2.1 Study region
We define Eastern Spain as the area comprising the Mediterranean coast of Spain between 36° – 41°N and 4.5°W – 4.5°E (Fig. 1), characterized by a complex topography: i.e., coastal plains, deltas, river mouths and valleys, flanked by inland mountain ranges such as the Betic, Prebetic and Iberian systems. The study area covers 82,165 km2 of territory and 3,017 km of coastline (from Malaga to Barcelona, including the Balearic Islands), and it is located in the Western Mediterranean basin. The climate is typically Mediterranean with mild winters and dry hot summers, modulated by the cooling effect of the SB, which represent the most dominant wind circulation at summertime (Olcina-Cantos and Azorin-Molina 2004). Because stable atmospheric conditions (i.e., high insolation and weak surface pressure gradient) are frequent throughout the year, SB can develop even during the cold semester (November-April), occurring in 2/3 of the days in the year (Azorin-Molina and Martín-Vide 2007). In addition, this local wind circulation can penetrate ~ 150 km inland (Kottmeier et al. 2000) through river valleys (Simpson et al. 1977; Arritt 1993), being responsible for most convection (Azorin-Molina et al. 2009) and summer precipitation (Azorin-Molina et al. 2014b) in mountainous areas inland.
We split the Eastern Spain into five regions according to the orientation of the coastline, which strongly influences SB dominant direction (Azorin-Molina and Lopez-Bustins 2008; see Fig. 1): (i) the Southern coast (coastline oriented from W to E) on the Betic system, with the highest mountain range peak (Mulhacén, 3,478.6 m a.s.l); (ii) the South-eastern coast (SW to NE) on the Prebetic system and irregular coastal relief (mostly headlands and cliffs); (iii) the Eastern coast (S to N) with the Iberian mountain range at west, large coastal plains at the coast, and the most important delta of the region: Ebro delta; (iv) the North-eastern coast (SW to NE) on the Catalan coastal range; and (v) the Balearic Islands (W to E), a group of islands with quite complex orography (altitudes reach 1,445 m a.s.l in the Serra de Tramuntana mountain range).
(Fig. 1 about here)
2.2 Observed wind speed data
Our dataset, including wind speed observations (daily mean and daily maximum wind gusts; in m s− 1) from 16 stations located over the Eastern Spain (see Fig. 1) for 1961–2019 represents the longest climatological SB study in the region. The Spanish Meteorological Agency (AEMET) provided the wind speed data, measured at standard 10 m height with different types of anemometers as detailed in Azorin-Molina et al. (2014a). We computed the daily averages from sub-daily observations at 0700, 1300, and 1800 UTC, discarding the nighttime one (i.e., 0000 UTC) where the breeze flow reverses, giving rise to the land-breeze. We selected those existing meteorological stations where the distance to the coastline does not exceed 20 km (given in Table 1), because the inland penetration of SB could be reduced in wintertime or inhibited by offshore synoptic flows (Salvador and Millán 2003; Azorin-Molina and Chen 2009; Azorin-Molina et al. 2011a). Five of the selected stations are sited in urban environments (see Table 1), while the rest are located in well exposed sites at airports, reducing the frequency of potential inhomogeneities due to surrounding changes in urbanization (Azorin-Molina et al. 2014a).
Table 1
Description of the meteorological stations with homogeneous wind speed measurements over the Eastern Spain (for locations see Fig. 1) for 1961–2019. In bold are shown the relocated stations used in this study. Weather stations sited in urban environments are marked with an *.
Nº
|
Id
|
Station
|
Lat.
(Decimal º)
|
Long.
(Decimal º)
|
Elevat. a.s.l. (m.)
|
Dist. Coast (km)
|
Orient.
