3.1 HW indices and trends
The results concerning HW indices and their time trends are shown in Figs. 3 to 8. Figures 3 and 4 show the time evolution of the four HW indices, for TX and TN. Trends and their statistical significances for TX, computed for 30-year moving window (30-YMW) subsets, are represented in Figs. 5 and 6 respectively, and those corresponding to TN are shown in Figs. 7 and 8.
Number of events per season, HWN
The time evolution of HWN_TX, (Fig. 3a), shows that, up to 1980, only in 1928 there is an outstanding value of four events, while since 1980 there are 13 years with a number between 4 and 7 events per season, 10 of these years corresponding to the period 2000–2020. In fact, since 2000, every year has suffered at least one heatwave episode. The HWN_TN index (Fig. 4a) depicts an outstanding value of 7 events in year 2003, and since 1985, all years have at least one heatwave episode.
The trends for different periods are summarised in Table 1. All trends are positive and significant at the 5% level. For the whole recording period (1914–2020), the HWN_TX trend is positive and equal to + 0.29 events/decade. Nevertheless, the maximum trend is reached for the last period 1990–2020, with + 1.38 heatwaves per decade. In agreement with this trend, the HWN increment would be of 11 episodes per season in 2100. If the reference is the trend of the whole period 1914–2020, the increment would be lower, 2.5 events per season. For TN, the trends obtained for the different periods are significant, as for the maximum temperature, though with lower values. The maximum trend is obtained again for the period 1990–2020, with a value of + 0.64 events/decade, being likely an increment of 5 HWs per season in 2100. Trends computed for 30-YMW subsets (Figs. 5a and 6a), since year 2000, are systematically greater than + 0.5 events/decade and statistically significant. Another period with significant positive trends is around 1970–1980. Instead, significant negative trends are obtained for 1940–1960. For HWN_TN, maximum positive significant trends are reached in 1990s (Figs. 7a and 8a).
Table 1
Time trends obtained for the four HW indices, for TX and TN and different time periods.
|
HWN
(ev./dec)
|
HWA
(oC/dec)
|
HWD (days/dec)
|
HWF
(days/dec)
|
1914–2020
|
TX90
|
+ 0.29
|
+ 0.28
|
+ 0.50
|
+ 1.48
|
TN90
|
+ 0.21
|
+ 0.19
|
+ 0.35
|
+ 1.08
|
1950–2020
|
TX90
|
+ 0.62
|
+ 0.71
|
+ 1.14
|
+ 3.05
|
TN90
|
+ 0.39
|
+ 0.47
|
+ 0.89
|
+ 2.22
|
1960–2020
|
TX90
|
+ 0.73
|
+ 0.77
|
+ 1.28
|
+ 3.59
|
TN90
|
+ 0.45
|
+ 0.50
|
+ 0.96
|
+ 2.64
|
1970–2020
|
TX90
|
+ 0.82
|
+ 0.77
|
+ 1.37
|
+ 4.00
|
TN90
|
+ 0.63
|
+ 0.77
|
+ 1.34
|
+ 3.52
|
1980–2020
|
TX90
|
+ 0.78
|
+ 0.43
|
+ 1.13
|
+ 3.81
|
TN90
|
+ 0.61
|
+ 0.70
|
+ 1.30
|
+ 3.70
|
1990–2020
|
TX90
|
+ 1.38
|
+ 1.02
|
+ 1.52
|
+ 6.17
|
TN90
|
+ 0.64
|
+ 0.69
|
+ 1.18
|
+ 4.01
|
Amplitude of the hottest event per season, HWA
The maximum amplitudes for TX (Fig. 3b), are close to 7oC, mainly concentrated in the last years. The maximum of 8.7oC should be carefully considered, as it would be a consequence of a forest fire occurred very close to Barcelona city in 1982. For TN (Fig. 4b), there are several amplitudes around 6oC spread out along the whole period and a notable maximum of 7.5oC (year 1931) and a second maximum of 6.6oC (year 1923). Significant positive trends from + 0.28 to + 1.02oC/decade (for TX) and from + 0.19 to + 0.77oC/decade (for TN) are obtained for the different analysed periods (Table 1). While for HWA_TX the maximum trend corresponds to the last period (1990–2020), as observed for HWN index, the maximum trend for HWA_TN is detected in the period 1970–2020. Amplitude increments varying from 1.6 to 8.2oC, depending on the period trend considered, could be expected in 2100. Consequently, the average increment would be close to 5oC with respect to the present values. As an example, taken as a reference the maximum temperature of 38.4oC reached in 2003 (Table 2), it could be likely to reach maximum temperatures around 40oC for the near future (2050) and of 44oC at the end of the 21st century in Barcelona. For the 30-YMW subsets (Figs. 5b and 6b), the maximum HWA_TX trend is reached in the 1970s with significant positive trends around + 1.5oC/decade. Since 2000s up to nowadays, the trends are significantly positive again. Instead, the maximum trends for HWA_TN are reached 20 years later, in 1990s, around + 1.0oC/decade (Figs. 7b and 8b).
