Assessment of Outdoor Thermal Comfort Using Landsat 8 Imageries with Machine Learning Tools over a Metropolitan City of India

Rapid urbanization causes potential changes in the urban landscape, resulting in significant changes in land surface temperature and outdoor thermal comfort. This urban growth has a detrimental impact on the health and comfort of residents. The comfort level experienced in any given region depends on various parameters, including atmospheric temperature, relative humidity, land use and land cover (LULC). In this study, we aim to examine the spatial variation of outdoor thermal comfort in the city of Hyderabad. To achieve this, we utilize medium-resolution Landsat 8 imageries along with in situ meteorological data. The classification of LULC is carried out using the maximum likelihood method. A machine learning tool known as Support Vector Machine (SVM) is implemented, with seven environmental indices such as normalized difference vegetation index (NDVI), normalized difference water index (NDWI), new built-up index (NBI), LST, brightness, greenness, and wetness to predict outdoor thermal comfort (THI). The study reveals significant variations in THI across different land covers. Barren lands exhibit the highest mean THI values (27.3), followed by built-up areas (26.9), vegetation (24.1), and water bodies (20.7). These findings indicate that barren and built-up areas are associated with higher levels of discomfort, while vegetated regions and water bodies provide more neutral to moderate comfort conditions. These results also highlight distinct spatial variations in THI across different regions of the city, demonstrating the influence of the urban landscape on outdoor thermal comfort. This research is vital for identifying specific areas within cities that require targeted mitigation strategies to enhance outdoor comfort.


Introduction
Rapid urbanization is influencing the increase in the land surface temperature (LST) over cities, and recent studies in India (Basu & Das, 2023;SSSSSC et al., 2022;Srikanth & Swain, 2022) showed evidence that the LST increment is associated with urbanization; subsequently, modifying the outdoor thermal comfort impacts city residents' health and well-being.According to ASHRAE Standard 55 (2017), thermal comfort is the satisfaction of humans with the outdoor environment.The meteorological parameters influencing the local or micro-climate are air temperature, relative humidity, mean radiant temperature, wind speed, etc. Significant change in LULC due to the increase in the urban landscape impacts local weather conditions.This change in local weather alters outdoor thermal comfort, creating poor public health conditions (Buchin et al., 2016).Xiong et al. (2015) reported that different kinds of health issues based on different thermal comfort levels might lead to an increase in dehydration, eye strain, dizziness, accelerated respiration, and cardiovascular problems.
Hence, it is important to understand the changes in the urban landscape and to evaluate its influence on outdoor thermal comfort.Imran et al. (2021) reported that water bodies and vegetation are lowering the LST, and the urban landscape is creating high LST, thus showing change in LULC significantly impacts the thermal comfort.Mijani et al. (2019) concluded that the downward surface shortwave and longwave radiation, brightness, greenness, and wetness of the surface also play a role in the local micro-climate, which eventually alters the thermal comfort levels.
However, most of the earlier studies are based on synoptic observations at a point location and by using standard thermal comfort formulas such as Thom (1959) and physiological equivalent temperature (PET), wet bulb globe temperature (WBGT), and Universal Thermal Climate Index (UTCI).

?
Few studies (e.g., Ali & Patnaik, 2018;Deevi & Chundeli, 2020;Kumar & Sharma, 2022;Rawal et al., 2022;Ziaul & Pal, 2019) reported the changes in outdoor thermal comfort over urban cities and rural regions based on single station synoptic observations.Based on these studies, it is understood that thermal comfort varies significantly from urban to rural regions, implying that thermal comfort is highly dependent on LULC parameters.So, these station data alone are insufficient for accurately modeling the thermal comfort changes in space and time.To estimate thermal comfort over different LULC, researchers have considered LST a proxy by estimating surface urban heat island (SUHI) for thermal comfort.
Recent literature (e.g., Kikon et al., 2016;Shahfahad et al., 2022;Sharma et al., 2021;Swain et al., 2017;Xu et al., 2017) reported a rapid increase in hotspots (zones of higher magnitude SUHI) using remote sensing data.However, it is essential to investigate the outdoor thermal comfort by incorporating the influence of other environmental parameters such as greenness, wetness, and brightness, along with LST for better estimation.Mushore et al. (2017) investigated the impact of seasonal land cover changes on human outdoor thermal comfort in Harare, Zimbabwe, using Landsat 8 and in situ air temperature data using the simple linear regression model.In continuation of their previous work, Mijani et al. (2020) implemented multiple regression models based on least square adjustment for the generation of thermal comfort index (THI) maps over the city of Tehran.These regression-based models can predict outdoor thermal comfort by considering several independent environmental parameters simultaneously.Based on the literature review, we observed that it is also important to understand the influence of different environmental parameters on outdoor thermal comfort with reasonably good spatial resolution.Multivariate regression with several variables to find thermal comfort is rare, and this kind of research is new in the Indian context.So, the main objective of the present research work is to estimate outdoor thermal comfort by considering the influence of various environmental parameters.Multi-station synoptic meteorological observations and a machine learning model (SVM) are utilized over a metropolitan city, Hyderabad.

