Spatial Distribution of Land Surface Temperatures in Kuwait: Urban Heat and Cool Islands

The global rise of urbanization has led to the formation of surface urban heat islands and surface urban cool islands. Urban heat islands have been shown to increase thermal discomfort, which increases heat stress and heat-related diseases. In Kuwait, a hyper-arid desert climate, most of the population lives in urban and suburban areas. In this study, we characterized the spatial distribution of land surface temperatures and investigated the presence of urban heat and cool effects in Kuwait. We used historical Moderate-Resolution Imaging Spectroradiometer (MODIS) Terra satellite 8-day composite land surface temperature (LST) from 2001 to 2017. We calculated the average LSTs of the urban/suburban governorates and compared them to the average LSTs of the rural and barren lands. We repeated the analysis for daytime and nighttime LST. During the day, the temperature difference (urban/suburban minus versus governorates) was −1.1 °C (95% CI; −1.2, −1.00, p < 0.001) indicating a daytime urban cool island. At night, the temperature difference (urban/suburban versus rural governorates) became 3.6 °C (95% CI; 3.5, 3.7, p < 0.001) indicating a nighttime urban heat island. In light of rising temperatures in Kuwait, this work can inform climate change adaptation efforts in the country including urban planning policies, but also has the potential to improve temperature exposure assessment for future population health studies.


Introduction
The rise of urbanization and urban development has been associated with energy-intensive land use that replaces natural land cover. Urban structures, such as buildings, pavements, and asphalt, have direct influence on surface temperatures [1]. The use of air conditioners [2], reduced air flow from narrow streets and tall buildings [3], and the absorbance of solar energy from dark surfaces [4] are all mechanisms that contribute to a well-studied phenomenon of urban heat islands, where urban areas are hotter than surrounding rural areas. Much of the research in previous decades has been dominated by the characterization, health effects, and policy implications of the urban heat island. Urban heat islands have been shown to have adverse health impacts. First, they have been linked to an increase in the frequency and magnitude of thermal discomfort, which increases heat stress and heat-related diseases [5][6][7]. Secondly, the urban warming also enhances the photochemical reaction that leads to 18,000 km 2 (nearly the size of New Jersey, USA) with a total population of about 4.12 million in 2017 [28]. It has a large petrochemical industry with associated urban land uses. Kuwait City is the capital and the main city in the country. A land-use survey in 2000 concluded that nearly 75% of the country's area was rangeland in the form of barren and open sandy fields, while the urbanized area did not exceed 5% of the total land use [29]. The country is divided administratively into six governorates; Al Ahmadi, Al Farwaniyah, Al Jahrah, Al Kuwait (Capital City), Hawalli and Mubarak Al-Kabeer ( Figure 1). Based on land cover, governorates that include barren lands were classified as rural governorates. There were two rural governorates: Al Ahmadi and Al Jahrah. We  exceed 5% of the total land use [29]. The country is divided administratively into six governorates; Al Ahmadi, Al Farwaniyah, Al Jahrah, Al Kuwait (Capital City), Hawalli and Mubarak Al-Kabeer ( Figure 1). Based on land cover, governorates that include barren lands were classified as rural governorates. There were two rural governorates: Al Ahmadi and Al Jahrah.

Data
We used Moderate-Resolution Imaging Spectroradiometer (MODIS) Terra satellite data from the National Aeronautics and Space Administration (NASA). The MOD11A2 (version 6) product from MODIS/Terra includes an 8-day average of land surface temperature and emissivity with a 1 km spatial resolution (each pixel is 1 × 1 km). The user guide explained that the 8-day compositing period was chosen because twice that period is the exact ground track repeat period of the Terra and Aqua platforms [31]. The MODIS/Terra LST products were repeatedly validated over a set of locations and time periods via several ground-truth and validation efforts [32][33][34]. Recent LST products (collection 6 or version 6) have addressed previous accuracy issues, and measurement errors in arid regions for bare soil and validation studies now recommend their application in these regions [35]. The daytime and nighttime LST data for the period from 2001 to 2017 were extracted for Kuwait using Google Earth Engine (https://earthengine.google.com). The data consists of one measure for every 8 days. For better interpretation, we converted the unit of temperature from Kelvin to Celsius. First, we multiplied by a given scale factor (from the user guide) of 0.02 to return the units to Kelvin [31]. Then, we subtracted 273. 15

Data
We used Moderate-Resolution Imaging Spectroradiometer (MODIS) Terra satellite data from the National Aeronautics and Space Administration (NASA). The MOD11A2 (version 6) product from MODIS/Terra includes an 8-day average of land surface temperature and emissivity with a 1 km spatial resolution (each pixel is 1 × 1 km). The user guide explained that the 8-day compositing period was chosen because twice that period is the exact ground track repeat period of the Terra and Aqua platforms [31]. The MODIS/Terra LST products were repeatedly validated over a set of locations and time periods via several ground-truth and validation efforts [32][33][34]. Recent LST products (collection 6 or version 6) have addressed previous accuracy issues, and measurement errors in arid regions for bare soil and validation studies now recommend their application in these regions [35]. The daytime and nighttime LST data for the period from 2001 to 2017 were extracted for Kuwait using Google Earth Engine (https://earthengine.google.com). The data consists of one measure for every 8 days. For better interpretation, we converted the unit of temperature from Kelvin to Celsius. First, we multiplied by a given scale factor (from the user guide) of 0.02 to return the units to Kelvin [31]. Then, we subtracted 273.15 to convert Kelvin to Celsius.
Shapefiles of first-level Kuwait Administrative Divisions (polygons of the six governorates in 2015) were available from the University of California, Berkeley, Museum of Vertebrate Zoology. The shapefiles can be downloaded from the Harvard Geospatial Library open portal (https://hgl.harvard. edu).

