An Analysis of Spatio-Temporal Trends of Land Surface Temperature in the Dhaka Metropolitan Area by Applying Landsat Images

Land surface temperature (LST) is a basic determinant of the global thermal behavior of the Earth surface. LST is a vital consideration for the appraisal of gradual thermal change for urban areas to examine the strength of the thermal intensity of the surface of urban heat island (SUHI) and to see how hot the surface of the Earth would be in a particular location. In this respect, the most developed urban city like Dhaka Metropolitan Area (DMA), Bangladesh is considered for estimation of LST, and Normalized Difference Vegetation Index (NDVI) changes trend in more developed and growing developing areas. The focus of this study is to find out the critical hotspot zones for further instantaneous analysis between these two types of areas. The trends of long-term spatial and temporal LST and NDVI are estimated applying Landsat images-Landsat 5-TM and Landsat OLI_TIRS-8 for the period of 1988 to 2018 for DMA and for developed and growing developing areas during the summer season like for the month of March. The supervised classification was used to estimate the land cover categories and to generate the LST trends maps of the different percentiles of LSTs over time using the emissivity and effective at sensor brightness temperature. The study found the change in land cover from 1988 to 2018. The findings of the study might be helpful for urban planners and researchers to take up appropriate measures to mitigate the thermal effect on urban environment.


Introduction
The rapid urbanization process plays a key role within the formation of urban heat islands (UHI), where the heating would have a further impact upon the urban life quality [1] [2] [3]. A UHI phenomenon creates in urban city areas with substantially warmer temperatures than adjacent rural areas causing huge thermal discomfort to all living entities within the city [4]. Natural landscapes of rural areas are transformed into modern land use and land covers like buildings, markets, roads, and other impervious surfaces, making urban landscapes fragmented and sophisticated and raising the urban temperature affecting the lives of urban dwellers. Urban land cover changes (ULCC) are mainly caused by the removal of natural vegetation cover, mainly responsible for the microclimate change of the town [5]. In general, the surface cover with vegetation and water provides lower surface temperature. In 1990, only 15% of the world's population lived in cities, while within the 20th century, this picture has been completely changed with about 50% population of the planet is estimated to reside in cities [6]. LST may be a controlling factor for many of the physical, chemical, and biological processes in the world and may be considered as a measuring indicator of global climate change [7]. For the urban environment, LST may be a crucial parameter for monitoring of the energy exchange between the land surface and thus the atmosphere in terms of the sensible and heat transformation fluxes [8]. This bears importance when discussing about the thermal effects of the cities on the regional climate. An understanding of LST is vital for urban climatology, global environmental change, and human-environmental interactions [9]. In fact, the observed geographical and ecological patterns and processes tend to be spatial variables. Therefore, the relationships between LST and its impact factors are often characterized by local changes [10]. It has been found that the relationship between land use land cover (LULC) type and LST is very strong with a strong positive correlation (r 2 = 0.9281) [11] [12] [13].
The DMA in Bangladesh is undergoing rapid urbanization transforming into a hotter city. The LST in Dhaka City has been increased substantially within the city area [4] [14] [15]. Most of the prevailing research has been accomplished based upon the LST changes or land cover (LC) changes in DMA [16] [17] [18] [19] [20] for the period of 1989-2009. The LSTs were retrieved to know the variation of temperature from rural areas to urban areas [21]- [26]. Another study in Dhaka city. It has also been found that the warming has been mostly connected to Dhaka city due to urbanization having a global phenomenon [27].
However, no studies were found on the trend of the variation of LST and LC changes in the developed and growing developing areas of DMA, Bangladesh during the summer season. With this respect, the study was aimed to provide a comprehensive assessment of the thermal environment using remote sensing data over time in the selected locations of DMA. The objectives were: 1) to determine trends in the frequency of extreme temperature indices; 2) to analyze the spatial patterns of extreme temperature related indices; and 3) to assess the statistical significance of the trends in indices of extreme temperature through parametric and non-parametric tests. The knowledge generated by this study will be an aid to assess and mitigate the socio-economic impacts of the increased thermal changes in the UHI of the DMA.

Study Area and Datasets
Dhaka City as Dhaka Metropolitan Area (DMA) is located almost in the geographical center of Bangladesh at 23˚43'0"N and 90˚24'0"E ( Figure 1 [32]. Usually, it was not necessary to conduct a geometric correction for Landsat level 1 products, as they were registered through a systematic process [33]. The main correction was radiometric and eliminates errors that affect the brightness values of the pixels [34]. These errors were mainly due to detection errors in the sensor system and environmental attenuation errors. The original image sizes were larger than the study area, so after pre-processing, they were edited using a shapefile of the DMA. In Table 1, the Landsat images used in the study are shown.

