Monitoring LST-NDVI Relationship Using Premonsoon Landsat Datasets

)e present study monitors the interrelationship of land surface temperature (LST) with normalized difference vegetation index (NDVI) in Raipur City of India using premonsoon Landsat satellite sensor for the season of 2002, 2006, 2010, 2014, and 2018. )e results describe that the mean LSTof Raipur City is gradually increased with time. )e value of mean NDVI is higher in the area below mean LSTcompared to the area above mean LST. )e value of mean NDVI is also higher in Landsat 8 data than Landsat 5 and Landsat 7 data. A strong negative LST-NDVI correlation is observed throughout the period. )e correlation coefficient is higher in the area above mean LSTand lower in the area belowmean LST.)e value of the correlation coefficient is decreased with time. )e mixed urban landscape of the city is closely related to the changes of LST-NDVI relationship. )ese results provide systematic planning of the urban environment.


Introduction
ermal infrared (TIR) bands of satellite images often regulate the biogeochemical actions of the Earth surface features [1][2][3][4]. Land surface temperature (LST) determination from TIR bands is very important as it depends on the land surface material and varies from time to time [5]. Fast urbanization rapidly changes the characteristics of the surface components [6]. Natural vegetation is one of the most significant features that control the variation of LST distribution [7]. e most commonly used vegetation index is normalized difference vegetation index (NDVI) which is significantly applied in the computation of LST [8][9][10][11]. ere are so many factors like climate, types of vegetation, land use, urbanization, etc., that influence the LST-NDVI correlation [12][13][14].
Many recent research works conducted on the Indian context describe the LST-NDVI correlation [49][50][51][52][53][54]. To discuss the spatial-temporal variation of LST-NDVI correlation in the premonsoon season, Raipur, a smart and rapidly growing Indian city, was selected. e study reflects the following specific objectives: (1) Describe the spatiotemporal distribution of LST and NDVI in the premonsoon months.
(2) Analyze the LST-NDVI correlation in the premonsoon months.

Study Area and Data
Raipur, the capital and the largest city of Chhattisgarh, India, was selected as the study area which extends between 21°11′22″N to 21°20′02″N and 81°32′20″E to 81°41′50″E with an elevation of 280-310 m ( Figure 1). e average annual temperature of the city is around 27-30°C. e entire work conducted on the hot and dry premonsoon months

Methodology
Geometric correction, radiometric correction, and resampling are the required preprocessing steps for using the  Landsat images. LST was determined by using the TIR bands (band 6 for TM and ETM + data, and band 10 for OLI and TIRS data).

Image Preprocessing and Atmospheric
Correction. e satellite data acquired from Landsat sensors were subset to limit the data size. e thermal infrared bands of Landsat sensors were resampled at 30 m resolution using the nearest neighbour algorithm to match the optical bands. Atmospheric correction of the satellite data was performed by the following steps.
For optical bands of Landsat data, the following equation is used to converting a Digital Number into spectral reflectance: where ρλ is the spectral reflectance at top-of-atmosphere (TOA) without correction for solar angle (Unitless), Q cal is the Level 1 pixel value in Digital Number (DN), Mρ is the reflectance multiplicative scaling factor for the band (REFLECTANCE_MULT_BAND_n from the metadata), and Aρ is the reflectance additive scaling factor for the band (REFLECTANCE_ADD_BAND_N from the metadata). e ρλ is corrected with local sun elevation angle θs by the following equation: For TIR band of Landsat data, a similar calibration equation is used: where Lλ is the spectral radiance at TOA in Wm −2 sr −1 mm −1 , Q cal is the Level 1 pixel value in Digital Number (DN), ML is the radiance multiplicative scaling factor for the band (RADIANCE_MULT_BAND_n from the metadata), and AL is radiance additive scaling factor for the band (RADIANCE_ADD_BAND_n from the metadata).

LST Estimation Using Landsat Satellite Sensors.
In the present study, the LST was determined by using the monowindow algorithm [55] in which the three necessary elements are atmospheric transmittance, ground emissivity, and effective mean atmospheric temperature. e original TIR bands of Landsat datasets were resampled into 30 m. e equations are as follows: where L λ is spectral radiance (Wm −2 sr −1 mm −1 ).
where T b is the at-sensor brightness temperature (Kelvin); K 2 and K 1 are calibration constants for Landsat datasets.     Advances in Meteorology where NDVI min is the minimum value (0.2) of NDVI for bare soil pixel and NDVI max is the maximum value (0.5) of NDVI for healthy vegetation pixel. dε is the geometric distribution effect for the natural surface and internal reflection. e value of dε may be 2% for mixed and elevated land surfaces.
where ε s is soil emissivity; F v is fractional vegetation; F is a shape factor (0.55); and ε v is vegetation emissivity.
where ε is land surface emissivity. e value of ε is calculated by the formula given below: Water vapour content is determined by the following equation: where w is water vapour content (g/cm 2 ); T 0 is near-surface air temperature (Kelvin); RH is relative humidity (%). e information on these parameters was provided by the Meteorological Centre, Raipur.
where τ is the total atmospheric transmittance. e effective mean atmospheric transmittance of Raipur City was determined as follows [33]:    T a � 17.9769 + 0.91715T 0 ,

Advances in Meteorology
where T a is mean atmospheric temperature and T s is land surface temperature; a � −67.355351 and b � 0.458606.
Here, only NDVI was applied to correlate with LST. NDVI can be used to estimate other LULC types along with green vegetation [33,[57][58][59][60]. Red and NIR bands of Landsat data are used to determine NDVI (Table 2).

