Evaluating the Spatial and Temporal Characteristics of Summer Urban Overheating through Weather Types in the Attica Region, Greece

Abstract

In this study, we investigated the association between weather type (WT) and urban heat island intensity (UHII) in the region of Attica (Greece). The application of the methodology resulted in ten WTs over the Attica region. The UHII was calculated for every hour of the day from 2008 to 2017, using a new air temperature dataset produced by Copernicus Climate Change Service. To obtain more definitive findings about the relationship between WTs and UHII, we also used the upper 5% of UHII (urban overheating, UO). UO was estimated for two time intervals (daytime and nighttime) and for the warm period (June–September). The UHII frequency distribution, as well as the spatiotemporal characteristics of the UO, were also investigated. It was found that UO was amplified under WT2 during the night, while WT10 was mainly related to increased UO magnitudes in the daytime in all months. Furthermore, analysis results revealed that the UO effect is more pronounced in Athens during the night, especially at the Athens center. The daytime hot spots identified were mainly in suburban and rural areas. Therefore, this methodology may help with heat mitigation strategies and climate adaptation measures in urban environments.

1. Introduction

The expansion of the built environment, as well as the extent of the population and density of a city, causes substantial alterations in the usage and cover of land in metropolitan areas. Nowadays, it is well established that these land cover changes significantly influence local climate and weather conditions between a city and its rural surroundings [1,2]. However, there is a propensity to apply temperature as a measure that most clearly separates a city from the nearby countryside. A large body of international literature has demonstrated that urban areas are, on average, warmer than their suburban and rural surroundings [3,4,5,6]. This effect is referred to as the urban heat island (UHI). The urban heat island is a local scale, anthropogenically generated phenomenon. It is very complex regarding the dynamics, space and time scales that it operates over. In general, the causes of an urban heat island include the increase of heat capacity in cities, reduction of evaporation and evapotranspiration due to buildings materials, complex urban infrastructure, paved surfaces, reduced vegetation, release of air pollutants associated with excess trapped heat, local morphological and surface geometry, city size, population density and the local weather conditions [7].

The combination of high-speed urbanization processes, increased urban population and urban heat island effects pose human health and well-being risks, especially for the elderly and other vulnerable groups related to outdoor activities (such as traveling, garden care and recreation) [8]. The effects of the urban heat island also include changes to the urban energy budget and social, economic and environmental issues [9,10]. Investigating synergies between urban heat islands and heat waves in Athens, Founda and Santamouris [11] found positive feedback between the factors, which may increase the thermal risk in cities.

Local meteorological phenomena, such as changes in precipitation rates, frequency of lightning strikes and formation of clouds and fog, may be strongly affected by urban heat islands [12,13]. Conversely, one of the key factors influencing the development and intensity of an urban heat island is the local weather conditions. In this context, numerous studies have been undertaken in order to find the influences of singular climate variables (wind speed, cloud cover, relative humidity and precipitation) on the magnitude of the urban heat island [14,15,16,17].

Since the focus of much urban overheating (UO) research is on the interaction of urban heat islands with individual meteorological variables (wind speed, temperature, relative humidity, cloud cover, etc.), the association between urban heat island intensities/patterns and large-scale prevailing weather conditions in a given location are largely absent from such univariate approaches. It is also necessary to have a deeper understanding of how synoptic-scale meteorological conditions affect the spatial and temporal patterns of urban overheating since these variables will affect local-scale conditions in urban areas. In this context, weather type categorizations have proven very useful tools in such research. Among others, in Lisbon’s Metropolitan Area, rainy days and sunny days (especially very cold winter days) were linked with lower and higher median urban heat island intensities, respectively [18]. Moreover, during the daytime, urban overheating is associated with humid, humid warm and warm conditions in Athens and Thessaloniki (Greece), and with warm, humid warm, dry and dry warm conditions during the nighttime [6]. In Sydney, Khan et al. (2021) utilized the gridded weather typing classification (GWTC2) and found that the high urban overheating magnitude was related to humid warm and warm air masses during extreme heat events and in all seasons [19].

