Introduction

Rapid urbanization in the world is quite alarming, especially in developing countries such as India (Kumar et al. 2007). Horizontal expansion of urban areas is continuously increasing and is consuming agricultural area of its periphery, thereby increasing input impervious surface through expansion and densification of these urban areas. Land cover alteration is replacing natural surfaces with man-made surfaces. Transformation of land use has promoted many meteorological problems like urban heat island and change in the climate at micro-level everywhere in the world (Ridd 1995). Population pressure, reduction in green areas, increase in impervious surfaces through land use transformation in various urban activities, and heavy traffic load have further added thermal stress on the city and changed the thermal character of the areas, making them warmer than the surroundings, which has a negative impact on the urban ecosystem and nearby region. Urban heat island is the result of this process. This phenomenon is present in almost all the big and small cities but of course with varying intensity and magnitude around the world. In developing countries, unplanned growth of the cities and reduction of greeneries are aggravating and complicating this problem. Million cities of India are expanding at a much faster rate than the rest, and almost all are facing the same problem coupled with other environmental problems. India’s urban population by 2036 is expected to reach 594 million, which was 377 million in 2011. As per 2011 census of India, 31% population of the country is living in urban areas, which is expected to rise up to 60% by 2050 (https://censusindia.gov.in/2011census/population_enumeration.html/).

The estimation of land surface temperature (LST) is essential to know the pattern of temporal changes in the temperature and also to identify those areas, which need immediate attention and require mitigation measures. Many studies have been carried out to study the variability between greenness and surface temperature and tried to assess the present status of land surface temperature (Weng et al. 2004, Raynolds et al. 2008; Mallick et al. 2008; Sheik 2019) and also suggested measures to cope up with this problem.

Land surface temperature (LST) is an important factor for the determination of several biophysical parameters and processes. Remote sensing images provide very useful information for extraction, determination, and measurements of land surface temperature (LST). It enables to monitor surface energy influx, evapotranspiration, crop production, near surface temperature etc. and also its association with the land use (Owen et al. 1998; Stisen et al. 2007; Xiao et al. 2008; Li et al. 2009; Amiri et al. 2009; Jiang and Tian 2010; Dontree 2010; Liu et al. 2017, and Igun and Williams 2018; Liu et al. 2020; Imran et al. 2021).

The results of various studies clearly conclude that there is a strong relationship between the LST (land surface temperature) and land use/land cover class. Availability of remote sensing data has opened up a new dimension in the study of land surface temperature (LST)/UHI. A wide range of sensors, like Landsat 4 and 5 (TM), 7 (ETM +), 8 (TIRS 1 and 2), Advanced Spaceborne Thermal Emission and Reflection (ASTER), Moderate Resolution Imaging Spectroradiometer (MODIS), Advanced Very High-Resolution Radiometer (AVHRR), and others are utilized for the extraction of land surface temperature (LST). Many scholars have studied heat islands and land surface temperature and have suggested mitigation measures.

Generally, biophysical components are defined as a set of indicators that are able to track the human impact on a given environment (Dietz et al. 2007; Sannigrahi et al. 2017). In most of the studies, NDVI has been calculated to assess UHI/LST. Several researchers used Landsat imageries to map land use/cover and calculated temperature as well and used NDVI to describe land surface vegetation coverage (Sundara et al. 2012; Xiong et al. 2012; Lu et al. 2013). Normalized difference water index (NDWI), normalized difference built-up index (NDBI), and normalized difference bareness index (NDBaI) are considerably used indices in LST-related research works (Zha et al. 2003; Essa et al. 2012; Guha and Govil 2021a; Taloor et al. 2021). To evaluate and compare the impact of biophysical components on urban environment, NDVI, NDBI, NDWI, MNDWI, NDBal, BSI, and SAVI have been calculated in this study.

In India, medium-sized cities like Gorakhpur are also under the process of expansion and intensification. Drastic urbanization and resultant change in LU/LC have changed the thermal character of the land surface. This kind of study will help to identify the present status and trend of the land surface temperature, so that preventive measures can be formulated and implemented before the situation worsens.

Materials and methods

Methodology adopted to quantify and to assess temporal change in the land surface temperature phenomenon of Gorakhpur city and its environs includes satellite data collection, classification of images, preparation of land use/land cover maps, preparation of different maps on selected indices, quantification of temperature, and correlation of land surface temperature with land use/land cover changes (Fig. 2). ArcGIS 10.2, ERDAS IMAGINE 2010, MS Office 2016, and SPSS 20 software have been used to achieve the objectives of the study.