Validation of the Result by Using Other Satellite Data.
Any satellite retrieved LST needs a proper validation with an in situ measurement or other satellite retrieved LST [2].
Here, MODIS data were used to validate the values of LST. As MODIS and Landsat sensors cannot pass over the same region in a particular date, MOD11A1 data (resolution-1 km) of the following particular dates (Table 4) were used to validate for the resulting LST. ese particular dates were free from any cloud coverage or precipitation. TIR bands of MODIS sensor (1 km) and Landsat sensors (120 m, 120 m, and 100 m for Landsat 5, 7, and 8 sensors, respectively). Landsat and MODIS sensors provide a slightly different value of LST due to water vapour content, resampling method, and different passing time of the sensors [56]. After performing the downscaling process, a significant correlation coefficient was found between the Landsat derived mean LST and corresponding MODIS derived mean downscaled LST (Table 4).

Variation in NDVI Distribution for Multitemporal
Landsat Data. Red and NIR bands are required to derive the formula of NDVI [60]. Table 5 represents the value of NDVI for the Landsat satellite images during the study period. Figure 4 presents the spatial and temporal status of NDVI for the whole study area, date-wise and year-wise.

Spatiotemporal Distribution of NDVI during the Entire
Period. Figure 6 shows the area above mean LST (pink part) and the area below mean LST (black part) for every single image during the period of study. Figure 7 and Table 6 represent the temporal variation in the NDVI distribution values for the area has more than mean LST during the entire period. e black portion of the maps shows the area below mean LST (Figure 7). Generally, it seems to have an increase in NDVI values with time. But, there is no such particular pattern of increase of NDVI. 2014 and 2018 have greater NDVI values (maximum and mean NDVI) than the earlier years. Figure 8 and Table 7 represent the temporal variation in the spatial distribution of NDVI values in the area below mean LST. e black portion of the maps shows the area above mean LST (Figure 8). A steady decreasing trend has been observed in the values of mean NDVI. e value of maximum NDVI is much higher in the area below mean LST than in the area above mean LST. Figure 9 shows the graphical presentation of the spatial and temporal change of mean NDVI for the area having less than mean LST, the area having more than mean LST, and the whole area of the city. e overall trend is increasing in nature. In 2006, the mean NDVI values were more than in 2002. e values were reduced again in 2010. From 2010 to 2014, a high slope was found in mean NDVI values. In 2014 and 2018, the trend line was quite stable. It is clear from Figure 9 that mean NDVI values are higher for Landsat 8   data, whereas these values were lower for Landsat 5 or Landsat 7 data. is variation of mean NDVI in different Landsat sensors is mainly due to the configuration of the sensors as the spectral resolution of NIR band of Landsat 8 data (0.851-0.879 μm wavelength) is different from the NIR band of Landsat 5 data (0.760-0.900 μm wavelength) or Landsat 7 data (0.770-0.900 μm wavelength). Further, the year-wise analysis of NDVI has been shown in Figure 10. e diagram shows that the values of maximum NDVI are gradually decreasing with time, whereas the values of minimum NDVI and mean NDVI are increasing. e result is quite significant as it reflects the loss of urban vegetation within the entire time period.

LST-NDVI Correlation in the Whole City, Above Mean
LST Areas, and Below Mean LST Area. LST builds a strong to moderate stable negative correlation with NDVI in the whole Raipur City during the study period. Figure 11 shows a date-wise correlation. e LST-NDVI correlation is moderate to strong negative for the whole area, whereas the correlation does not show any specific pattern for below mean LST zones and above mean LST zones, separately. Figure 12 shows a year-wise correlation. e negativity was almost gradually decreased with time. In the area above mean LST, this correlation is moderately negative and it is stable as these regions mainly cover a high proportion of urban vegetation. In the area below mean LST, LST builds a

Conclusion
e present study monitors the LST-NDVI correlation using different Landsat satellite sensors of the premonsoon season for a specific time interval. Mean LST was a significant measurement for the study as it performs in the area above mean LST as well as in the area below mean LST along with the whole Raipur City. From 2002 to 2018 premonsoon months, the LST was increased by 11.62°C. e LST-NDVI correlation was negative for the study area throughout the period. For Landsat 8 data, mean NDVI values are higher than the other Landsat sensors and thus, the mean NDVI values become higher in recent times. e area below mean LST has a weaker correlation than the area above mean LST. e strength of the correlation was reduced gradually with time. Future urban and environmental planning in the premonsoon season can be implemented using the spatiotemporal variation of LST-NDVI relationship.

Data Availability
e data used to support the findings of this study are included within the article.

Conflicts of Interest
e authors declare that they have no conflicts of interest.