**Canopy Urban Heat Island and Its Association with Climate Conditions in Dubai, UAE**

#### **Afifa Mohammed 1,\*, Gloria Pignatta 1, Evangelia Topriska <sup>2</sup> and Mattheos Santamouris <sup>1</sup>**


Received: 25 May 2020; Accepted: 24 June 2020; Published: 26 June 2020

**Abstract:** The impact that climate change and urbanization are having on the thermal-energy balance of the built environment is a major environmental concern today. Urban heat island (UHI) is another phenomenon that can raise the temperature in cities. This study aims to examine the UHI magnitude and its association with the main meteorological parameters (i.e., temperature, wind speed, and wind direction) in Dubai, United Arab Emirates. Five years of hourly weather data (2014–2018) obtained from weather stations located in an urban, suburban, and rural area, were post-processed by means of a clustering technique. Six clusters characterized by different ranges of wind directions were analyzed. The analysis reveals that UHI is affected by the synoptic weather conditions (i.e., sea breeze and hot air coming from the desert) and is larger at night. In the urban area, air temperature and night-time UHI intensity, averaged on the five year period, are 1.3 ◦C and 3.3 ◦C higher with respect to the rural area, respectively, and the UHI and air temperature are independent of each other only when the wind comes from the desert. A negative and inverse correlation was found between the UHI and wind speed for all the wind directions, except for the northern wind where no correlation was observed. In the suburban area, the UHI and both temperatures and wind speed ranged between the strong and a weak negative correlation considering all the wind directions, while a strong negative correlation was observed in the rural area. This paper concludes that UHI intensity is strongly associated with local climatic parameters and to the changes in wind direction.

**Keywords:** subtropical desert climate; urban overheating; cluster analysis; air temperature; wind speed and wind directions; synoptic conditions

#### **1. Introduction**

As observed in many cities globally, rapid urbanization has produced negative effects on the climate and the local microclimate. Currently, the urban population exceeds 50% of the total of the world's population and by 2050 it is expected to rise above 60%. This means that with urban development worldwide and at the current rate of population growth, another 2.5 billion people will be living in urban areas by 2050 [1]. Consequently, this urban expansion will exacerbate the hostile impact that human activities are already having on environmental systems.

Rapid urbanization has boosted regional climate change. In particular, the increase in urban heat island (UHI) intensity is strictly connected with the urbanization growth given the derived increase in anthropogenic heat emissions, land-use change (with the associated decrease in the vegetation and the albedo of the built area), and changes in the advection.

The UHI phenomenon has been documented in more than 400 cities around the world [2]. Its impact is closely related to land cover which controls the energy budget on the earth's surface. The surface energy budget difference between the urban and rural zone, caused by various thermal-optical surface characteristics, leads to the occurrence of the UHI phenomenon [3]. Thus, urban areas are hotter than their undeveloped surroundings. This phenomenon has a significant impact not only on the environmental quality of cities, but also on energy, thermal comfort, and health [4–7]. In particular, the UHI increases the demand for peak-time electricity and the consumption of cooling energy in buildings, intensifies the concentration of various harmful pollutants, increases the ecologically harmful footprint of cities, and has a significant impact on health [8]. Most cities are sources of pollution and heat, released from buildings and roads. The manifestation of the UHI phenomenon is influenced by several factors including climate variables (i.e., air temperature, air relative humidity, wind speed) and the related local synoptic weather conditions [9,10], thermal-optical characteristics of the materials, the magnitude of the anthropogenic heat released, and the existing heat sources in the areas [2].

On the other hand, the construction and building industries are indispensable to everyday life and are essential to future social and technological developments. However, they are responsible for a great deal of pollution, generation of waste, and the consumption of energy and natural resources accounting for 30%–40% of the overall consumption on the planet [11,12]. Moreover, these industries have negative effects on the local and global climate [12]. The International Energy Agency estimates that the growth in energy consumption will be around 38.4 PWh in urbanized countries by 2040, while energy consumption globally was about 23.7 PWh in 2010 [13]. These industries account for 38% of all greenhouses gas emissions. They also contribute significantly to increase the temperature of cities by producing the urban heat island (UHI) phenomenon [8].

Population growth and rural depopulation will lead to greater energy consumption, particularly in hot and dry cities which require more cooling loads; in addition, this increases the number of gas emissions and pollution that negatively interact with the phenomenon of the UHI [14]. Furthermore, the expansion of urban areas will require a more complex energy infrastructure to meet demand [12,15].

Recently, many studies have demonstrated that per degree of temperature increase, the peak electricity demand increases by between 0.45% and 4.6% due to the UHI [16]. Moreover, it is expected that by 2050, there will be a massive increase of about 750% and 275% globally in the cooling requirements of the residential and commercial sectors, respectively [17]. Furthermore, the UHI has a significant impact on human health, as several studies have shown that high temperatures lead to a great increase in the number of people suffering from heat exhaustion and heatstroke in urban areas [18]. Moreover, it has been pointed out that there is a strong correlation between the increase in human mortality and the rise in urban temperatures [19]. Several studies in various Asian cities have shown that temperatures above 29 ◦C can increase the mortality rate between 4.1% and 7.5% per 1 ◦C increase in temperature [20]. Hence, from the aforementioned discussion, it is clear that cities need specific controls and strategies to mitigate the negative impacts of the UHI phenomenon [12].

Among the performed studies on UHI, three different definitions of UHI are usually considered, such as the boundary UHI for mesoscale analysis [21], the canopy UHI for microscale analysis [22], and the surface UHI [23]. Different methods have been reported [24] to identify the different UHI types, including direct and indirect methods, numerical modeling, and estimates based on empirical models.

In contrast with the numerous studies on UHI performed in temperate regions, only a few studies have been concentrated in the observation of the UHI intensity in desert regions [25,26]. Some of those studies state that the UHI presents a diurnal and seasonal cycle in the Gulf area. The canopy level of the UHI of Muscat, Oman, reaches the peak approximately 7 h after sunset based on summer meteorological observations. Muscat city is characterized by low ventilation, many business activities, multi-storied buildings, heavy traffic, and topography factors [27]. In another study in Bahrain, numerical modeling was performed and the results show an increase in the simulated average air temperature of about 2 ◦C–5 ◦C when assessing the impact of the urbanization and that the canopy UHI magnitude is enhanced by various urban activities such as construction processes, vegetation shrinkage, and sea reclamation [28].

Given the lack of information about UHI intensity concepts in the Middle East in general [28] and specifically in Dubai, United Arab Emirates (UAE), this paper aims to quantify and analyze the intensity of the canopy UHI in Dubai, highly influenced by the urban geometry and physical properties of the built environment, and its relation with the main climatic parameters continuously monitored for a period of five years.

#### **2. Materials and Methods**

#### *2.1. Geographical Location, Population, and Climate of Dubai*

Dubai, one of the seven emirates and the second largest city of the United Arab Emirates (UAE), has an area of 4114 km2, accounting for 5% of the overall area of the country. Dubai city is recognized as the economic capital of the UAE. Situated on the Tropic of Cancer, between 25◦16 N and 55◦18 E, it has a coastal length of 72 km on the eastern coast of the Arabian Peninsula and faces the South-West of the Arabian Gulf [29,30].

According to the census conducted by the Statistics Centre of Dubai, Dubai is the city with the highest population density in the UAE. The population of Dubai increased by 27% from 2,327,350 to 3,192,275 between 2014 and 2018 inclusively [31].

The Dubai region has a subtropical desert climate, with hot and humid summers and warm winters [30]. The air temperature ranges between 10 ◦C to 30 ◦C in the winter season and increases up to 48 ◦C in the summer season [29]. The hot period starts with average daily temperatures over 37 ◦C from 18 May to 23 September, lasting for 4.1 months; the cool season starts with average temperatures of less than 27 ◦C from December 4th to March 8th and lasts for 3.1 months [32]. Winter is characterized by rainfall and fog, while in summer, the relative humidity reaches 80%–90% [29]. The average rainfall ranges between 13 and 17 mm [32].

#### *2.2. Meteorological Stations and Data Analysis*

In this study, climate data (i.e., air temperature, relative humidity, wind speed, and wind direction) were collected by three meteorological stations in Dubai (i.e., Dubai International Airport station, Al-Maktoum International Airport station, and Saih Al-Salem Station) belonging to the UAE National Center of Meteorology (NCM). Hourly data of each microclimate parameter were collected from each meteorological station and analyzed in the MS Excel tool for a period of five years from 2014 to 2018 (i.e., 43,824 values recorded in total for each climate variable). Data of cloud coverage were also collected by the two stations of Dubai Airport and Al-Maktoum Airport, while the solar radiation that equally affects the three areas of Dubai was not measured by any meteorological station. In this study, no further tools were used for the data analysis except for MS Excel.

Table 1 summarizes the geographical information of the three meteorological stations [33], while Figure 1 shows their locations in Dubai. All three stations are installed at a height of about 10 m from the ground, allowing to perform the microscale analysis of the UHI phenomenon, investigating the canopy UHI and its association with the climate parameters measured at the same height.

Station 1 (Table 1) is located within Dubai International Airport, one of the largest and most modern airports in the world, located in Al Garhoud City, which is a commercial and residential region of Dubai, approximately 5 km (2.9 miles) East of the Dubai's CBD (central business district), and around 19 m above sea level [34]. A total of 15,000 solar panels were installed on the roof of the airport building 2 km away from the meteorological station [35].

Station 2 (Table 1) is located within the Al Maktoum International Airport, 37 km (23 miles) South-West of Dubai. The airport is surrounded by a mixed-use area, made of residential and commercial buildings; it is the major part of the Dubai World Central in the Jebel Ali zone with a total area of around 14,000 hectares (35,000 acres). This area could be described as an open area. It is almost at a 16 km (10 miles) distance from the sea and has no surrounding buildings or major obstacles that could interfere with the wind flowing from different directions [36].

Station 3 (Table 1) is located within Saih Al Salem, a village in Dubai with a population of 589 residents, accounting for 0.02% of the total population of Dubai, as reported by the Dubai Statistic Centre in 2018 [37]. The meteorological station is surrounded by the Marmoom Desert Conservation Reserve, which constitutes 10% of the total area of Dubai. It is the first unfenced desert reserve in the UAE. The total area exceeds 40 hectares of shrubland and contains many lakes with a total area of around 10 km<sup>2</sup> [38]. The amount of vegetation in the areas surrounding the three stations is negligible.

In Table 1, the population density of the three locations is calculated as the ratio between the population and the land area of the sectors of Dubai to which the weather stations belong. Considering as references the population densities of Sector 1 (i.e., 10,561/km2) classified as "High" and of the Sector 9 (i.e., 5/km2) classified as "Low", then Sector 2, the sector where the weather station of Dubai International Airport is located, is classified as an urban area with a "Medium" population density (3611/km2). Al Maktoum International Airport is instead located within Sector 5, presenting a population density of 825/km<sup>2</sup> and therefore classified as a suburban area with a "Medium/Low" population density. The area of Saih Al-Salem presents a low population density of 7/km<sup>2</sup> and being located within Sector 9, was defined as a rural area for this study [39].

**Table 1.** Geographical information and description of the urban, suburban, and rural meteorological stations.


**Figure 1.** Location of the weather stations in Dubai as shown in Google Maps (1. Dubai International Airport station, 2. Al Maktoum International Airport station, 3. Saih Al Salem station).

