Introduction

Urban Heat Island (UHI) problem now becomes a global phenomenon affecting the city’s climate and its environmental quality to a larger extent (Voogt and Oke 2003; Chudnovsky et al. 2004). The magnitude of air and earth surface temperature difference between cities and its suburban is collectively defined as UHI (Landsberg 1981; Weng at al. 2004) or urban heat archipelago if the structure is multicellular in nature which could probably occur at the first part of the night (Bottyan and Unger 2003). Oke (1987) classified the UHI by two types: canopy layer UHI (consist with the rough air temperature of tree canopy and building with an upper boundary situated just below the roof level) and boundary layer UHI (it is located above the canopy layer, where the lower boundary influencing urban surface temperature) (Weng et al. 2004). Highly urbanized impervious surface could alter the biophysical compositions which could modify the physical process of earth-atmosphere energy exchange (Oke 1987). UHI studies have directed into two distinct way since the beginning of its application on urban climatic studies: (1) quantifying UHI phenomenon of a particular air parcel based on automobile transect and in situ weather network measurement, and (2) measuring earth surface temperature through the use of satellite remote sensing observation (Streutker 2003).

The changing dynamics of land use and land cover brought by human activity at any ecosystem scale seems to have incremental and unconscious, could lead to an undesirable modification of the native environment (Carlson and Arthur 2000). The unusual changes in land use and expansion of urbanized surface could substantially decrease half of the diurnal temperature range and could warm the mean earth surface temperature by 0.27 °C per century (Kalnay and Cai 2003). In addition, the sensitivity of land use and land cover changes and earth surface temperature found to be zero or close to zero in globally, however, the dynamics of land surface temperature, precipitation, and other climate components can be as large as or greater than those that result from the anthropogenic increase of well-mixed greenhouse gases (Pielke 2005). Moreover, the immediate effects of environmental changes are more prominent at regional and local scale than that of globally averaged values (Pielke 2005). It is also evident that long run land use changes by the human may have decreased the air temperature by 1–2 °C at mid-latitude agricultural region and could be warming by 1–2 °C in the deforested tropical region and extratropical region due to teleconnection process (Feddema et al. 2005). UHI phenomenon could be the extreme case of land use alteration at regional and local scale, where the unprecedented changes of vegetation cover, impervious surface area, and complex geometry of building in an urban landscape lowering the evaporative cooling, store the excess heat and hence, warm the earth surface (Foley et al. 2005). Likewise, Yue et al. (2007), Chen et al. (2006) reported that UHI intensity level is closely associated with the landscape pattern, i.e. composition and configuration of land use and land cover (LULC), the abundance of green cover, extent of impervious (Deng and Wu 2013b) surface and their changes and correlation of spatial metrics with urban heat island is helpful to determine the spatial change of land use classes and associated changes in surface temperature. Therefore, the detail investigation of long run LULC changes at any ecosystem scale could help us to explain the discrepant and unprecedented modification of earth surfaces, and also contribute to describing the directions and degree of other human-related environmental changes (Xiao and Weng 2007; Zeng et al. 2015a, b).

There is a substantial evidence of using remote sensing based biophysical indices to examine the thermal heat island phenomenon (Streutker 2003; Weng et al. 2004, 2007; Chen et al. 2006; Zhang et al. 2009; Amiri et al. 2009). NDVI was used in conjunction with soil-vegetation-atmospheric-transfer (SVAT) model to ascertain the surface soil moisture variability and fraction of vegetation cover (FVC) (Carlson et al. 1994), which can used for estimating leaf area index (LAI), surface impervious fraction, biomass and crop yield (Carlson and Ripley 1997). Weng et al. (2004) studied UHI intensity by investigating the applicability of vegetation fraction estimated from the linear spectral unmixing model as potential indicators of vegetation abundance. Recent studies have been giving thrust on dynamics of the biophysical indicator in response to changing urban heat island to measure the sustainability of an urban area and living conditions within the city. Besides, the selected biophysical descriptors are found highly correlated with land surface temperature (LST) in pearl river delta (PRD) in Guangdong Province, southern China, where NDVI, LSWI, and NDBaI were negatively associated with LST, NDBI was positively correlated with temperature (Chen et al. 2006). However, the correlation between LST and NDVI values related to different land-use types are significantly different (Yue et al. 2007). Yuan and Bauer (2007) analyzed the usability of NDVI and the impervious surface fraction as indicators of SUHI effects by examining the correlation between the LST, impervious surface fraction, and the NDVI. The similar observation has been made by Zhang et al. (2009), where the correlations between NDVI and LST are rather weak than that of the relationship between percent ISA, NDBI, and LST. This suggests that percent ISA, combined with LST, and NDBI, can quantitatively describe the spatial distribution and temporal variation of urban thermal patterns and associated LULC conditions.

