Introduction

Urbanisation is the most important anthropogenic activity after greenhouse gas emissions that impact climate (Kalnay and Cai 2003). Urbanisation on one hand benefits economic welfare of society; on the contrary, it threatens the biophysical health of the city itself. It causes detriment to environmental quality, including biodiversity (McKinney 2006; Delgado-V and French 2012; Threlfall et al. 2012), soil fertility (Chen 2007), water quality (Kaushal et al. 2008; Paul and Meyer 2001), and impacts other natural resources (Huang et al. 2010) along with ecosystem services (Alberti 2005; Bolund and Hunhammar 1999). Urban land transformations are most complex and dramatically irreversible land use changes and are thus one of the most studied phenomena (Wenhui 2012; Jiang and Tian 2010; Taubenböck et al. 2009; Souch and Grimmond 2006). Urban land use changes such as loss of vegetation (Scolozzi and Geneletti 2012), increased built-up and open areas (Ng et al. 2011) along with increased expanse of urban fabrics such as concrete and asphalt altogether alter the local climate of a city by modifying various biological and physical characteristics of the environment. These include vegetation cover, impervious built-up covers, presence of moisture, surface and air temperatures, soil properties and others, which in turn are governed by land use characteristics (Voogt and Oke 2003).

With urbanisation taking over the world at an unprecedented pace, Indian cities are no exception. By 2050, India is expected to inhabit 0.9 billion urban population with a projection from 29.7 % in 2009 to about 54.2 % by 2050 (UN 2010). Delhi, Mumbai and Kolkata already fall in the category of the mega-cities, and Chennai is soon going to join these along with Hyderabad and Bangalore (Joshi et al. 2011). Apart from megacities, a number of smaller urban agglomerations are showing tremendous growth since the past decade (Taubenböck et al. 2009). Surat, for instance, has great potential of transforming into a megacity. Surprisingly, not enough work has been done to study its urban sprawl and related impacts. The city has been expanding ever since 1951 when it had a population of 0.24 million that gradually increased to 0.49 in 1971, to 1.52 in 1991 and finally crossing 2.8 in 2001 and reaching 4.46 million in 2011 (Census of India 2011). It is expected to inhabit 5.57 million people by 2025 (UN 2010).

Land surface temperature (LST) is the most studied biophysical parameter related to urban health. It has a two-way relationship with environmental parameters as it influences some of them and in turn gets influenced by others. Some literature is available on former part of this interaction. Baur and Baur (1993) found that urbanisation-related changes in LST resulted in local extinction of land snails in Basel. Whitford et al. (2001) stated LST as one of the four ecological performance indicators. LST also directly impacts surface energy budget (Bastiaanssen et al. 1998) and thus influences air temperature of the area. This results in emergence of two types of urban heat Islands (UHI) in same area, viz., surface UHI and atmospheric UHI. This paper examines formal half of LST and environment interaction using satellite-derived equivalents of various parameters. This study thus employs remote sensing to demonstrate how different environmental parameters like vegetation cover and health, moisture intensity and bareness influence LST.

With the advent of geospatial technology, it has become possible to remotely monitor the biophysical variables and changing land use patterns and to analyse their interactions (Buyantuyev and Wu 2012). Though a number of studies have been carried out to examine the variability among greenness, surface temperature (Son et al. 2012; Julien and Sobrino 2009; Raynolds et al. 2008; Julien et al. 2006; Weng et al. 2004; Sandholt et al. 2002; Owen et al. 1998; Gillies et al. 1997; Goetz 1997) and land use associations (Amiri et al. 2009; Zhou et al. 2011; Jiang and Tian 2010; Xiao et al. 2008; Xiao and Weng 2007), little work has been done in assessing the variability of other biophysical factors in context to LST (Uddin et al. 2010; Chen et al. 2006). Limited research material could be found on the relationship of these parameters with respect to each other and how this relationship differs for different land uses. Uddin et al. (2010) and Chen et al. (2006) have attempted classification using four main indices, viz., Normalised Difference Vegetation Index (NDVI), Normalised Difference Water Index (NDWI), Normalised Difference Built-up Index (NDBI) and Normalised Difference Bareness Index (NDBaI). In the present study, an additional modified water index was developed, Modified Normalised Water Index (MNDWI), for extracting water. All these factors were considered as these could account to assess spatio-temporal footprints of urbanisation.