Study Area
For the present study, a metropolitan city, Hyderabad (17°23'13''N, 78°29'30''E), at 570 m above mean sea level, has been considered (shown in Fig. 1).It is one of the biggest metropolitan cities in India, with an area of 625 km 2 .It has tropical wet and dry weather climatic conditions (Norman, 1995) with an annual mean temperature of 26.6 °C.May is the hottest month, with a mean temperature of 33 °C, and December is the coldest month, with a mean temperature of 22 °C (Canty & Associates LLC, 2013).It has the highest relative humidity in August, with a mean of 69%, and the lowest in April, with a mean of 28%.The rainfall is highest in August (207 mm) and lowest (4.9 mm) in December.In the city, most of the area is covered by an urban landscape with a population of about 9,482,000 as of 2018 (The World's Cities in 2018).In the middle of the city, a large water body called Hussain Sagar is situated at 17°25'25''N, 78°28'25''E.A city water source from the Musi River passes through the city providing a small vegetation belt.

Datasets
The Landsat 8 satellite multispectral bands (collection-1, level-1) are collected for 3 consecutive years, 2018, 2019, and 2020, with  The daily observed meteorological data consist of maximum and minimum values of temperature T a (°C), relative humidity RH (%), and wind speed V (ms -1 ) in 16 mandals in Hyderabad city (the mandal names are given in Fig. 1).The meteorological data were collected from an open data source: https://data.telangana.gov.in/2.3.Methodology

Calculation of environmental parameters using satellite imagery
The multispectral bands (blue, green, red, NIR, SWIR 1, SWIR 2, TSIR 1, TSIR 2) of Landsat 8 are used to calculate the three indices, Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and New Built-up Index (NBI).Tasseled cap transformation (Baig et al. 2014) is used to retrieve the parameters, i.e., brightness, greenness, and wetness.Thermal bands are used to estimate the satellite brightness temperature.The brightness, temperature, and NDVI are used to retrieve the emissivity-corrected LST.The supervised land use land cover (LULC) classification was done based on multispectral bands using QGIS (http:// www.qgis.org/).The different estimated indices are given below.Some preprocessing of various atmospheric corrections is applied using the procedure given in https://www.usgs.gov/landsat-missions/using-usgslandsat-level-1-data-product,as given below.
Landsat Level-1 data can be converted to TOA spectral radiance using the radiance rescaling factors in the MTL file: L k is the TOA spectral radiance (Watts/ (m 2 *srad*lm)), M L is the band-specific multiplicative rescaling factor from the metadata, A L is the band-specific additive rescaling factor from the metadata, and Q cal is the quantized and calibrated standard product pixel value (DN).Reflective band DNs can be converted to TOA reflectance using the rescaling coefficients in the MTL file: where q k ' is the TOA planetary reflectance, without correction for solar angle.Note that qk' does not contain a correction for the sun angle, M q is the band-specific multiplicative rescaling factor from the metadata, A q is the band-specific additive rescaling TOA reflectance with a correction for the sun angle is then: where q k is the TOA planetary reflectance, and h SE is the local sun elevation angle.The scene center sun elevation angle in degrees is provided in the metadata (SUN_ELEVATION), and h SZ is the local solar zenith angle; The vegetation index, NDVI, was calculated from Landsat 8 data, the typical range of NDVI is [-1, 1], 0.2 to 0.5 represents small grasslands, and 0.6 to 1 represents dense vegetation.The estimation of NDVI is given by Kriegler (1969), Crippen (1990) given below.
The water bodies identified by the NDWI index were estimated using the equation given by Gao (1996).The index range lies between [-1, 1], and water bodies are identified when the value is [ 0.5.Build-up values lie between 0 and 0.2.
The NBI index was first proposed by Chen et al. (2010), which is used to identify built-up and bare lands.
The LST estimation for Landsat 8 was taken from Sultana andSatyanarayana (2018, 2020) and Weng et al. (2004) as follows: where BT is the brightness temperature 1321:0789 As the LST is estimated using the thermal band with a resolution of 100 m, it is resampled into 30-m resolution to generate THI maps.