(Decimal º)
|
Period
|
1
|
6155A
|
Málaga airport *
|
36.67
|
-4.48
|
5
|
1.9
|
255–75
|
1961–2019
|
2
|
6325O
|
Almería airport
|
36.85
|
-2.36
|
21
|
1
|
255–75
|
1968–2019
|
3
|
7031
|
San Javier airport
|
37.79
|
-0.80
|
4
|
4.5
|
223 − 43
|
1961–2019
|
4
|
8019
|
Alicante airport
|
38.28
|
-0.57
|
43
|
4.4
|
223 − 43
|
1961–2019
|
5
|
8025
|
Alicante Ciudad Jardín *
|
38.37
|
-0.49
|
81
|
2.8
|
223 − 43
|
1961–2019
|
6
|
8414A
|
Valencia airport
|
39.48
|
-0.47
|
56
|
12.7
|
193 − 13
|
1961–2019
|
7
|
8416
|
Valencia Viveros *
|
39.48
|
-0.36
|
11
|
3.2
|
193 − 13
|
1961–2019
|
8
|
8500A
|
Castellón-Almassora *
|
39.96
|
-0.07
|
43
|
6.7
|
193 − 13
|
1976–2019
|
9
|
9981A
|
Tortosa
|
40.82
|
0.49
|
50
|
18
|
193 − 13
|
1961–2019
|
10
|
0016A
|
Reus airport
|
41.15
|
1.16
|
71
|
7.5
|
229 − 49
|
1961–2019
|
11
|
0076
|
Barcelona airport
|
41.29
|
2.07
|
4
|
1.5
|
229 − 49
|
1961–2019
|
12
|
0200E
|
Barcelona Fabra
|
41.42
|
2.12
|
408
|
8
|
229 − 49
|
1961–2019
|
13
|
B893
|
Menorca airport
|
39.85
|
4.22
|
91
|
3
|
270 − 90
|
1965–2019
|
14
|
B278
|
Palma Mallorca airport
|
39.56
|
2.74
|
8
|
3.7
|
270 − 90
|
1972–2019
|
15
|
B228
|
Palma Mallorca port*
|
39.56
|
2.63
|
3
|
0
|
270 − 90
|
1978–2019
|
16
|
B954
|
Ibiza airport
|
38.88
|
1.38
|
6
|
2.8
|
270 − 90
|
1961–2019
|
2.3 Homogenization of wind speed data
Wind speed series were created by applying a robust quality control and homogenization protocol, following the method used by Azorin-Molina et al. (2019). Reconstructions of wind speed series were made by checking for discontinuities, due to the installation of automatic weather stations (after the 1980s) and relocation (see Table 1). All stations were homogenized to avoid disturbances in climate trends due to non-climatic break-points in wind series. Although Azorin-Molina et al. (2014a) suggest the use of some regional climate models (i.e., MM5) as reference series for homogenization due to their reliability to reproduce mesoscale circulations (such as SB), we rather used wind speed data from the National Center for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis (https://psl.noaa.gov/data/gridded/data.ncep.reanalysis.html; last accessed February 24, 2022), as it covers the entire study period. Moreover, high correlated (r = 0.4, p < 0.05; up to 25 km) neighboring stations were also used to detect inhomogeneities, according to Azorin-Molina et al. (2014a). We applied the CLIMATOL package (http://www.climatol.eu/, last accessed February 24, 2022), which implements a variation of the Alexandersson’s Standard Normal Homogeneity Test (SNHT; Alexandersson 1986) in all 16 stations on a daily basis, which represent one of the most used tests to detect inhomogeneities in climate series. In summary, the procedure did not detect any clearly outlier value, but corrected 68 and 76 break-points with SNHT > 25 for SB speeds and gusts respectively.