Length of the longest event per season, HWD
The maximum values reached by HWD index are 15 and 14 days in 2003 for TX and TN, respectively (Figs. 3c and 4c). Others years reached values greater or equal 10 days. For TX,
values of 10–12 days are reached in years 1947, 1982, 1987, 2005 and 2009. For TN, values of 11–12 days occur in 1923, 1947 and 2006. Table 1 shows significant positive trends for all the periods considered, with values from + 0.50 to + 1.52 days/decade for HWD_TX and from + 0.35 to + 1.34 days/decade for HWD_TN. It would imply that increases from 3 to 12 days in the maximum HW duration could be expected in 2100. By comparing these trends with those obtained by Perkins-Kirkpatrick and Lewis (2020) for the Mediterranean region (+ 0.61 days/decade) in the period 1950–2020, it is observed that the trends derived for Barcelona are slightly greater (+ 0.89 days/decade for TN and + 1.14 days/decade for TX). When the 30-YMW trends are analysed (Figs. 5c, 6c, 7c and 8c), clear decreasing trends until 1960 and an abrupt change to higher positive trends since 1960 until the end of the period, are observed. This behaviour is detected in both temperatures TX and TN, but with a higher sharp increase for TX. For HWD_TX, maximum trends are of 1.5 days/decade for the 1970–1980 years and since year 2000. For HWD_TN maximum trends are slightly higher, 2.0 days/decade, during 1990s.
Number of heatwave days per season, HWF
The year 2003 is the most outstanding, with 47 and 55 days heatwave days for TX and TN, respectively (Figs. 3d and 4d). Table 1 shows that trends for the different studied periods range from + 1.5 to + 6.2 days/decade for TX and from + 1.1 to + 4.0 days/decade for TN, again with greater values for TX. Consequently, an extreme increment of 48 days per season for HWF_TX and of 32 days for HWF_TN in 2100 would be possible. The HWF-trend obtained for the Mediterranean region by Perkins-Kirkpatrick and Lewis (2020) was + 2.61 days/decade for the period 1950–2017, in agreement with the values summarised in Table 1 (+ 3.05 days/decade for TX and + 2.22 days/decade for TN) for an almost coincident period. (1950–2020). Whit respect to the 30-YMW trends (Figs. 5d and 7d), a maximum of 6 days/decade is obtained for both TX and TN in the mid 2000s and 1990s, respectively. The Mann-Kendall test coefficients (Figs. 6d and 8d) are significant for the periods 1965–1982 and 1995–2005 for TX, and for 1985–1997 and 2002–2005 for TN. These different time periods detected for TX and TN manifest the discrepancies between the time behaviour of maximum and minimum temperatures.
Table 2
Date, duration and maximum temperature of the HWTX and HWTN extreme events (exceeding 98th percentiles of calendar day TX and/or TN).