The application of thermal remote sensing is considered the most feasible way to examine and characterizing the urban ecosystem at any ecosystem scale. Amiri et al. (2009), Chun and Guldmann (2014), Deng and Wu (2013a, b, c), Ishola et al. (2016), Ayanlade, (2016), Kayet et al. (2016), Sahana et al. (2016) have been used the multi-temporal Landsat 4 multi-spectral scanner (MSS), 5 thematic mapper (TM) and Landsat 7 enhanced thematic mapper (ETM+) sensors for estimating spatiotemporal LST and UHI. Buyantuyev and Wu (2010) was used the spectral and thermal data of advanced spaceborne thermal emission and reflection radiometer (ASTER) to examine the spatially explicit patterns of Phoenix’s SUHI. Cao et al. (2010) have also been used moderate resolution ASTER and fine resolution IKONOS data to quantify park cool island intensity (PCI). Therefore, the best advantage of incorporating the geospatial information to evaluate the status of an urban climatic phenomenon is its easy availability and economic viability.

In the present study, Greater Hyderabad Municipal Corporation (GHMC) is our study boundary, despite the uses of surroundings area. The planning for future sustainable area come under the administrative boundary and moreover, Hyderabad is the capital city of Andhra Pradesh, and the population growth is continuously increasing. Therefore, the study insists on assessing biophysical controls on LST and UHI in an urban ecosystem by implementing biophysical indices (Fig. 1). The discrepant impact of LULC changes and biophysical descriptors on UHI intensity of an urban ecosystem has been evaluated using thermal remote sensing data. Five distinct UHI intensity classes were generated using the relative brightness temperature approach and were deployed to assess the spatial cluster of UHI hot spot and cold spot. The simple linear and multiple regression models were demonstrated the complex and nonlinear behavior of the UHI and LST with the biophysical components. Therefore, the spatial coherence among the biophysical indices with LST ensembles the necessity of urban greenery and parks within the urban counterpart to mitigate the outdoor thermal discomfort to a reasonable extent.

Fig. 1
figure 1

Location of Greater Hyderabad Municipal Corporation (GHMC), with elevation and major transportation of the city

Full size image

Methods and data

Land surface temperature (LST) is the radiative temperature of the earth surface which plays a significant role in the physics of earth surface processes through the exchanges of water and energy with the atmosphere (Zhang et al. 2009). The thermal infrared spectrum (10.44–12.42 µm) of 16 days Landsat and 8 days composite of MODerate Resolution Imaging Spectroradiometer (MODIS) satellite data were used for the retrieval of land surface temperature in 2002, 2008, 2011 and 2015, respectively (Weng et al. 2004). The thermal band of Thematic Mapper (TM) having the spatial resolution of 120 m was further resampled onto 60 m for spatial adjustment with the enhanced thematic mapper (ETM+) data. The ETM+ data has been well calibrated in different training locations across the world with a negligible bias (Arvidson 2002). Landsat-based LST was estimated as follows: (1) converted the digital number (DN) of thermal bands into absolute units of top-of-atmospheric (TOA) radiance; (2) converted the top-of-atmospheric radiance to at-satellite brightness temperature; and (3) converted the at-satellite brightness temperature to earth surface temperature (Barsi et al. 2003; Li et al. 2011). Firstly, the actual DN’s of each thermal band was converted to spectral radiance as follows:

$${L_y}=(\frac{{({L_{\max }} - {L_{\min }})}}{{(Qca{l_{\max }} - Qca{l_{\min }})}} \times Qcal-Qca{l_{\min }})+{L_{\min }}$$
(1)