Data collection

This study has incorporated remotely sensed Landsat image data and applied remote sensing/digital image processing techniques to quantify the impact of urban growth on an land surface temperature of Gorakhpur city by using the data of last 28 years. Cloud-free Landsat time series satellite data of 1990, 2000, 2010, and 2019 have been downloaded from USGS Earth Explorer website (www.earthexplorer.usgs.gov). A Landsat-5TM + , 7 ETM + , and Landsat-8, OLI/TIRS image scenes (Row 142/Path 41) were selected for this study (Table 1). The multispectral image data consists of eight spectral bands (three visible, one NIR, two MIRs, and two thermals) and has a spatial resolution of 30 m for the reflective bands and 60 m for the thermal bands. Thermal infrared spectral provides data for land surface temperature measurement.

Table 1 Details of Landsat data used
Full size table

All the data have been first pre-processed and projected to the Universal Transverse Mercator (UTM) projection system. Topographic sheets of 1:50,000 scale were used for geometric corrections. The following methodologies have been adopted to quantify land surface temperature and various biophysical parameters.

Land use/land cover class extraction

The images were geometrically and radiometrically corrected, and land use/land cover maps were extracted using supervised classification with the maximum likelihood classification algorithm, which is supposed to be the most popular and useful parametric classifier. The maximum likelihood (ML) is a supervised classification method derived from the Bayes theorem, which states that the a posteriori distribution P (i|ω), i.e., the probability that a pixel with feature vector ω belongs to class i, is given by

$$P\left(i|\omega \right)=(P\left(\omega |i\right)P(i))/P(\omega )$$
(1)

where P (ω|i) is the likelihood function; P (i) is the a priori information, i.e., the probability that class i occurs in the study area; and P(ω)is the probability that ω is observed, which can be written as

$$P\left(\omega \right)=\sum\nolimits_{i=1}^{M}P\left(\omega |i\right)P(i)$$
(2)

where M is the number of classes. P (ω) is often treated as a normalization constant to ensure sums to 1:

$$\sum\nolimits_{i=1}^{M}P(i|\omega )$$
(3)

In this study, a land use/land cover classification scheme has been developed keeping in mind the classification system proposed by the National Remote Sensing Centre, Hyderabad, Telangana, India (NRSC, 2014). The seven land use/ land cover classes, viz., built-up, agriculture, forest, water bodies, sand, vacant land, and miscellaneous, are included in this classification system, depending on the availability of type of land use/land cover, and the area under these classes has been calculated.

Accuracy assessment

Accuracy assessment or validation is a key component of any project employing spatial data (Congalton 2001). The accuracy assessment/validation of the spatial distribution of different land use/land cover classes was done by comparing the results of classification with high-resolution images available on Google Earth, personal experience, and field samples.

Detailed field knowledge and survey conducted during studies in this area served as the bases for the estimation of accuracy of land use/land cover classification for 4 years.

Overall accuracy assessment

$$OA=\frac{Number\;of\;true\;positive\;+\;Number\;of\;true\;Negative}{Pixels\;in\;the\;ground\;truth}\times100\%$$
(4)
  1. i.

    User’s accuracy assessment

    $$UA= \frac{{Row elements}_{diagonal}}{{Row}_{total}}$$
    (5)
  2. ii.

    Producer’s accuracy assessment

    $$PA= \frac{{Column elements}_{diagonal}}{{Column}_{total}}$$
    (6)
  3. iii.

    Kappa coefficient (K)

    $$K= \frac{N\sum_{1=1}^{r}{x}_{ii-{\sum }_{i=1}^{r}{x}_{i+{x}_{+i}}}}{{N}^{2}-\sum_{i=1}^{r}{x}_{i+}{x}_{+i}}$$
    (7)

where:

  • r it the number of rows/columns in confusion matrix.

  • Xii is the number of observations in row i and column i.

  • Xi + is the total number of row i.

  • X + i is the total number of column i.

  • N is the number of observations.

The value of K with more than 0.81–1 is considered as almost perfect agreement (Landis and Koch 1977 Fleiss 1981).

Biophysical components

These components are defined as a set of indicators that are able to track the human impact on the given environment (Dietz et al., 2007). To evaluate and compare the impact of biophysical components on urban environment, NDVI, NDBI, NDWI, MNDWI, NDBal, BSI, and SAVI have been calculated and mapped.

Normalized difference vegetation index (NDVI)

NDVI gives the intensity of greenness in the area and quantifies the vegetation coverage. Range of NDVI product is from − 1 to + 1. Higher values refer to healthy and dense vegetation, while lower values indicate sparse vegetation. Water and built-up areas or non-vegetative areas are represented by near zero or negative values. This index is widely used by the researchers for calculation of land surface temperature (LST) or UHI. Landsat satellite image has been used for the generation of NDVI from the red (3rd) and near-infrared (4th) bands, using the following equation:

$$NDVI= \frac{NIR-R}{NIR+R}$$
(8)

Normalized difference built-up index

This index serves as an indicator of surface urban heat island (SUHI) (Li et al., 2009). The normalized difference built-up index is a multispectral index established on the ratio between higher reflectance in the shortwave infrared (SWIR) region to the near-infrared (NIR) region (Lu and Weng, 2006). It is the important index to extract built-up area, i.e., impervious surfaces. This index has been used to extract built-up areas in the study region using following equation:

$$NDBI= \frac{MIR-NIR}{MIR+NIR}$$
(9)