However, the majority of existing studies on the associations between urban overheating and synoptic weather conditions have evaluated the urban heat island effects by using already developed weather typing classifications and one (or more) urban and rural stations in order to calculate temperature differences between urban and rural places. In this study, a definition of weather types (WTs) for the 41-year period (1980–2020) was conducted by applying statistical methods to the recently released high-resolution grid point meteorological data over the Mediterranean and Attica region (Greece). In addition, high temporal and spatial resolution data from the Copernicus dataset were applied to investigate urban heat island intensity and examine the influence of WTs, which were defined based on the urban overheating magnitude in the Attica region.

2. Materials and Methods

2.1. Study Area

The undertaken research is focused on the geographical region of Attica in Greece, which encompasses the entire Athens Metropolitan area, the country’s capital and largest city. Attica is a highly crowded, built-up region that lacks open spaces and green areas and covers about 3808 km². The permanent population increased to 3.8 million inhabitants in 2011, with a density of 1250 inhabitants/km², while more than 95% are inhabitants of the Athens metropolitan area (Hellenic Statistical Authority; https://www.statistics.gr/statistics/pop (accessed on 1 February 2023)). In addition, the Attica region is a triangular peninsula jutting into the Aegean Sea and is watered by the gulfs of the Aegean Sea. In the center of the peninsula, there is a large basin where the entire metroplex of Athens–Piraeus has been constructed. This basin is surrounded by four mountains: Hymettus, Parnitha (the highest mountain of Attica), Egaleo and Penteli (Figure S1), providing significant amounts of green. According to the Hellenic National Meteorological Service (HNMS), Attica is one of the warmest regions in Greece, with mean monthly temperatures and precipitation heights ranging from 8.8 °C to 28.3 °C and from 1.6 mm to 12.5 mm, respectively (http://www.emy.gr/emy/en/climatology/climatology (accessed on 2 February 2023)).

2.2. Meteorological Data

For the purpose of this study, two sets of data were obtained from the NCEP/NCAR Reanalysis (https://iridl.ldeo.columbia.edu/SOURCES/.NOAA/.NCEPNCAR/.CDAS-1/.DAILY/ (accessed on 3 September 2022)), covering a 41-year time period (1980−2020) [20]. Since the NCEP/NCAR database is the longest reanalysis database and uses the full set of available observations, it was chosen above other existing datasets. The first set of data contains 2.5° × 2.5° intrinsic daily grid point values of 500 and 1000 hPa geopotential heights, 850 hPa temperature, and 850 hPa specific humidity over the main geographical domain of the Mediterranean region (30° N–45° N, 10° Ε–35° E) (Figure 1). The second set comprises daily diagnostic gridded meteorological variables at 1° × 1° (2 m temperature, specific humidity, 2 m zonal and meridional wind, total cloud cover, precipitation rate and convective precipitation rate) for the geographical domain of central Greece (37° Ν–39° Ν, 22.5° Ε–24.375° E), including the region of Attica (Figure 1). The geopotential height data describe the atmospheric circulation conditions over the region, whereas the above meteorological parameters that we used provide crucial information on the meteorological and climatic conditions prevailing over the Attica region [21].

2.3. Copernicus Urban Climate Dataset

To achieve a detailed spatial distribution of the urban heat island in the Attica region, we used the dataset presented by the Copernicus Climate Change Service [22]. This dataset contains hourly gridded values (air temperature, specific humidity, relative humidity, and wind speed) for 100 European cities from 2008 to 2017, with a spatial resolution of 100 m. The data were produced using the UrbClim urban climate model. This model was created with a geographical resolution of 100 m (at the scale of a city neighborhood) to simulate and research the urban heat island and other urban climatic factors and requires two types of input data: large-scale meteorological data (downscaling of ERA 5 reanalysis) and terrain data (land use, vegetation, and soil sealing).

Therefore, in this study, hourly air temperature gridded data at the height of 2 m above the surface were obtained in NetCDF-4 format for Athens between 2008 and 2017. June, July, August, and September were only selected, representing summer season. The temperature data were divided into daytime (06:00–18:00 LT) and nighttime (23:00–06:00 LT) to explore the urban heat island effects in the day (ΔΤdaytime) and night (ΔΤnighttime) for Athens. Land–sea mask and rural–urban mask data were also obtained to isolate land area from sea area and separate urban from rural area.