#### 2.2.1. Cluster Analysis of Climate Data

Clustering techniques were used to investigate the level of correlation between the microclimate parameters, i.e., air temperature and wind speed, monitored in the urban and suburban locations, and the canopy UHI intensity under the different ranges of wind directions. The cluster analysis was performed in the MS Excel tool only for the urban and suburban locations because those are the areas where the urbanization may produce an impact on the UHI intensity, while the rural area was considered as the reference location for the calculation of the UHI intensity. The data collected by each meteorological station for the entire period of interest (i.e., 5 years) were divided into six clusters based on the directions of the wind (i.e., either from the seaside or from the desert side, or from the coastal side which is in between, and divided according to four wind direction ranges). The characteristics of the six clusters are summarized in Table 2 and Figure 2.


**Table 2.** Clusters of wind direction.

**Figure 2.** Representation of the six clusters of wind direction.

In Table 2, it can be observed that the dimensions of the six clusters (i.e., number of measurements included in each cluster) are consistent between the urban (i.e., station 1) and the suburban (i.e., station 2) area due to their close proximity. From the cluster analysis, the resulting predominant wind is the one coming from the sea, mainly seen in cluster 1 (i.e., 13,689 and 13,305 for station 1 and station 2, respectively) and cluster 5 (i.e., 4298 and 4862 for station 1 and station 2, respectively). The second predominant wind is the one coming from the desert, mainly seen in cluster 2, but also in cluster 4.

#### 2.2.2. Canopy UHI Magnitudes

The hourly canopy UHI magnitude was calculated at 10 m above the ground for the urban and suburban locations for a period of five years (i.e., 2014–2018). Saih Al Salem (i.e., station 3) was chosen as a reference station, because it is located in a rural area not too far from the downtown city and close to the desert. Of the three stations, this was the least affected by the UHI due to its low level of urbanization. Since this reference station is located outside of the built-up area, it has natural desert coverage and negligible anthropogenic heat.

The canopy UHI magnitude (UHI intensity T1–T3) was determined as the air temperature difference between Dubai International Airport (T1 Urban Average) which is station 1 and Saih Al Salem (T3 Rural Average) which is station 3, according to the following equation [40], Equation (1):

$$\text{UHI intensity } \,\,\,\text{T\,s}\,\,\,\text{T\,C}\,\,\text{I}\,\,\text{T\,C}\,\,\text{I}\,\,\text{T\,s}\,\,\text{Arverage\,[}\,\text{°C]}\,\,\text{-}\,\text{T\,3}\,\,\text{Rural\,Average\,[}\,\text{°C]}\,\,\text{}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\text{V}\,\,\text{A}\,\,\,$$

While the canopy UHI magnitude (UHI intensity T2–T3) was determined by calculating the air temperature difference between Al Maktoum International Airport (T2 Suburban Average) which is station 2 and Saih Al Salem (T3 Rural Average) which is station 3, according to the following equation, [40] Equation (2):

$$\text{UHI intensity } \text{T2-T3 [}^{\circ}\text{C]} = \text{T2 Suburban Average [}^{\circ}\text{C]} - \text{T3 Rural Average [}^{\circ}\text{C]} \tag{2}$$

The canopy UHI calculations, performed according to the above equations, and the data post-processing presented in the result section, were based on the climate data analyzed in MS Excel.

#### **3. Results**

#### *3.1. Microclimate Analysis*

This section presents the microclimate analysis for the three investigated locations in Dubai.

Table 3 summarizes the results of the statistical analysis performed on the main weather parameters (i.e., air temperature, relative humidity, and wind speed) recorded by the three urban, suburban, and rural meteorological stations for the entire period of investigation (i.e., 2014–2018).


**Table 3.** Statistical data of the weather parameters for the three meteorological stations for 5 years (i.e., 2014–2018).

Considering the entire 5 year period from 2014 to 2018, the average, maximum, and minimum air temperatures (measured at 10 m above the ground) were 29.6 ◦C, 48.6 ◦C, and 12.3 ◦C for Dubai International Airport station 1, 28.0 ◦C, 48.5 ◦C, and 7.1 ◦C for Al Maktoum International Airport station 2, and 28.3 ◦C, 50.8 ◦C, and 4.7 ◦C for Saih Al Salem station 3, respectively. Station 3 (i.e., Saih Al Salem station), located nearest to the desert, showed the lowest minimum and the highest maximum temperatures (i.e., highest thermal excursion) compared to the other two stations which, being located near the sea, present a local microclimate that is influenced by the Arabic Gulf. The minimum temperatures were recorded by station 3 in February 2014 and 2018 and were lower by about 7.6 ◦C and 2.4 ◦C than those recorded by Dubai Airport station and Al Maktoum Airport station, respectively. In July 2014 and 2017, the maximum temperatures recorded by station 3 were higher by about 2.2 ◦C and 2.3 ◦C than those recorded by the stations at Dubai Airport and Al Maktoum Airport, respectively.

Figure 3 summarizes, with the box plot representation, the air temperature measured by the three weather stations for the entire period of investigation (i.e., 2014–2018) and for each year separately.

**Figure 3.** Box plot of the air temperature for the urban, suburban, and rural areas in Dubai for 5 years (i.e., 2014–2018).

As expected, each station has an air temperature profile slightly different from the other two, despite their proximity to each other. The different locations of the stations with respect to the desert, the built area, and the coast, produced significant differences in terms of air temperature. The synoptic conditions (i.e., the hot air coming from the desert, the cool air coming from the sea or sea breeze) that characterize the entire area interact in a different way in the proximity of each meteorological station.

According to the collected relative humidity data, which affects the local microclimate, the Summer season is characterized by around 90% humidity. The maximum value recorded in all three stations is 100%, while the minimum and average values range from 1% to 4% and from 48% to 53%, respectively, where as expected, the highest values are experienced in the urban and suburban areas that are located closer to the sea, with a negative impact on human activities and outdoor thermal comfort.

A limited presence of clouds was observed in the urban and suburban areas of Dubai during the monitored period, with a clear annual and seasonal variability of the cloud coverage. In the 5 years of monitoring, the cloud cover rarely reached the maximum value of 9 octas (i.e., sky obstructed from view). Only for 14% and 12% of the time, the sky was between 4 octas (i.e., sky half cloudy) and 9 octas during the monitoring period in the urban and suburban areas, respectively.

In contrast, the total absence of cloud cover (i.e., 0 octas) was recorded very frequently along the monitored period (i.e., 67% and 69% of the time in the urban and suburban areas, respectively) leading to high values of the incident solar radiation, and as a result, to a large amount of the heat absorbed by the built environment during the daytime. This absorbed heat is then released into the atmosphere during the night time. Under these conditions, the nocturnal radiative cooling will be less effective in the built areas with respect to the desert rural area that easily cools down. This contributes to the night-time UHI phenomenon, where the temperature difference between the urban and rural areas is positive at night.

When the sky is completely obstructed from view (i.e., 9 octas), the night-time UHI intensity T1–T3 and T2–T3 reach the maximum (minimum) value of 7.1 ◦C (0.3 ◦C) and 3.4 ◦C (−0.9 ◦C), respectively, when measured in correspondence to station 1, and of 9.4 ◦C (0.3 ◦C) and 7.8 ◦C (−3.0 ◦C), respectively, when measured in correspondence to station 2. When the sky is completely clear (i.e., 0 octas), the night-time UHI intensity T1–T3 and T2–T3 reach the maximum (minimum) value of 11.5 ◦C (−3.3 ◦C) and 13.0 ◦C (−7.7 ◦C), respectively, when measured in correspondence to station 1, and of 11.5 ◦C (−3.3 ◦C) and 13.0 ◦C (−8.9 ◦C), respectively, when measured in correspondence to station 2. Thus, the night time UHI variability range is higher under clear sky conditions (i.e., 0 octas) than under cloud-covered sky conditions (i.e., 9 octas). The same result is obtained for the UHI intensity value, as can be observed in Figure 4, which shows the existing relation between the UHI intensity T1–T3 and T2–T3 and the cloud cover for the entire monitored period of 5 years and for the urban and suburban areas.

**Figure 4.** Comparison between the urban heat island (UHI) and the cloud cover relation: UHI intensity T1–T3 and cloud cover from station 1 (**top left**), UHI intensity T2–T3 and cloud cover from station 1 (**top right**), UHI intensity T1–T3 and cloud cover from station 2 (**bottom left**), UHI intensity T2–T3 and cloud cover from station 2 (**bottom right**).

#### *3.2. Daytime and Night-Time UHI Intensity and Frequency Distribution*

Figure 5 shows the daytime (i.e., 6 a.m.–9 p.m.) and night-time (i.e., 9 p.m.–6 a.m.) boxplot of the canopy UHI intensity T1–T3 and T2–T3 for the entire period of five years considered in this study.

**Figure 5.** Box plot of the daytime and night-time UHI intensity for 5 years (i.e., 2014–2018): (**a**) daytime UHI intensity T1–T3, (**b**) daytime UHI intensity T2–T3, (**c**) night-time UHI intensity T1–T3, (**d**) night-time UHI intensity T2–T3.

The daytime and night-time UHI T2–T3 show higher maximum values of 11.7 ◦C and 13 ◦C compared to the daytime and night-time UHI T1–T3 values which are 11.3 ◦C and 11.5 ◦C, respectively (Figure 5 and Table 4). The average and median values recorded for the daytime and night-time UHI intensities T1–T3 and T2–T3 are shown in Table 4 together with the maximum and minimum values.


**Table 4.** Daytime and night-time UHI intensities T1–T3 and T2–T3 values for 5 years (i.e., 2014–2018).

It is worth noting that, considering the period of interest and as summarized in Table 4, the urban and suburban areas of Dubai were in average warmer at night than the rural area by 4.6 ◦C and 1.3 ◦C, respectively. On the contrary, during the day, the rural area was in average warmer than the urban and suburban areas (i.e., 0.4 ◦C and 1.1 ◦C, respectively).

The frequency distribution of the daytime and night-time values for the UHI intensity T1–T3 and T2–T3 given in (Figure 6) help to understand the distribution of the values representing the UHI intensity during the period of interest.

**Figure 6.** Frequency distribution of the daytime and night-time UHI intensity for 5 years (i.e., 2014–2018): daytime UHI intensity T1–T3 (**top left**), night-time UHI intensity T1–T3 (**top right**), daytime UHI intensity T2–T3 (**bottom left**), night-time UHI intensity T2–T3 (**bottom right**).

Figures 5 and 6 show that the UHI intensity is higher at night than during the daytime in both urban and suburban areas. The urban area presents higher values of average UHI intensity with respect to the suburban area both during the night (i.e., 3.3 ◦C of difference) and the day (i.e., 0.7 ◦C of difference).

#### *3.3. UHI Intensity T1–T3 and Climate Parameters Collected by Station 1*

The air temperature and wind speed measured by station 1, located in the Dubai International Airport, were investigated for the six clusters of different ranges of wind direction in relation with the canopy urban heat island intensity (UHI) calculated as the difference between the air temperature measured at the Dubai International Airport (i.e., station 1) and at Saih Al Salem (i.e., station 3), where station 1 is considered as located in an urban area and station 3 is considered as located in a suburban area (Figure 7).