Surat—the Diamond City

Surat city is the commercial capital city of Gujarat (India) and also serves as administrative capital of Surat district. It is situated on the banks of Tapi river which perennially flows northeast to southwest finally joining the Arabian Sea, situated 22 km west to the city. The city is situated at 21.25° N and 72.87° E (Fig. 1). The region experiences hot summers with temperature ranging from 38 to 45 °C. Winters are mild, but the months of December and January are coldest with temperatures varying between 10 and 15.5 °C. The average annual rainfall is 1,143 mm (Surat Municipal Corporation 2011).

Fig. 1
figure 1

Location of the study area

Full size image

Being a commercial hub, the region has witnessed tremendous urbanisation over the past few decades. It ranks 36th in the list of the world’s largest cities with a population of over 2.5 million in 2001 (population density of 1,376 persons/km2) and the third cleanest metropolitan region in India. The city has total population of more than 6 million of which 80 % is urban population and the remaining 1.24 million is rural (Census of India 2011). Surat has immense business and job opportunities that trigger high immigration rates to the city resulting in increased population. The two major economic activities are agriculture and diamond cutting. A large number of small-scale diamond cutting industries have given the name ‘Diamond City’ to this place.

The study area extends spatially from 21.29° N and 21.12° N to 72.74° E and 72.94° E, spanning across 386.28 km2 covering the city (city limits of 326 km2) and peripheral urban–rural fringe. The study area mainly falls in the Chorasi taluk of Surat district but also covers parts of adjoining taluks of Olpad in the north, Kamrej in the east and Palsana in the south-east. It is bound by Navsari on the south.

Materials and methods

Satellite data

Landsat 5/4 Thematic Mapper (TM) satellite images (Path/Row, 148/45) dated October 19, 1990 and October 23, 2009 were used. Landsat is a medium-resolution (30 and 120 m) data with seven bands that are most commonly used for environmental studies. Landsat TM data consist of seven bands of which the first three are visible bands, the fourth is near-infrared, bands five and seven fall in shortwave infrared regions and sixth is the thermal band. The geometrically and radiometrically corrected images rectified to a common Universal Transverse Mercator were procured from the USGS Earth Resource Observation Systems Data Center. The details and characteristics of the satellite data used are given in Table 1.

Table 1 Landsat 5 TM sensor system characteristics
Full size table

Spectral enhancement was done using band ratioing to compute various indices. This performed a twofold function: firstly, these served as variables in classification, and secondly, these were used to study dynamics of different biophysical parameters (viz., greenness, wetness, bareness and built-up intensity) with respect to each other and that to thermal response of environment. Red (TM 3) and infrared (TM4) bands helped to enhance greenness (Maxwell and Sylvester 2012; Purevdorj et al. 1998), infrared (TM4) and shortwave infrared (TM5) were used to enhance built-up areas (Bridhikitti and Overcamp 2012; Ma et al. 2010; Zhang et al. 2009) and enhance canopy vegetation water content (Jackson et al. 2004; Serrano et al. 2000; Gao 1996). Shortwave infrared (TM5) and thermal infrared (TM6) were employed for enhancing bare lands (Nasipuri and Chatterjee 2009). LST retrieval was performed using thermal information of TM6 band. For more accurate mapping and extraction of water bodies from image, MNDWI was computed by transforming blue (TM1) and shortwave infrared (TM5) bands. Figure 2 briefly describes the methodology followed.