Normalized difference water index

The NDWI, a remote sensing image-based indicator, is developed by Gao (1996) to enhance the water-related features of the landscapes and is used in estimating the leaf content at canopy level. According to Gao, NDWI is a good indicator for vegetation water content and is less sensitive to atmospheric scattering effects than normalized difference vegetation index. This index is derived from the near-infrared and shortwave infrared bands using the following equation:

$$NDWI= \frac{G-NIR}{G+NIR}$$
(10)

The range of NDWI product varies between − 1 and + 1, depending not only on the leaf water content but also on the vegetation type and cover. High NDWI (in blue) value is related to high vegetation water content and high vegetation cover, while low NDWI values (in yellow) correspond to low vegetation water content and low vegetation fraction cover. NDWI value will decrease during the period of water stress.

Modified normalized difference water index (MNDWI)

MNDWI is useful in the estimation of water bodies significantly without any built-up area and vegetation noise (Xu 2006). It is more suitable for enhancing and extracting water information for a water body with a background dominated by built-up land areas, because of its advantage in reducing and removing built-up noise over NDWI:

$$MNDWI= \frac{G-MIR}{G+MIR}$$
(11)

Normalized difference bareness index (NDBal)

This index can be used to derive the distribution of bare land as well as temporal study to assess the change in bare land. It can be used to monitor the anthropogenic impact:

$$NDBal= \frac{SWIR1-TIR}{SWIR1+TIR}$$
(12)

Bare soil index (BSI)

This indicator is useful in the identification of bare soil areas, vacant lands, and urban expansion as well. BSI has been calculated using the following formula:

$$BSI= \frac{\left(SWIR1+R\right)-(NIR+B)}{\left(SWIR1+R\right)+(NIR+B)}$$
(13)

Soil-adjusted vegetation index (SAVI)

The soil-adjusted vegetation index was developed as a modification of NDVI to correct the influence of soil brightness, when vegetative cover is low. Here, L is the soil brightness correction factor. Lower values indicate lower amount of green vegetation:

$$SAVI= \frac{(NIR-R)(1+L)}{(NIR+R+L)}$$
(14)

Land surface temperature (LST)

The spectral radiance, converted from pixel DN values, is used to compute brightness temperature (i.e., blackbody temperature) under the assumption of unit emissivity and using pre-launch calibration constants (Landsat Project Science Office, 2002). Using the radiance, rescaling factors are provided in the metadata file (USGS, Landsat 8 handbook, 2016). Thermal band (6th) of the Landsat Thematic Mapper 5 (TM) satellite images have been used to extract the land surface temperature, near-infrared, and red (R) for estimating vegetation health and other related information. The following methodology has been adopted to calculate land surface temperature of the study region.

Conversion of digital number (DN) to spectral radiance (Lλ)

The spectral radiance (Lλ) can be given (USGS 2016) as-

$$L\lambda ={L}_{min\lambda }+\left[\frac{({L}_{max\lambda }-{L}_{min\lambda )}}{({Q}_{CAL max}-{Q}_{CAL min})}\times {Q}_{CAL}\right]$$
(15)

where L is the sensor-derived spectral reflectance (W m − 2 sr − 1 μm − 1) and Lmaxλ and Lminλ are the minimum and maximum spectral radiances for band 6, respectively. QCAL is the digital number (DN) of each pixel. QCALmin is the minimum DN value of the image, here QCALmin = 0; and QCALmax is the maximum DN value of the image, here QCALmax = 255.

Conversion of spectral radiance (Lλ) to brightness temperature (Tβ)

The emissivity can be computed by using the formula of Artis and Carnahan (1982):

$$T\beta =\frac{{K}_{2}}{In\left(\frac{{K}_{1}}{{L}_{\lambda }}+1\right)}-273.15$$
(16)

where Tβ is the brightness temperature (K), Lλ represents spectral radiance of sensor (W m − 2 sr − 1 μm − 1), and K1 and K2 are the calibration constant (K1 = 60.776 mW cm − 2 sr − 2 μm − 1 and K2 = 1260.56 K for Landsat band). An absolute zero (approximately − 273.15 °C) should be added to revise the temperature in terms of degree Celsius.

Emissivity correction through NDVI method

This is done using following formula:

$$Pv=\left(\frac{{NDVI-NDVI}_{Soil}}{{NDVI}_{veg}+{NDVI}_{Soil}}\right)$$
(17)

where NDVIsoil and NDVIveg are the threshold values of soil pixel and the pixel of vegetation. The threshold values of NDVIs are 0.2, and NDVIV is 0.7.

Land surface emissivity calculation (ϵ)

The land surface emissivity (є) is calculated as

$$\in \lambda = {\in }_{veg\lambda }{P}_{v}+{\in }_{soil\lambda }\left(1-{P}_{v}\right)+{C}_{v}$$
(18)

where ϵveg and ϵsoil are the vegetation and soil emissivity, respectively, and C is the representation of surface roughness.