**Figure 7.** Representation of the UHI calculation and the station (i.e., urban station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

Figure 8 shows a comparison of the relation between the hourly canopy UHI intensity (UHI intensity T1–T3) and the hourly air temperature measured by station 1 under a different cluster of wind directions. A moderate negative correlation between the temperature and the UHI intensity was found for cluster 1 (i.e., wind direction between 260◦ and 330◦ from the seaside) (Figure 8a) where the UHI ranges between 11.5 ◦C and −12.8 ◦C, cluster 3 (i.e., wind direction between 250◦ and 210◦ from the coastal area) (Figure 8c) where the UHI ranges between 10.1 ◦C and −7.8 ◦C, and cluster 4 (i.e., wind direction between 200◦ and 150◦ from the desert) (Figure 8d) where the UHI ranges between 11.4 ◦C and −11.2 ◦C. The temperature for these three clusters fluctuates between 47 ◦C and 12.3 ◦C. A weak negative correlation between the temperature and UHI intensity was found for all the remaining clusters except for cluster 5 (i.e., wind direction located between 20◦ and 340◦) (Figure 8e) where no correlation was found. The results show that the UHI intensity varies with different wind directions. When the wind is blowing from the desert (i.e., cluster 2) (Figure 8b), the temperature and the UHI are almost independent. Despite the positive impact of the sea breeze, coastal cities suffer from the UHI [41]. Different experimental and numerical investigations have shown the impact of the sea on the development of the UHI in coastal cities [42,43]. The temperature in station 1 is affected by UHI when the wind is blowing from the seaside. It seems that, when there is a sea breeze combined with high temperatures, the negative UHI value is high (Figure 8a). This is due to the fact that the high temperatures recorded by station 1 are mitigated by the sea breeze more than the temperatures recorded by rural station 3, which is closer to the desert and typically reaches higher temperatures during hot days. This pattern is due to advection from the sea breeze cooling mechanism and the flow of air, which is affected by buildings. These results align with those of similar studies in Athens and Sydney [2,7,44,45].

Figure 9 shows a comparison of the relation between canopy UHI intensity (UHI intensity T1–T3) and the wind speed measured by station 1 under different clusters of wind directions.

For the wind speed, as for the air temperature, a negative slope of the regression line (*p*-value < 0.05) is evident in all clusters except in cluster 5 (Figure 9e) where there is no correlation. In this case, the correlation coefficient (R) between the UHI and the wind speed is negative and weak (ranging between −0.2–−0.4 for all clusters) regardless of the wind direction. The results show that the wind speed is lower when the wind blows from the coastal area (i.e., clusters 3 and 6) (Figure 9c,f) with average wind speeds of 10 km/h and 11 km/h, respectively. It reaches higher values when the wind blows from the sea (i.e., cluster 1) (Figure 9a) with average wind speeds of about 18 km/h due to the energy produced by the sea breeze. The results show that wind speed has an impact on the magnitude of the UHI. Results show that the low values of wind speed are more conducive to the development of a UHI [46]. Thus, the data from station 1 shows an inverse relation between the wind speed and UHI

intensity when the wind speed increases for all clusters except cluster 5. This relation is consistent with several studies conducted in various regions [2,47–49]. In cluster 5 (Figure 9e), with the wind from the North, no correlation was found between the wind speed and UHI intensity [2].

**Figure 8.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

**Figure 9.** Comparison between the UHI and the wind speed relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.4. UHI Intensity T1–T3 and Climate Parameters Collected by Station 2*

The air temperature and wind speed measured by station 2 were investigated for the six clusters of wind direction in relation to the UHI calculated as in the previous section and as shown in Figure 10.

A strong negative correlation between the air temperaturemeasured by station 2 and (UHIintensity T1–T3) was found for cluster 1 (Figure 11a) and cluster 3 (Figure 11c), where the UHI values range from 10.7 ◦C to −12.8 ◦C and the temperatures range between 48.2 ◦C and 12.4 ◦C for both clusters. The values of the correlation coefficient comprise about 40% of the data. Regarding the other clusters, there is a weak and negative linear relation between the temperature and the UHI intensity as shown in Figure 11.

As evident in Figure 12, the canopy UHI intensity has a moderate negative correlation with wind speed, averaging around 14 km/h, in cluster 5 (Figure 12e) where the wind comes from the seaside direction. The other clusters in the same station show a weak and negative correlation with close (R) values varying between −0.2 and −0.5. The results indicate that the high values of wind speed decrease the differences in the temperatures for all clusters, as observed in the study of Sydney [2] as well as in [50]. Additionally, the impact on UHI intensity in terms of all wind directions is minimized due to the effects of the two parameters. An increase in wind speed in the urban area enhances the heat flux, whose heating impact on the urban environment is stronger than the enhanced cooling effect [49,51]. It could also be that an urban atmosphere can absorb the heat from solar radiation during the daytime. These outcomes are supported by the results of previous studies [52,53]. As shown, the station recorded a maximum ambient temperature of 48.5 ◦C in July 2015, while the reference station recorded 50.2 ◦C for the same year.

**Figure 10.** Representation of the UHI calculation and the station (i.e., the suburban station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

**Figure 11.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

**Figure 12.** Comparison between the UHI and the wind speed relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.5. UHI Intensity T2–T3 and Climate Parameters Collected by Station 1*

The air temperature and wind speed measured by station 1 were investigated for the six clusters of different ranges of wind direction in relation with the canopy urban heat island intensity (UHI) calculated as the difference between the air temperature measured at the Al Maktoum International Airport (i.e., station 2) and at Saih Al Salem (i.e., station 3) (Figure 13).

**Figure 13.** Representation of the UHI calculation and the station (i.e., the urban station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

Figure 14 shows a moderate negative linear relation between the temperature and the (UHI intensity T2–T3) for cluster 1 (Figure 14a). The temperature for this cluster ranges between 46.1 ◦C and 13 ◦C with an average value close to 30.8 ◦C and the UHI ranges from 12.4 ◦C to −13.5 ◦C. For the five other clusters, the canopy UHI intensity shows a weak negative correlation with temperature, with few variations. Here, the synoptic climate conditions associated with the advection and convection phenomena play a role in this mechanism as does the additional anthropogenic heat in the area. These findings are supported by several other studies [2,7,54,55]. Other previous studies have reported that the daytime intensity of the heat island is reduced under specific synoptic conditions [56,57].

**Figure 14.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

In Figure 15, the (UHI intensity T2–T3) ranges between 11.6 ◦C and −13.4 ◦C showing a moderate negative correlation when the wind speed ranges from 39 km/h to 0 km/h, with an average value close to 15 km/h when the wind is coming from the North as in cluster 5 (Figure 15e). The data of the cluster demonstrate a 22.8% variation in the UHI intensity compared with the other clusters which have a very weak correlation with the UHI intensity as they are close to zero. Moreover, in regard to western and north-western wind directions, there is a negative relation between the UHI and both the temperature and wind speed, similar to the findings of other studies [2].

**Figure 15.** Comparison between the UHI and the wind speed relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.6. UHI Intensity T2–T3 and Climate Parameters Collected by Station 2*

The air temperature and wind speed measured by station 2 were investigated for the six clusters of wind direction in relation to the UHI calculated as described in the previous section and as shown in Figure 16.

**Figure 16.** Representation of the UHI calculation and the station (i.e., the suburban station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

Cluster 1 (Figure 17a) and cluster 3 (Figure 17c) show a moderate negative correlation between the temperature and (UHI intensity T2–T3). The temperature in both clusters ranges between 48.2 ◦C and 12.2 ◦C. However, in the other clusters, there is a negative weak correlation between the temperature and the UHI intensity, as shown in Figure 17.

**Figure 17.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

Figure 18 shows a weak negative correlation which means little relation between the wind speed and UHI intensity; it is close to zero in all clusters [2] for Al Maktoum International Airport (station 2). The results show no significant difference between station 1 and station 2 regarding the climate parameters and the UHI intensity phenomenon. In the surrounding area of station 1, the prevailing wind is coming from the sea. It is clear that when the wind is coming from the seaside or the desert, due to the higher energy embedded in the synoptic conditions, the wind speed will be higher than when the wind direction is parallel to the coast (cluster 3) (Figure 18c) These conclusions are supported by previous studies undertaken in Asian and Australian cities [2,40,49,58].

**Figure 18.** Comparison between the UHI and the wind speed relation under different cluster of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.7. UHI Intensity T1–T3 and Climate Parameters Collected by Station 3*

As done previously with both stations 1 and 2, the same procedure was followed to determine the temperature and wind speed for Saih Al Salem (i.e., station 3) and their relation with the UHI magnitude, calculated as the difference between the temperature measured at the Dubai International Airport (i.e., station 1) and at the Saih Al Salem area for all clusters of wind direction (Figure 19).

Figure 20 shows that for all clusters, there is a strong negative correlation between the UHI intensity T1–T3, ranging from 11.5 ◦C to −12.8 ◦C, and the air temperature, fluctuating between 50.8 ◦C and 4.7 ◦C, with an average of 28.3 ◦C. To recap, cluster 1 (Figure 20a) is from the sea direction, cluster 2 (Figure 20b) from the desert area, and cluster 3 (Figure 20c) from the coastal area. In terms of this parameter, there is a variation from 55% to 65% in UHI intensity among these three clusters. When the temperature increases, so does the intensity of the UHI.

**Figure 19.** Representation of the UHI calculation and the station (i.e., the rural station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

**Figure 20.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

The slope of the regression line (*p* value < 0.05) shown in Figure 21 indicates that for all clusters, there is a moderate negative relation between the wind speed and the UHI intensity. For all six clusters, the results indicate a relation ranging from moderate to strong between the UHI intensity and the two climate parameters in Saih Al Salem station. Synoptic weather conditions could have an impact on the parameter mechanisms as could the open desert space criterion. Numerous previous studies have reported the same results [2,47,54].

**Figure 21.** Comparison between the UHI and the wind speed relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.8. UHI Intensity T2–T3 and Climate Parameters Collected by Station 3*

The temperature and wind speed measured by station 3 were investigated for the six clusters of wind direction as in the previous sections. The UHI was calculated as the difference between the temperature measured by station 2 (i.e., Al Maktoum International Airport) and station 3 (i.e., Saih Al Salem) as shown in Figure 22.

**Figure 22.** Representation of the UHI calculation and the station (i.e., the urban station under the yellow circle) selected for the investigation of the climate parameters, i.e., air temperature (T) and wind speed (WS).

The scatter plots in Figure 23 show strong negative correlations for the different clusters between the UHI intensity T2–T3 and the temperature in cluster 1 (Figure 23a) from the sea direction, cluster 3 (Figure 23c) from the coastal area, cluster 4 (Figure 23d) from the desert side, and cluster 5 (Figure 23e) from the sea. The data demonstrates about 47%, 57%, 41%, and 37% variation in the canopy UHI, respectively. For the other two clusters, there is a weak and negative linear association between the temperature and UHI intensity; also, the temperature had independent effects on UHI intensity [2,7,55,59].