2.4 Identification and validation of sea breeze days
SB selection algorithms consist in detecting past events from well-defined criteria, such as certain thresholds for wind or thermal contrast between land-sea air (ΔT), among many others. However, there is no universal method to accurately detect SB days. Instead, SB identification depends mostly on the criteria used in each study site, physical complexity of the area and data availability (see Table 1 in Azorin-Molina et al. 2011b). In general, there is a global lack of historical observed data for many essential variables to automatically detect SB events from long-term climate series. Most historical observed data presents (i) large gaps; (ii) low quality; and (iii) limited temporal resolution to create with reliability a historical and high-quality SB database. Even when procedures such as quality control, missing data infilling, reconstruction and homogenization can be applied to entire datasets (i.e., essential climate variables for SB detection) on a sub-daily basis (Guijarro 2018; Curci et al. 2021), this implies huge computational cost and effort. To address the need to improve the processing time and accuracy in SB selection methods for long-term studies, we propose a robust automated method to identify potential SB days using alternative variables (e.g., Laird et al. 2001; Azorin-Molina and Lopez-Bustins 2008), which can be used worldwide at any coastal region with few adjustments.
The automated selection algorithm presented here is based on six alternative filters, classified in (a) synoptic and (b) local-scale conditions (see Fig. 2). The first main group (filters 1 to 3) employs the synoptic flow regime at 850 hPa (in terms of regional teleconnection patterns, synoptic weather types and geostrophic winds) as alternative criteria to reject days with large-scale synoptic disturbances for developing SB. The second group (filters 4 to 6) uses objective criteria in local weather to confirm SB passage (by onshore winds and weak wind gusts), and to ensure pure SB events (days without precipitation). After running the data through these filters, we identified 9,312 potential SB days at the regional level across Eastern Spain (Fig. 3) and between 6,741 and 8,315 SB days at the coastal level (Table 2) for 1961–2019. We distinguished between SB days at station level (i.e., exact number of events identified for each weather station), coastal level (i.e., encompass all the SB episodes presented in one same coast-region) and regional level (all SB days identified for all stations).
(Fig. 2 about here)
Table 2
SB days identified for each weather station and sub-regional coast for 1961–2019. SB days at station level means the exact number of days identified for each weather station, while days at coastal level encompass all the episodes occurred in one region.
Nº
|
Id
|
Station
|
Sea breeze days
at station level
|
Coast
|
Sea-breeze days at coastal level
|
1
|
6155A
|
Málaga airport
|
7,457
|
Southern
|
8,315
|
2
|
6325O
|
Almería airport
|
6,271
|
3
|
7031
|
San Javier airport
|
6,822
|
South-eastern
|
8,090
|
4
|
8019
|
Alicante airport
|
7,005
|
5
|
8025
|
Alicante Ciudad Jardín
|
7,547
|
6
|
8414A
|
Valencia airport
|
6,394
|
Eastern
|
8,209
|
7
|
8416
|
Valencia Viveros
|
6,520
|
8
|
8500A
|
Castellón-Almassora
|
6,849
|
9
|
9981A
|
Tortosa
|
4,356
|
10
|
0016A
|
Reus airport
|
4,811
|
North-eastern
|
7,179
|
11
|
0076
|
Barcelona airport
|
6,180
|
12
|
0200E
|
Barcelona Fabra
|
5,934
|
13
|
B893
|
Menorca airport
|
2,959
|
Balearic
Islands
|
6,741
|
14
|
B278
|
Palma Mallorca airport
|
5,642
|
15
|
B228
|
Palma Mallorca port
|
5,783
|
16
|
B954
|
Ibiza airport
|
5,199
|
A validation is presented below, whereas thresholds and descriptions of each criterion are described as follows:
Filter 1: To assure a surface pressure gradient weak enough to develop local winds, we used the neutral phase (interval range between − 1 and 1) of the Western Mediterranean Oscillation Index (WeMOI, details in section 2.6). Azorin-Molina and Lopez-Bustins (2008) proposed it as the first criterion to detect potential SB passages associated with weak surface pressure conditions over the Western Mediterranean basin.
Filter 2: Eight weather types (A, E, NE, AE, ANE, C, N, and SE) were filtered from the automated classification by the Jenkinson and Collison´s method (1977; JC), as favorable synoptic conditions for SB (Azorin-Molina et al., 2011a). The method uses daily air pressure from NCEP-NCAR reanalysis to compute the 27 weather types defined by Lamb (1950) for the British Islands. A detailed description of the method can be found in Azorin-Molina et al. (2011a).