|
HWTX98
|
HWTN98
|
|
Date
(first day)
|
D
(days)
|
TXmax
(oC)
|
Date
(first day)
|
D
(days)
|
TNmax
(oC)
|
1
|
|
|
|
02/09/1930
|
3
|
22.8
|
2
|
11/06/1931
|
3
|
35.3
|
10/06/1931
|
4
|
26.6
|
3
|
|
|
|
30/08/1944
|
4
|
24.0
|
4
|
|
|
|
23/07/1947
|
6
|
26.8
|
5
|
12/06/1981
|
5
|
35.1
|
|
|
|
6
|
05/07/1982
|
5
|
39.8
|
|
|
|
7
|
|
|
|
13/08/1987
|
3
|
27.2
|
8
|
16/09/1987
|
5
|
32.4
|
16/09/1987
|
4
|
23.6
|
9
|
|
|
|
30/07/2001
|
3
|
24.8
|
10
|
11/06/2003
|
5
|
34.9
|
11/06/2003
|
5
|
25.2
|
11
|
18/06/2003
|
5
|
35.4
|
19/06/2003
|
3
|
24.0
|
12
|
|
|
|
10/07/2003
|
3
|
24.4
|
13
|
02/08/2003
|
13
|
38.4
|
03/08/2003
|
11
|
27.8
|
14
|
|
|
|
18/06/2005
|
3
|
23.7
|
15
|
|
|
|
10/07/2006
|
3
|
24.8
|
16
|
30/06/2009
|
3
|
34.8
|
|
|
|
17
|
16/08/2009
|
4
|
37.7
|
|
|
|
18
|
12/09/2011
|
4
|
30.4
|
|
|
|
19
|
18/08/2012
|
5
|
35.2
|
19/08/2012
|
4
|
26.3
|
20
|
03/07/2015
|
3
|
35.0
|
04/07/2015
|
3
|
26.0
|
21
|
03/09/2016
|
3
|
32.4
|
03/09/2016
|
3
|
23.4
|
22
|
|
|
|
10/06/2017
|
6
|
23.7
|
23
|
02/08/2018
|
3
|
37.2
|
01/08/2018
|
4
|
26.8
|
24
|
26/06/2019
|
3
|
37.7
|
26/06/2019
|
4
|
27.5
|
25
|
|
|
|
14/09/2020
|
3
|
22.7
|
3.2 Extreme HWs and hotspots
Table 2 shows date, duration and maximum, TX or TN, for the extreme heatwave episodes occurred all along the period considered (1914–2020). During these extreme events, in three or more consecutive days, TX and/or TN exceeded the corresponding calendar-day 98th percentiles. Most of these extreme episodes occurred after the year 1980 (21 out of 25) and especially since 2000 (17 out of 25). It is outstanding the extremely severe episode of august 2003, with maximum TX of 38.4oC and maximum TN of 27.8oC, and durations of 13 and 11 days respectively. A second outstanding episode occurred in June 2019 with extreme TX and TN of 37.7 oC and 27.5 oC, respectively. This second episode was shorter than the first (3 and 4 days), but also very remarkable since it occurred in June, when is rather unusual to suffer extreme heatwaves with so higher extreme TX and TN.
It is also worth noting that 10 out of the 25 HW events along the period considered are compound TX and TN (daytime-nighttime) heatwaves. A majority of these compound episodes (8 out of 10) have occurred since 2000, three of them in 2003 and the remaining five since 2012. An eventual increase in the frequency of occurrence of these daytime-nighttime HW events could be a matter of concern, given that the exposure of population to daytime hot conditions with little or no relief from nighttime cooling may result in increased heat-stress related hazards to human health (Meehl and Tebaldi, 2004; Li et al., 2017; Wang et al., 2020).
Figures 9 and 10 depict the spatial distribution of the land surface temperature, LST, obtained from the Modis satellite for day (LSTd) and night (LSTn), for six extreme HW episodes. The differences between the spatial distribution of LSTd and LSTn are evident. While for diurnal maps the greater LST values are reached in the inner valleys and basins (Llobregat, Vallès and Penedès), for LSTn the higher values are attained mainly in Barcelona city, notably suggesting the evidences of the UHI effect. The exception is the episode of June 28th 2019, when the LSTn highest values spread all over both coastal and inner locations. The LSTd of this outstanding episode, represented in Fig. 9, depicts a pattern similar to the rest of the LSTd maps, but showing the highest maximum LST values, greater than 50oC, in some locations of the Vallès valley. Even though the LST values are not equal than the air temperatures at 2-metres height, these maps are a first insight into the location of the hotspots of the BMR.