Estimation of Thermal Index Using Meteorological Data
The thermal indices are classified into two categories: (1) simple indices, mostly based on meteorological parameters such as atmospheric temperature, vapor pressure, and wind speed, etc.; (2) energy budget models, which not only depend on meteorological but also on thermo-physiological parameters.In the present study, the thermal index (THI) used was estimated following Suping et al. (1992), which was developed to consider the parameters of ambient air temperature (T a ) and relative humidity (RH) along with the mean wind speed (V).This is also called effective temperature, given as The above index is used in this study because it is well correlated to the energy budget UTCI index.The details of different indices and their utility are given by Farajzadeh et al. (2015) and are very useful for evaluating outdoor thermal comfort at any particular location.THI of thermal comfort ranges from very uncomfortable to more comfortable conditions over Hyderabad is characterized and given in Table 1.

THI Mapping Using Machining Learning Model at 30-m Resolution
Initially, available meteorological variables such as air temperature, relative humidity, and wind speed at 16 mandal locations over Hyderabad for the two seasons (summer and winter) in the reference year 2019 are considered.From these data points, THI values of each mandal location are computed following Eq.8.
The environmental parameters are estimated using Landsat 8 imageries (as explained in Sect.2.4) at the finer resolution of 30 m.Then, the median of each of these parameters is computed to get the representative value for each mandal.These parameters are normalized for use in the machine learning tool, namely SVM regression, following Vapnik (2000).THI over at any location depends upon the combined impact of these environmental parameters.
In the present study, we have implemented the supervised linear SVM algorithm to find coefficients such as a o, a 1 , a 2 , a 3 , a 4 , a 5 , a 6 , and a 7 .The linear SVM consists of one dependent variable, i.e., THI, and the independent variables are NDVI, NDWI, NBI, LST, brightness, greenness, and wetness.
The method of SVM regression is as follows.
Now to find THI with a minimum norm value subjected to constrained It is quite difficult to construct a function with the above assumptions; thus, fn, f Ã n are chosen at every point.These are called slack variables.
where f n [ 0 and f Ã n [ 0 and is satisfied when applying the box constraint C, which helps reduce overfitting in regression models.One advantage of Support Vector Machines (SVM) over normal linear regression is their ability to minimize overfitting, making them highly suitable for cases with a low number of observations and a high mean deviation.SVM utilizes Convex Optimization, which ensures optimality in results by finding a global minimum instead of a local minimum (Smola & Scho ¨lkopf, 2004).
In SVM, support vectors are employed, and their errors are constrained to be less than a predefined value.The hyperplane is constructed between support vectors to best fit the line within a specified threshold.Unlike other regression models that aim to minimize the error between actual and predicted values, SVM regression utilizes only support vectors to construct the hyperplane within the threshold value.The coefficients obtained from SVM regression are then used to generate THI maps for the entire city.LULC classification of Hyderabad is conducted using Landsat 8 imagery (Otukei & Blaschke, 2010), and the corresponding THIs at different LULCs are analyzed.The flow chart outlining the overall workflow is depicted in Fig. 2.

Error Estimation
The error was estimated from MAE (mean absolute error) value Here, N is the number of mandals.MAE is extremely useful, especially in regression analysis, and it turns learning problems into optimization problems.It also estimates the quantifiable measurement of errors for regression problems.3. Results

Assessment of Influence of Environmental Parameters on THI
The present study considers seven important environmental parameters.For this purpose, the one-to-one correlation coefficients of these parameters and THI are computed.Details of these coefficients (given in Table 2), the NDVI, NDWI, greenness, and wetness, show an inverse correlation, and LST, NBI, and brightness show a significant positive correlation.The higher the correlation is, the higher the discomfort, and vice versa.
Notably, LST has a higher positive correlation with THI, signifying that the higher the LST, the higher the discomfort.This influence of the LST on THI has been reported in (Imran et al., 2021;Taleghani, 2018).Similarly, the wetness parameter has a higher negative correlation, and hence it plays an important role in the assessment of more comfortable conditions.