where \({L_y}\) is the at satellite spectral radiance, \({L_{\max }}\) and \({L_{\min }}\) are the spectral radiance of band 6 at maximum, \(Qca{l_{\max }}\) and \(Qca{l_{\min }}\) are 255 and 1 respectively and \(Qcal\) is the DN value of thermal band. For TM, \({L_{\max }}\) and \({L_{\min }}\) are 15.303 [Watts/(m2 sr μm)] and 1.238 [Watts/(m2 sr μm)], and for ETM+, the values are b61: 17.040 [Watts/(m2 sr μm)] and 0 [Watts/(m2 sr μm)]; b62: 12.650 [Watts/(m2 sr μm)] and 3.2 [Watts/(m2 sr μm)]. For Landsat 8, the equation can be followed:

$${L_y}=ML \times Qcal+AL$$
(2)

where \(ML\): (3.3 × 10−04) and \(AL\): (0.1) are the band-specific multiplicative rescaling factor and band-specific additive rescaling factor of b10 and b11, \(Qcal\) is the calibrated pixel values (DN). Afterward, at-satellite spectral radiance was converted into at-satellite brightness temperature as follows:

$${T_b}=\frac{{{K_2}}}{{\ln [({K_1}/{L_y})+1]}}$$
(3)

where \({T_b}\) is the at-satellite brightness temperature (K), \({K_1}\) and \({K_2}\) are the prelaunch calibrations constant in [Watts/(m2 sr μm)] and Kelvin, respectively. For Landsat TM, \({K_1}\) and \({K_2}\) are: 607.76 [Watts/(m2 sr μm)] and 1260.56 K. For Landsat ETM+, \({K_1}\) and \({K_2}\) are 666.09 [Watts/(m2 sr μm)] and 1282.71 K. For Landsat 8, \({K_1}\) and \({K_2}\) are 774.89 [Watts/(m2 sr μm)] and 1321.08 K for band 10 and 480.89 [Watts/(m2 sr μm)] and 1201.14 K for band 11 respectively. (Chander et al. 2009; Zhang et al. 2009; Chander and Markham 2003).

Finally, LST was retrieved after converting the at-satellite brightness temperature (K) to land surface temperature (LST) as follows:

$${T_s}=\frac{{{T_b}}}{{[1+(\lambda {T_b}/\rho )\ln \varepsilon ]}} - 273.15$$
(4)

where \({T_s}\) is at surface temperature (°C), \(\lambda\) is the wavelength of radiance (for which the peak response attain) (11.5 μm was used here), \(\varepsilon\) is the surface emissivity.

$$\rho =h \times (c/\sigma )$$

where \(h\) = Planck’s constant (6.626 × 10−34 Js), \(c\) = velocity of light (2.998 × 108 m/s−1), \(\sigma\) = Boltzmann constant (1.38 × 10−23 J/K). \(\rho\) = 14,380 used in this study (Weng et al. 2004; Markham and Baker 1985; Snyder et al. 1998).

$$\begin{gathered} {\varepsilon _{TM}}=0.004{P_v}+0.986 \hfill \\ {\varepsilon _{ETM+L8}}=0.02644{P_v}+0.96356, \hfill \\ \end{gathered}$$
(5)

where \({P_v}\) = fractional vegetation cover can be extracted as follows:

$${P_v}={\left[ {\frac{{NDVI - NDV{I_{\min }}}}{{NDV{I_{\max }} - NDV{I_{\min }}}}} \right]^2}$$
(6)
$$NDV{I_{TM,ETM+}}=\frac{{{\rho _4} - {\rho _3}}}{{{\rho _4}+{\rho _3}}}$$
(7)
$$NDV{I_{L8}}=\frac{{{\rho _5} - {\rho _4}}}{{{\rho _5}+{\rho _4}}}$$
(8)

where \(NDVI\) = normalized difference vegetation index, \({\rho _3}\), \({\rho _4}\) and \({\rho _5}\) are the spectral bands of Landsat TM, ETM+ and Landsat 8, respectively. \(NDV{I_{\max }}\) = 0.5 and \(NDV{I_{\min }}\) = 0.2. (Sobrino et al. 2004).