2.4. Formation of the Prevailing WTs

In the present study, in order to identify the WTs over the Attica region, we employed a combination of two main statistical methods: k-means cluster analysis (CA) and components analysis (PCA). In climatology, both statistical methods are often employed [23,24,25].

Components analysis is a dimensionality-reduction technique that is frequently used to decrease the dimensionality of big data sets. It does this by splitting a large number of variables (X₁ + X₂..., X_i) into a smaller set of variables called “principal components” (PCs), which are formed in such a way that they are uncorrelated [26]. In components analysis, rotation of the factor axes (dimensions) is a method that rotates the eigenvectors (factors) in an effort to produce a simple structure [27]. Rotation methods are either orthogonal (equamax, orthomax, quartimax, and varimax) or oblique. Orthogonal rotation methods assume that the factors in the analysis are uncorrelated. The varimax rotation was utilized in this study, which is the most common orthogonal rotation method that maximizes high- and low-value factor loadings and minimizes mid-value factor loadings. The optimal number of principal components is indicated by the SCREE plot graph, along with the physical meaning of the principal components. The eigenvalues are shown on the SCREE plot as a descending slope, ranging from greatest to lowest [28].

A set of variables with similar properties are grouped together into objectively defined groupings (clusters) using the statistical approach known as cluster analysis. In exploratory data analysis, cluster analysis is frequently employed because it can reveal patterns or relationships in the data that might not be immediately apparent [29]. For cluster analysis, a variety of techniques are employed, including k-means, hierarchical clustering, and density-based clustering. In the present study, the non-hierarchical method k-means was utilized to find clusters, while the Euclidean distance was used to assess case similarity. Additionally, by employing the distortion test, which is based on the typical distance between observations and cluster centers, the ideal number of clusters was indicated [30].

As the first stage of the particular approach utilized here, component analysis is applied to the correlation matrix that includes the two sets of variables. More specifically, the first matrix includes 500 and 1000 hPa geopotential heights, 850 hPa temperature, and 850 hPa specific humidity over the wider domain of the Mediterranean region, and the second matrix contains the meteorological values over the sub-domain, namely 2 m temperature, specific humidity, 2 m zonal component and meridional component wind, total cloud cover, precipitation rate, and convective precipitation rate. In order to group the days into clusters with homogenous climatic features, the K-means cluster analysis algorithm was applied in the second step to the set of principal components that were retained. On this basis, local WTs could be located, and our findings could be directly compared to those of other research carried out in the same area.

2.5. Urban Heat Island Intensity Calculation

The difference in temperature between urban and rural regions at a given time is known as urban heat island intensity (UHII) [31,32]. The non-urban points must be located outside of the built-up urban region or any surface that has been altered by construction (asphalt, cement, etc.) since this site has traditionally been linked with a rural area. Hence, the urban heat island intensity was calculated for every hour of the day from 2008 to 2017 according to the following equation:

ΔΤi = T_urb − T_{rul (mean)}

(1)

where ΔΤi represents the urban heat island intensity, T_urb is the 2 m temperature for each urban grid point, and T_{rur (mean)} is the 2 m temperature averaged over the rural grid points. The ΔΤi frequency distribution at daytime and nighttime for each month was also examined. Additionally, in order to find associations between local WTs and urban overheating and have a clearer view of these associations, we have also used the higher 5% of the daily maximum ΔΤi for both daytime and nighttime.

However, the city’s bounds are frequently erroneous since the urban continuum occasionally lacks distinct boundaries, suggesting a shift toward normally rural land usage [33]. In our case, the rural area is represented by the rural classes of CORINE (coordination of information on the environment) covering grassland, cropland, shrubland, woodland, broadleaf forest, and needleleaf forest (https://land.copernicus.eu/pan-european/corine-land-cover (accessed on 15 July 2022)). More specifically, according to the land–sea mask, land area receives value 1 and is represented by the land classes of CORINE, sea area receives missing value NaN and is represented by the sea classes of CORINE, while according to the rural–urban mask, rural area receives value 1 and is represented by the rural classes of CORINE, urban area receives missing value NaN and is represented by the urban classes of CORINE (Figure S2).

