Around 50% (or 3.5 billion) of world population is now living in urban areas, which is expected to be 60% (4.9 billion) by 2030 [8
]. This increasing trend in global urbanization influenced many researchers to investigate the potential impacts of man-made activities on urban thermal environment such as the LST and UHI effect [9
]. Research on LST is, therefore, indispensible in order to inform the development of sustainable urban policy. This section reviews literature on the LST component of UHI effect, its character, and key measures.
2.1. UHI and Its Key Characteristics
The concept of UHI was described by Luke Howard in 1833 and, since then, research on this topic gradually intensified [10
]. In a broader geographical context, UHI is a metropolitan area, which is significantly warmer than its surrounding rural areas. However, UHI can also exist within a city region—which implies that a portion of urban area is significantly warmer than its surrounding urban areas [11
]. The temperature difference is usually larger at night, in winter, and under weaker wind condition compared to daytime, summer, and stronger wind condition, respectively [12
]. However, a different scenario might also exist depending on spatio-temporal context of an analysis.
Ohashi and Kida (2002) have shown that under clear skies and light wind conditions, cities are typically warmer than surrounding rural environments by up to 10 °C [13
]. Souch and Grimmond (2006) summarized their findings as: the UHI (1) is primarily a nocturnal phenomenon, (2) can occur throughout the year, (3) is dependent on weather conditions such as wind, and (4) generally consists of higher temperatures in the urban core and commercial locations, with lower temperatures in residential and rural sites [14
2.2. Causes and Consequences of UHI
Unplanned and haphazard urbanization coupled with the poor building design are the biggest causes of heat islands in cities. Research has shown that UHI is primarily caused by the built environment in urban areas, in which natural areas are replaced with non-permeable and high temperature surface of concrete and asphalt [15
]. These surfaces absorb the sun’s heat more then they reflect it, causing surface temperature and overall ambient temperature to rise. In addition, tall building and narrow streets can heat the air trapped between them. The UHI effect is the result of a number of factors including but not limited to urban geometry (i.e
., size shape, height, and arrangement of buildings), thermal characteristics of urban surfaces, scarce vegetation, extensive use of air conditioning during hot weather, industrial process and automobile use, and the amount of built-up area that exists within a city [16
]. The built-up area of a city (e.g., building roofs and walls, road, parking lots, industrial area, and commercial area) has less moisture content, and therefore, evaporates less water into the air causing high surface and air temperature. The opposite is true in the case of vegetation in urban areas, which works as a natural cooling system [17
In summary, there are four major urban elements which may influence the micro-climate of a city. They are built-up area, vegetation, bare soil, and water bodies [18
]. UHI can impact local weather and climate, by altering local wind patterns, spurring the development of cloud and fog, and influencing the rate of precipitation [11
The effects of urban heat island are manifold. First, the higher temperature in a city brought about by UHI has an adverse impact on air quality [19
]. Compared to rural areas, cities experience higher rates of heat related illness and death. The heat island effect is one factor among several that can raise summertime temperature levels that pose a threat to public health [20
2.3. Measurement of LST
Researchers have identified UHI using a number of different measures depending on the type of components of comparison made (e.g., spatial, temporal) in their studies. For example, when comparisons are made between two time periods, UHI was identified by measuring LST differences between the periods of an identical location. This method is often applied to measure the impact of urban growth on UHI effect (e.g., temperature differences between pre-urban and post-urban conditions) [21
]. Others have identified UHI through measuring rural-urban temperature differences [21
]. Therefore, the comparison is made between geographical units rather than between time periods. However, most of these studies are based on observed temperature of places.
More recently, researcher started utilizing airborne or satellite remote sensing data to derive UHI based on LST [23
]. This technology also allows monitoring land cover changes over time. These two types of opportunities (e.g., derivation of surface temperature and monitoring land cover changes), therefore, enabled researchers to investigate the links between land cover changes and LST changes of a given site between two time periods, i.e.