Fig. 2
figure 2

Flowchart describing methodology. (Where, L Max and L Min (watts per square metre per steradian per micrometre) are spectral radiances for thermal band (band 6) at digital numbers 1 and 255, respectively, i.e. QCalMin and QCalMax values, while QCal represents the DN value; K1 (607.76 W/(m2/sr-μm)) and K2 (1260.56 K) are pre-launch satellite calibration constants; a and b are constants; T a refers to effective mean atmospheric temperature, and C and D are computed using emissivity and atmospheric transmittance derived from water vapour (grams per square centimetre), where ε is emissivity and τ is transmittance)

Full size image

Data processing

Estimating biophysical parameters

NDVI, NDBI and MNDWI were used to map land use and land cover (LULC) classes. These metrics along with NDBaI and NDWI were used to investigate correlation of various biophysical parameters (intensity of moisture, greenness, build-up density and extent of bareness) with that of thermal response of urbanised areas in contrast to vegetated areas. Table 2 presents a brief review of various indices used, their computation and uses along with references.

Table 2 Description of various image transformation used with references
Full size table

Land surface temperature retrieval

Landsat TM band 6 has been extensively exploited to study thermal dynamics of various earth surface features (Qin et al. 2001; Schott et al. 2001). LST was derived using thermal infrared (TM6) band (10.40–12.50 μm), with effective wavelength of 11.457 μm, has relatively lower radiometric sensitivity and coarser spatial resolution of 120 × 120 m. Qin et al. (2001) mono-window algorithm (Sun et al. 2010) was used to retrieve LST from thermal DN values, and NDVI was used here for emissivity correction to obtain final LST images.

Characterising land use patterns

Five broad categories of LULC were identified based on visual image interpretation keys; (1) Built-up comprised of urban features and impervious structures like asphalt and concrete roads; (2) River bed included sediments along river and other sandy and silt deposition structures; (3) A broad class named ‘Vegetation’ included land under agriculture as well as urban vegetation and other canopy; (4) Agricultural land without any vegetation cover including agricultural fallow was classified as Fallow land; and (5) Water body covered Tapi river and other small water storage tanks.

Knowledge-based classification technique was employed to categorise satellite images into LULC classes. It is a type of hierarchical decision tree algorithm based on hypothesis testing that evaluates various rules and conditions defined by the user. Rules are condition (IF) and action (THEN) statements. Hypothesis, rules and condition represented in linear dendritic decision tree are inferred and processed to generate outputs (Table 3). Indices computed earlier were fed into knowledge engineer as variables for different LULC classes (the hypothesis), and their threshold values defined the rules for each hypothesis. Due to high efficiency of MNDWI in identifying water pixels, it was used alone for water extraction. The decision tree thus generated was executed to map LULC. To assess the classification accuracy, 150 randomly generated points were overlaid on the satellite image. Statistically valid sampling strategy was adopted to assess commission, omission and overall accuracy (Stehman 1996).

Table 3 Hypothesis, rules and variables for 1990 and 2009 satellite images
Full size table

Statistical analysis

Biophysical parameters

Urban land transformations have great impacts on the biophysical health of the environment. To examine the complexity and dynamics of this relationship, 150 experimental and 150 control points were randomly generated. All the points were well distributed among the LULC classes and the entire study area. At least 30 points per LULC were generated to extract values of each of the parameters under study. These were then used to mine out the values of biophysical parameters two times for analysing the influence of LULC change on the biophysical environment. Correlation and regression techniques were employed to exemplify how the environment responds to stimulus of urbanisation.

Directional profiles of LST

Based on the spatial patterns of urban expansion in the area, four directional profiles for LST were selected: north–south (N–S), east–west (E–W), northeast–southwest (NE–SW) and northwest–southeast (NW–SE) profiles. LST values for these spatial profiles were analysed across the background of expansion in the built-up area over the study period.

Results and discussion

LULCC analysis

1990 and 2009 LULC maps were analysed to study expansion of urban area. The classes mapped were built-up, vegetation, fallow, river bed and water. Area statistics for LULC ascertained massive expansion (nearly three times) in urban expanse at the cost of vegetation and fallow lands. The urban extent increase has been equal to 42.74 km2 from 1990 to 2009. With improved connectivity through roads, highways and bridges, the city has expanded in almost all directions (Fig. 3) resulting in newer settlements in Amroli, Nana Varachha, Choriyasi, Athwa, Vishal Nagar, Sima Nagar and Jain Wadi. Udhana has been the centre for highly denser sprawls that could be attributed to its high economic importance for diamond cutting and polishing works. Overall classification accuracy of LULC map for 1990 was 90.4 % and for 2009 was 84.4 % (Table 4). The expert classification accuracy is remarkably good as compared to hard classification methods (Punia et al. 2011; Wentz et al. 2008). Due to similar spectral responses of long fallow and less dense built-up areas, 2009 image has poorer accuracy for fallow and built-up classes.