Filter 3: Weak geostrophic winds (≤ 12 m s− 1) exclude days dominated by moderate-strong synoptic flows, which disturb SB development (Ramis and Alonso 1988). We calculated the daily 850 hPa geostrophic wind (in m s− 1) at 1200 UTC in terms of geopotential, which assumes a horizontal constant pressure gradient. Geopotential data were taken from ERA5 reanalysis dataset (available online at https://cds.climate.copernicus.eu/; last accessed February 24, 2022).
Filter 4: Winds from the sea were selected based on the coast orientation (see section 2.1 and Fig. 1) and the wind direction at 1200 UTC. Further, to include sideshore SB, we extended orientations at Table 1 in ± 15°. As some recent studies suggest the good representation of near-surface winds provided by ERA5 (Ramon et al. 2019; Minola et al. 2020), we used 10-m height zonal (u) and meridional (v) wind components at 1200 UTC to compute the wind direction.
Filter 5: Daily wind gusts ≤ 13.9 m s− 1 were required to exclude moderate-strong onshore flows, as SB gusts in this region do not exceed 14 m s− 1 (Azorin-Molina et al. 2011b). For this filter, daily observations were provided by AEMET in section 2.2.
Filter 6: Daily total precipitation < 0.1 mm d− 1 was needed to reject days with atmospheric instability (Arillaga et al. 2020; Grau et al. 2021). Daily precipitation series were provided by AEMET for each station described in Table 1. Further, we used daily total precipitation from ERA5 to full length (since 1961) of the shortest observed-precipitation series (see Table 1) plus few missing data therein.
We evaluated the accuracy of the proposed method (PM) in this study with a reference manual method (RM) developed by Azorin-Molina et al. (2011b) in Alicante (Spain) for 2000–2005. Comparisons between databases were made by computing probabilities of detection of potential SB and non-SB episodes (NSB) first described in Laird et al. (2001), and used by Azorin-Molina et al. (2011b). Probabilities of potential detection of SB days were calculated as follows: (i) identical SB days; (ii) identical NSB days; (iii) different SB days; and (iv) different NSB days; for more details see Table S1. Our PM identified 804 SB events for Alicante Ciudad Jardín weather station (Table 3) for 2000–2005, which is more than half of the number of SB days (1,414) detected by the RM. From 37% of episodes identified for 2000–2005, the validation results for equal SB (48%) and NSB (47%) detections suggest that our method is capable to create a robust long “potential” SB database by taking only pure SB days, i.e., discarding most situations in which SB is combined with synoptic winds. Overall, the methods differed from each other in 11% SB days while 89% of NSB days are due to differences in the type of method (i.e., manual or automatic detection), test criteria, and data sources (Table 3).
(Fig. 3. about here)
Table 3
Comparison of sea-breeze (SB) and non-sea-breeze (NSB) days detected by the reference method (RM) and ours (PM) in the Alicante Ciudad Jardín station for 2000–2005. Below are described the probabilities of occurrence of SB and NSB days between PM and RM. Probabilities range from 0 to 1, where 1 indicates the equal or different detections of SB episodes. Details of the calculation are described in Table S1.