Retrieval of land surface temperature

After the calculation of land surface emissivity, land surface temperature is retrieved using the following formula:

$$LST=\frac{T\beta }{[1+\{(\lambda \bullet T\beta /\rho )In\bullet \epsilon \lambda \}]}$$
(19)

where land surface temperature means land surface temperature in °C, Tβ is the sensor brightness temperature (°C), \(\lambda\) is the wavelength of emitted radiance in meter (\(\lambda\)= 10.895 µm), and є \(\lambda\) is the emissivity:

$$\rho =h\frac{C}{\sigma }=1.438\times {10}^{-2}mK$$
(20)

where ϭ is the Boltzmann constant (1.38 × 10−23 J K−1), h is the Planck’s constant (6.626 × 10−34 J K−1), and C is the velocity of light (2.998 × 108 m s−1).

Multiple linear regression

Multiple linear regression has been derived in the following manner:

$$y={\beta }_{0}+{\beta }_{1}{w}_{1}+{\beta }_{2}{w}_{2}+{\beta }_{3}{w}_{3}+{\beta }_{4}{w}_{4}+{\beta }_{5}{w}_{5}+{\beta }_{6}{w}_{6}+{\beta }_{7}{w}_{7}$$

where:

  • y is the dependent variable.

  • β0 is the constant.

  • β(1–7) is the unstandardized coefficient for each predictor variables.

Predictor variables w1–w7:

  • w1, BSI (°C)

  • w2, MNDWI

  • w3, NDBal

  • w4, NDBI

  • w5, NDVI

  • w6, NDWI

  • w7, SAVI

Hot spot analysis (Getis‑Ord Gi* statistics)

Hot spot analysis is used to identify the location of statistically significant hot spots and cold spots in the data by aggregating points of occurrence into polygon as points that are in proximity to one another based on a calculated distance. It produces Z scores (Gi scores) and P values (Gi P values). Getis-Ord Gi* is calculated (Getis and Ord 1992) as

$${G}_{i}^{*}=\frac{{\sum }_{j=1}^{n}{\varphi }_{i,j}xj-\overline{x}{\sum }_{j=1}^{n}{\varphi }_{i,j}}{\sqrt[n]{\frac{n{\sum }_{j=1}^{n}{\varphi }_{i,j}^{2}-\left({\sum }_{j=1}^{n}{\varphi }_{i,j}\right)}{n-1}}}$$
(21)

where xj is the attribute value for the feature j, φ,j is the spatial weight between feature i, and j,n is equal to the total number of the feature:

$$\overline{x }=\frac{{\sum }_{j=1}^{n}xi}{n}$$
(22)

and

$$S= \sqrt{\frac{{\sum }_{j=i}^{n}x{j}^{2}-{(\overline{x })}^{2}}{n}}$$
(23)

Gi* in hot spot analysis classifies the data into range from − 3 (cold spot, 99% confidence) to 3 (hot spot, 99% confidence) with 0 being non-significant. A high Z score and small P value indicate a significant hot spot, while low negative Z score and small P value represent a significant cold spot. The higher or lower Z score shows more intense clustering of high and low value, respectively. A Z score near zero means no spatial clustering.

Study area

Gorakhpur city, one of the most important and fast expanding cities of eastern Uttar Pradesh (India), is situated in the fertile tract of Saryupar Plain, the subdivision of the Middle Ganga Plain. Gorakhpur, the administrative headquarter of North Eastern Railway (since 1954) as well as the district and divisional headquarter, is situated at the confluence of Rapti and Rohin rivers. Towards the central western end of the city, River Rohin meets River Rapti, and both these rivers form the western boundary of the city. It is situated between 26°42ʹ N to 26°47ʹ N latitude and 83°20ʹ E to 83°25ʹ E longitude (Fig. 1). Railway line passes through the middle of the city and divides it into two distinct parts northern and southern. Gorakhpur has a number of large perennial lakes, formed in most of the cases by the abandoned channels of rivers; Ramgarh lake is the biggest lake in the area. Chilika Lake and Sumer Sagar are also worth mentioning water bodies.

Fig. 1
figure 1

Flow chart of methodology for measuring the land surface temperature

Full size image

The city has been expanding since 1951, when it had a population of 1,32,436, which has increased to 2,30,911 in 1971, to 3,07,501in 1981, to 5,05,566 in 1991, to 6,22,701 in 2001, and finally reaching to 6.73 lakh in 2011 (source: census of India). The highest increase (64.41 per cent) was recorded between 1981 and 1991 decade, when in 1982, 47 villages were incorporated within the municipal limit. The area of interest (AOI) selected for the study area (Fig. 2) includes the surrounding non-urban area of Gorakhpur city also, so that the variation and comparison of the temperature between built-up and other land uses can easily be observed.