Fig. 3
figure 3

Expansion in built-up from 1990 to 2009

Full size image
Table 4 Accuracy assessment for 1990 and 2009 classification
Full size table

Analysis of biophysical parameters

For quantification of relationship between biophysical parameters under study (Fig. 4), correlation analysis was carried out using 300 point values. The analysis demonstrated that LST shoots up with a fall in NDVI or greenness. A similar relationship was observed for NDWI also. Lesser will be the water content, lesser will be evaporation and hence reduced cooling will result in higher temperature. This explanation also supports a positive correlation between NDVI and NDWI (Fig. 5). NDBI represents built-up intensity of land. The higher the built-up intensity, the more impervious and the lower the moisture content. Thus, LST–NDBI behaves asynchronously. NDBI and NDWI exhibited a perfectly negative relationship as the two indices have exactly the same numerical values but antonymous signs.

Fig. 4
figure 4

Biophysical parameters in 1990 and 2009 (NDWI, NDVI, NDBI and NDBaI)

Full size image
Fig. 5
figure 5

Scatter plots and coefficient of determination for different parameters in 1990 and 2009

Full size image

Bareness index did not exhibit a strong correlation with any of the parameters. Similar results have been observed by Essa et al. (2012). NDBaI exemplified a negative correlation with NDVI and NDWI and a positive one with LST and NDBI for both years. But this relationship shows inconsistency when analysed separately for control and experimental points showing positive values for year 1990 (0.032 for control and 0.183 for experimental) and negative for year 2009 (−0.09 for control and −0.20671 for experimental). Uddin et al. (2010) also state that in some isolated cases, NDBaI and LST have a positive variability with respect to each other (Fig. 5).

LST characterisation

Experimental and control point values of LST were analysed to assess thermal response of different LULC classes. The average temperature of experimental built-up points is relatively higher than that for vegetation areas. LST for vegetation to built-up change areas increased by 8.8 ± 2.6 °C, and fallow land to built-up increased by 6.6 ± 2.8 °C, while 8.6 ± 2.8 °C change was observed for increase in built-up density.

Increase in average temperature for unchanged vegetated areas was assessed using control points and was found to be 3.3 ± 3.4 °C. Hence, the net increase in the temperature for changes from vegetation to built-up is around 5.5 ± 2.6 °C, vegetation to fallow is 6.7 ± 3 °C, fallow land to built-up is 3.5 ± 2.9 °C and built-up to dense built-up is 5.3 ± 2.8 °C. Temperature change for vegetation to fallow is higher than vegetation to built-up which could be due to land being fallow for long and thus being deficient in moisture. Also, as compared to fallow land, which is completely bare, built-up areas are characterised by shadow effects from building structures.

Temporal analysis of LST demonstrated that built-up areas have a higher temperature as compared to vegetation thus forming the foundation for phenomenon of urban heat island in the area (Schwarz et al. 2012; Liu and Zhang 2011) (Fig. 6). Built-up areas that have come up by replacing vegetation show elevated temperatures as compared to other areas. Urban areas contain impervious surfaces that decrease local infiltration, percolation and soil moisture (Brun and Band 2000). Due to their thermal properties, heat island phenomenon is triggered. LST is one of the key factors that control physical, chemical and biological processes in the environment and in turn is governed by them. Urbanisation in particular changes the thermal environment due to physical properties of its urban fabric (Pu et al. 2006) and encourages the development of urban heat island.

