinued (since only JJA are considered for each year), so averages of each warm season were calculated and then tested for trends (Fig. 5). There is a definite increasing temperature trend during these 58 years at these three stations (Fig. 5). Since the seasonal maximum temperatures show distinct differences comparing 1961–1989 vs. 1990–2018 periods (see e.g., Figs. 3 and 4), separate trends were calculated for these two periods (Fig. 5).
The positive trend coefficients are all statistically significant at the 0.01 level and they are quite large for all three locations for the second part of the period, while there are small insignificant and slightly negative trends in the first part of the period (1961–1989). Average seasonal maximum temperatures for Rijeka are characterized by larger variability than for Dubrovnik and Split and the largest trend coefficient is 0.68 °C per decade (R2=0.55). The bathymetry is shallower in the northern Adriatic compared to the middle and southern regions and wind patterns exhibit pronounced variability (Orlić et al. 1994). Although average seasonal maximum temperatures for Dubrovnik are lower with a narrower amplitude, the trend coefficient (0.45 °C per decade; R2=0.50) is similar to the trend coefficient for Split (0.44 °C per decade; R2=0.41). This signifies that the effects of global and regional warming are increasing in time and consequences on the frequency and intensity of HWs will be further discussed in the later text (Section 5).
Since the air and sea are in the coupled climate system, these very high trend coefficients in the seasonal averages of the maximum air temperature suggest examining SST trends in this region and relating them to the maximum air temperature trends. Shaltout and Omsted (2014) estimate SST trends in the Mediterranean using the 0.25° AVHRR daily measurements for 1982–2012. On average, they obtained increases of 0.38 °C per decade for the Adriatic. Summer trends were somehow lower (0.30 °C per decade) compared to spring trends (0.48 °C per decade). Although there is a difference in the period considered, both the SST and BSSDMAT data show significant regional warming of the sea and air in the Adriatic.
3.5 Analysis of heat wave events
3.5.1 Autocorrelation analysis of BSSDMATs
While considering HWs, it is important to estimate both the length of each event and separation between consecutive events.
To assure statistical independence of consecutive HWs which can allow the use of probability density functions and modeling, most studies consider that HW events can be assumed to be independent if the separation between consecutive events is greater than some specified number of days, for example, five days or more (Curriero et al. 2002; Keellings and Wylen 2012).
However, in our study, an additional analysis of one- and two-day exceedances was also included to provide more insight into the general structure of the BSSDMAT extremes. Note that heat stress and eventual mortality rates might be significant in the first part of a HW (Zaninović and Matzarakis 2014) when the human body is still not well adapted to the new and severe heat stress conditions. To provide more insight into length of events, an autocorrelation analysis was conducted for all years and all locations. Autocorrelation function shapes can be useful for stochastic modeling of time series of climate variables (Pandžić 1984; Willks 2006).
The average autocorrelation coefficients as functions of a time lag (in days) are shown in Fig. 6.
Autocorrelation function properties are similar among all locations, especially for lags up to 4 days or so. They compare well with the autocorrelation function for the Markov process with r1=0.8 for temperature (Wilks 2006; Eq. 8.6). Note that the correlation coefficients are equal to about 0.5-0.3 for usual lags of 3-5 days, which is usually considered sufficient separation for taking HW events as independent and can be applied to a HW analysis.
To further test the behavior of the autocorrelation function, an analysis was conducted for all years and all locations (Fig. 7). It appears that the properties of the autocorrelation functions are quite complex, as shown by examining individual years.
In contrast to the average values, there is a large scatter of autocorrelation functions for each location and year. The spread is very large comparing individual years and locations. Even for one-day lag the correlation coefficient varies from 0.9 to 0.6 considering a spectrum of results for every year and every location. For the usual 5 days, the spread of the coefficients covers values from 0.7 to zero. There is some tendency of the autocorrelation coefficient in the first part of the period to drop at a faster rate compared to the second part of the period, but the variability of the coefficients in the second part of the period is quite large. A separation of 3-5 days reported in the literature can be taken as a general value, considering the usual periodicity of synoptic systems (which, of course, can be longer in the summer months and changing under global and regional warming). However, for detailed analysis, the autocorrelation functions could be considered separately for individual years.
3.5.2 95th and 99th percentile criteria for determining HWs
Two approaches for determining HW events were applied: a) a maximum daily temperature in excess of the 95th percentile will be estimated as a basis for determining heat waves; b) same as a) but for more severe conditions when the maximum daily temperature exceeds the 99th percentile.