Fig. 2
figure 2

Location of the study area

Full size image

Results

Land use/land cover changes

Land use classification scheme has been adopted and supervised land use/land cover classification maps for the year 1990, 2000, 2010, and 2019 have been prepared, using maximum likelihood classification algorithm, and seven categories, viz., built-up, agriculture, forest, water bodies, sand, vacant land, and unclassified, have been identified. These classified images are important evidence of urban extent and also clearly identify the growth, pattern, direction, LU/LC dynamics, and spatio-temporal distribution. It is observed that Gorakhpur city has experienced rapid transformation in the land use/land cover in the last 28 years (Fig. 3, Table 2). The total area of the rectangular area of interest (AOI) is 483.97 sq. km.

Fig. 3
figure 3

Land use/land cover changes (1990–2019)

Full size image
Table 2 Land use/land cover changes (area in sq. km 1990–2019)
Full size table

An analysis, based on interpretation of land use/land cover maps of selected years, shows obvious changes in the area under built-up and vacant land. In 1990, built-up areas were limited to the middle part of the city with the expansion of 56.50 sq. km., while, in 2019, it has expanded beyond the municipal boundaries with more density and increased up to 171.56 sq. km. Thus, 23.77% increase has been noticed during this period (Table 2). Increasing demand of land for residential, industrial, educational, and other urban activities has forced the other land use classes to shrink, and the tendency of encroachment of urban land use on other types of land uses is increasing. As a result, the size of urban built-up kept on increasing. Land under agriculture, forest, sand, and water bodies has shrunk, and negative growth in all these land use classes has been noticed for the same time span. Agricultural land has experienced negative growth of about 21% and because of this tendency, agricultural land has been reduced from 313.89 sq. km in 1990 to 212.72 sq. km. in 2019.

Vacant spaces have also experienced positive growth during the study period because of the urban expansion and resulting clearing of agricultural land. This area has been left vacant sometimes for speculation and will gradually be converted into built-up area in the coming years. The built-up areas in the study area are not only densifying but also expanding almost in every direction, except the western part, where the presence of Rapti River and embankment to protect the city from annual floods has forced any built-up area to develop and less urban growth is observed particularly in the south-western part of the city. In other directions, the city growth has crossed municipal boundaries of Gorakhpur. It is also clear from the Fig. 4 that transformation of land use is mostly into built-up area from almost every land use classes during 1990–2019.

Fig. 4
figure 4

Land use/land cover change trajectories (1990–2019)

Full size image

After classifying the images into different land use classes, classification accuracy has also been achieved. Table 3 shows producer’s, user’s, and overall accuracy and kappa coefficient of the land use/land cover classification, which has been derived using 100 control points selected by high-resolution image (Google earth images), field survey, and field knowledge of the study area. Overall accuracy is found to be between 92 and 96%, while kappa coefficient (K) value ranges between 0.864 and 0.936, which is considered an almost perfect agreement.

Table 3 Classification accuracy assessment
Full size table

Biophysical components analysis

NDVI analysis

Natural vegetation provides a cooling effect on the surroundings. This clearly influences the micro-climate of any urban area. For Gorakhpur city and in its vicinity, vegetation cover is continuously decreasing due to increasing population pressure, resulting in urban expansion and land use transformation. These vegetative areas are either completely removed or converted into non-vegetative areas or their density has decreased.

In 1990, north-eastern, eastern, and south-western part of the city had high NDVI values (Fig. 5). These areas had high density of vegetation but due to urban expansion, other activities started surfacing, and gradually, non-vegetative areas started increasing, forcing the vegetative areas to shrink, and by 2019, the entire northern, north-eastern, and eastern parts (except the eastern part where reserve forest is located) were converted into the impervious surface. Only the south-western part of the city, beyond River Rapti, has sparse vegetation. In the eastern part (the reserve forest) of the city, the density of the trees has been decreasing over the years and so is the value of NDVI. Hence, the result of this is clearly seen on the land surface temperature of the area.

Fig. 5
figure 5

NDVI and SAVI Index

Full size image

Soil-adjusted vegetation index (SAVI)

SAVI index maps for the years 1990 to 2019 show almost the same results as NDVI maps of the same years (Fig. 5). Lower values are found in the areas of the city, where lower NDVI values have been observed. Spatial distribution and intensity of lower values are increasing with the time and following the distribution pattern of NDVI.

Normalized difference built-up index

It is the most important and widely used technique to measure the spectral signature of the urban built-up and to extract built-up area class, i.e., impervious surface in the land use classification. City expansion trend can also be assessed by applying this index on temporal data. The difference in impervious surface cover is determined by raster subtraction for monitoring the changes.

NDBI images for the study region are prepared for the year 1990, 2000, 2010, and 2019 (Fig. 6). In 1990, the city expansion was limited to the mid-western part of the city, and large stretch of agriculture land was available on the north-eastern part and to some extent in the eastern part of the city. In 2008, the built-up expansion of the city started to encroach this green area. In 2010, almost every part of the land within the city limit, except the south-western corner, had been converted into the impervious surface, and in 2019, it has expanded beyond the city municipal boundary, particularly in the north-east region and along the southern boundary.