3. Results and Discussion

3.1. Definition of WTs over the 41-Year Time Period (1980–2020)

The application of components analysis on the two data sets of the wider domain and the sub-domain for the entire 41-year time period resulted in five principal components in each case (representing 88.1% and 77.3% of the total variance, respectively). The number of principal components was based on the results of the SCREE plots (Figure S3), as mentioned above. The application of cluster analysis on all 10 resultant principal components together led to 10 distinct and homogeneous clusters according to the distortion test (Figure S4). Considering that each of the 10 clusters defines a specific WT, Figure 2 displays the average values of the meteorological variables for each of the 10 WTs. The WT frequency was also investigated in the Attica region from 1980 to 2020 (Figure 3). WT 2 and WT 10 were the most frequent WTs in all seasons (average around 19% of the time), followed by WT1 (12%), WT8 (11%), WT4 (11%), WT9 (8%), WT7 (7.2%), WT6 (5.3%), WT3 (4.1%), and WT5 (3%).

The 41-year period’s long-term average patterns of 1000 hPa geopotential height for each WT are displayed in Figure 4. WT2 and WT10 are typically warm period WTs, due to the fact that they mainly prevail between May and September, and their frequencies are maximum in August and July, respectively (Figure 4). The Azores’ high-pressure system over southern Europe and the expansion of the Asiatic Low into the eastern Mediterranean are both responsible for these characteristic summer circulations. These atmospheric patterns are also responsible for the formation of the Etesian winds (northerly direction) over the Aegean Sea [34,35]. More specifically, WT10 was the maximum frequency in July and was associated with high temperatures, partly cloudy conditions, and limited precipitation. The weak pressure gradient over the Aegean causes low wind speeds, mainly from a northerly direction. According to Kassomenos and Katsoulis (2006), this atmospheric situation favors the formation of local circulations, such as sea breezes, carrying humid air masses from the sea [36]. On the contrary, the maximum WT2 frequency appears in August, when the Etesian wind outbreaks are most frequent [37] and associated with the stronger northerly flow. Air temperature is high, and cloud cover and precipitation are generally low, while according to Lolis and Kotsias (2020), this WT is generally dry [21].

WT3, WT4, and WT9 are cold period WTs, correlated to low-pressure systems across the southeast Mediterranean, centered on the southeast Aegean Sea and Cyprus (Figure 4). During the winter, these atmospheric circulation patterns are typical and associated with low temperatures, cloud development (Figure 2), and rainy weather conditions over the Aegean Sea [38]. They prevail between October and May, causing a northeasterly flow over the southern Balkans, transferring cold air masses from eastern Europe, and can be related to prolonged sunshine in North Greece [39].

WT5, WT6, WT7, and WT8 are also cold WTs, which generally prevail from October to May. These WTs correspond to low-pressure systems over the Ionian Sea, whilst anticyclonic conditions are present in the northeastern Balkans (Figure 4), causing S–SW flow over the southern Balkans and, in the case of WT7, ENE flow. These types of circulations are associated with low temperatures and extensive cloud cover (Figure 2) that can be followed by precipitation events [40], especially over the western part of the country [41].

WT1 could also be considered a cold period WT prevails mainly from November to May, with a depression over the northeast Balkans and over the Middle East, with an anticyclone over the central Mediterranean (Figure 4). The temperature was relatively low, with cloud cover characterized by low values (Figure 2). Such atmospheric circulation causes westerly airflow and is associated with dry weather conditions and low precipitation [39].

3.2. Urban Heat Island Intensity Frequency Distribution

The suburban–urban temperature difference (ΔΤi) frequency distribution throughout the day and at night was examined for all months during the summer from 2008 to 2017 (Figure 5). As mentioned above, the ΔΤi represents the urban heat island intensity in the Attica region. In June, the daytime ΔΤi was positive for about 53% and negative for 40%. As for the nighttime in June, ΔΤi appeared positive for 58% and negative for 31%. In July, the daytime and nighttime ΔΤi calculated was positive for about 62%, in contrast to the negative daytime and nighttime ΔΤi, which reached a percentage close to 28%. In August, the daytime ΔΤi was reported positive for 52% and negative for 32%. Contrarily, during the August night, ΔΤi was found positive for 62% and negative for only 27%. Finally, in September, the daytime ΔΤi was positive for 46% and negative for 40%, while the nighttime ΔΤi was positive for 60% and negative for 31%.