Estimation of Land Surface Temperature
The land surface temperature (LST) was derived from atmospherically corrected Journal of Geographic Information System Landsat 5 TM (band 6) and from Landsat 8 TIRS (band 10) through the raster calculation in ArcGIS. In this study, the supervised classification was used to estimate the land cover categories and to generate the LST trends maps of the different percentiles of LSTs over time using the emissivity and effective at-sensor brightness temperature. All equations were considered from USGS website. The additional input parameters such as atmospheric water vapor content and near surface air temperature from ground-based observations were also required to calculate the LST. They were usually unavailable [37]. All the digital numbers (DN values) of thermal bands were converted into spectral radiance using Google Earth Engine. A three-step process was followed to derive the LST from Landsat image in both developed and growing developed areas [38]. The trend of different percentile of LST over time like 50 th , 75 th , and 90 th percentile for the whole DMAs was estimated by different year LST groups. The thermal infrared observations are easily contaminated by cloudiness, leading to many gaps during the LST estimation [39]. So, it is indispensable to observe the LST trend in the different locations of urban areas for further adaptation of the microclimatic changes or to mitigate the UHI. To mitigate global warming, LST is a very cohesive factor that increases UHI. The final step of retrieving the LST or the emissivity-corrected land surface temperature was computed using the emissivity and effective at-sensor brightness temperature, where images were further used to derive LST using Equation (1) where LST is in ˚C, BT is the at sensor brightness temperature (˚C), λ is the average wavelength of the specific band, ε is the emissivity and ρ = hc/σ, σ = Boltzmann constant (1.38 × 10 −23 J/K), h is Planck's constant (6.626 × 10 −34 J/s), c is the velocity of light (2.998 × 10 8 m/s).

Calculation of Brightness Temperature
The Equation for calculating the brightness temperature Equation (2) where BT is the effective satellite temperature (brightness temperature) in ˚C, K 1 is that the band-specific conversion constant, and K 2 is another calibration constant in Kelvin. Therefore, values of K 1 and K 2 are constant for OLI/TIRS, but the values of bias and gain values could also be different for various satellite images.

Calculation of Top of Atmospheric (TOA) Spectral Radiance
Here, M L represents the band-specific multiplicative rescaling factor from the metadata, Q cal is the Quantized and calibrated standard product pixel values, correspond to band 10, A L is the band-specific additive rescaling factor from the metadata, the pixel values of satellite images (DN) were converted to Kelvin and further to Celsius.

Estimation of Emissivity
Surface emissivity is important for calculation of land surface temperature by remote sensing. There have been several studies on emissivity. Among these, we adopted the frequently used method of the estimation of emissivity using simplified normalized difference vegetation index (NDVI) thresholds derived from the spectral reflectance in the red and near-infrared bands. It is assumed that the surface is flat and homogeneous. The conditional Equation (4) for estimation of emissivity [42] is as follows: where εv and εs are the vegetation and soil emissivity.

Estimation of Normalized Difference Vegetation Index (NDVI)
In this study, NDVI was used to express the vegetation land cover changes in urban area specifically, in the more developed and growing developing areas within the study area and differences in the spatial resolution of the images, which are within 30 m. GIS tools were then applied to the data using visual analysis, reference data, as well as local knowledge to split and recode these covers. It was done to closely reflect their true classes. Conversely, assessment of NDVI for a specified pixel always results in a number that ranges from −1 to +1. The Normalized Difference Vegetation Index (NDVI) was calculated by the equation (6). Three main types were identified: low vegetation, medium vegetation, and high vegetation. In this study, NDVI values were selected less than 0 to 0.05 for low vegetation, 0.05 to 0.2 for medium vegetation, and 0.2 to 1 for high vegetation. The satellite data from 1988 to 2018 were studied using spectral and Journal of Geographic Information System spatial profiles to ascertain the digital numbers (DNs) of different categories prior to classification. The supervised classification of NDVI was reclassed because sometimes urban landfill were merged with other classes, which were not possible to separate them due to their similar spectral properties. So, reclassification was used to improve the accuracy of the classification by the physical field survey method. In this study, classification accuracy referred to communication between the remotely sensed data and reference physical information of those pixel values. In order to assess the accuracy of land cover maps extracted from Landsat data, a total of 20 stratified random pixels were generated for the year 2018. NDVI will be computed temporally to understand the change of land cover during the study period and for the proportion of vegetation (Pv). That is why Pv is highly related to the NDVI and emissivity (ε).