**Figure 23.** Comparison between the UHI and the temperature relation under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

As shown in Figure 24, the slope of the regression line (*p* value < 0.05) indicates a moderate negative correlation between the wind speed and the canopy UHI magnitude in cluster 3 (Figure 24c), cluster 4 (Figure 24d) and cluster 5 (Figure 24e). The UHI intensity for these three clusters is between 13 ◦C and −13.5 ◦C, while the wind speed ranges from 51 km/h to 0 km/h, with an average value close to 9 km/h. The results indicate moderate to strong relations between the UHI magnitude and the two

climate parameters for this station. These results may be due to the presence of different synoptic weather systems of heating mechanism, a desert wind with a cooling mechanism, and coastal wind, which are created by the combination of climate parameters. These conclusions are compatible with those of previous studies conducted in various regions [2,47,55].

**Figure 24.** Comparison between the UHI and the wind speed relation under different cluster of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

#### *3.9. Correlation between UHI Intensity T1–T3 and T2–T3 and Climate Parameters*

Figure 25 shows a comparison between the UHI intensity T1–T3 and UHI intensity T2–T3 and the hourly air temperature measured by station 1 and station 2 under the respective clusters of wind directions.

**Figure 25.** *Cont.*

**Figure 25.** Comparison between the UHI and the air temperature relation measured in the urban (green) and suburban (orange) areas under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

A strong and negative correlation coefficient (R) between the temperature and UHI intensity T1–T3 and T2–T3 was found in Al Maktoum International Airport when the wind was coming from the sea. Hence, the temperature only demonstrates about 32% variation in the urban heat island intensity. Whereas, when the wind is coming from the desert side (cluster 2) (Figure 25b), there is no correlation between the temperature and the intensity of UHI at station 2 (i.e., Al Maktoum International Airport), which is close to zero. These results are aligned with those of other studies conducted in different regions [2,49,58].

Figure 26 compares the correlations between the wind speed and the canopy UHI intensity for both stations through the same procedure used for measuring the parameters of temperature and UHI intensity.

**Figure 26.** *Cont.*

**Figure 26.** Comparison between the UHI and the wind speed relation measured in the urban (green) and suburban (orange) areas under different clusters of wind directions: (**a**) cluster 1, (**b**) cluster 2, (**c**) cluster 3, (**d**) cluster 4, (**e**) cluster 5, and (**f**) cluster 6.

Results show that when the wind direction is coming from the sea, there is a weak negative correlation between the wind speed and the UHI intensity. For cluster 1 (Figure 26a), the wind speed is responsible for only a 12% variation in the UHI magnitude. However, when the wind is coming from the desert side as in cluster 2 (Figure 26b), there is no correlation between the wind speed UHI intensity. These results are supported by those of other studies [2].

#### **4. Discussion**

Urban areas are facing several climate issues, one of which is the urban heat island (UHI) phenomenon. Thus, to understand the characteristics of UHI and its effects on the built environment, this study investigated the correlation between the canopy UHI and two meteorological parameters, air temperature and wind speed, with given specific ranges of wind direction. This was done by collecting and analyzing the hourly data from three meteorological stations situated in different areas of Dubai (i.e., urban, suburban, and rural areas).

It has been observed that the UHI intensity is larger at night in both urban and suburban areas and that the urban area presents the highest values of average UHI intensity. The main reason for the negative UHI at night is the presence of the heat that, given the intense incident solar radiation, remains trapped in the urban and suburban areas according to the geometry and density of the built environment and the thermal-optical properties of the urban surfaces. During the day, the built areas in Dubai act as urban cool islands showing negative UHI intensity values.

The analysis results show that canopy UHI intensity varies with wind direction and meteorological conditions. This has been confirmed by other studies carried out in cities with different climatic conditions [60,61]. Moreover, in regard to UHI intensity T1–T3, the data from the urban meteorological station indicated that air temperature and UHI were almost independent when the wind was coming from the desert (i.e., from the East and South-East), while the temperature is affected by UHI when the wind is flowing from the sea (i.e., West and North-West) thanks to the cooling mechanism of the sea breeze and the short distance of the seashore to the station 1 [41,43]. It could also be influenced by the building's orientation. These results are consistent with the findings of studies conducted in Sydney where the eastern suburbs benefited from the significant cooling mechanism of the sea breeze and the variations in UHI were a function of the prevailing climate conditions [2]. Moreover, it appears that a negative UHI intensity exists during the day and the night (i.e., the temperature in the urban area is lower than the temperature in the rural area), which is also observed in many cities. Several factors could be the cause of negative UHI values. In this case, two of the main reasons could be the hot air that, coming from the desert, increases the air temperature in the rural area more than in the urban area and the presence of the sea breeze [9,10] mitigates the urban overheating when the cold wind blows from the seaside. This confirms the many experimental investigations about the effect of the sea breeze on the magnitude of the UHI [43]. In addition, several studies have demonstrated that the flow of the sea breeze across the city is delayed by the UHI because there is a stagnation zone over the built environments [45,62]; other studies conclude that UHI could accelerate the front of the wind blowing from the seaside and moving across a city, more than in rural areas where it has little or no effect [63,64].

The results of this study, which focuses on wind speed, proved that wind speed has an impact on the magnitude of UHI [58] depending on the prevailing climatic conditions and the wind direction. Similar to the study of Sydney in 2017, a relation between the wind speed and UHI intensity was identified in this study [2]. Moreover, other studies in Chicago, Salamanca, and Granada explain the impact of UHI under ambient climatic conditions such as cloud cover and wind [50,65,66].

Regarding the cloud cover, it was observed that the night-time UHI intensity in the urban and suburban areas reaches the highest values in the presence of clear sky days, due to the larger amount of heat that is absorbed by the built area during the day and then released at night.

In the urban station (i.e., station 1), a reverse correlation between the wind speed and UHI was observed. For most wind directions, when the wind speed increased, the UHI decreased. However, when the wind is coming from the North, the correlation is close to zero. These results are consistent with the findings of previous studies that have examined the impact of wind speed on UHI magnitude in various areas such as Ibadan and Ulaanbaatar [3,47,48].

Moreover, a significantly strong and negative correlation was observed in the suburban area (i.e., station 2) between the levels of the air temperature and wind speed and the magnitude of the UHI. It was found that the effect of UHI intensity in all wind directions is decreasing or even minimized with the increase in the temperature and especially of the wind speed. A possible explanation is that when the wind speed increases, the temperature variations (i.e., between 48.6 ◦C and 12.3 ◦C with an average value of about 29.5 ◦C) gradually decrease since higher wind speeds are connected with

the advection of air masses that tend to minimize the temperature differences. This relation has been observed in several cities around the world [51,67], as well as in the urban station (i.e., station 1).

It was observed that when the wind speed is low or zero, the UHI has a positive value because the temperature in the urban area (where the urban surface easily absorbs solar radiation) is higher than in the rural area (where the natural surface has less absorption). When the wind speed is very high, the UHI effect disappears due to the heat dissipation produced by the wind. Similarly, the study of Athens found that the high temperatures resulted from solar radiation when the wind speed was low [52].

The data collected by reference station 3 (i.e., Saih Al Salem), which is located in a rural desert village far from the seacoast, show a moderate to a strong correlation between the UHI and the temperature and wind speed, for all wind directions. This is similar to the relation found in the studies of Ibadan and Sydney [2,47].

In the case of UHI intensity T2–T3, the UHI magnitude varies between 11.5 ◦C and −12.8 ◦C. It is found that the results for the urban station are similar to those observed in the suburban station through the correlation methodology used for the UHI magnitude and the climate parameters. It was also observed that there were no significant differences between the two stations during the 5 year period of interest. This indicates that when the wind comes from the sea (i.e., western and north-western) directions, there is a moderate negative relation between the UHI and both the temperature and wind speed. The negative results could be due to the additional anthropogenic heat and the cooling rates caused by evaporation. Data analysis showed an inverse relation between the wind speed and the UHI intensity where when the former increased, the latter decreased. In contrast, for all other wind directions, the analysis indicated no correlation between the UHI intensity and both temperature and wind speed. These results are aligned with those of previous studies conducted in Asian and Australian cities [58].

The results show that, in the case of station 3, there is a significant correlation between the meteorological parameters and UHI. A moderate to a quite strong relation is found between this UHI magnitude and the two parameters (i.e., temperature and wind speed) for all the wind directions compared to the previous case of the UHI intensity T1–T3. These results could occur with synoptic weather and specific conditions created by the combination of climate parameters, which is consistent with the findings from the previous studies carried out in various regions [2,47].

The results of this study indicate that wind speed and temperature have an impact on the urban heat island depending on the prevailing weather conditions on whether the wind comes from the sea or from the desert. However, the urban area is the one most affected. In fact, the relations, found in this study, between the meteorological parameters and UHI, have several implications for future urban planning. Strategies are needed to reduce the temperatures and have better cooling systems within urban areas in order to control urban heat islands. There is a need to establish guidelines for urban planning and design so that urbanization is sustainable. Worth noting is the fact that one of the challenges faced in this study was the limited availability of meteorological parameters and adequate monitoring data covering a greater number of years.

#### **5. Conclusions**

The aim of this paper was to examine the correlation between the intensity of the canopy urban heat island and the various meteorological parameters (e.g., temperature, wind speed, and wind direction) when the wind is coming from different directions such as from the seaside and the desert side. Hourly data of climate parameters were derived from five years (2014–2018) of monitoring from three meteorological stations located in an urban, suburban, and a rural area of Dubai. Six clusters of wind directions were identified to perform a cluster analysis on the available set of climatic data.

It was found that the temperature and wind speed results differed between the western (seaside) and eastern (desert side) parts of Dubai. The UHI intensity varies with the wind direction and meteorological conditions. Regarding the UHI intensity T1–T3, it was found that when the wind in

the urban area was coming from the desert, the temperature and the UHI were almost independent. However, a relation exists when the wind is coming from the seaside (i.e., sea breeze). Therefore, the wind speed influences the UHI intensity depending on the synoptic climate conditions and wind directions. A reverse correlation was found between the wind speed and UHI intensity for all wind directions, except the North direction where no correlation was found in the urban area. In the suburban area, an inverse relation was observed between the wind speed and the magnitude of UHI, for all wind directions. At the same time, a moderate to strong correlation was found between the UHI and both temperature and wind speed for all wind directions in the rural area.

In contrast, in the case of the UHI intensity T2–T3, it was observed that in the urban and suburban areas, there was a similar correlation between the UHI and the climate parameters when the wind was blowing inland from the sea due to additional anthropogenic heat and evaporation. No correlation was observed for all other wind directions. A moderate to a strong association was found between the UHI and the two climate parameters for all the wind directions in the open zone of the desert, where the reference station (i.e., rural area) is located, due to a combination of climate parameters and synoptic conditions.

From that perspective, the wind that flows from the seaside or the desert side is completely different as the air coming from the sea is cooler, which decreases the temperature and minimizes the UHI. Thus, an onshore wind cools the air temperature of the area depending on the wind speed; hence, the higher the wind speed, the lower will be the magnitude of the UHI. On the other hand, the wind coming from the desert is warmer, producing dry hot weather conditions and increasing the temperature. However, occasionally the wind blowing from the desert has different effects on temperature depending on other factors.

This study, investigating the canopy UHI phenomenon in the coastal and desert areas of Dubai, contributes to a deeper understanding of the local microclimate and urban overheating in desert regions that is essential to inform the climate-resilient urban design and planning. It was observed that the sea breeze plays a vital role in the urban zone and the seashore area, contributing to the mitigation of the summer urban and suburban warming. The results of this study confirmed that additional mitigation strategies for heat reduction should be implemented in desert cities to reduce the thermal stress in the urban ecosystem and avoid many issues that can be caused by high urban temperatures such as heat-related illness and mortality and uncomfortable outside areas.