Fig. 6
figure 6

Normalized difference built-up index (NDBI)

Full size image

Normalized difference water index

NDWI shows the water content of leaves and is considered a better indicator than NDVI. This index for the study area shows that the highest value, which should be below one, is continuously decreasing. In 1990, it was 0.33, which has reduced to 0.17 in 2019. This trend is the clear indicator that vegetation cover is shrinking due to expansion of built-up areas. The areas of high value are also reducing, which is very clearly seen in the map of 2010 and 2019 (Fig. 7).

Fig. 7
figure 7

NDWI and MNDWI index

Full size image

Modified normalized difference water index (MNDWI)

This index clearly identifies and demarcates the water bodies, situated within the built-up areas. Comparison of the maps prepared through this index, which shows the effect of urban expansion, is clearly visible on the water bodies. The biggest water body is located in the south-eastern part of the city. The western margin of this water body shows the reduction in size, which has been either converted into built-up areas or filled and left as vacant land for future development (Fig. 7). Even houses located at the bank of north-west margin are continuously filling the land and converting the water bodies into residential areas due to poor monitoring of the administration.

Normalized difference bareness index (NDBal)

Bareness index identifies the distribution and changes in bare land and is associated with high values. Bare land shows increasing trend (Fig. 8). Particularly the outer areas of old city. north-eastern, and southern parts of the city have exhibited increasing values and spatial expansion with the time.

Fig. 8
figure 8

NDBal and BSI index

Full size image

Bare soil index (BSI)

Urban expansion results in the removal of vegetation or encroachment of agricultural land and convert these areas into vacant land. BSI for Gorakhpur city for the year 1990, 2000, 2010, and 2019 shows that the areas of high BSI are continuously increasing. In 1990, the northern part of the city having the low BSI value has been transformed into high BSI values in 2019 (Fig. 8).

Land surface temperature pattern and statistics

From brightness temperature (TB) and emissivity images, the final land surface temperature (LST) maps for four years have been generated by using algorithms of ERDAS Imagine 10. The spatial and temporal distribution trend of land surface temperature during 1991–2019 in the study area is shown in Fig. 9, which indicates that remarkable increase in land surface temperature has been observed in the northern part of within the municipal boundary and in the eastern part outside the city limit.

Fig. 9
figure 9

Land surface temperature (1990–2019)

Full size image

Multiple linear regression

The multiple linear regression analysis has been used to find out the relationship between independent and dependent variables, and their effect on the pattern of land surface temperature has also been assessed. Multiple linear regression can also be used to predict future pattern of land surface temperature. In this study, variables like BSI, MNDWI, NDBal, NDBI, NDVI, NDWI, and SAVI have been used for the selected four time periods to assess the status of land surface temperature with the changeability of biophysical parameters.

The correlation coefficient of the year 1990 is R = 0.732, and the coefficient of determination R2 = 0.536, i.e., 53.60%, and the predicted R2 = 50.1% (Table 6). It means, this year land surface temperature was not much affected, as the built-up area was less, and significant area existed under vegetation cover and water bodies. But the results of later years indicate significant correlation coefficient among the variables (60% to 82%). The correlation coefficient of the year 2000 is R = 0.829, and the coefficient of determination R2 = 0.688, i.e., 68.80%, and the predicted R2 = 66%. Similarly, in 2010, the correlation coefficient is 0.873, and the coefficient of determination R2 = 0.763, i.e., 76.30%, and the predicted R2 is 74.5%. The R value for the year 2019 is 0.905, and R2 is 0.82, i.e., 82%, and the predicted R2 is 80.6%, which show that the selected biophysical variables are significantly controlling and defining the patterns and distribution of land surface temperature. Thus, the study explains that rapidly increasing built-up areas and decreasing vegetation cover and water bodies are gradually increasing the land surface temperature in the study area.

Hot spot analysis

The identification of the hot spots helps to identify areas of priority for the implementation of mitigation and adaptation strategies. Hot spot maps using Getis-Ord-Gi* statistics for the selected years to highlight the hot spot and cold spot areas in Gorakhpur Urban Area have been produced. It is based on mean land surface temperature derived from remotely sensed data. From the statistical results, the land surface temperature pattern has been divided into seven categories.

The expanding heat island is clearly visible during the 18-year change analysis. Significant hot spots of high land surface temperature were recorded in built-up areas, industrial areas, and bare surfaces, while significant cold spots were seen on vegetated surfaces and water bodies. In general, the land surface temperature, hot spots, and cold spots were shown to have been greatly influenced by the LU/LC of the area.