Figure 11a shows the spatial distribution of TX and TN recorded on July 5 and July 29, 2015, using the temperature data of 48 thermometric stations in the BMR (Serra et al., 2020). For July 5th, pertaining to the extreme HW of 2th -7th July, the two hotspots just observed in the LST maps, are again noticed, with a smoothed resolution, the highest TX observed in inner locations and the highest TN in Barcelona city. Instead, for an example out of the HW episode, July 29th, the maximum values are more spread out for both TX and TN.
A Principal Component Analysis, PCA, is applied to TN recorded in July and August 2015. Figure 11b depicts the spatial distribution of the factor scores corresponding to the two rotated principal components, RPCs, obtained. The maximum values of factor scores for July are located at Barcelona for both RPCs, but the UHI effect is clearly manifested in RCP2, with an explained variance of 28%. The RPC2 is highly correlated with the days pertaining to the HW of the first days of July 2015. For August, the maximum factor scores are detected along the littoral fringe for RPC1 with an explained variance of 54.1%. The RPC2 factor scores for August show higher values (40.5% of variance) in the inner regions of the BMR.
3.3 PCA of HW indices
Table 3 shows the correlations among the four indices HWN, HWD, HWA and HWF for TX and TN. High correlations are obtained among the TX indices (0.762–0.926), with the highest values corresponding to the correlation between the number of episodes per season, HWN, and the total number of heatwave days per season, HWF. The correlations among the TN indices are slightly lower (0.648–0.927). When the correlations among TX and TN indices are computed, the coefficients are lower, with values between 0.49 and 0.77. The possible redundancies in the information provided by the four indices are confirmed by the values of the factor loadings for the RPCs shown in Table 4. Applying the eigenvalue equal to 1.0 criterion, two principal components are retained, thus explaining a total variance of 85.7%. While the first rotated principal component, RPC1, is highly correlated with the TN indices, the second, RPC2, is notably related to the TX indices, and each one of the RPCs explain almost the same variance, 43.3% and 42.4%, respectively. Although the HW indices considered in this research are those proposed by Fisher and Schär (2010), after applying the RPCA, these eight indices can be replaced by two rotated principal components, one for TX and the other for TN. The time evolution of the RPC1 and RPC2 factor loadings is illustrated in Fig. 12. It is remarkable the increasing values since 1980s, with an outstanding maximum in 2003 for RPC1 and in 1982 for RPC2. Two secondary maxima are reached in 2009 and 2013 for RPC2.
3.4 HW and WeMO
With the aim of investigating the possible relationship between the WeMO index and the occurrence of HW episodes, Fig. 13a shows the time evolution of the average summer WeMO index from 1914 to 2020. It is outstanding the negative trend during all the period. The evolution of the annual WeMO analysed by the Climatology Research Group of the University of Barcelona (http://www.ub.edu/gc/wemo/) for the period 1821–2020 reflects a clear positive trend from 1821 to 1910 approximately, and a negative trend from 1910 until 2020. Then, the period studied in this paper is in the phase of decreasing values of WeMOi. A first step has consisted on the study of the correlation between WeMOi and daily TX and TN, which leads to non-conclusive results, given that small correlation coefficients have been obtained. Nevertheless, some non-negligible negative correlations have been obtained between annual summer WeMOi and HW indices (Table 5). Figure 13b illustrates the summer WeMOi, SWeMOi, histogram, where the highest frequency corresponds to values between − 0.5 and + 0.5. However, Fig. 13a shows that, from 1980, the indices are predominantly negative in concordance with the increasing frequency and intensity of HWs from 1980s. The histograms of WeMOi values corresponding to the first day of the studied HWs are represented in Figs. 13c and 13e. More than 80% of the values are negative and slightly lower for TX than for TN. When the last-day WeMOi of a HW episode is represented (Figs. 13d and 13f), the histograms show higher values. This fact suggests that some signs of a forthcoming HW would start with lower WeMOi values and would finish with higher WeMOi values. Figure 14 depicts two examples of the WeMOi evolution for summer 2003 and 2015. In both cases the HW episodes are coincident with a beginning of WeMOi low values, then increasing along the HW. The synoptic situation associated with the summer HWs (Figs. 15 and 16) usually shows a low pressure in the southwest of Spain and high pressures in central Europe, implying an advection from the south along the Spanish Mediterranean coast. This low-pressure system shifts towards the Iberian Peninsula centre and remains some days there. At 500 hPa, a ridge comes from Africa and extends towards Western Europe.