Analysis
Over the entire study period and across all pixels, LST was summarized by mean, standard deviation, median, and interquartile range. We plotted LST against time (8-day intervals) for the study period from 2001 to 2017. The descriptive analyses were repeated for both daytime and nighttime LST and were stratified by each governorate. Before visualizing the raster data, we created a normalized land surface temperature (NLST) by scaling the LST values for a given pixel between the minimum (LST min ) and maximum (LST max ) values in each image: where the NLST ranges between 0 and 1 and enables the comparison between images calculated from the equation below [36]. An NLST value of 1 represents the maximum temperature and a value of 0 represents the minimum temperature, for a given season or image. We compared images of NLST spatial distribution in winter, spring, summer and fall. The differences between the average daytime and nighttime LST between each governorate were calculated using analysis of variance (ANOVA) and Tukey's honest significant difference (HSD) method, a Studentized range statistic. The Tukey's HSD sets confidence intervals on the differences between the means of the levels of a factor with the specified family-wise probability of coverage. Surface urban heat and cool islands were determined by calculating the average LST of the urban/suburban (T u/s ) governorate pixels minus the average LST of the rural and barren (T r ) pixels. A negative difference (T u/s < T r ) indicates an urban cool effect, while a positive difference (T u/s > T r ) indicates an urban heat effect.
Extraction, processing and exporting of raster satellite data was done using Google Earth Engine code editor (hands-on tutorials are available from: https://developers.google.com/earth-engine/ tutorials). All other analyses were conducted using R software (version 3.6.0) (R Foundation for Statistical Computing, Vienna, Austria).

Results
Over the entire study period from January 2001 to March 2017, the time trends of 8-day composite LST in Kuwait showed clear seasonal variability (Figure 2). Remotely sensed daytime land surface temperatures were higher than nighttime temperatures, but both showed similar patterns over time. In general, the spatial distribution of NLST in Kuwait showed similar patterns across seasons, but with lesser intensity in fall and winter seasons ( Figure 3).
When summarizing the average daytime LST by governorates, we found that urban and suburban governorates had lower temperatures compared to rural governorates, except for Al Farwaniyah. The nighttime LST showed opposite findings, where urbanized areas had higher LST. The mean daytime LST in the capital city (Al Kuwait governorate) was 33.7 ± 3.1 • C, and the nighttime mean was 22.4 ± 0.5 • C. The mean LSTs in Al Jahrah governorate, where most of the large rangeland and open sandy fields are, were 37.2 ± 1.1 • C in the day and 18.3 ± 0.9 • C at night ( Table 1).
The results of tests for statistical differences between each governorate are presented in Table 2. The mean daytime LST in the capital city was higher than in Al Jahrah governorate by 3.5 • C (95% CI: 3.1, 3.9, p < 0.001), and higher than in Al Ahmadi governorate by 3.2 • C (2.8, 3.6, p < 0.001). The nighttime effect was in the opposite direction, Al Jahrah governorate was warmer than the capital city by 4.1 • C (3.8, 4.5, p < 0.001), while Al Ahmadi was warmer than the capital city by 3.8 • C (3.4, 4.8, p < 0.001). We then combined all pixels of urban and suburban governorates (T u/s ) and rural governorates (T r ) and tested for the presence of surface urban heat and cool islands (Table 3). In daytime, the difference of urban and suburban governorates versus rural governorates (T u/s -T r ) was −1.1 • C (−1.2, −1.00, p < 0.001) indicating an urban cool effect. At night, the difference (T u/s -T r ) became 3.6 • C (3.5, 3.7, p < 0.001) indicating an urban heat effect.

Results
Over the entire study period from January 2001 to March 2017, the time trends of 8-day composite LST in Kuwait showed clear seasonal variability (Figure 2). Remotely sensed daytime land surface temperatures were higher than nighttime temperatures, but both showed similar patterns over time. In general, the spatial distribution of NLST in Kuwait showed similar patterns across seasons, but with lesser intensity in fall and winter seasons (Figure 3).      LST; land surface temperate (in degrees Celsius). The 95% confidence intervals were constructed from analysis of variance (ANOVA) and Tukey's honest significant difference (HSD) with specified family-wise probability of coverage.

