A COMPARATIVE STUDY OF LAND SURFACE TEMPERATURE WITH DIFFERENT INDICES ON HETEROGENEOUS LAND COVER USING LANDSAT 8 DATA

The temperature rise in urban areas has become a major environmental concern. Hence, the study of Land surface temperature (LST) in urban areas is important to understand the behaviour of different land covers on temperature. Relation of LST with different indices is required to study LST in urban areas using satellite data. The present study focuses on the relation of LST with the selected indices based on different land cover using Landsat 8 OLI (Operational Land Imager) and TIRS (Thermal Infrared Sensor) data in Varanasi, India. A regression analysis was done between LST and Normalized Difference Vegetation index (NDVI), Normalized Difference Soil Index (NDSI), Normalized Difference Built-up Index (NDBI) and Normalized Difference Water Index (NDWI). The non-linear relations of LST with NDVI and NDWI were observed, whereas NDBI and NDSI were found to show positive linear relation with LST. The correlation of LST with NDSI was found better than NDBI. Further analysis was done by choosing 25 pure pixels from each land cover of water, vegetation, bare soil and urban areas to determine the behaviour of indices on LST for each land cover. The investigation shows that NDSI and NDBI can be effectively used for study of LST in urban areas. However, NDBI can explain urban LST in the better way for the regions without water body.


INTRODUCTION
Urban Heat Island (UHI) refers to the urban areas which has temperatures higher than the surrounding suburban or rural areas due to increase in urbanization.Urbanization has led to the replacement of natural land covers with the artificial built up surfaces made up of impervious materials and decrease of vegetation which helps in reducing temperatures due to evapotranspiration (Voogt and Oke, 2003).Various previous studies have shown that bare soil and built up areas have higher temperatures whereas waterbodies and vegetated land have lower temperatures (Song et al. 2014).The study of LST of urban areas and also different land covers has become important for the requirement of sustainable planning and environmental protection.The factors responsible for resultant temperature rise must be identified in order to attenuate the UHI effect (Kaufmann et al. 2007, Mountrakis andLuo. 2011).* Thermal bands in remote sensing satellite are the most important source for determining LST.Various satellites contain thermal bands like Landsat, Aster, AVHRR, MODIS etc. Fine spatial resolution is required for the study of LST variation in urban areas due to greater heterogeneity in urban areas.Landsat and Aster data has been widely * Corresponding author used for study of UHI effect in urban areas due to its fine spatial resolution (Tsou et al. 2017, Yu et al. 2014).
Various scholars have studied on the correlation between LST and some land use (LU) /land cover (LC) indices for different study area and found that the correlation of LST with LU-LC indices was different (Deilami and Kamruzzaman, 2017;Ma et al. 2016, Mathew et al. 2016, Nie et al. 2016, Tran. et al. 2017).So, this study focuses on the relation of LST with the selected indices i.e.NDVI, NDBI, NDSI and NDWI based on different land cover using Landsat 8 data in Varanasi, India.In order to understand the relation, 25 pure pixels for each land cover of water, vegetation, bare soil and urban areas were chosen to determine behaviour of LST and indices with the land covers.

STUDY AREA
The study area, Varanasi city, is a very old city located at the banks of River Ganges in the Eastern part of the state of Uttar Pradesh, North India.The city lies at the coordinates 25.28 0 N and 82.96 0 E at an elevation of 80.71 meters and covers an area of 82.10 km 2 .Varanasi experiences a humid subtropical climate.Location map of study area is shown in Figure 1.

Figure1. Location map of study area
This image of Varanasi is the false color composite image obtained from Landsat data.The city was chosen for this study due to the availability of land covers like waterbody, urban areas, bare soil region as well as vegetated land covering sufficient area which makes this useful to study the behaviour of each land cover.

DATA USED AND IMAGE PRE-PROCESSING
Landsat satellite images were generated and distributed by the U.S.

Calculation of Indices
The indices used in the study were obtained from the visible, Near Infra-Red and Short Wave Infra-Red reflectance bands from the Landsat satellite data.The index maps were obtained using the equations as shown in Table1.

Indices Equations
Normalized Difference Vegetation Index (NDVI) The digital number of thermal band was first rescaled into Top of atmosphere radiance (TOA) using Equation (1).
where ML = multiplicative rescaling factor AL = additive rescaling factor DN = digital number of thermal band.
The TOA radiance includes a mixed signal that contains emission from land as well as atmosphere.Hence, atmospheric correction was performed to eliminate the contribution from the atmosphere.Therefore, TOA radiance was converted into surface leaving radiance (LT) using Equation (2).

= transmission ɛ = emissivity
The atmospheric parameters like transmission (τ), upwelling radiance (Lμ) and downwelling radiances (Ld) values were obtained from an atmospheric correction tool available at http://atmcorr.gsfc.nasa.gov/only for Landsat 4-5, 7 and 8 satellite images developed by Barsi et al. (2005) and the emissivity values were estimated using NDVI (Van de Griend and Owe, 1993) as shown in the Then, the surface leaving radiance was converted into LST using Planck's law as given in Equation ( 3) where K1, K2 = thermal constants Ts = Land Surface Temperature.

RESULTS AND DISCUSSIONS
In order to study the behaviour of LST in the city, the LST image was generated as shown in Figure 2. It was observed that the River Ganges show very low LST, the dry sandy area the bank of River Ganga shows very high LST.The urban areas of the city show LST higher than the vegetated areas and waterbody but lower than the bare soil or dry sandy areas.

CONCLUSIONS
The study focuses on the relation of LST with the selected indices i.e.NDVI, NDBI, NDSI and NDWI using Landsat 8 OLI and TIRS data.A regression analysis was done between LST and the selected indices.The non-linear relations of LST with NDVI and NDWI were observed, whereas NDBI and NDSI were found to show positive linear relation with LST.The correlation of LST with NDSI was found better than NDBI.Further analysis was done by choosing 25 pure pixels from each land cover of water, vegetation, bare soil and urban areas and plotted to determine the behaviour of indices on LST for each land cover.The investigation shows that NDSI and NDBI can be effectively used for study of LST in urban areas.However, NDBI can explain urban LST in the better way for the regions without water body.
for estimating different Indices 4.2 Calculation of LST Thermal band (band10) of Landsat 8 images was used for determining LST image.Various steps are involved in the estimation of LST from Landsat image (Essa et al. 2012).

Figure2.
Figure2.LST image of VaranasiIn order to study the relation of LST with the respective indices, scatter plots were obtained for LST with NDVI, NDBI, NDSI and NDWI and are shown in Figures3 -6respectively.

Figure 3 .
Figure 3.Scatter plot of LST with NDVI

Table 2 .
Land Surface Emissivity estimation from NDVI In case of NDBI, water pixels shows higher NDBI than vegetated pixels but LST is lower for water pixels than vegetated pixels which results in lower R-square.In case of NDSI, water, vegetation, soil as well as urban pixels shows positive relation with LST making it more suitable for study of LST in urban areas.NDBI can also explain urban LST better in regions without water body.
Figure7.LST-NDVI relation with land cover Waterbodies has lower LST but shows negative values of NDVI and other land covers shows positive values.The vegetative land covers shows higher positive values than bare soil and urban regions but lower LST values.This explains the non-linear behaviour of LST with NDVI.Since, urban areas also include soil, vegetation and water, NDVI cannot explain LST in urban regions due to greater heterogeneity in land cover.But agricultural region contains only bare soil and vegetation, so NDVI can explain LST better in agricultural regions.