Four major biophysical indices, i.e. NDVI, LSWI, NDBI, and NDBaI were considered to examine the impact of urban landscape composition on surface heat island intensity and thermal changes. As the built-up and green surface have very high reflectance within the near infrared spectrum and very low reflectance in visible spectrum, the extent of built-up and green surface coverage have been measured using the reflectance values of three infrared bands (b4, b5, and b6) and one visible band (b3) (Xiao et al. 2001) as follows:

$$LSWI=\frac{{{\rho _4} - {\rho _5}}}{{{\rho _4}+{\rho _5}}}$$
(9)
$$NDBI=\frac{{{\rho _5} - {\rho _4}}}{{{\rho _5}+{\rho _4}}}$$
(10)
$$NDBaI=\frac{{{\rho _5} - {\rho _6}}}{{{\rho _5}+{\rho _6}}},$$
(11)

where LSWI land surface water index, NDBI normalized difference built-up index, and NDBaI normalized difference bareness index (Zha et al. 2003; Jackson et al. 2004; Maki et al. 2004; Zhao et al. 2005; Xu 2006; He et al. 2010).

Urban heat island effects considered as a relative phenomenon of thermal discrepancies between the urban and its suburban counterpart. The earlier study established a good correlation between the infrared and surface temperature over urban areas could be attributed reliable predictors of SUHI. The relative brightness temperature approach proposed by Xu et al. 2013 was used in this study to quantify the UHI in GHMC region as follows:

$${T_R}=\frac{{{T_i} - {T_a}}}{{{T_a}}},$$
(12)

where the relative brightness temperature (K) is \({T_R}\), \({T_i}\) is the brightness temperature at one place, and \({T_a}\) is the average brightness temperature of the region.

The spatial coherence between the estimated biophysical indices and LST has been assessed through standard model validation statistics: coefficient of determination (R2), correlation coefficient (r), root mean square error (RMSE) and bias.

$${R^2}=1 - \frac{{\sum {{{({X_i} - {Y_i})}^2}} }}{{\sum {{Y^2} - \frac{{\sum {{Y^2}} }}{N}} }}$$
(13)
$$r=\frac{{\sum {xy} }}{{\sqrt {\sum {{x^2}} \sum {{y^2}} } }}$$
(14)
$$RMSE=\sum\limits_{i=1}^n {({y_i} - } {\widehat y_i}{)^2}$$
(15)
$$Bias=\frac{{\sum\nolimits_{i=1}^n {({y_i} - {{\widehat y}_i})} }}{n},$$
(16)

where \(x\) and \(y\) is the explanatory and response variables, \(\widehat y\) is the y predicted, \(n\) is the total no. of observation.

Results

LULC dynamics from 1973 to 2015

LULC of the study area were analyzed for the year 1973, 2002, 2011 and 2015 are presented in Fig. 2. For the period of 1973 to 1991, maximum changes (37.38–11.45%) has been observed in the urban green space class. This category of LULC has a large negative change (−25.93% percentage point/pp), followed by farmland (17.12–37.38%) urban built-up (31.2–33.92%), aquatic vegetation, (7.65–9.73%), fallow land (4.12–5.66%) and water body (2.53–1.86%), respectively. Second largest change in this period is in farmland. This category of LULC has a large positive change (+20.26% pp) (Fig. 2; Table 1). Looking at the period of 1973 to 2015 together, aquatic vegetation cover changed from 7.65 to 10.18% (+2.53 pp), followed by water body [2.53–1.33% (−1.2 pp)], urban built-up [31.2–62.87% (31.67 pp)], farmland [17.12–5.17% (−11.95 pp)], fallow land [4.12–14.3% (10.18 pp)] and urban green space [37.38–6.15% (−31.23 pp)], respectively. The largest change in this period is in urban built-up. This category of LULC has a large positive change (−31.67 pp) in the period of 1973 to 2015. Second most significant change in this period is in urban green space. This category of LULC has a large negative change (−31.23 pp) (Fig. 2; Table 1). The multidirectional expansion of built-up urban surface from 1973 to 2015 derived from Landsat images are shown in Fig. 3. The maximum expansion is accounted towards the south-east and northwest direction of the study area. After that, the three major socioeconomic variables, i.e. population distribution, population density and settlement density have been evaluated and tried to correlate with UHI intensity. The highest population density was observed in 4, 5, 8, 9 circles, while settlement density is highest in 7 and 8th circles of the city (Fig. 4). LULC transformation between different classes has been assessed for two different periods, 1991 to 2015 and 2002 to 2015, respectively (Fig. 5).