Discussion
In this study, we used publicly available historical satellite data to characterize temperature differences between urban and non-urban areas in Kuwait. We found a daytime urban cool island and a nighttime urban heat island. On average, the magnitude of the difference estimate was higher for the nighttime heat effect. This work can inform urban planning policies and help establish spatial temperature exposure data for future population health studies.
To the best of our knowledge, we have identified only one study that investigated urban heat islands in Kuwait. Nasrallah et al. [26] analyzed 23 years (between 1951 and 1980) of maximum and minimum air temperature data from selected monitoring stations in and near Kuwait City, Kuwait. They concluded that there is a general lack of a well-developed heat island in Kuwait. The authors hypothesized that the similarities in the urban and rural landscape of Kuwait City, and its close proximity to a large water body, were the possible explanations for the lack of heat island development. In this investigation, we now show with more spatial granularity a clear development of a daytime urban cool island and a nighttime urban heat island using a different measure of temperature and a different study period. The development of urban heat islands is commonly found in big metropolitan cities in many regions in the world, [25,[37][38][39] while the daytime urban cool island seems to be a distinctive feature of arid cities. Our findings of a daytime cool island and nighttime heat island aligned with previous literature from neighboring cities that have similar arid desert climates such as Dubai and Abu Dhabi, United Arab Emirates [40,41], Erbil, Iraq [13], and Tehran, Iran [36].
In general, urban structures and tall buildings absorb heat, reduce air flow, and generate hot air from air conditioner usage [1][2][3][4]. Manmade dark impervious surfaces absorb shortwave radiation and store heat during the daytime and then release longwave radiation slowly at night contributing to the nighttime urban heat island [25,42]. Urban areas also have higher concentrations of cars and other heat-generating activities released from burning fuel [43]. A study by Al-Hemoud et al. [44] identified a surface temperature inversion in urban areas in Kuwait that occurs immediately after sunset and diminishes at daylight after sunrise; this natural phenomenon could add to the occurrence of nighttime surface urban heat islands. On the other hand, daytime urban cool islands can be attributed to several other factors. Soil in urban lands is moist, and daytime evaporation could potentially reduce the surface temperature compared to the dry soil in barren lands in arid areas [13]. Additionally, large water bodies have greater specific heat capacity and provide a potential cooling effect during the daytime [45]. In contrast to the rural areas, the urban areas in Kuwait are located on the Arabian Gulf, which showed significant daytime cool island effect. The temporal variation in LST between coastal and inland pixels can be strongly driven by sea breeze circulation in coastal cities, especially during clear-sky days in hot summers [46]. It is possible that sea breeze is the main driver of lower surface temperatures around the coastal areas in Kuwait. However, we observed daytime urban cool island effects throughout the four seasons and not just in the summer. Additionally, there was an east-west oriented belt of high LST in the inland part of the city (Al Farwaniyah) in the daytime of winter, fall, and faintly in spring (Figure 3). In other words, although a number of arid and desert countries reported an urban cool effect, it seems that the concept is more complicated than is explained by a simple concept of a cool island. More research that can detect high spatial and temporal LST patterns from an aircraft or new-generation geostationary satellite would help urban planners understand urban cool islands better [46]. This study can inform policymakers and urban planners in Kuwait when considering how to relieve the urban heat effect, especially at night. There are several mitigation strategies that have been proposed worldwide to tackle urban warming. Replacing low-albedo surface material (asphalt and concrete) with high-albedo and high-emissivity surfaces may keep the surfaces cooler when exposed to solar radiation [47][48][49]. Vegetation can intercept solar energy, provide shade to surfaces and has higher albedo than pavement; plants absorb and accumulate less heat, while the evapotranspiration process helps to cool the environment [50][51][52]. Green roofs may also contribute to the mitigation of urban heat islands. The albedo value of bitumen, tar, and gravel roofs typically ranges from 0.1 to 0.2, while the albedo value of green roofs is between 0.7 and 0.85 [53]. Experimental evidence showed that large-scale applications of green roofs reduced ambient temperatures by 0.3 to 3 • C [54].
There are a number of limitations to this study. First, we did not compare the satellite data to observed data from ground facilities, as we did not have surface temperature data to reference. Secondly, although satellites provide extensive historical and geographical coverage, we were limited to a 1 km spatial resolution and we could not obtain historical LST data from Google Earth Engine before 2001 to compare our results to the previously published study in 1990 in Kuwait. The spatial resolution of 1 km may not describe finer details of hot or cool spots within the urban city. Additionally, satellites may not be able to accurately capture temperatures from surfaces that are obscured by trees or tall buildings. Data were only available for clear weather conditions and only during the day and night times when the satellite passes over Kuwait. However, we had data approximately every 8 days, which would reduce the amount of missing data. Finally, we did not investigate how land use land cover change influences the intensity of urban heat or cool effects, and we did not investigate spatial differences in LST trends over each year.

Conclusions
Urbanization in Kuwait has converted natural landscapes of open soil and undisturbed desert areas to manmade engineered surfaces and infrastructure. About 17 years of historical satellite data suggest the presence of a nighttime urban heat island, but also an urban cool island in the daytime. This work can inform climate change adaptation efforts, especially urban planning policies. Our spatial analysis of land surface temperatures in Kuwait can be used to improve temperature exposure assessment for population health studies.