Statistically significant hot spots and their spatial distribution show that high hot spot areas are scattered in small clusters in the northern part of the municipal area (Fig. 10), and their areal distribution shows gradually expanding trend. In 1990, this hot spot was confined to the north-west part of the city only, but in later years, it has almost covered the northern part of the municipal area. Airport, situated in the south-eastern part of the city, also falls in the very high hot spot category. Railway workshop, situated in the middle part, shows diminishing trend because of increasing greener cover. Hot spots and warm spots are also distributed in the vicinity of very high spot. Water body (Ramgarh lake) shows up as a very cold spot. Other small water bodies scattered in the area are in cold and cooler spots. These water bodies have positive impact in reducing the urban thermal environment.

Fig. 10
figure 10

Hot spots of year 1990, 2000, 2010, and 2019

Full size image

Discussion

It is obvious from the results of the present study that LU/LC and biophysical parameters are significant to address the dynamic pattern of LST of an urban landscape due to anthropogenic activities. The analysis of images of land surface temperature in Gorakhpur city shows that high temperature areas correspond to areas with no vegetated cover, while vegetated surfaces and water bodies recorded a relatively low temperature. The northern part of the city exhibits increasing trend in the temperature and spatial distribution as well. In 1990, the northwestern part, which is an industrial area of the city, railway station, and nearby railway workshop area, and the densely populated areas of the city have recorded the highest temperature.

Spatial distribution pattern of land surface temperature (LST) in 2010 and 2019 shows that the northern part of the city was under high temperature zone. The eastern part where airstrip is located inside the forest area and central part is also among the high-temperature areas.

The number of the highest land surface temperature areas kept on increasing, and the whole northern and eastern areas, at the periphery of the municipal boundary, are included in this category, whereas Ramgarh Lake is the coolest area of the city. Thus, this area is generating cooling effect on the urban micro climate. Central, western, and the south-western parts also have higher temperature. Thus, high temperature is associated with industrial activities, impervious surface, and other urban activities, while water bodies and vegetative areas are directly related to the lower land surface temperature. Vegetation and evapotranspiration help in lowering both surface and air temperatures (Adeyeri and Okogbue 2014).

The average land surface temperature of the study area is continuously rising during 1990–2019 (Fig. 11 and Table 4). The highest change has been observed within the municipal areas. In 1990, the average land surface temperature was 18.22 °C, but in 2019, with the increase of more than 2 °C, it reached up to 20.52 °C. Thus, in the last 28 years, overall, there is an increase of 2.3 °C in land surface temperature of Gorakhpur city. Moderate change in the temperature occurred in the neighboring urban areas, located outside the municipal limits. Here, 0.59 °C temperature increase has been observed. Not much difference in the land surface temperature is noticed in the rural areas. Here, only 0.44 °C increase in the temperature is recorded. Thus, land surface temperature pattern is strongly correlated with the land use/land cover class distribution. The temperature profiles and average increase in land surface temperature clearly demonstrate the urban heat island effect in some parts of the study area.

Fig. 11
figure 11

Average LST in Gorakhpur city (1990–2019)

Full size image
Table 4 Multiple linear regression for land surface temperature and biophysical parameters, 1990 to 2019
Full size table

Four transacts, three from west to east and one passing through north-west to south-east part of area of interest (AOI), were drawn (Fig. 12). These lines pass through almost all types of land use/land cover classes. Along these lines, temperature profiles are created, and it has been observed by the analysis of these transacts that in 2010, the highest temperature of about 22.86 °C exists on urban built-up area, which represents intense concrete areas; 24.56 °C on vacant land, which has been converted from agricultural land; and 24.55° C on industrial and other impervious surfaces. In 2019, agricultural areas show relatively low temperature of about 16.6 °C; the lowest temperature 15.06 °C has been observed at Ramgarh Lake areas due to the availability of vast stretch of water. Thus, peaks of the temperature correspond to the built-up and vacant land.

Fig. 12
figure 12

Land surface temperature: cross-profile of 1990, 2000, 2010, and 2019

Full size image

A linear regression using Pearson’s correlation analysis has been carried out to evaluate the correlation of land surface temperature (LST) with the seven indices with 100 sample points (Fig. 12). Land surface temperature is positively correlated with some indices, while it is negatively correlated with others depending on the type of LU/LC classes (Sannigrahi et al. 2017). Land surface temperature has strong positive correlation with BSI and NDBI, which show the contribution of impervious surface on heat island or hot spot formation. The value of relationship of these two parameters ranges between 0.646 in 1990 to 0.842 in 2019 and 0.689 in 1990 to 0.830 in 2019, respectively (Table 5), which is in accordance with the results of Wuhan city, China (Chen et al. 2013). Land surface temperature has moderate positive correlation with NDBal and NDWI. The values of correlation of coefficient also show a gradually increasing trend. Nimish and Bharath (2020) established a correlation between LST and a NDBaI in Kolkata Metropolitan Area, India. The correlation of land surface temperature with NDVI, MNDWI, and SAVI exhibit a negative trend.