Further studies should be performed to better understand the overall UHI phenomenon in Dubai, for example also investigating the surface UHI and the boundary UHI with the aim to cover different scales of investigation.

**Author Contributions:** Conceptualization, G.P. and M.S.; methodology, M.S.; formal analysis, A.M.; investigation, A.M., G.P. and M.S.; resources, E.T.; data curation, A.M. and G.P.; writing—original draft preparation, A.M. and G.P.; writing—review and editing, A.M., G.P., E.T. and M.S.; visualization, A.M. and G.P.; supervision, G.P. and M.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors express their gratefulness to the UAE National Center of Meteorology (NCM) for providing the data used in this study.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **On the E**ffi**ciency of Using Transpiration Cooling to Mitigate Urban Heat**

#### **Kai Gao 1,\*, Mattheos Santamouris 1,2 and Jie Feng <sup>1</sup>**


Received: 19 April 2020; Accepted: 29 May 2020; Published: 1 June 2020

**Abstract:** Trees are considered to be effective for the mitigation of urban overheating, and the cooling capacity of trees mainly comes from two mechanisms: transpiration and shading. This study explores the transpiration cooling of large trees in urban environments where the sea breeze dominates the climate. In the experiment, sap flow sensors were used to measure the transpiration rate of two large trees located in Sydney over one year. Also, the temperature difference between the inside and outside of the canopy, as well as the vertical temperature distribution below the canopy, were measured during summer. In this experiment, the temperature under the canopies decreased by about 0.5 degrees from a 0.5 m height to a 3.5 m height, and the maximum temperature difference between the inside and outside of the canopy was about 2 degrees. After applying a principal component analysis of multiple variables, we found that when a strong sea breeze is the primary cooling mechanism, the sap flow still makes a considerable contribution to cooling. Further, the sea breeze and the transpiration cooling of trees are complementary. In conclusion, the characteristics of synoptic conditions must be fully considered when planting trees for mitigation purposes. Since the patterns of sea breeze and sap often do not coincide, the transpiration cooling of trees is still effective when the area is dominated by sea breeze.

**Keywords:** transpiration cooling; coastal cities; sap flow

#### **1. Introduction**

Climate change and the urban heat island effect have caused a worldwide issue of urban overheating [1]. Urban overheating will increase the total energy consumption and peak electricity demands, thereby deteriorating outdoor and indoor thermal comfort, as well as the health of the public [2–4]. In order to mitigate urban overheating, researchers have proposed a variety of options: integrating cool materials, such as reflective materials [5] and radiative cooling surfaces, into facades and roofs [6]; using cool pavement on roads [4,7]; and expanding green spaces in the urban environment [8].

In the past few decades, the mitigation effect of trees has been proven effective by a large number of studies [9–11]. The difference between the ambient temperature and the temperature in the canopy is usually used as an essential indicator of plant cooling capacity. For instance, the daily average difference of the air temperature between the shady area of a street tree and an open space is 0.9 ◦C in Melbourne, Australia [12]; 1 ◦C in Munich, Germany [13]; and up to 2.8 ◦C in Southeast Brazil [14]. The cooling effect of trees mainly comes from shading and transpiration. The cooling effect caused by shading at a macroscopical level is very limited, as the albedo of the tree canopy is very similar to that of the ground surface, indicating that the total energy balance is not significantly affected by the presence of trees. Transpiration, on the other hand, alters the energy balance of the whole area by converting sensible heat flux to latent heat flux and has a cooling effect both locally and on a large scale. When sufficient water is supplied, the transpiration rate can be maximized, providing a considerable cooling effect. Plenty of studies have shown that the air temperature over vegetated land is significantly reduced after applying irrigation [15–17].

Since research in forestry and agronomy usually focuses on large-scale crops and woodland, the heat flux over the land is easier to quantify to determine the transpiration cooling effect. However, in cities, where the surface is highly heterogeneous, the heat flux method is not suitable. Existing studies on urban tree cooling often use the sap flow method to measure the transpiration rate of a single tree or a small number of trees.

When switching to microscale studies, the transpiration cooling effect of urban trees can be explored more in depth. Researchers have found that the cooling effect of trees is closely related to their species [18–20] and hydraulic architecture. The leaf area index (LAI) is considered to be a particularly critical parameter that determines the cooling capacity of trees. Trees with a high LAI can achieve more efficient shading and exhibit higher transpiration potential.

However, microscale studies also face difficulties. The temperature difference indicator reflects the cooling capacity of a plant to a certain extent, but this indicator can be environmentally sensitive. Remoting sensing research [21] has found that in the core areas of several U.S. cities, the urban tree cooling effect was significantly affected by the surrounding thermal conditions. Compared with studies in forestry, experiments conducted in cities usually found larger temperature decreases under the canopy. For instance, an experiment in the Munich urban area [22] reported that under the canopy, temperature reduced by 3.5 degrees, from 0.5 m above the ground to 4 m above the ground. However, in a *Picea abies* L. forest, research [23] found that the temperature only reduces by one degree every 4 m vertically under the canopy. Possibly, the warming effect of the paved street prevented the downward penetration of the tree cooling effect in the experiment conducted in Munich [22]. This possibility emphasizes the influence of the energy balance of the local environment on the magnitude of the vertical temperature difference. Moreover, since a tree cooling mechanism differs from that of short vegetation, the vertical temperature cannot be used as the only indicator for transpirational cooling. Short vegetation directly provides cooling on the ground, while the transpiration cooling effect of trees needs to penetrate several meters to reach the ground from the canopy. The overall cooling performance is determined only after determining both the vertical temperature difference and the horizontal temperature difference between the inside and outside of the canopy.

At the same time, weather conditions could also affect the cooling effects of trees [24]. The research in [21] observed the variation of the urban tree cooling effect under different climatic conditions, and in [25], more effective cooling was observed from trees in the dessert compared to trees on the coast. In scenarios where strong advection exists, the temperature difference under the canopy in the vertical direction and that between the inside and outside of the canopy may be affected. On the one hand, if the wind is dry and hot, the vapour pressure deficit (VPD) in the canopy will increase, and transpiration cooling will be stimulated [26]. However, continuous dry and hot wind will also blow away the cold air that is mass-produced by the trees [27]. On the other hand, if the wind is cold and humid, such as a sea breeze, transpiration can be weakened. At the same time, the canopy can directly block the wind, making it receive less cold from the wind compared to that received by the outside. If the space outside the canopy is also shaded, the temperature under the canopy can be even hotter than that outside. This possible phenomenon indicates that the effect of the wind may completely offset the cooling effect of transpiration. However, the effect of sea breeze on the cooling of trees has not been sufficiently studied, and many issues remain unclear. For example, in cities where the sea breeze is the dominant cooling mechanism, does planting more trees have an obvious cooling effect? How much tree cooling contributes to local temperature distribution? Answering such questions is crucial for mitigating overheating in coastal cities.

At the same time, some cities are affected by both desert and marine climates, such as Sydney, Australia. In Sydney, a synoptic condition exists where the east side of Sydney is adjacent to the Pacific Ocean and is affected by strong humid and cold sea breeze, while the west side of Sydney faces a vast

desert and is influenced by the dry and hot desert winds. If a large number of trees are planted to alleviate urban overheating in the city, how should we arrange these trees spatially to achieve the most effective cooling? An analysis of the impact of sea breeze on tree cooling can help answer this question.

In this paper, when discussing the effect of the transpiration cooling of big trees, weather station data are used to include the impact of Sydney's synoptic conditions, and the temperature difference between the inside and outside of the canopy was measured. The experimental site in this paper is about 2 km away from the coastline and is significantly affected by the sea breeze. As the elevation of the experimental site is higher than that of the eastern coast, the sea breeze can easily reach this area. Thus trees here are significantly affected by the sea breeze. After obtaining the transpiration rate, temperature difference, and wind speed, a principal component analysis (PCA) was used to determine the contribution of transpiration to the temperature difference inside and outside the canopy. At the same time, this research also studies the factors that affect the transpiration rates of plants. The results show that the sea breeze does play a crucial role in ooling, and sap flow also makes a considerable contribution to cooling. The analysis also shows that the daily peaks of sea breeze and sap flow do not overlap, indicating that the introduction of trees in coastal cities can complement the cooling of the sea breeze. This result answers the above questions and affirms the plant's cooling potential under sea breeze conditions.

In the study, attention was also paid to the vertical temperature gradient below the tree, and the influence of the ground surface on the vertical temperature distribution was discussed according to the measurement results. This result further emphasizes the effect of the ground surface on the cooling capacity of trees and highlights that when measuring the cooling effect of trees in a microenvironment, the vertical temperature difference cannot be used as the only criterion to assess the cooling capacity of plants; the assessments should instead be made after correcting the possible impact of the surface. We believe that this study makes an important contribution to rationalizing the layout of urban greenery to achieve a better mitigation effect.

#### **2. Methods**

#### *2.1. Experiment Design*

In order to analyse the transpiration cooling effects of large urban trees, the transpiration rate and surrounding temperature of two large Melaleuca trees were measured. The transpiration rate was measured using sap flow sensors. The studied site was located on campus at the University of New South Wales in Sydney.

The recording of sap flow and soil moisture began in early 2018 and continued until May 2019. To obtain an understanding of the trees' cooling effect, multiple temperature sensors were placed under the canopies and outside the canopies of Maleluca from January 2019 to March 2019. The temperature sensors were placed outside canopy. Since aluminum foil was also proven to be an effective solar protection measure in [28], we used it to shield the sensors from direct solar radiation. At the same time, the vertical air temperature distribution under the canopies was recorded for 2 weeks in the middle of February 2019. The weather data were extracted from the nearest weather station.

#### *2.2. Measurement Set Up*

#### 2.2.1. Site Details

Two almost-identical Melaleuca trees were investigated, as shown in Figure 1. The lawn at the university of New South Wales (UNSW) campus has dense grass and is regularly irrigated. Hence, the soil in the studied site has remained moist for the entire period of measurement, ensuring that the transpiration of trees is not limited by soil water content.

**Figure 1.** (**a**) site plan of the studied area; the red circles are the canopies of the researched trees. (**b**) weather station information.

The essential information of the studied trees is measured and listed in Table 1. The canopy of the trees starts at around 3 m of height. Site weather data were extracted from the nearest weather station (Randwick weather station, 33.92◦ S, 151.2◦ E), which is in similar condition to the experimental site. The vapour pressure deficit used in our experiment was calculated from the vapour pressure and saturated vapour pressure of the weather station observations. The air temperature, wind direction, and speed were also extracted from the station observation. The site soil surface moisture and soil surface temperature was monitored at 5 cm beneath the soil surface using an Edaphic TEROS-11 model sensor.


#### 2.2.2. Sap Flow Method

In this study, the transpiration rate of trees was reflected by the sap flow rate, and we used Edaphic SF3 model heat pulse sensors to record the sap flow rate. The sap flow rate was recorded in a 30 min time interval, which provided a balance between battery life and experimental accuracy. A sensor was installed on the north side of each tree.

The sensor used analog signals to record the time, and the sensor received the heat pulse signal after the heat pulse was generated from the heat source. Both the sap flow data and soil moisture were recorded using a CR100 datalogger from Campbell Science.