Comparing the frequency distribution, the higher frequencies of positive ΔΤi were observed during the night, especially in July and August (62%). Moreover, in all cases, the positive ΔΤi outweighed the negative ΔΤi, while in August, we observed the highest difference between positive and negative ΔΤi (62% against 27%, at nighttime).

It is evident that the urban heat island is more pronounced during the night than the day, in agreement with previous studies of urban heat island formation, which found that the urban heat island is primarily a nocturnal phenomenon due to the fact that in the late afternoon and evening, urban areas fail to cool as quickly as the surrounding rural regions [4,18]. Similarly, previous studies for Athens reported that the highest urban heat island intensities are related mostly to the nighttime hours [42,43]. For instance, Giannaros et al. (2013), who studied the urban heat island over Athens by using a WRF model, found significantly higher nighttime temperatures for the Athens metropolitan area compared to the nearby non-urbanized areas [44]. The surface of the city serves as an urban heat sink throughout the day, while during the night, the city’s surface seems to be warmer than its surroundings, as noted by Keramitsoglou et al. [45].

3.3. Association of WTs with Urban Overheating Magnitudes

To assess the relationship between the daily WTs and the urban overheating in summer, the upper 5% of daily daytime and nighttime maximum ΔΤi (UO) was calculated. Figure 6 depicts the higher 5% of the maximum ΔΤi (UO) in the Attica region for summer under different weather types. It is apparent that WT2 and WT10 are the only weather conditions in the daytime, enhancing the urban overheating magnitude. More specifically, WT10 contributed about 52.4%, and WT2 for 47.6% of the time. The median value of the urban overheating magnitude in the daytime during WT10 was estimated to be equal to 7.4 °C, comparatively higher during WT2 conditions (6.9 °C).

While examining the association between the urban overheating magnitude and WTs at nighttime, it was observed that urban overheating was associated with more than two WTs. As illustrated in Figure 6, WT2, WT10, WT8, WT6, WT5, and WT1 are related to urban overheating during the night in summer. However, WT2 and WT10 were the most dominant weather conditions. In terms of WT frequency, during the night, WT10 contributed 44.8% and WT2 for 50.9% of the time, followed by WT8 (1.7%), WT5 (1.5%), WT6 (0.6%), and WT1 (0.3%). Furthermore, estimates of the median value of the urban overheating magnitude during the night were 8 °C for WT2 and 8.1 °C for WT10, slightly higher compared to those in the daytime. It appears that daytime urban overheating was amplified under WT10, while during the night, urban overheating was associated mainly with WT2.

These results do not come as a surprise, as WT2 and WT10 are typical summer circulations and the most frequent WTs in all seasons. According to Kassomenos et al. (2022), the warm WTs were more frequently documented in Athens, notably in recent years and during summertime, and the highest urban overheating levels were correlated with warm and humid conditions during the day and dry warm conditions at night [6]. Additionally, Kassomenos and Katsoulis (2006) stated that high urban heat island classes in Athens were associated with similar conditions as those of WT2 and WT10 (a combination of the Azores Anticyclone and SW Asia thermal low) [36]. Similarly, Mihalakakou et al. (2002) studied the impact of synoptic-scale atmospheric circulation on the urban heat island over Athens and identified an anticyclonic synoptic category characterized by a weak flow regime, which favors the development of the urban heat island phenomenon [46].

3.4. Spatial and Temporal Characteristics of Urban Overheating during Summer

The spatial characteristics of mean urban overheating values are illustrated in Figure 7 for each month during the summer (daytime and nighttime) from 2008 to 2017. It is apparent that, in June, during the day, urban overheating extended to the whole Athens metropolitan area, including the Athens urban area (also known as Greater Athens), Piraeus, northwest suburban areas, and rural areas of East Attica. The urban overheating magnitude at these sites varied between 8 and 11 °C on average. During the night in June, the urban overheating was mainly limited to the Athens urban area, Piraeus, and some suburban areas in southeast Attica. In July, urban overheating hot-spots (8–11 °C) were found mainly near suburban areas such as Mount Hymettus at East Attica, north near Mount Penteli (Mesogeia), and west near Mount Parnitha and Mount Egaleo (Petroupoli and Nikaia). During the night hours in July, the highest magnitude of urban overheating was related to the central municipalities of the Athens Greater Area (City of Athens, Piraeus, Moshaton, Kallithea, N. Smyrni). Remarkably, there was a reduction in terms of urban overheating magnitudes during the day in August. The daytime hot-spots were located mainly west of Athens urban area (Egaleo), while nighttime hot-spots in August appeared at the central municipalities of the Athens Greater Area. Similar patterns were observed in September, although during the night, urban overheating hot-spots expanded to the east suburban and rural areas (Mesogeia).