Sea-breeze days (2000–2005)
|
SB day (PM)
|
SB day (RM)
|
NSB day (PM)
|
NSB day (RM)
|
804
|
475
|
1,414
|
783
|
Sea-breeze probabilities (2000–2005)
|
Identical SB days
|
Identical NSB days
|
Different SB days
|
Different NSB days
|
0.48
|
0.47
|
0.11
|
0.89
|
2.5 Trend analysis and statistics
Long-term trends for SB speeds, gusts and occurrence were estimated for 1961–2019. As a first step, we create daily time series of wind speed and wind gust for SB days in each weather station. From daily time series, we derive monthly averaged and monthly occurrence series. Weighted averages were performed by grouping all the SB days found in one same region to create regional and coastal series (see Table 2). After this step, we calculated the anomalies with respect to 1981–2010 to assure that no weather station dominated the averaged series (Zhang et al. 2020). Then, we estimated trends with a weighted linear regression in meters per second per decade (m s− 1 decade− 1). Trend significance was assessed by applying the non-parametric modified Mann-Kendall test (Hamed and Rao 1998), which corrects the variance for highly autocorrelated series. Statistical significance was also assessed from a “process and importance” perspective, following McVicar et al. 2010 and Minola et al. 2021. This perspective uses three p-level thresholds, defined as (i) highly significant (p < 0.05); (ii) significant (p < 0.10) and (iii) not significant (p > 0.10). To evaluate the trend persistence, we performed running trends (Brunetti et al., 2006), which detect weakened or stronger trends sub-periods (Morán-Tejeda et al. 2016) with a minimum window length of 30 years, which runs from 1961 to 2019. We also computed a 15-year Gaussian low pass filter to illustrate the multidecadal variability of SB, and the coefficient of determination value (R2) to estimate the variance between SB speeds and gusts. Lastly, calculations are performed over three different time scales: annually, monthly and boreal seasons: December, January and February (DJF) for winter; March, April and May (MAM) for spring; June, July and August (JJA) for summer; and September, October and November (SON) for autumn.
2.6 Atmospheric circulation indices
We selected four atmospheric circulation indices to assess their relationship (via Pearson’s correlation coefficient (r) with the observed SB variability over the 58-year study: (i) The daily WeMOI (Martin-Vide and Lopez-Bustins 2006) provided by the Group of Climatology at the University of Barcelona (http://www.ub.edu/gc/wemo/, last accessed February 24, 2022). This index is based on the regional sea-level pressure difference between the barometric dipole San Fernando (Cádiz) and Padua (Italy), a representation is shown in Fig. 1. We used it to detect potential SB passages for the Eastern Spain (Azorin-Molina and Lopez-Bustins 2008); (ii) The MOI (Palutikof 2003) provided by the Climate Research Unit (CRU), available online at https://crudata.uea.ac.uk/cru/data/moi/ (last accessed February 24, 2022); (iii) The NAOI (Jones et al. 1997) retrieved from CRU (https://crudata.uea.ac.uk/cru/data/nao/, last accessed February 24, 2022); and (iv) the SNAOI (Folland et al. 2009) calculated by averaging the NAOI values for July-August. These indices are well known to drive most of the climatic variability in the Eastern Spain (Corell et al. 2020; Martinez-Artigas et al. 2021), and some authors have analyzed their influence on wind speed variability for this region (Azorin-Molina et al. 2014a, 2016).
2.7 Local-scale drivers of SB activity
To estimate the possible relationship between SB trends, variability and local-physical factors, we computed the Pearson’s correlation between the land-sea temperature difference and horizontal pressure gradient (PG) and our SB parameters (i.e., speeds, gusts and occurrence). To calculate the land-sea thermal contrast we obtained 2m air temperature data over land and ocean surfaces at 1200 UTC from ERA5 Reanalysis from 1979 onwards (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=overview; last accessed February 24, 2022) and its preliminary back extension dataset reaching back 1950 (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels-preliminary-back-extension?tab=overview; last accessed February 24, 2022), to cover the 1961–2019 period. The grid points defined to estimate the temperature differences are given in Table S3. Further to compute the horizontal PG, we obtained the ERA5’s surface pressure from the Copernicus Climate Data Store, which is expressed as:
$$\text{P}\text{G}= \left(\frac{{\text{S}\text{P}}_{land} - {\text{S}\text{P}}_{sea}}{\text{D}}\right)*100$$
where SP refers to surface pressure (in hPa) over land and sea surfaces, and D is the distance (in km) between grid points (described in Table S3).