In order to obtain more accurate sap flow data, two methods were used to calculate the sap flow velocity. The Marshall heat pulse method [29] was applied to calculate the slow sap flow rate, while the T-max method [30] was used to calculate the high sap flow rate. The T-max method is written as

$$V\_{ll} = \frac{\sqrt{\mathbf{x}^2 - 4kt\_m}}{t\_m},\tag{1}$$

where *Vh* is the heat pulse velocity(cm/hr), *x* is the distance to the axial temp, *k* is the thermal diffusivity, and *tm* is the time it takes the sensor to reach its max temperature after a heat pulse is generated by a heater.

To calculate the slow sap flow rate, the heat pulse method was used, which is written as

$$V\_h = \frac{k}{\chi} \ln \left( \frac{\Delta T\_d}{\Delta T\_u} \right) \tag{2}$$

where Δ*Td* and Δ*Tu* are the temperature changes recorded by the upstream sensors and downstream sensors, respectively. Later, the heat pulse velocity was transformed into sap flow velocity. Since we were unable to obtain a trunk sample from the studied trees, the calculated sap flow rate was not calibrated. Therefore, the analysis of sap flows only focused on patterns.

#### 2.2.3. Temperature Measurement

The Log tag sensors were used to record the air temperature. Two Log tag temperature sensors were set up and fixed at around 1.5 m height under the canopies; they were attached to the trunk on the south side to avoid direct sunlight. Two other Log tag sensors were placed outside the canopy at the same height as the sensors inside canopy. All Log tags were shaded from direct sunlight, and the measurement interval was 15 min.

The vertical temperature was measured using globe thermal meters. The measurements were conducted under the canopy for two weeks in February 2019, from 12:30 to 16:20 every day, with a measurement interval of 10 min. The height from 0.5 m to 3.5 m was measured with a space interval of 0.3 m, and the vertical temperature distribution was measured manually. The globe thermometers were fixed on supports (tripods). Each time after adjusting the heights of the globe thermometers, the researchers walked away to avoid interference of the human body in the measurements. The readings were recorded after they became stable. To reduce random errors, the measurement of the temperature at each height was repeated 8 times from eight different points, which were all about 1 m away from the trunk in different directions. The average value of the 8 measurements was considered to be the value of the temperature at the specific height.

All sensors were calibrated, and the error of the temperature sensors was negligible (±0.2 ◦C).

#### **3. Results**

#### *3.1. Site Analysis*

In Sydney, summer weather is dominated alternately by the wet and cold sea breeze from the east side and the hot and dry desert wind from the west side. Desert wind brings heat, but, at the same time, it facilitates evapotranspiration and improves the cooling effect of evapotranspiration. In contrast, the wet and cold sea breeze has a cooling effect but can also hinder evapotranspiration. The extra moisture produced by the sea breeze can also increase evapotranspiration resistance. In our case, the effect of the sea breeze was substantial because the study area was very close to the coastline. Figure 2 shows an analysis of the wind direction from an observation of the nearest weather station. It can be seen in Figure 2d that most of the wind came from the southeast starting at noon every day during summer 2019.

**Figure 2.** Wind direction distribution of the studied area in summer (**a**) 2018 and (**b**) 2019. Average daily wind direction distribution in summer (**c**) 2018 and (**d**) 2019.

The wind speed of the sea breeze gradually increased in the afternoon, as shown in Figure 3. The wind speed reached its maximum between 17:00 and 19:00 in the summer.

**Figure 3.** Daily average wind speed variation of the studied area in summer (**a**) 2018 (**b**) 2019.

#### *3.2. Diurnal Sap Flow Patterns*

The Diurnal period in this study is defined as the time between 6:00 and 20:00, and we analysed the average sap flow during this period. The canopy temperature is the average temperature recorded by two log tag sensors under the canopies.

As mentioned above, the UNSW campus is regularly irrigated, and the measured Melaleuca canopy is a large tree with a deep root system. Thus, the pattern of sap flow is not limited by the soil moisture content. The analysis shows that the correlation coefficient between the sap flow rate and soil moisture was smaller than 5%. Moreover, the multiple linear regression analysis shows that the diurnal sap flow patterns are positively correlated to both the temperature and the VPD, as shown in Figure 4.

**Figure 4.** Multilinear regression for sap flow, canopy temperature, and VPD.

A non-linear model, expressed as f(x) = c − a × exp(−b × x), was used to fit the relationship between the VPD and diurnal sap flow, as shown in Figure 5. When considering 95% confidence bounds, the fit results are a = 61.27 (58.54, 63.99) and b = 0.05848 (0.0534, 0.06357), while c is determined as 58.4534. At the same time, a linear regression relationship was built between the sap flow and the canopy temperature, which has an R<sup>2</sup> of 0.65, as shown in Figure 6.

**Figure 5.** The non-linear fit between the sap flow rate and VPD.

**Figure 6.** The linear fit between the sap flow rate and canopy temperature.

The Pearson correlation coefficient [31] was also calculated. The correlation coefficient between the diurnal sap flow and diurnal canopy temperature is 0.8, while the correlation coefficient between the diurnal sap flow and diurnal VPD is 0.69.

#### *3.3. Air Temperature Distribution*

#### 3.3.1. Vertical Temperature Distribution

Measurement of the vertical temperature distribution was performed between 12:30 and 16:30 since the highest daily radiation and temperatures usually occur within these hours. The measurement results show that the maximum temperature difference vertically was only about 0.5 ◦C. Figure 7 shows the vertical temperature change recorded during the experiment.

**Figure 7.** 0.5–3.5 m height vertical temperature distribution under the canopy.

3.3.2. Temperature Difference between the Inside and outside of the Canopy

A linear regression of the temperature difference between the inside and outside of the canopy showed that a stronger sap rate was associated with a larger temperature difference between the inside and outside, as shown in Figure 8a. However, the same relationship was also found in the analysis of VPD and wind speed, as shown in Figure 8b,c.

**Figure 8.** Canopy temperature difference between (**a**) sap; (**b**) VPD; (**c**) wind speed. The R squares are 0.30, 0.15, and 0.14, respectively.

At the same time, most of the variables in this study are highly correlated with each other, especially when strong wind conditions are considered. The variation of the temperature difference inside and outside the canopy can be explained only when the correlation between the observed variables is eliminated. Therefore, a PCA analysis was performed using the VPD, sap flow rate, and wind speed as input observational groups. Ultimately, three groups of components were created. The results showed that principal components 1 and 2 account for 92.04% of the variability in the data set. As shown in Figure 9, in the configuration of principal components 1 and 2, the contributions of wind speed and sap flow rate dominate with similar contributions. The details of the principle components are shown in Table 2.

A multiple linear regression with canopy temperature difference (CTD) was also performed using principal components 1 and 2, and the results are shown in Figure 10. The R square is 0.32, and principal component 1 shows a negative correlation with temperature difference.

**Figure 9.** Recentered component configuration after applying PCA to the observations. The lengths of the vectors indicate the contribution of the specific observation.



**Figure 10.** Multi-linear regression of the temperature difference and principle components 1 and 2.

#### **4. Discussion**

#### *4.1. Horizontal Cooling E*ff*ect*

In this experiment, the temperature difference between the outside and inside canopy was measured. The analysis indicates that the synoptic conditions in Sydney may have made a significant contribution to the results.

Changes in the transpiration rates of large trees depend largely on direct radiation changes, while regional synoptic conditions rule the wind in the studied area. The study area is close to the coastline and faces open spaces in both the eastern and the western directions, making the sea breeze the dominant factor in the surface energy balance despite the impact of downward shortwave radiation input. Further, the cooling effect of transpiration is negligible compared to the cold brought by the sea breeze. Studies conducted in Munich [13] reported that the sap flow has a significant relationship with the difference between the temperature under the canopy and that outside the canopy. Munich has a mild climate, and in the researched area, no climatic conditions (such as a stiff sea breeze) were observed. This indicates that the presence of a windy environment must be considered before using plants for transpiration cooling. The difference between the results of this study and the results in [13] indicates the high dependence of tree transpiration cooling on the environment, especially on local wind environments and under regional synoptic conditions.

In order to make a comparison between variables, the daily variation of several variables was calculated, as shown in Figure 11 (daily variation is calculated as the average at a specific hour equal to the sum of all the data measured that hour during the two weeks divided by the total number of data at that specific hour). The maximum daily temperature difference appears after around 17:00 (Figure 11d), which is earlier than the peak of the wind speed (Figure 3b) but later than the peak of the sap flow rate (Figure 11a). This indicates that the wind may exert influence on the temperature difference in the afternoon and delay the time of the peak temperature difference. At the same time, starting from 11 am, the difference between the canopy temperature and the air temperature remained at a similar level (Figure 11c). In our experiment, this also indicates that the wind is likely to dominate the energy balance and negate the effects of other factors in the local space under the canopy. If the wind is not the dominant factor, the variation in sap flow and solar radiation will lead to a significant peak on the curve in 11c after 11 am.

It should be noted that the temperature sensor outside the canopy was located in the downwind direction for most of the time in the afternoon, as shown in Figure 12. The northeast wind was significantly blocked by the large thick canopies and the layout, causing the wind in the downwind direction in the local environment to significantly weaken. This could be the reason for the results of our experiment: The wind cooled the canopy more than it cooled the outer areas in the downwind direction. Furthermore, if a sensor was placed in the upwind direction outside the canopy, the results indicate that the sensor was able to record a lower air temperature.

The PCA helped us eliminate the correlation between variables, showing that transpiration made similar contributions to cooling compared to the wind in the experiment. As mentioned above, the peaks of the daily sap flow, temperature difference, and wind appeared in sequence during the afternoon. This pattern suggests that in the early afternoon, when the daily maximum air temperature is most likely to occur, transpiration cooling plays a leading cooling role in the cooling mechanism. The transpiration cooling and cooling of the sea breeze complement each other in different periods under specific synoptic conditions, providing the city with a considerable cooling effect in the daytime and further emphasizing the critical impact of trees on the local climate.

However, it should be noted that after the principal value analysis, the correlation between the temperature difference and the principal component remained limited because the transpiration cooling of trees only accounts for a part of the cooling effect, and the shading effect is also an essential cooling mechanism of trees. Though the shading effect was not able to be analysed in the experiment, the effect of shading should not be ignored. In the daytime, changes in the incoming radiation intensity and direction can affect the temperature difference between the outside and inside of the canopy.

**Figure 11.** Daily variation of the temperature difference.

**Figure 12.** Canopy temperature sensor location.

Moreover, a sharp peak in the soil temperature is shown in Figure 11a. This is because the soil was receiving direct solar radiation at this moment. During the time when the the soil temperature peak occured, the recorded soil surface was not shaded by the canopy. Hence, a sharp rise in the soil surface temperature was observed. This result also underscores the importance of shading effects in tree cooling mechanisms.

#### *4.2. Vertical Cooling E*ff*ect*

In the studied area, a vertical temperature of 0.5 ◦C from 0.5 m to 3.5 m was observed under the canopy. This value is relatively small compared to the values in urban tree cooling studies, as mentioned above [22]. The surrounding surface conditions made the largest contributions to this result.