95th percentile criterion – main characteristics
The determined number of HWs varies by location and threshold selected (Tables 3 and 4).
Table 3. Number of HWs for each duration in days (Dur) and their total duration in days for Rijeka (RI), Split (ST), and Dubrovnik (DU) determined by the 95th percentile criterion for JJA 1961–2018. No criterion on minimum separation between HWs was applied.
p95
|
RI
|
33.8°C
|
|
ST
|
34.5°C
|
|
DU
|
28.6°C
|
Cases
|
Days
|
Dur
|
HWs
|
Days
|
Dur
|
HWs
|
Days
|
Dur
|
HWs
|
Days
|
total
|
total
|
0
|
|
5077
|
0
|
|
5091
|
0
|
|
5080
|
|
15248
|
1
|
75
|
75
|
1
|
75
|
75
|
1
|
103
|
103
|
253
|
253
|
2
|
14
|
28
|
2
|
24
|
48
|
2
|
16
|
32
|
54
|
108
|
3
|
11
|
33
|
3
|
10
|
30
|
3
|
13
|
39
|
34
|
102
|
4
|
8
|
32
|
4
|
8
|
32
|
4
|
7
|
28
|
23
|
92
|
5
|
4
|
20
|
5
|
3
|
15
|
5
|
3
|
15
|
10
|
50
|
6
|
5
|
30
|
6
|
2
|
12
|
6
|
1
|
6
|
8
|
48
|
7
|
1
|
7
|
7
|
2
|
14
|
7
|
3
|
21
|
6
|
42
|
8
|
2
|
16
|
8
|
1
|
8
|
8
|
0
|
0
|
3
|
24
|
9
|
2
|
18
|
9
|
0
|
0
|
9
|
0
|
0
|
2
|
18
|
10
|
0
|
0
|
10
|
0
|
0
|
10
|
0
|
0
|
0
|
0
|
11
|
0
|
0
|
11
|
1
|
11
|
11
|
0
|
0
|
1
|
11
|
12
|
0
|
0
|
12
|
0
|
0
|
12
|
1
|
12
|
1
|
12
|
|
122
|
259
|
|
126
|
245
|
|
147
|
256
|
395
|
760
|
In 53 of the 58 years, HWs occurred at least at one location and in 30 years they were shown at all locations in the same year. There were a total of 4.7% of HW days in the whole period. The number of total HWs among the locations ranged from 122 to 147 with 245-259 total event days. Note that for the period of 58 years, the number of cases and especially the total number of days with HWs are similar for all locations indicating regional-scale characteristics. The maximum duration of determined HWs were 9, 11, and 12 days for Rijeka, Split, and Dubrovnik, respectively. The most frequent one-day exceedances are for Dubrovnik, possibly due to a narrow range of values where small changes in the temperature could become greater than the threshold and vice versa. Most likely are event durations of 3-4 days. Events of more than 7 days duration are quite rare.
99th percentile criterion – main characteristics
Table 4 shows the basic statistics of the HWs according to the 99th percentile criterion. The maximum duration length for all locations was 5 days.
Table 4. Number of heat waves (HWs) for each duration (Dur) and their total duration in days (Days) for Rijeka (RI), Split (ST), and Dubrovnik (DU) determined by the 99th percentile criterion for JJA 1961–2018. No restriction on minimum separation between the HWs was applied (as discussed in the beginning of Section 4 – autocorrelation analysis).
p99
|
RI
|
35.8°C
|
|
ST
|
36.1°C
|
|
DU
|
30.1°C
|
Cases
|
Days
|
Dur
|
Cases
|
Days
|
Dur
|
Cases
|
Days
|
Dur
|
Cases
|
Days
|
total
|
total
|
0
|
|
5282
|
0
|
|
5284
|
0
|
|
5285
|
|
15851
|
1
|
17
|
17
|
1
|
19
|
19
|
1
|
27
|
27
|
63
|
63
|
2
|
7
|
14
|
2
|
10
|
20
|
2
|
5
|
10
|
22
|
44
|
3
|
3
|
9
|
3
|
1
|
3
|
3
|
3
|
9
|
7
|
21
|
4
|
1
|
4
|
4
|
0
|
0
|
4
|
0
|
0
|
1
|
4
|
5
|
2
|
10
|
5
|
2
|
10
|
5
|
1
|
5
|
5
|
25
|
|
30
|
54
|
|
32
|
52
|
|
36
|
51
|
98
|
157
|
The 99th percentile limit indicates a total of »1% extreme conditions (days) within the whole period. Note that this stricter criterion provides numbers of events and associated days among the locations similar to the 95th percentile case. Consequently, the results for the 99th percentile criterion indicates similar uniform regional characteristics of HWs as for the 95th percentile criterion over the coastal Adriatic.