Impact of Land Surface Temperature on Urbanization
The

Spatial Trends of Different Percentile of Land Surface Temperature (LST) and Vegetation and Land Cover Classification
The LSTs of the study area were calculated from the Landsat images as discussed in the methodology. The common cloud cover for the selected months

Spatial Trend of Area Coverage by Different Percentile of LST Groups and Different NDVI Ranges for the Developed and Growing Developing Areas
The land area cover patterns by different temperature groups based on the 50 th , 75 th , and 90 th percentile showed a similar gradient to the temperature values in Figures 4(a)-(e). Furthermore, vegetation coverage mapping of the study area would provide information on the identification and estimation of trends of vegetation index changing over the past 30 years. The area changing scenarios of vegetation in these time frames for the months of March of the selected areas are also presented in Table 3. The land cover patterns with vegetation, which were identified by 3 categories used for the year 1988-2018, as well as land cover changes along the time, are listed in Table 3.
Along the time scale (1988-1997, 1998-2007, 2008-2018) Figure 4, it was clearly understandable that the more developed urban areas show higher percentile groups of LST values compare with growing developing areas. One of the reasons for having high temperature values for developed areas was that the bares lands were in places where there was on-going development taking place. As a result, the vegetation cover was reducing in the developed and growing developing areas Table 3. Therefore, the background climate change was weaker than surface warming in urban areas.
This implied that, in the months of March, 50 th percentile group of LST coverage areas were decreasing at the rate of 67% -96%, the 75 th percentile group of LST  (Table 3). It was a well-known fact that more greenery areas led to more cooling effects. So, it was to be confirmed that vegetation coverage such as urban rooftop agriculture (URTA), gardens, forests, parks, and grasslands were well established tools for reducing urban thermal environment enhancing urban cooling through evapotranspiration and shading effect by green activities. The study exposed that the urban LST and vegetation changes in the three periods examined in this study did not occur evenly in all directions. New developments were observed along with the urban areas as well as in the areas that

Temporal Trend of Different Percentile Groups of LST
The   However, due to the warming effect of climate change, from the above discussion, it is clear that LST is the major issue and factor of climate change as well as global warming. This phenomenon is more reflected in the city center of urban areas, especially the more developed areas.  Table 3 shows that the peaks of the LST were found usually in the built-up areas, while the troughs were found in the vegetation areas. The peaks of NDVI were found in vegetated areas. Thus, NDVI and LST showed a transparent negative correlation. In other words, the NDVI values were diminutive (or, even adverse) where LST was high and vice-versa. From Table 3, it was found that vegetation coverage within the developed area was decreased by 7%

Comparison between Vegetation Coverage and Different Percentile of LST with Developed and Growing Developed Areas in DMA
to 10% in comparison with the growing developed areas. The connection between LST and concrete land covers was investigated with an identical correlation. The very best LST was found in Motijheel which was above Demra by 37.932˚C. It was a contrast for early developing and densely built-up areas. In Demra, the medium vegetation range area was increased by 13.74% compared with the developed areas, like Motijheel. A vegetation coverage area was decreased by 6.74% within the growing, developing areas compare with developed areas in 1988-2019. So, the studies used the typical satellite images at two or three different dates within the same month. All available clear-sky images within several years were studied for the selected areas to avoid the cloud contamination and the less accuracy. It was obvious that the urban vegetation landscape played a vital role in reducing the UHI effect in the city centers. Urban planners to come forward to increase the urban green spaces through planning as mitigation tools to reduce urban heating in Dhaka City.

Conclusions
This study analyzed the land surface temperature and urbanization trend of the DMA and its impacts on the creation of UHI exhausting remote sensing and GIS tools. The results of this study clearly indicated a significant warming trend for the most developed or the most built-up areas that are facing increased UHI ag-Journal of Geographic Information System gravating climate warming. It is certain that the warming trend would further deteriorate the urban ecosystem and modify the major hydro-ecological processes over the study area. An increased vegetation activity is capable to deter the current heated local environment from the urban areas eliminating climate change impacts. The spatial and temporal trends of vegetation and their effects on LST changes as per percentiles of 50 th , 75 th , and 90 th led to conclusions that hot spot zones are assembled in the most developed areas where vegetation coverage is lower than the growing and developing areas. LST in the developed areas is more sensitive to climate changes than the growing, developing areas, and adaptation approaches or vegetation increases are needed to overcome LST increases in hotspot regions.
It is imperative to carry out analysis of urban heating impacts using multi-source satellite images/radar data for curbing of future UHI effect of DMA.
The integrated use of ArcGIS, RS, and socio-economic data would be effective to understand the spatial and temporal dynamics of the major changes in both LST and NDVI.
These results are supposed to provide valuable information for urban planners and researchers to take up appropriate green actions like roof top agriculture to mitigate the UHI effects. It will pave the way of achieving sustainable urban cities. Furthermore, with these findings, the planners can predict the possible changes in urban growth configurations.