Fig. 2
figure 2

Land use land cover change dynamics in Hyderabad city during 1973–2015

Full size image
Fig. 3
figure 3

Spatiotemporal dynamics of impervious surface in a 1973, b 1991, c 2002, d 2011 and e 2015 extracted from Landsat Satellite imagery

Full size image
Fig. 4
figure 4

Spatial variation of the socioeconomic variables; i.e. a distribution of population, b population density and c settlement density in Hyderabad city

Full size image
Fig. 5
figure 5

LULC conversion during 1991–2015 and 2002–2015 in GHMC

Full size image
Table 1 LULC dynamics in the GHMC during 1973–2015 (unit = %)
Full size table

LST and UHI dynamics across the study region

Spatio-temporal variation of mean air temperature was analyzed for the month of January in the years 1991, 2002 and 2015. There is a very clear pattern of higher mean air temperature in the southeasterly direction to lower mean air temperature in the NW direction in the study area. In the study period (1991–2015), there is a gradual increase in the coverage of higher mean air temperature from southeast direction to the central part and towards northwest direction. Looking at the LULC map, urban built up class reveals the similar pattern of increase in coverage over time (Fig. 6).The estimated land surface temperature was measured for the year 2002, 2011 and 2015 in the month of October are shown in Fig. 7. For the year 2002, the observed minimum temperature for different LULC class varies between 16.1 °C (built-up land) to 18.9 °C (water body). In the same year, for various LULC classes, maximum temperature varies between 28.3 °C (water body) to 34.8 (built-up land). Furthermore, in the year of 2011, considering different LULC classes, the minimum temperature varies between 12.8 °C (built-up area) to 21.9 °C (water body) while, the maximum temperature ranges from 29.1 °C (water body) to 34.1 (farmland). Besides, in 2015, for different LULC class minimum temperature varies between 18.7 °C (built-up land) to 19.7 °C (farmland), while concerning different LULC classes, maximum temperature ranges from 26.8 °C (water body) to 33.6 (built-up land) across the study region. There is a trend of increase in minimum temperature in the period of 2002 to 2015. The result showed that the estimated minimum temperature of different LULC (aquatic vegetation (16.3–19.1 °C), built-up land (16.1–18.7 °C), farmland (17.1–19.7 °C), fallow land (16.5–19.4 °C) and green space (16.6–19.0 °C)] changes sharply. An exception to this trend is water body, which remained at 18.9 °C in this period but with very high fluctuation in temperature (21.9 °C) in the year 2002 (Fig. 7; Table 2). However, results showed that there is a general trend of decrease in maximum temperature in the period of 2002 to 2015, found maximum over the areas occupied by aquatic vegetation cover, (34.3–31.2 °C), followed by water body (28.3–26.8 °C), built-up land (34.8–33.6 °C), farm land (33.1–30.9 °C) and fallow land (33.5–31.8 °C) (Fig. 7; Table 2). The exception to this trend is urban green space, which increased from 32.9–33.4 °C in this period but with a peak temperature of (34.0 °C) in the year 2002. The UHI Intensity for the year of 2002 (−0.61 to 0.37), 2011 (−0.51 to 0.45) and 2015 (−0.28 to 0.28) have been calculated and are shown in Fig. 8. The resulting five distinct UHI intensity classes, i.e. green island, weak heat island, medium heat island, strong heat island, and very strong heat island are being discussed in Fig. 9 and Table 3 with addressing the spatial heterogeneity and temporal discrepancies of transforming the thermal cluster of the region from cool spot or atoll to hot spot or island. Only southeastern and eastern part of the area converted to the cool island with time, while, the most of the region are being transformed into strong and extremely heat island during 2002–2015. Four different transects (North, South, East and West), are drawn from the center up to 10 km to examine the UHI sensitivity and thermal differences between the urban CBD and its rural counterpart for the year 1991, 2002, 2011 and 2015, respectively (Fig. 10).