Estimation of THI with the Combined Influence of Environmental Parameters Using a Machine Learning Model
After the correlation analysis, the SVM regression model was applied to train the data set, i.e., data of the summer and winter of THI for the year 2019.The THI model coefficients corresponding to NDVI, NDWI, NBI, LST, brightness, greenness, and wetness are 0.0168, -0.0467, 0.7986, 1.2296, 0.8508, 0.0256, and -1.0991, respectively.The coefficients show that the LST (a 4 ¼ 1:2296Þ has the highest contribution in assessing THI, followed by brightness and NBI.The highest inverse contribution is wetness, followed by NDWI.The remaining parameters, NDVI and greenness, are influenced reasonably in estimating THI.
The accuracy of the model for 3 different years and two seasons is given in Table 3.The lowest error was observed in the summer of 2019, and the highest error was in the summer of 2020.

Spatial Variation of THI and LST over Different Land Covers During Summer and Winter
The spatial variation of LST for summer and winter during 3 years, 2018, 2019, and 2020, are presented in Fig. 3a-c and d-f, respectively.During these 3 years, the LST varies from 30 to 41 °C in summer and 20 to 30 °C in winter.Clearly, the southern part of Hyderabad has a higher mean LST than the northern part.Higher vegetation cover has been noted in the northeastern part of the city.Previous studies such as Sannigrahi et al. (2017) and Sultana & Satyanarayana (2018) reported a similar result of LST over Hyderabad.The spatial variations of THI for summer and winter during the 3 years 2018, 2019, and 2020 are presented in Fig. 3a-c and  d-f, respectively.The mean THI varies from 20 to 35 °C in summer and 14 to 28 °C in winter.In the recent study by Mijani et al. (2020), the mean THI in Tehran city in summer varies from 13.7 to 28.8 °C and in winter from 7.3 to 17.9 °C, and these values show a linear relationship between THI and LST.
The region over the southern part of the city has shown a slight decrease in LST in the summer months of 2018 to 2020, whereas during winter, it showed an increase from 2019 to 2020.Interestingly, the northwestern part of the city has shown no considerable change in LST variation during the summer, whereas during winter, a decrease in LST in the same region has been seen (Fig. 3a-f).
After analyzing the spatial variation of LST and THI, the southern part exhibits a discomfort region compared to the northern part because of the presence of more urban and bare lands.More than urban areas, bare land has slightly higher LST and THI magnitudes and causes discomfort for pedestrians.The bare lands absorb the high direct solar radiation; hence, the LST will be somewhat higher than for the urban structures.Bare land was found in areas with high LST, possibly because of the low thermal inertia of the bare soil as reported by Abir and Saha (2021).During the study period, variation in minimum and maximum LST was noticed, whereas there was no significant variation in THI over the city.In the summer season of 2019 (Fig. 3b), a maximum LST of [ 41 °C was observed, whereas, during the summer of 2018 (Fig. 3a) and 2020 (Fig. 3c), it was 38 °C and 39 °C, respectively.However, the maximum thermal comfort value is 32 °C (Fig. 4a-c) during the study period.Hence, even though the LST influence is more for THI, the influence of the remaining six environmental parameters has been clearly portrayed.
Hence, one can conclude that the environmental parameters in combination influence the microclimate of the region.For clarification, one can compare the spatial variation of LST and THI wherever the locations of higher LST zones do not exhibit higher THI zones (compared in terms of area/patches).

LULC Classification and Corresponding Mean THI over Different Land Covers
To investigate the variation of THI over different regions of the city in detail, we need to classify the LULC.The LULC maps are generated from the maximum likelihood method suggested by Ahmad (2012).The estimated classification accuracy of LULC maps is reasonably based on kappa statistics (Fleiss & Cohen, 2016).Over the study region, during the study period, the LULC maps (shown in Fig. 5) showed four primary classes, i.e., vegetation, water bodies, barren lands, and built-up.The percentage was covered with vegetation, water bodies, built-up areas, and barren lands, which are, respectively, 19%, 2%, 51%, and 28%.No significant changes in the LULC were noted during the study period.So, for analysis purposes, winter 2019 was taken as the base of LULC.The accuracy of these ranges was estimated with kappa statistics, and the overall kapaa accuracy was 91% in winter 2019.The estimated kappa value is 0.84.
Figure 5 shows that the city of Hyderabad is more dominant in the urbanized area.Figure 6a depicts the mean THI and mean LST magnitudes over different land covers during the study period along with interquartile ranges (IQR) to understand their variation over four different LULC classes.The mean magnitudes of THI are 20.73 °C, 27.28 °C, 24.21 °C, and 26.9 °C, and mean LST is 27.39 °C, 31.94°C, 30.28 °C, and 31.89°C over water bodies, barren land, vegetation, and built-up regions, respectively.The regions of water bodies, vegetation, and built-up and barren land exhibit more comfortable, neutral, moderate, and higher discomfort conditions, respectively.Figure 6b shows the mean THI and mean LST magnitudes along with IQR during summer and winter of study period.The present results are similar to that of reported ranges of values by Toy et al. (2007) where a THI value between 15 and 20 °C is considered the most comfortable condition and [ 30 causing the most discomfort.