, the monitoring of UHI effect due to land cover changes [25
In the recent past, researchers used the National Oceanic and Atmospheric Administration (NOAA) data to derive LST and consequently to measure UHI for studies conducted at the regional scale [26
]. However, in recent years, the Landsat Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) thermal infrared (TIR) data have often been utilized for smaller scale studies [27
]. Other studies have used mixed type of images. For example, Liu and Zhang (2011) analyzed Landsat TM and ASTER images in order to derive UHI intensity in Hong Kong city [11
Different types of land cover indices have also been developed in order to investigate the correlations between land cover changes and LST. Amongst the various indices, studies have found that normalized difference vegetation index (NDVI), normalized difference built-up index (NDBI), normalized difference water index (NDWI), and normalized difference bareness index (NDBaI) correlate strongly with LST [5
]. NDVI is used worldwide to express the density of vegetation, monitor drought, monitor and predict agricultural production, assist in predicting hazardous fire zones, and map desert encroachment [28
]. The NDVI process creates a single-band dataset that mainly represents healthy biomass. This index output values between −1.0 and 1.0, where any negative values are mainly generated from clouds, water, and snow, and values near zero are mainly generated from rock and bare soil. Very low values of NDVI (0.1 and below) correspond to barren areas of agriculture, rock, sand, or snow. Moderate values represent parks, shrub, and grassland (0.2 to 0.3), while high values indicate temperate and tropical rainforests (0.6 to 0.8) [30
NDWI implies the water content within vegetation and water state of vegetation [31
]. NDBI is sensitive to built-up area. It has been recently used as an indicator to represent the extent of built-up areas [5
]. NDBaI is useful to classify different barren lands [34
]. These indices are used to classify different land cover types by setting the appropriate threshold values [5
]. The relationship between LST and NDVI has been extensively documented in literature [35
]. The urban thermal environment is related to the reduction in vegetation coverage [28
]. NDBI is used for the extraction and mapping of built-up areas [33
]. NDWI and NDBaI are being analyzed for delineating water content in vegetation and identifying bareness of soil respectively [5
]. The intensity of LST is directly related to the rate of urbanization, land use patterns, and building density. LST is related to patterns of land use/cover changes, e.g., the composition of built-up area, vegetation, and water bodies [5
Rinner and Hussain showed that LST is extreme where commercial and industrial land uses are located in Toronto, i.e.
, correlation exists between NDBI and LST [27
]. Similarly, Chen et al.
have shown that LST has become more prominent in areas that experienced rapid urbanization in Guangdong Province, southern China. This study also analyzed dynamics in the spatial distribution of heat islands. The distribution changed from a mixed pattern in 1990 (where bare land, semi-bare land, and land under development were warmer than other surface types) to UHI in 2000 [5
]. Lo and Quattrochi found that land use changes resulted in significant UHI effect at urban boundary in Atlanta Metropolitan Area in Georgia [6
Xiong et al.
found that high temperature anomalies were closely associated with built-up land, densely populated zones, and heavily industrialized districts. They analyzed Landsat TM/ETM+ images, NDVI and NDBI indices for UHI analysis of Guangzhou in South China [28
]. Xiao et al.
reported that impervious surface is positively correlated with LST in Beijing, China. They used Landsat TM and QuickBird images to analyze the correlations [38
]. Weng and Yang argued that low vegetation coverage is one of the main reasons for UHI effect in Guangzhou city, China. They generated land cover and LST maps from Landsat TM images [39
Based on the above review, the following observations can be made in terms of measuring LST. First, although LST is influenced by a number of factors related to urban land cover, most studies have examined just one factor in establishing the correlation. As a result, the influence of other factors has not been disentangled from the correlations presented in these studies. Second, in all cases, researchers have studied only the past or present dataset in measuring LST. Little attempt has been made so far to analyze changes in LST based on future land cover simulations. This article aims to address these gaps in the literature.
2.4. Simulation Studies on Land Cover Changes
An urban area is a complex dynamic system. The growth of a city depends on numerous driving factors like social, economic, demographic, environmental, geographic, cultural, and other phenomena. Modeling such dynamic systems is not an easy task [20
]. Different tools have been developed over the years to underpin the modeling of urban growth and land use changes. Some popular packages include Geomod [40
], SLEUTH [41
], Land Use Scanner [42
], Environment Explorer [43
], SAMBA [44
], Land Transformation Model [45
], and CLUE [46
]. These tools, again, utilize a number of methods in order to model land cover change such as Markov Chain [47
], Cellular Automata [48
], Logistic Regression [49
], and Artificial Neural Network (ANN) [50
Each tool has its own advantages and disadvantages. For example, Geomod is designed to simulate only a one-way transition from one land cover category to another land cover category [40
]. Again, each method has its own strengths and weaknesses. For example, Markov Chain is better applicable when the trend of land cover changes is known. However, this method lacks spatial dependency and spatial distribution [47
]. The Cellular Automata method models the state of each cell in an array depending on the previous state of the cells within a neighborhood, according to a set transition rules [48
]. The purpose of this research is not to discuss the strengths and weaknesses of all the models used in the literature as they have been discussed elsewhere [51
]. This research used the Multi-layer Perceptron (MLP) Neural Network method in order to model/simulate land cover changes [52
The MLP Neural Network method makes its own decisions about the parameters to be used and how they should be changed to better model the data. It undertakes the classification of remotely sensed imagery through a Multi-Layer Perceptron neural network classifier using the back propagation algorithm. The MLP also performs a non-parametric regression analysis between input variables and one dependent variable with the output containing one output neuron, i.e.
, the predicted memberships [53
]. One of its main advantages is that it is distribution-free, i.e.
, no underlying model is assumed for the multivariate distribution of the class-specific data in feature space [54
]. Neural networks are non-linear and can be conceived as a complex mathematical function that converts input data (e.g., remotely sensed imagery) to a desired output (e.g., a land cover classification) [20