Fig. 6
figure 6

Spatio-temporal variation of mean air temperature (°C) in a 1991, b 2002, and c 2015

Full size image
Fig. 7
figure 7figure 7

Spatio- temporal variation of land surface temperature (LST) on a 27th Jan 1991, b 17th Jan 2002, c 18th Jan 2011, d 13th Jan 2015, e 7th April 2002, f 21st May 2015, g 14th September 2002, h 1st October 2011, i 12th October 2015 in the study area

Full size image
Fig. 8
figure 8

Shows the nature of Urban Heat Island (UHI) Intensity on a September 2002, b October 2011 and c October 2015 in GHMC

Full size image
Fig. 9
figure 9

Spatial distribution of UHI intensity classes in GHMC observed in a 2002, b 2011 and c 2015

Full size image
Fig. 10
figure 10

LST retrieved from four transects having 2 km interval from the urban hotspot cluster for 1991, 2002, 2011 and 2015, respectively

Full size image
Table 2 Changes of LST (°C) in different LULC classes from 2002 to 2015
Full size table
Table 3 Spatiotemporal UHI class dynamics during 2002 to 2015 in Hyderabad city
Full size table

Dynamics of biophysical indices across the study region

Spatio-temporal variation of LSWI was analyzed for the month of September. There is a gradual change in the coverage of medium-higher values of LSWI to lower values of LSWI. This shift is prominent in the central to south-central part of the study area. This change is also evident in central to northern part of the study area, but in a scattered manner (Fig. 11). NDVI and NDBI have followed the similar tendencies as the minimum values are highly concentrated over built-up areas.

Fig. 11
figure 11

Spatiotemporal changes of LSWI, NDBI, and NDVI on Sep 2002, Oct 2011 and 2015 in the study region

Full size image

The coefficient of determination and person correlation coefficient test was done between the explanatory (biophysical variables) and response (LST). The coefficient of regression between NDVI for farmland (−152.23, 76.32), A. vegetation (12.24), urban built up (−8.01), fallow land (−7.74, 3.7) and urban green space (−4.72) and LST are in decreasing order (absolute value) and was found to be negative considering all LULC classes. The similar tendencies were observed while testing the spatial and temporal coherence between LSWI for farmland (−26.01), urban built up (−17.66), A. vegetation (−11.11), urban green space (−10.10) and fallow land (−9.42) with LST. The absolute value found to be decreased with time. The correlation was done between LST and NDBaI for farmland (30.00), A. vegetation (−27.78), urban green space (−25.98), urban built up (−25.62) and fallow land (9.12) are also found in decreasing order (absolute value) (Table 4). However, the coefficient of regression was found to be positive between LST and NDBI for all the LULC classes, while the absolute value follows the declining trend [farmland (24.89), urban built up (17.45), A. vegetation (10.94), fallow land (10.75) and urban green space (4.81)] (Figs. 12, 13). The sensitivity between the LST and the biophysical indices (NDVI, LSWI, NDBI, and NDBaI) are shown in Fig. 14. Fallow land is found to be most sensitive in the response of the incremental LST for all indices, followed by farmland, urban built-up, urban green space and aquatic vegetation class, respectively.

Fig. 12
figure 12

Shows the spatial coherence between the biophysical indices; i.e. a NDVI, b LSWI, c NDBaI and d NDBI and LST

Full size image
Fig. 13
figure 13

Shows the sensitivity between LST and the biophysical variables (NDVI, LSWI, NDBaI and NDBI) in different LULC categories across the study region

Full size image
Fig. 14
figure 14

shows the coherence between LST and biophysical indices (LSWI, NDBI, NDVI, and NDBaI) for four different LULC categories, viz. aquatic vegetation, farmland, fallow land and urban built-up, respectively

Full size image
Table 4 Biophysical dynamics of different LULC and its relationship with LST (°C) in Hyderabad city
Full size table