3.5.3 Simultaneous occurrence of HWs on an annual basis
Heat waves at at least one location were determined in 53 out of 58 years considering the 95th percentile criterion (Fig. 8). Annual variations in terms of an index (yes/no) of HW occurrence in each year are shown in Fig. 8. Indices 0, 1, 2, and 3 represent numbers indicating how many simultaneous locations had HWs present for a particular year. For example, index 3 means that HWs occurred for all three locations in a particular year, while index 2 means that HWs were determined for two locations.
Regarding the 95th percentile criterion, the extreme temperatures and HWs occurred in 34, 45, and 44 out of 58 years for Rijeka, Split, and Dubrovnik, respectively. Considering all stations, there were only 5 years (within the period 1976–1997) in which there were no HWs. In 30 years (24 years in and after 1990), HW occurred at all three locations. After 2000, for almost all years (except 2014) HWs occurred at all locations.
Regarding the 99th percentile criterion, HWs were predominantly absent before 1990. The extreme temperatures and HWs occurred in 17, 15, and 19 out of 58 years for Rijeka, Split, and Dubrovnik, respectively. In 31 years, there were no HWs at any station, while in 9 years there were HWs at all locations. Two or more events per year occur only on and after 1998. All these results further confirm previous conclusions that the second part of the period is characterized by much larger occurrence of HWs indicating the effects of global and regional warming processes.
3.5.4 Frequency and duration
The total number of HW events and duration of each event per year in the period 1961–2018 for Rijeka, Split, and Dubrovnik using the 95th and 99th percentile methods is shown in Fig. 9.
The total number of HW days significantly increased over time (Fig. 9, right panels). For both parameters and both criteria, the values generally fall below the 1:1 line (higher values correspond to later years). For later years the intervals between the highest and lowest values (frequency; total event days) are increasing, i.e., higher values are more likely in the later years. The total number of HW days significantly increased in time. This is especially pronounced for Rijeka and Split and to a lesser extent for Dubrovnik. Only after 1989 do HW lengths of 8 days or more (3 or more) occur for the 95th (99th) percentile criteria. The annual frequency of HW days follows increasing BSSDMAT trends over time as shown in Figs. 2 and 5. Figure 9 shows a significant increase in events per year and corresponding duration days after 1989. More than 10 (4) days annual duration of HWs for the 95th (99th) percentile criterion, respectively, occurred in and after 1994.
3.5.5 Heat wave intensity – Heat-wave index (HWI)
Heat wave intensity can be examined in terms of ratios between the peak vs. threshold temperatures (95th and 99th percentile criteria) during the length of a HW. Distributions of peak maximum temperatures for each event show distinct differences in amplitudes between the northern Adriatic with shallow bathymetry and the southern region with the cooler influence of the deeper Adriatic Sea (Fig. 10).
The peak temperatures within HWs increase over time with larger amplitudes compared to the average values. There is clear clustering, with increasing values after the 1990s. Note that the extreme values occurred at all locations and for both criteria in the last decade (2009-2018). The ratios between the peak and threshold temperatures can be assumed as a heat-wave index (HWI) for both criteria (Fig. 11). Besides an increasing number of events in time, the indices are increasing with greater variability in time. Note that the majority of greater indices are in the second part of the period.
In accordance with the previous results, this figure also confirms that a significant number of HW events occurred in the second part of the period and that the HWI becomes greater in time (7 to 18% in all cases using both thresholds). The greatest HWI values are in the last decades of the period. The maximum HWIs with respect to the threshold temperatures using the 95th (99th) percentiles are 1.18 (1.12) for Rijeka, 1.12 (1.07) for Split, and 1.17 (1.11) for Dubrovnik. The smaller ratios of the maximum compared to the threshold temperature for the 99th criterion are caused by the higher threshold and smaller range of the BSSDMAT values.
Note that the ratio between the peak and threshold temperatures can be used as an intensity index (HWI) for future comparison studies for other locations and other times.