Mandal-Wise Analysis of THI over Different Land Covers
The percentage of the area coming under different classes of THI during summer and winter, along with LULC of each mandal region, are depicted in Fig. 7a-d.During winter, most of the area in each of the mandal regions has moderately comfortable to comfortable conditions (Fig. 7a).However, during summer, about 50 to 70% of the area experiences moderate discomfort to neutral conditions, as depicted in Fig. 7b, except Secunderabad (17.43 N, 78.49E).Higher percentages of area in the mandals, such as Tirumalgiri (17.47 N, 78.51 E) and Charminar (17.36 N,78.47 E), have discomfort conditions in summer because of the presence of a higher percentage of urban and bare lands, whereas the regions of Khairatabad (17.41 N,78.46 E) and Himayatnagar (17.40 N,78.48 E) all have classes of discomfort conditions because of the presence of moderate urbanization with noticeable vegetation, as shown in Fig. 7c.
Figure 7d shows the percentage of area having extreme discomfort and high comfort conditions during summer and winter, respectively.Mandal regions such as Golkonda (78.39 E,17.38 N) and Tirumalgiri (17.47 N,78.51E) have extreme discomfort conditions, with an area of about 15%.However, in winter, the mandal regions Secunderabad (17.43 N,78.49 E),Musheerabad (17.41 N,78.49 E) and Golkonda (78.39 E,17.38 N) have the highest comfort zones, covering 35% of the area.
From the spatial distribution of various classes of outdoor thermal comfort, it is observed that over a few regions, the thermal comfort conditions vary from summer to winter in such a way that a few discomfort condition zones become comfortable, and vice versa.Hence, to find the areas which are changing from zones of discomfort (summer) to comfort (winter), and vice versa, we plotted the difference in THI magnitude between summer and winter, as shown in Fig. 8.A difference [ 7 means the THI class is changing considerably from summer to winter, and difference \ 3 means the THI class is almost the same in summer and winter.We noted locations near Bandlaguda (78.46 E,17.30 N),Bhaduarapura (78.42 E,17.34 N),Golkonda (78.39 E,17.3 8N) and Begumpet airport (78.47 E,17.45 N) showed a maximum difference in THI from summer to winter, experiencing more discomfort zones in summer but comfort zones in winter.The large gray area in Fig. 8 shows that the variation from summer to winter is only between 4 and 7 °C, which means only a few regions largely change discomfort levels from season to season.
The variation in the percentage of mean THI among the comfort/discomfort locations is presented in Fig. 9.In summer, most of the area above 8% has extreme discomfort, about 60% moderate discomfort, 20% neutral, 6% moderately comfortable and only about 6% highly comfortable.In winter, nearly 70% are moderately comfortable, followed by 25% highly comfortable, with 5% remaining covered by remaining classes.In the summer, moderate discomfort conditions are observed, but thermal comfort is mostly moderately comfortable in the winter.The occurrence of extreme comfort in both summer is less, and the occurrence of extreme discomfort in winter is almost negligible.