For all cases, during the daytime, hot-spots located mainly in suburban and rural areas, such as Egaleo and Mesogeia, reached their maxima in June and September. On the other hand, a thorough examination during the night in Figure 7 reveals that nighttime urban overheating hot-spots related to urban areas, including the Athens center and central municipalities of the Athens Greater Area. As for urban overheating magnitudes, nighttime hot-spots exhibited similar behavior, reaching their maximum values in early/later summer. These findings support those of Giannaros et al. (2013), who found the existence of three daytime hot-spots, at Megara, Elefsina, and Mesogeia, while in the nighttime, the city center was warmer than its surroundings [44]. Furthermore, this behavior is well known in Athens, as already reported in Keramitsoglou et al. [45]. The combined impact of the topography and surface cover features determines the urban overheating hot-spots of Athens throughout the day. For instance, Mount Hymettus is east of Mesogeia, which is bordered to the north by Mount Penteli’s slopes, including areas with sparse low vegetation, such as olive trees and vineyards, industrial zones, as well as areas covered with bare soil, such as the Athens International Airport “El. Venizelos’’. Such semi-rural environments start heating very fast during the daytime and cooling quickly after sunset due to their open exposure and low thermal surface materials. Conversely, the higher release of stored heat from urban surfaces such as concrete and asphalt causes the cooling rate of the densely populated urban environment to be greatly lowered. It is worth mentioning that high surface temperatures during heat waves may reduce the definition of these hotspots.

4. Conclusions

In the present study, a classification of WTs was conducted and implemented for studying the urban heat island effect over the Attica region (Greece) during the summer, for both day and night. An association between urban overheating and daily WT conditions, as well as the spatial characteristics of urban overheating, were also investigated. For this purpose, a brand new, publicly accessible dataset on urban climate from Copernicus Climate Change Service was used. The analysis above allowed for the following conclusions to be formed:

• The warm period WTs (WT2 and WT10) were reported to occur with a higher frequency between 1980 and 2020.
• As previous studies have shown, urban heat island intensities in the Attica region were exacerbated during the night, suggesting that the urban heat island is mainly a nighttime phenomenon.
• WT10 was mainly responsible for exacerbated urban overheating magnitude at daytime in all months, while urban overheating at nighttime was mostly attributed to WT2 conditions.
• WT10 is associated with the weak pressure gradient over the Aegean, causing low wind speeds mainly from northerly directions, in contrast with WT2, which is associated with stronger northerly flow.
• During the day, high urban overheating magnitudes occur mainly in sub-urban and rural areas (Mesogeia, Penteli, Aspropyrgos) of the Attica region. Conversely, urban overheating hot-spots are consistently located in urban areas, including the Athens center and central municipalities of the Athens Greater Area. In both cases, the urban overheating reached their maxima in June and September.

These findings highlight that studying the urban heat island effect under the influence of different atmospheric circulation patterns provides essential information on the prevention and control of the negative urban overheating effects in a densely populated city like Athens. Urban heat island effects become even more important when considering that in a changing climate, the frequency of warm weather types will increase with their frequency [47]. To minimize the impacts of urban heat islands, it is crucial to increase urban greening and improve suburbanization levels. Planting trees and vegetation in strategic locations in the Athens metropolitan area, installing green roofs, and creating pavements with improved reflectiveness (increased albedo), permeability, and water retention, are some mitigation measures that can be taken to reduce urban heat island effects.

Future research, however, would benefit from extending the investigation to the winter and looking at how urban heat islands differ depending on the weather throughout all aspects of their daily cycle. These findings can also be applied by the local government for heat mitigation techniques and climate adaptation methods in metropolitan areas.