Discussion

Correlation between urban biophysical composition and LST and UHI

As it can be seen in Fig. 10, the spatial coherence between NDVI, LSWI, NDBaI and NDBI was found very high over the areas occupied by the built-up urban surface, which substantially indicates the importance of surface moisture content, amount, and nature of vegetation cover and thermal properties of soil on surface energy exchanges (Deng and Wu 2013c). Also, the high negative correlation between LST and biophysical composition observed across the city exhibits the physical significance of vegetation to moderate UHI and LST (Weng et al. 2004). Though the sensitivity and degree of association between LST and biophysical composition varied among different LULC (Fig. 14; Table 1). The similar has been reported earlier, where the presence of UHI cluster within the city controlled by the surface moisture content, fraction of vegetation cover, impervious faction to a larger extent (Guo et al. 2015). Moreover, irrespective of LULC, Guo et al. 2015 have been found that both NDVI (negatively) and NDBaI (positively) are significantly correlated with LST and UHI. However, in our study, we observed that, among the five major LULC classes, NDBaI mitigates the UHI and LST over aquatic vegetation and green space cover areas, whereas, the rest three LULC classes were strengthening the UHI. This nonlinearly can be attributed due to the high variability of surface moisture dynamics across the study region. As it has been reported by Deng and Wu 2013c, land class with very low NDVI values could be composed by a variety of non-vegetated components, i.e. dark and bright impervious soil made of concrete basalts, dark and moist soil with high organic matter content etc. which is very difficult to be characterized only by NDVI and having distinct reflectance properties within the thermal spectrum (Yuan and Bauer 2007; Li et al. 2011; Deng and Wu 2013c). In this study, the correlation between NDVI and estimated LST ranged from R2 = 0.6–0.7, similar to the R2 = 0.49 (Yue et al. 2007), r = 0.67 (Guo et al. 2015), r = 0.69 (Zhu et al. 2013), R2 = 0.83 (Li et al. 2011). While, the NDBI (r ≥ 0.8) and NDBaI (r ≥ 0.6) were being found to strengthen the UHI mostly, also been observed by Deng and Wu (2013c) using physical spectral unmixing and thermal mixing (SUTM) model depicts that in urban areas, NDBI and NDBaI based impervious fraction performs consistently and explained maximum variances than that of less impervious (rural) areas.

The correlation between vegetation and moisture abundance of each LULC and LST was examined using Pearson correlation coefficient test and regression analysis (Table 1) with 0.05 and 0.01 significance level. It can be seen in Table 1, which shows LST was negatively correlated with NDVI for all LULC classes with a different intensity. The highest negative association between NDVI and LULC types were accounted over the areas occupied by aquatic vegetation class (r = −0.86), followed by urban green space (r = 0.82), urban built-up (r = −0.8), farmland (r = −0.58) and follow land (r = −0.52), results is similar with the Indianapolis, USA (Weng et al. 2004), whereas, maximum negative correlation between NDVI and LST at 30 m resolution scale was found in cropland (r = −0.73), forest (r = −0.72), urban built-up (r = −0.62) and grassland (r = −0.36), respectively. Concerning the coherence between LSWI and LST, the highest values has been observed in built-up urban areas (r = −0.84), followed by aquatic vegetation cover (r = −0.83), fallow land (r = −0.82), farmland (r = −0.78) and urban green spaces (r = −0.75), respectively. It is worth to notice that, among all indices, NDBI explain the maximum model variance with the highest correlation value (r = 0.84), followed by LSWI (r = −0.8), NDBaI (r = 0.79) and NDVI (r = −0.7) for all LULC types strongly indicates the impact of surface imperviousness on thermal anomalies and surface energy balance of a city region.