Different ground surfaces have different thermal properties, optical properties, and evapotranspiration capabilities. Therefore, under the same environmental conditions, different ground surfaces will have different surface temperatures and lead to a diverse vertical temperature and humidity distribution near the ground. Studies that distinguish vertical temperature differences are usually conducted on built surfaces like paved streets [22]. The warming effect of built surfaces enlarges the vertical temperature difference. This warming effect is considerable, especially when the trees are sparsely distributed. However, in our experiment, the trees are located on a dense grassland, where the soil surface remains moist most of the time. The transpiration of grass and direct evaporation from the soil cause a considerable cooling effect, and the cooling effect of the lawn and canopy work together, resulting in a more uniform vertical temperature distribution.

To confirm the cooling effect of the lawn on our measurements, a regression of the surface soil temperature and air temperature was developed, and the results are presented in Figure 13. According to the fitting equation, the projected soil temperature is significantly lower than the air temperature. When the air temperature is 30 ◦C, the projected soil surface temperature is 28.9 ◦C at the experimental site, and the temperature difference is 1.1 ◦C. When the air temperature is 35 ◦C, the projected soil surface temperature is 33.1 ◦C, and the temperature difference reaches 1.9 ◦C. This means that the cooling effect of the lawn did contribute to the vertical temperature distribution.

**Figure 13.** Correlation between the air temperature and soil temperature.

#### *4.3. Limitation of the Research*

Due to the complexity of the urban environment and the experimental conditions, the following limitations exist in this study. In this experiment, the sap flow rate was recorded but not calibrated. Hence, only the sap flow pattern was analyzed. The transpiration cooling ability is thus unable to be quantified since the transpiration cooling power cannot be calculated. Part of the data used in this experiment was taken from the nearest observation station. The distance between the experimental site and the nearest station was around 1.1 km, and the conditions of the station were similar to those of the experiment site. Since distance remained between the observation stations and the experimental site, the correlation coefficient between the station observation and site data inevitably decreased, although some studies have shown that it is feasible to use data from nearby observation stations in this kind of research. Considering the analysis results above, the influence of wind may cause a

significant difference in the air temperature from the east and west sides of the canopies. However, this difference was not quantified in the present measurements since only one side of the canopy was measured. Also, the absence of radiation data can also have an impact on the multi-linear fiting results of the principle components. The R square value of the current result may be lower than the fitting result using principle components that consider radiation.

Due to the complexity of the urban environment, the results of this experiment may not be suitable for all places that have a strong coastal breeze because the local wind environment is also greatly affected by the terrain and building layout. If there are large buildings present in the upwind direction, the contribution of the wind field to the cooling of the local environment could be significantly weakened, so the role of the transpiration cooling of the plants in such scenarios is more critical. However, the shading effect produced by the possible appearance of buildings may also obfuscate the shading effect of plants. These scenarios also need to be investigated. The results of this experiment are only suitable for occasions where the local wind environment is profoundly affected by weather conditions.

#### **5. Conclusions**

The results of this study emphasize the importance of climatic conditions for the transpiration cooling effects of trees in urban overheating. We found that transpiration clearly contributes to decreasing the temperature, even under strong coastal wind conditions. Therefore, in cities, where the climate is dominated by the sea breeze, the side of the city that is far from the coastal line is less affected by the sea breeze, but transpiration still plays a critical role in the cooling of these areas. At the same time, the experiment also showed that the patterns of sea breeze and sap often do not coincide and that trees and the wind can complement each other in terms of urban overheating mitigation at different times of day. Hence, even in the areas near the coast, transpiration cooling could still be useful.

Furthermore, for cities in the desert, the situation may be opposite to the situation in this article. Dry and hot desert wind can bring heat to the city. However, the desert wind will also increase the VPD of the local space, thus amplifying the transpiration rate. Therefore, in such cities, the mitigation potential of trees planted in well-ventilated areas can be maximized. The canopy barrier can reduce the invasion of the desert wind into the city and further enhance the value of the trees in alleviating urban overheating. At the same time, it should be noted that sufficient irrigation of trees in these areas may be a precondition for effective mitigation.

At the same time, the measurement results and analysis of the vertical temperature distribution emphasize the influence of the microenvironment on the cooling effect of the trees. In a highly built urban environment, the warming effect of surfaces, such as concrete sidewalks or asphalt roads, is considerable at noon in the summer, and the warm air flow near the ground surface can be substantial. If the canopies of the trees are not large and dense enough, the warming effects of the surroundings can easily conceal the shading and transpiration cooling effects of the trees, leading to a decreased cooling effect. Therefore, in urban landscape design, to achieve effective vegetation cooling in a highly built environment, trees with a high LAI and a large canopy projection area must be selected.

**Author Contributions:** Conceptualization, K.G. and M.S.; Data curation, K.G. and J.F.; Formal analysis, K.G., M.S. and J.F.; Investigation, K.G.; Methodology, K.G. and M.S.; Writing—review & editing, J.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** This research was supported in part by the City of Parramatta with the research contract "Parramatta Urban Overheating. Thanks also for the assistance from Riccardo Paolini in the experiment preparation.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **Probability Risk of Heat- and Cold-Related Mortality to Temperature, Gender, and Age Using GAM Regression Analysis**

#### **Andri Pyrgou and Mattheos Santamouris \***

Faculty of Built Environment, University of New South Wales, Sydney 2052, Australia **\*** Correspondence: m.santamouris@unsw.edu.au

Received: 7 February 2020; Accepted: 11 March 2020; Published: 11 March 2020

**Abstract:** We have examined the heat and cold-related mortality risk subject to cold and heat extremes by using a generalized additive model (GAM) regression technique to quantify the effect of the stimulus of mortality in the presence of covariate data for 2007–2014 in Nicosia, Cyprus. The use of the GAM technique with multiple linear regression allowed for the continuous covariates of temperature and diurnal temperature range (DTR) to be modeled as smooth functions and the lag period was considered to relate mortality to lagged values of temperature. Our findings indicate that the previous three days' temperatures were strongly predictive of mortality. The mortality risk decreased as the minimum temperature (Tmin) increased from the coldest days to a certain threshold temperature about 20–21◦C (different for each age group and gender), above which the mortality risk increased as Tmin increased. The investigated fixed factors analysis showed an insignificant association of gender-mortality, whereas the age-mortality association showed that the population over 80 was more vulnerable to temperature variations. It was recommended that the minimum mortality temperature is calculated using the minimum daily temperatures because it has a stronger correlation to the probability for risk of mortality. It is still undetermined as to what degree a change in existing climatic conditions will increase the environmental stress to humans as the population is acclimatized to different climates with different threshold temperatures and minimum mortality temperatures.

**Keywords:** heatwave; diurnal temperature range; time-series; relative risk; health

#### **1. Introduction**

The relationship between hot and cold temperatures and mortality from respiratory and cardiovascular causes is well established. Governments and scientists are concerned with the increased frequency of temperature extremes as they are associated with increased morbidity and mortality [1,2]. Exploration of time series data in different countries has revealed a different temperature threshold of their population [3], necessitating different adaptation measures for the avoidance of the climate-change impact and to increase the countries' capacity to function at a forthcoming temperature [4–8]. It is of utmost importance for the threshold temperature to be determined per country or prevalent climate as there is a gap in the identification of the correct course of adaptation.

Mortality risk with respect to temperature has been assessed by scientists in northeastern Europe [9–12], the USA [13,14], and China [4] by considering fixed variables such as gender and age, resulting in contrasting findings of the threshold temperature and minimum mortality temperature (MMT). MMT is defined as the temperature at which there is the lowest risk of mortality according to a probability risk assessment. The question of whether fixed factors are affecting the MMT has to be addressed carefully in the eastern Mediterranean region because of the limited number of studies showing high MMT; 29–32 ◦C [15]. Generally, MMT has been found to be lower for populations

living in colder climates and higher for populations living in warmer climates [11,14,16] with the temperature–mortality relationship described as a J-, V-, or U-shaped curve.

People acclimatize to new temperatures at varying rates and to a certain extent based on physiological parameters such as age, gender, and other prevalent health conditions. The environmental stress even within a day may be a factor of increased cardiovascular and respiratory mortality. Moreover, the effect of exposure to extreme cold or heat conditions is not limited to the specific day, but may be delayed in time [2,14,17]. The diurnal temperature range (DTR) is defined as the difference between the daily minimum and maximum temperatures, with some studies correlating a high DTR with an increase of mortality risk [17].

To fill the above research gaps, we addressed the issue of temperature-related mortality via the use of the generalized additive model (GAM) regression technique. Using the GAM, we investigated the effect of same day temperatures and the weighted average temperature of the preceding three on mortality rates. The use of the GAM model to examine the short-term mortality relationship to regional minimum temperatures, maximum temperatures, and DTR variations revealed that MMT should be calculated using daily minimum temperature values.

#### **2. Methods**

#### *2.1. Study Area and Datasets*

Hourly weather data (temperature [◦C]) and daily mortality data for 2007 to 2014 inclusive were collected for two meteorological stations [18] in the urban (35.17◦N, 33.36◦E) and rural (35.05◦N, 33.54◦E) areas of Nicosia, Cyprus. Nicosia is the capital of the island of Cyprus, located in the eastern basin of the Mediterranean Sea with a hot summer Mediterranean climate and hot semi-arid climate (in the northeastern part of island), according to the Köppen climate classification signs Csa (Mediterranean hot summer climates) and BSh (Hot semi-arid climates) [19], with warm to hot dry summers and wet winters.

The daily mortality counts were gathered only for circulatory and respiratory causes of death and included ischemic heart diseases (I20–I25), cerebrovascular diseases (I60–I69), other heart diseases (I30–I51), other circulatory diseases (I00–I15, I26–I28, I70–I99), influenza (J00–J99), pneumonia (J12–J18), chronic lower respiratory diseases (J40–J47), and other respiratory causes (J00–J06, J20–J39, J60–J99), according to the ICD-10-CM (International Classification of Diseases, Tenth Revision, Clinical Modification). The daily mortality data were provided by the Health Monitoring Unit of the Ministry of Health of Cyprus.

#### *2.2. Log-Linear Regression of Mortality–Temperature Relation*

The log mortality based on temperature was assumed to be smooth, but not necessarily linear and a generalized additive model (GAM) was used. This GAM offered a high quality of prediction of the dependent variable (log mortality rate) from the various distributions by estimating unspecific (non-parametric) functions of the predictor variables xj, which were connected to the dependent variable (mortality rate) via a link function.

As the death on a given day is not only a function of the same-day exposure to temperature but is also affected by exposure during a certain lag period, we also used the weighted average temperature of the preceding three days prior to the death. The lag period was determined using the cross correlation function (CCF) in RStudio software. Cross-correlation analysis showed the similarity of two series as a function of the displacement of one relative to the other. Figure 1 shows the cross correlation of mortality rate with minimum daily temperature for a cold period (months NDJFMA) and a hot period (months MJJASO). Similar analysis was also done for the mean and the maximum daily temperatures. The mortality rate decreased with increasing temperature during the cold period (Figure 1a) with a lag period of four days (that is, the highest peak of mortality four days after the coldest temperature), whereas the mortality rate reached a peak on the same day (no lag period) as

the maximum temperature (Figure 1b). The lag period of cross correlation function (CCF analysis) that had an effect on mortality rate was four days during the cold period and 0 days during the hot period. In an individual, the lag period should not be regarded as a well-defined interval as it may vary according to the magnitude of the temperature and individual characteristics such as acclimatization habits to heat, genetic background, and income.