Fig. 6
figure 6

LULC and LST for 1990 and 2009

Full size image

Spatial profile analysis

For a detailed analysis of spatial distribution of LST in the study area, four directional profiles were studied (Fig. 7). The profiles were constructed based on directions in which built-up area has expanded. N–S profile ran from Kosam and Sherdi villages in north of the river to Udhana industrial centre, in south of Surat city. Spatial N–S profile of LST for 1990 and 2009 improved our understanding that industrialised area increased thermal environment. The curve became flatter for Surat city in 2009, and new peaks were observed in Udhana centre which were not there in 1990. This could be explained based on the expansion of built-up in Udhana area. E–W profile starts with agricultural lands of Bhesan and Melgama villages, passing built-up area of Jain Wadi crossing the river to industrial areas of Bharat Nagar, through the city of Surat, and finally ending in the agricultural lands of Kosmada and Ladvi villages. E–W profile exhibited results similar to N–S profile with low thermal response of agriculture. In 1990, only one major peak was observed in built-up area in and around Surat, while in 2009, profile flattened with higher temperature values illustrating impacts of increased built-up on LST.

Fig. 7
figure 7

Directional profiling of LST distribution in 1990 and 2009

Full size image

The NW–SE profile sets out from agricultural areas of Talad and Sarol Gam villages; crossing across the river, it passes through Katargam, Surat and Shakti Nagar areas finally terminating through agricultural areas of Bonand and Kharsava villages. Temperature values (Fig. 7) for agricultural areas, in the beginning and end of the profile, were generally lower with sporadic peaks due to fallow land. Built-up areas of Katargam, Surat and Shakti Nagar gave high temperature responses. Profile corresponding to river showed a steep depression in temperature. Such an observation was recorded for all four profiles regarding encountering river in between. NE–SW profile begins at fields of Abrama crossing the river passing through Kodiyar Nagar and Surat city through Krishnaganj and Athwa to Vesu village. A peak was observed for the new built-up in 2009, towards the northern bank of the river. In 1990, fewer peaks were observed for Surat, Kishanganj and Athawa areas, but more and higher peaks were observed in 2009.

Conclusion

The study was taken up to assess spatio-temporal footprints of urbanisation in Surat city. Over a span of two decades, the city has spread over an area of 42.75 km2, with multi-directional expansion of its built-up. The city has densely spread to the south in Udhana industrial hub and comparatively less dense northward spreads in Amroli and Kosad Navi Vasahat. Choriyasi, a small satellite town, manifests the eastward spread of the city, while Athawa manifested the same for the southwestern spreads. Older parts of Sima Nagar and Jain Wadi served as centres of urban growth for the west.

The urban expansion has altered the state of various biophysical parameters (including surface temperatures, moisture contents and vegetation cover) which govern the health of the environment. In this study, we qualitatively studied the multifaceted interactions of biophysical parameters and quantified relationships among them. The selected parameters retrieved from satellite data are LST (thermal behaviour of surface), NDVI (intensity of greenness), NDBI (intensity of built-up), NDWI (moisture status of surface) and NDBaI (bareness). Of these, LST–NDVI and LST–NDWI were found to vary inversely, while LST–NDBI and NDVI–NDWI had a strong positive correlation. Of these, NDBaI was one parameter that could not give consistent results and was characterised by very weak correlations with the other parameters. Although NDBaI has been successfully employed by Chen et al. (2006), in this work, some inconsistent results were obtained for this index. Thus, we found that this index gave site-specific results, and there is further scope of developing a more robust index to measure the bareness intensity of the earth surface.

In face of urban heat island (UHI), LST parameter was more intensely analysed with respect to its changing spatial distribution against the background of changing LULC patterns. LST through UHI can have severe impacts on human and environmental health by increasing the frequency and intensity of heat waves (Tan et al. 2010). It was observed that with southward expansion of Udhana industrial area, new peaks for LST came up in 2009. Similar results were found for Athwa, Jain Wadi and Surat city areas. Apart from urban areas, some peaks came up for agricultural fallow lands.

Global urbanisation needs accurate information on the expansion of impervious surfaces and associated parameters. The parameters discussed in this study are of prime importance to assess the impact of urbanisation-linked development. Eco-planning in urban sector could use such information to achieve the goals of sustainable practices and planning. The information generated can also aid in understanding the contribution of local effects on the global phenomena of climate change and associated changes. These should be explored to understand the health of urban ecosystem and linkages with human well-being.