Discussion
In the past decade, the impact of urbanization on the urban heat island (UHI) phenomenon and its consequences has gained significant scientific interest.Recent reviews, like the one by Ren et al. (2023), confirm that UHI has the potential to greatly influence outdoor thermal comfort.Consequently, the utilization of machine learning tools has emerged as a promising approach for analyzing outdoor thermal comfort.Notably, recent studies by Boccalatte et al. (2023) 2023).These elements have a profound impact on outdoor thermal comfort.
For accurate estimation of outdoor thermal comfort, precise variable selection is required.The current study incorporates seven independent environmental parameters to model outdoor thermal comfort.These variables were selected based on previous research and correlation analysis.Recent studies, such as, e.g., Taleghani & Berardi (2018) and Xu et al. (2017) emphasized the importance of meteorological variables and urban morphology parameters, including albedo, in estimating land surface temperature (LST).They employed multivariate regression to analyze thermal comfort using various environmental parameters individually.Furthermore, Mijani et al. (2019Mijani et al. ( , 2020) ) demonstrated the significance of brightness, greenness, and wetness parameters in modeling outdoor thermal comfort.While the aforementioned studies employed ordinary least-square regression methods, our approach aimed to achieve more precise thermal comfort estimation using modern tools like machine learning models.One major advantage of these regression models is the potential to apply the derived coefficients to model thermal comfort in other cities as well.
The results of our study indicate significant variations in outdoor thermal comfort between summer and winter in Hyderabad.Many areas in the city experience moderate discomfort during summer but exhibit moderate to high comfort levels in winter.The classification of extreme discomfort to high discomfort is based on recent studies by Binarti et al. (2020) and Toy et al. (2007), along with relevant literature.The specified ranges are provided in Table 1, although they may slightly differ across cities.On average, the THI in Hyderabad during summer is approximately 7 °C higher than in winter.Similar studies have also found a similar temperature increase of 6-7 °C in other regions of India, as revealed in a recent study by Manavvi and Rajasekar (2022).
Compared to the mean THI, bare land and builtup areas experience an excess THI of approximately 2.5 °C and 2.1 °C, respectively.A recent study on Chennai by Rajan and Amirtham (2021) also confirmed that urban areas have 2.5-4.5 °C higher temperatures than rural areas.
The primary factor influencing higher THI in built-up and bare lands is the higher land surface temperature (LST) and surface brightness.discomfort.These results indicate that a significant portion of Hyderabad is covered with built-up areas, leading to discomfort for pedestrians.

Conclusion
The city of Hyderabad is characterized by different land cover types, with 51% being built-up areas, 28% barren land, 19% vegetation cover, and 2% water bodies.This distribution highlights the dominance of urban in the city's landscape.
Within the city, significant variations in land cover change can be observed.This study reveals that areas with built-up and barren land exhibit higher land surface temperatures (LST) compared to vegetated areas and water bodies.Specifically, the LST values over built-up areas were recorded at 30.28 °C, barren lands at 31.89 °C, vegetated areas at 31.94 °C and water bodies at 27.39 °C, indicating the impact of urbanization.These elevated LSTs can have implications for the outdoor thermal comfort of the surroundings, significantly affecting the well-being of the city's residents.
The findings suggest that the presence of water bodies and vegetation in urban spaces contributes to a substantial reduction in thermal loads.On the other hand, built-up areas and barren lands tend to be hotter, causing maximum discomfort for pedestrians and residents in urban areas.The mean thermal comfort (THI) of Hyderabad exceeds 25 °C during summer, indicating that almost all classes experience discomfort then.Different regions within the city exhibited varying degrees of outdoor thermal comfort, influenced by the percentages of different land cover classes.
Water bodies have a cooling effect, especially during summer, and a slight cooling effect during winter, contributing to better thermal comfort in urban spaces.The southern part of Hyderabad had higher temperatures compared to the northern part.Barren lands are identified as highly uncomfortable zones in summer, while built-up areas consistently exhibit discomfort conditions throughout the year.These results underscore the need to understand and quantify outdoor thermal comfort in response to rapid urbanization witnessed in Indian metropolitan cities over the past few decades.

Figure 1
Figure 1 Demographic representation of Hyderabad Figure 2 Flow chart of the methodology Figure 5 Supervised LULC classification of Hyderabad during the study period

Figure 6
Figure 6 Box plots of mean LST and THI with IQR over a different land covers and b different seasons

Figure 7
Figure 7 Percentage area of various classes of outdoor thermal comfort during a winter, b summer, c LULC and d extreme discomfort and high comfort over different mandal regions of Hyderabad

Figure 8
Figure 8 Identification of the regions changing from extreme hot to cold during summer to winter, and vice versa

Table 1
Classification of thermal comfort levels

Table 2
Correlation between environmental parameters and THI Spatial variation of LST over Hyderabad from Landsat 8-derived data 3628 P. S. H. Prasad and A. N. V. Satyanarayana Pure Appl.Geophys.