Table 5 Average LST in Gorakhpur city, 1990–2019
Full size table

NDVI values tend to become weaker with the increase in built-up area. These areas normally have high NDBI and BSI.

Study on the impact of land use changes on land surface temperature in Lagos, Nigeria, between 2002–2013 showed that land surface temperature has increased with urban growth and density of urban area. Non-vegetative areas had higher land surface temperature (Imnoff et al., 2010). This has a negative impact on the ecosystem and human health. In the past, many studies found a negative correlation between land surface temperature and NDVI (Kustas et al., 2003; Agam et al., 2007; Inamdar et al., 2008) (Table 6).

Table 6 Correlation coefficient for the biophysical parameters and LST for the year 1990 to 2019
Full size table

Mallick et al. (2008) estimated land surface temperature over Delhi using Landsat-7 ETM + , while a comparative study carried out in Mumbai and Delhi revealed that the intensity of the UHI has increased due to the alteration of vegetation cover caused by urban development. The study concludes that there is a strong negative correlation between NDVI and UHI. Mumbai had more UHI than Delhi. The validation of the heat island is done in relation to the normalized difference vegetation index pattern. The study concludes that there exists a strong negative correlation between NDVI and UHI. Mumbai had more UHI than Delhi (Grover and Singh, 2015).

Weng et al. (2004) derived the vegetation fraction from Landsat ETM + images and found a stronger negative correlation between land surface temperature than NDVI. Yuan and Bauer (2007) made a similar analysis, but they correlated the fraction of impervious surface area (ISA) that shows strong positive correlation to the land surface temperature in urban areas. The impervious nature of LU/LC class and surface roughness also has a role to play in the UHI creation (Li et al., 2011; Yuan and Bauer 2007; Zhou et al. 2011; Sharma et al. 2015; Wang et al. 2018).

In order to analyze the correlation between land surface temperature and selected seven indicators, scatter plots are drawn (Fig. 13). The trend analysis of land surface temperature against different land use/land cover (LU/LC) indicates that land surface temperature increases with a decrease in vegetation cover, while it increases as the built-up area increases. Thus, parks, playgrounds, and water bodies are the land surface temperature reducing surfaces. This observation is very similar to previous studies (Liu and Zhang 2011; Guo et al., 2016; Guha and Govil 2021b).

Fig. 13
figure 13

Scatterplot of LST and biophysical parameters

Full size image

Conclusion

The study is aimed to evaluate the nature of urban growth and its impact on land use transformation and ultimately its contribution on the overall increase in the land surface temperature of Gorakhpur city and its environs. During the study period, urban built-up area has increased from 11.68to 35.45%, and thus, 23.77% overall increase has been observed. Vacant land has also experienced increasing trend. The rest of the land use classes showed negative growth. It is necessary to understand spatial distribution of land surface temperature and its relation with different types of land use/land cover to mitigate the urban heat island effect. On one hand, it will help to know the extent of hot spots and their intensity, while on the other hand, it will also help to plan strategies accordingly and more effectively. Getis-Ord-Gi* statistics has been utilized to understand the spatio-temporal dynamics of land surface temperature. The expanding heat island is clearly visible in the northern and outside the north-eastern boundary of the city.

The study clearly indicates that land surface temperature significantly varies with the variation of biophysical variables. High correlation is found between biophysical parameters and land surface temperature. The highest correlation is observed in BSI and NDBI with land surface temperature (R = 0.842 and 0.830, respectively). Vegetative and non-vegetative areas play important role in the modification of land surface temperature. An effective way to buffer environmental temperature in a city is to increase vegetated surfaces (Anyanwu and Kanu, 2006; Adeyeri and Okogbue, 2014). Areas having high land surface temperature have been clearly identified by the study. The northern part of the city (an industrial area of the city) and the middle part need immediate attention to decrease the temperature.

This study concludes that high temperature areas are located in scattered pattern and related to certain types of land use such as built-up and vacant. Rapidly increasing built-up areas, industrial activities, diminishing agricultural activities, decreasing vegetation cover, and shrinking water bodies are the main factors responsible for the present increasing trend of land surface temperature. This has adversely affected various biophysical parameters. Therefore, this kind of study and implementation of policies according to the plan will ensure the healthy and sustainable development of urban areas. Tree plantation drive and conservation of green spaces will be useful in reducing the land surface temperature and local climate as a whole. Comprehensive greening plan needs to be taken on zonation basis, and first priority should be given to high hot spot areas. Reduction of excessive heat surfaces, anthropogenic heat, and increase in shading and evapotranspiration will help in the improvement of local thermal environment. Efforts should be made to increase greenery in thickly populated urban areas through regulations. The results of this study will provide information and guidance to urban planners, administrators, and other authorities to understand the ground situation in a better way. So that they can draft their plan to reduce the land surface temperature in future on priority basis. Better urban planning and regulations can prevent further increase of land surface temperature and heat island phenomena in Gorakhpur city.