LULC dynamics and its impact on surface UHI and LST

The spatiotemporal distribution and changes of different LULC categories since 1973 to 2015 are shown in Fig. 2. It can be seen in Fig. 4, that the estimated LST has been varied distinctly over the central built-up urban areas in comparison to other classes. This is an indication of the presence of hot objects especially in the central part of the region displaying the more importance of influencing surface UHI patterns (Weng et al. 2007). The similar has been observed while built-up impervious surface (hot objects) was found positively correlated with LST and vegetation (cold objects) was negatively associated with LST ensembles the impact of thermal fractional components (hot objects and cold objects) in conjunction with the biophysical descriptors (NDVI, LSWI, NDBI, NDBaI) to changes the spatiotemporal pattern of LST cluster in a city (Lu and Weng 2006). However, the uses of the biophysical descriptor, especially NDVI to examine the SUHI phenomenon are suggested to restrict for all season as NDVI suffers from apparent seasonal variation (Yuan and Bauer 2007). The changing dynamics of different UHI classes from 2002 to 2015 are shown in Table 2. The maximum pace of changes among the UHI classes had seen in the extremely strong urban heat island categories (72.22), followed by strong heat island (38.3), medium heat island (11.84), weak heat island (−5.04) and green island (−6.94), respectively. It can also be noted that built-up urban area was the dominant land use types in extremely heat island zones, whereas, vegetation and water bodies characterize the green island and weak heat island areas. The similar has been observed Liu and Weng 2008, where the spatial configuration of dominant LULC types are controlling the spatial configuration of the different temperature and UHI clusters in that zone. The temperature and vegetation index (TVX) space was used in several studies to examine the individual and multiplicative impact of different LULC types on LST and UHI dynamics (Lambin and Ehrlich 1996, 1997; Gillies et al. 1997; Sandholt et al. 2002; Goward et al. 2002). The spatial distribution and temporal dynamics of different UHI classes for the year of 2002, 2011 and 2015 are shown in Fig. 6. This could be attributed to the discrepant nature of surface LST across the city region with uneven spatial expansion of extremely strong and strong heat island at the expense of natural and semi-natural surface cover leads to undesirable thermal anomalies in GHMC due to uncontrolled urban expansion. The similar has been observed in PRD, in Guangdong Province, southern China, is one of the regions experiencing rapid urbanization that has resulted in remarkable UHI effect, which is affecting the regional climate, environment, and socio-economic development to a larger extent (Chen et al. 2006). However, the estimated LST being varied highly in intra LULC classes than that of inter LULC variation. The biophysical sensitivity between the indices and LST for different LULC was assessed in this study and are drawn in Fig. 12. In built-up urban areas, NDVI is found highly sensitive with LST, followed by NDBI, LSWI, and NDBaI, respectively, indicates the cumulative importance of urban green cover to moderate UHI and thermal discomfort of a city. However, fallow land shows the same sensitivity for LSWI, NDBI, and NDBaI, and less sensitivity between NDVI and LST exhibited the less green coverage in this zone. Farmland is being found less sensitive with LST, followed by aquatic vegetation cover and urban green cover, respectively.

Conclusion

In this study, the discrepant effects of urbanization and biophysical changes on UHI have been investigated in Hyderabad city, the sixth largest urban agglomeration of India. The spatially explicit coherence between the selected indices and UHI of different LULC classes are found highly correlated across the region. Therefore, the main conclusions of the study are as follows: (1) LST was negatively correlated with NDVI for all LULC classes with a different intensity. The highest negative association between NDVI and LULC types were accounted over the areas occupied by aquatic vegetation class (r = −0.86), followed by urban green space (r = 0.82), urban built-up (r = −0.8), farmland (r = −0.58) and follow land (r = −0.52), respectively. (2) Among all indices, NDBI explain the maximum model variance with the highest correlation value (r = 0.84), followed by LSWI (r = −0.8), NDBaI (r = 0.79) and NDVI (r = −0.7) for all LULC types strongly indicates the impact of surface imperviousness on thermal anomalies and surface energy balance of a city region. (3) Among the five major LULC classes, NDBaI mitigates the UHI and LST over aquatic vegetation and green surface areas, whereas, the rest three LULC classes were strengthening the UHI. (4) The biophysical sensitivity between the indices and LST for different LULC were varied distinctly. In built-up urban areas, NDVI is found highly sensitive with LST, followed by NDBI, LSWI, and NDBaI. (5)Among the five UHI classes, extremely strong urban heat island categories changes rapidly (72.22) during 2002–2015, followed by strong heat island (38.3), medium heat island (11.84), weak heat island (−5.04) and green island (−6.94), respectively. (6) The high negative correlation between LST and biophysical composition observed across the city exhibits the physical significance of vegetation to moderate UHI and LST.