**Figure 1.** Cross correlation for (**a**) minimum daily temperature (x) and mortality (y) for months November to April; (**b**) Minimum daily temperature (x) and mortality (y) for months May to October). Dotted blue horizontal lines show the 95% significance limits.

To address the lagged dependence of mortality on temperature, the same-day maximum (Tmax) and minimum (Tmin) temperatures, the same-day diurnal temperature range (DTR), and the average temperature of the preceding three days (Tmax[–3] and Tmin[–3], respectively) and weighted average DTR of the preceding three days (DTR[–3]) were used in order to closer examine the lag period of four days found during the cold period. Diurnal temperature range is the temperature difference of the daily maximum value and the daily minimum value [20].

The advantage of GAM was to limit the error in the prediction of the dependent variable—mortality rate—by defining the model in terms of smooth function. In the GAM, the degree of smoothness of the estimated mortality–temperature relative risk curve is controlled by its number of degrees of freedom (df). Many degrees of freedom were preferred to allow highly nonlinear shapes.

$$\text{Relative risk of Mortality} = \beta + \mathbf{f}\_1(\mathbf{x}\_1) + \mathbf{f}\_2(\mathbf{x}\_2) + \dots + \mathbf{f}\_m(\mathbf{x}\_m) \tag{1}$$

The functions fj may have a parametric or non-parametric form. xj represents the temperature term, which could be Tmin, Tmax, Tmax[–3], Tmin[–3], DTR or DTR[–3], as explained above. The relative risk of mortality was calculated using the gam function of the mgcv package in RStudio software [21]. The limitation of this short-term temperature analysis was that longer-term population characteristics were not considered such as health behaviors (smoking, drinking), comorbidities (hypertension, diabetes, cancer, etc.), medications, and access to health care.

#### **3. Results**

#### *Temperature–Mortality Relative Risk Analysis*

Figure 2 shows the temperature–mortality relative risk function estimated for Nicosia using GAM analysis. We examined six temperature parameters: the maximum daily temperature (Tmax), the weighted average of the maximum daily temperatures of the preceding three days (Tmax[–3]), the minimum daily temperature (Tmin), the weighted average of the minimum daily temperatures of the preceding three days (Tmin[–3]), the diurnal temperature range (DTR) of the day, and the weighted average of the DTR of the preceding three days (DTR[–3]). A smooth function of time with 50 df over the investigated years was used for the model. Similar findings were also found for smaller and larger df.

**Figure 2.** Temperature–mortality relative risk functions for Nicosia, Cyprus, 2004–2014. Tmax, Tmax[–3], Tmin, Tmin[–3], DTR, and DTR[–3] are shown in (**a**–**f**), respectively.

The relationship between maximum daily temperature and probability risk of mortality seems linear, with a steeper slope when the average of the preceding three days' maximum temperature is considered. The analysis of the minimum daily temperature with respect to the mortality's probability risk (Figure 2c,d) showed a U shape for same-day relationship and an inverse J shape when Tmin[–3] was used. That is, mortality risk decreased as the minimum temperature increased from the coldest temperatures and began to rise as the temperature increased from a certain threshold temperature

(approximately 22 ◦C). The applied model offers flexibility and agrees with Curriero et al. [14] as the population in a warm region (such as Nicosia) tends to be more vulnerable to cold rather than those residing in cold climates who are most sensitive to heat.

The analysis of DTR (Figure 2d,f) showed no significant results (mortality relative risk close to zero), insinuating that the human body could adapt to any DTR within the same day. These results thus need to be interpreted with attention, as previous studies have reported noteworthy findings of DTR with the probability risk of mortality, therefore a more focused analysis sub-grouped by age could elaborate on better assumptions.

Further analysis focused on the Tmin of the current day and the weighted average of the preceding three days (Tmin[–3]) by subgrouping the risk by gender and age groups. The variables gender and age were related using GAM with these minimum temperatures and the results are shown in Figure 3a,b. According to Figure 3a,b, cold temperatures impose a greater risk than hot temperatures, but other factors such as respiratory epidemics, usually present in winter, made unclear the exact role of temperatures on increased mortality.

**Figure 3.** Mortality relative risk functions for Nicosia, Cyprus, 2007–2014 with (**a**) minimum temperature of the same day (Tmin), (**b**) Weighted average minimum temperature of the preceding three days (Tmin[–3]).

On the other hand, high minimum temperatures and heatwaves were also associated with increased mortality. Heat waves have gained more attention due to the urban warming attributed to greenhouse gases and other anthropogenic sources. Other studies have shown that different cities have different sensitivities to extremes in temperature and that the latitude and local climate are factors to consider [14,22]. Air conditioning and human behavior can substantially modify the adverse effects of high temperatures, but even in the hot city of Nicosia, where people are more accustomed to higher temperatures and use air conditioning frequently, the effect of heat on health showed increasing deaths during heat waves. Thus, adaptation should be readdressed and governments should aim for adequate people awareness.

Figure 4 shows the relative risk of mortality for the DTR of the same day and the weighted average DTR of the preceding three days. Kan et al. (2007) hypothesized that large diurnal temperature change might be a source of additional environmental stress, leading to a greater risk factor for death [17]. In contrast, our results showed a greater risk at DTR in the range of 6–8 ◦C, and smaller risk at larger DTRs. Men and women aged 50–64 (magenta lines) were not affected by the variations in DTR throughout the day or the preceding three days. Men 65 years and over had a greater relative risk for DTR smaller than 5 ◦C, whereas for larger DTR, there was a negative relative risk, showing that large DTR did not impose a risk factor for death in Nicosia. Overall, we found that DTR was independently associated with daily mortality in Nicosia and that fluctuations in DTR appeared to mostly affect people over 80, probably because they have reduced ability to regulate body temperatures, thus making them marginally more vulnerable.

**Figure 4.** Mortality relative risk functions for Nicosia, Cyprus, 2007-2014 with (**a**) DTR of the same day, (**b**) Weighted average DTR of the preceding three days (DTR[–3]).

#### **4. Discussion**

The mortality risk by respiratory and circulatory causes for different age groups and per gender in relation to temperature has been examined. We employed two regression techniques to evaluate the impact of fixed and time-dependent factors on the vulnerability to temperature–mortality associations for 6882 deaths resulting from cardiovascular and respiratory causes between the years 2007 and 2014, in Nicosia, Cyprus. To model the relationship between temperature and mortality, we used the generalized additive model (GAM) to assess the interaction between a variable and the observation time and to interpret quantitative results. The GAM model revealed significantly increased mortality on hot (no lag period) and cold days (with a lag of 3–4 days), agreeing with a similar study in Estonia [4,9]. The adverse effects of heat on health are usually more direct with increased mortality on the same day or a couple of days after a heatwave [2,10]. As proposed by Curriero et al. [14], adaptation of populations to their local climate is evident by the increased health risk in relation to cold temperatures in warmer climates and on the contrary in relation to high temperature in colder climates. In warmer climates such as Nicosia, the people are more acclimatized to high temperature conditions and therefore, a steeper increase of the relative risk of mortality in colder conditions was observed.

The most important result to emerge from the analysis is that in the investigated area, the threshold temperature was about 21 ◦C, supporting the assumption that people in the area are more acclimatized to higher temperatures. MMT has been calculated in over 400 locations using the mean daily temperatures showing values of 29–32 ◦C in the eastern Mediterranean region and over 23 ◦C in the rest of the Mediterranean [15]. The probability risk of mortality using GAM revealed a stronger relation with the minimum daily temperature, so future studies should focus on the investigation of MMT using daily minimum temperatures. A similar GAM analysis in Shanghai [17] found smaller relative risk for elderly people with values of 0.3, therefore in Nicosia, where the population is more accustomed to higher temperatures, they are more vulnerable to lower temperatures with relative risk up to 0.5.

Our cross-correlation results agreed with previous studies where the effects of heat on mortality rate shortly after temperatures start to increase, whereas the effects of cold may take longer to emerge, and, depending on the latitude and the local climate, these periods may vary [2,9,23]. The results are in complete agreement with previous studies [24], finding no difference in heat-mortality risk between men and women, and weak evidence of a higher association of cold-mortality risk for men. The results are also in line with a study in Nicosia and a study in Stockholm [11], which found a difference in heat wave duration effects by age groups [2], while the increasing susceptibility to cold temperatures in the elderly has not been shown before.

This study has not confirmed previous research on DTR. In fact, in contrast with what was previously thought, we found that DTR does not have a noteworthy effect on mortality risk. This serves to allow for more focused research on environmental stress factors and whether prolonged duration of extreme heat or cold conditions is more important than within day variations of temperature.

A limitation of this study is the poor correlation of indoor and outdoor temperatures due to a number of modifying factors such as air conditioning, ventilation, and clothing. Time indoors affects the individual's exposure as well as workplace conditions and other comorbidities. Another limitation is that we did not adjust the analysis for micro-level socio-economic or demographic variables or other comorbidities that could have a potential confounding or modifying effect on the mortality–temperature relationship.

#### **5. Conclusions**

The main concern of the paper was to examine the mortality risk in relation to high or low temperatures of different age groups and compare them between the two genders. Mortality risk has only been evaluated for respiratory and circulatory causes. Particular attention is paid to elderly people, over 65 years old, as the results have shown a great vulnerability to ambient air temperature. We have addressed not only minimum and maximum daily temperatures, but also the diurnal temperature range (DTR) in order to examine the sensitivity for within the same day air temperature variations.

The originality of our approach lies in the fact that we have combined the cross correlation analysis to identify the effect of the preceding days' temperature with the generalized additive model (GAM) regression technique. From the research that has been performed, it is possible to conclude that there was increased mortality on extremely hot and cold days. The effects of the heat had no lag period, whereas cold effects had a lag effect of 3–4 days. The existence of these responses implies that in warmer climates, people are more acclimatized to high temperatures, and therefore a higher mortality risk was observed at colder temperatures with a lag of three days.

The approach used in this paper is applicable to several environmental areas such as air quality analysis where the results may be subjective to a delay period and may slowly diminish human health and well-being. The results of this study should alert organizations and governments on the possible impacts of climate change on public health by not considering adaptation. The identification of a threshold temperature per latitude and local climate will assist in the evaluation of the adaptation capacity of a specific population. This threshold temperature, according to our results, should be calculated using the daily minimum temperature. Nevertheless, even if humans become fully acclimatized to high temperatures, their health may still be negatively affected as a result of the poorer air quality associated with extremely high temperatures [2,25].

On the basis of the promising findings presented in this paper, work on the remaining issues is continuing to examine whether socio-economic or demographic variables or other comorbidities could have a potential confounding or modifying effect on the mortality–temperature relationship.

**Author Contributions:** M.S. conceived the research topic. A.P. obtained the datasets, created the figures, and analyzed the results. Both authors (A.P. and M.S.) contributed in the discussion of the results and reviewed the manuscript. All authors have read and agreed to the published version of the manuscript.

**Acknowledgments:** The authors are grateful to the Ministry of Agriculture, Rural Development, and Environment (MADRE) of the Republic of Cyprus for the Department of Meteorology historical meteorological data, and the Health Monitoring Unit of the Ministry of Health of Cyprus for the mortality data. The ideas and opinions expressed herein are those of the authors. Endorsement of these ideas and opinions by the Ministry of Health of Cyprus is not intended nor should it be inferred.

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
