Environmental Science and Pollution Research (2022) 29:36376–36391

https://doi.org/10.1007/s11356-021-17480-9

RESEARCH ARTICLE

Urban green space mediates spatiotemporal variation in land surface

temperature: a case study of an urbanized city, Bangladesh
Mehedi Hasan1 · Leion Hassan1 · Mamun Abdullah Al2,3,4 · Muyassar H. Abualreesh5 · Mohd Hanafi Idris6 ·
Abu Hena Mustafa Kamal6

Received: 28 June 2021 / Accepted: 8 November 2021 / Published online: 21 January 2022
© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021

Abstract
Rapid urbanization with an increasing rate of urban built-up area is decreasing urban green space resulting in changing urban
microclimate conditions showing increasing land surface temperature. A better understanding of these effects is important
to formulate effective strategies in addressing the impact of increasing built-up area. Land surface temperature patterns in
an urbanized city in Bangladesh (Mymensingh district) were investigated using Landsat satellite sensor data from 1988 to
2016. A total of nineteen Landsat satellite images were used to retrieve land surface temperature (LST), normalized dif-
ference vegetation index (NDVI), and normalized difference built-up index (NDBI). The radiative transfer equation (RTE)
model was applied to derive LST for the years 1988, 1992, 1999, 2004, 2008, 2012, and 2016. Further, the Landsat-derived
LST results were compared with MODIS Terra satellite outputs (MOD11A1) for the validation of our study results. Our
results showed NDVI higher in 2008 and lower in 2004, LST maximum in 1988 and minimum in 2008, and NDBI higher
in 2004 and lower in 2012. Seasonally, summer was characterized by higher LST and winter by lower LST, while NDVI
was higher in autumn and lower in winter, however, NDBI was higher in winter and lower in autumn. Spatially, a relatively
higher LST and NDBI was observed in the southwest, followed by central, and northern regions, whereas the trend was
opposite for NDVI. Using Pearson’s correlation, results showed a strong significant negative correlation between LST and
NDVI and a positive significant correlation between LST and NDBI. Further, simple linear regression analysis revealed
that LST decreased with increasing NDVI most quickly in 2012, followed by the years 2016, 2008, 1992, 1988, 1999, and
2004. On the other hand, LST increased with increasing NDBI most quickly in 1999, followed by the years 2016, 1988,
1992, 2012, 2004, and 2008. Thus, long-term observation suggested that urbanization had driven a decrease in green space
while simultaneously increasing the land surface temperature within an urbanized area. This study has concluded that the
protection of urban green spaces is needed as an effective step toward addressing adverse effects of regional climate change
and desertification.

Keywords Urbanization · Urban green space · Land surface temperature (LST) · Built-up area · Mymensingh district ·
Bangladesh

Abbreviations
DN Digital number
ETM+ Enhanced Thematic Mapper Plus
Responsible Editor: Philippe Garrigues

* Mamun Abdullah Al Chinese Academy of Sciences, 1799 Jimei Road,

mamun.imsfcu16@gmail.com Xiamen 361021, People’s Republic of China
4
1 University of Chinese Academy of Sciences, Beijing 100049,
Department of Oceanography, Faculty of Marine Sciences
People’s Republic of China
and Fisheries, University of Chittagong, 4331 Chittagong,
5
Bangladesh Department of Marine Biology, Faculty of Marine Sciences,
2 King Abdul Aziz University, Jeddah 21589, Saudi Arabia
Institute of Marine Sciences, University of Chittagong,
6
Chittagong 4331, Bangladesh Faculty of Fisheries and Food Science, Universiti Malaysia
3 Terengganu, Kuala Nerus, 21030 Terengganu, Malaysia
Aquatic Eco‑Health Group, Key Laboratory of Urban
Environment and Health, Institute of Urban Environment,

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36377

LSE Land surface emissivity techniques worldwide (Ghobadi et al. 2015; Kumar and
NSIDC National Snow and Ice Data Center Shekhar 2015; Kumari et al. 2018). LST can be retrieved
OLI Operational Land Imager easily using remote sensing techniques over large geographi-
RTE Radiative transfer equation cal areas. LST-NDVI relationships depend on various cli-
TIRS Thermal Infrared Sensor matic and environmental factors, for example, vegetation
TM Thematic Mapper density, soil moisture, water bodies, and precipitation of
USGS United States Geological Survey the season or year at spatial scale (Tariq et al. 2020; Nse
et al. 2020; Guha et al. 2020). Due to freely accessible and
easy data acquisition originating from diverse satellite sen-
Introduction sors, namely, Landsat TM/ETM+/OLI, MODIS, AVHRR,
ASTER, and Sentinel, satellite imageries are widely used to
Land surface temperature (LST) is characterized as the inter- obtain LST. LST can be freely downloaded from the United
face temperature between the ground surface and the atmos- States Geological Survey (USGS) website (Macarof et al.
phere, as generated from infrared radiation from the sun 2018; Zhang et al. 2018; Alademomi et al. 2020) and can
(Ahmed et al. 2013). It is an essential parameter controlling also be estimated directly from in situ measurement. How-
physical processes such as balance and exchange of surface ever, due to factors of cost and difficulty, single-channel
radiation, energy, water, and carbon dioxide (Vlassova et al. algorithm, radiative transfer equation, and mono-window
2014; He et al. 2019). Biochemical processes on the surface algorithm methods are widely accepted for the retrieval of
of the earth can be evaluated by measuring LST (Guha and LST from Landsat satellite imageries (Zhang et al. 2018).
Govil 2021). In addition, LST also has application to studies Numerous researchers have studied LST-NDVI and LST-
of agricultural meteorology, oceanography, hydrology, urban NDBI relationships for their significance in urban monitor-
planning, forestry, and global climate change including other ing. Joint analyses of LST and NDVI have been conducted
diverse environmental studies (Vázquez et al. 1997; Macarof but only on short periods of time, such as for one or two sea-
et al. 2018). Population and economic growth leading to sons. The relationship between LST and NDVI was assessed
intensified rapid urbanization are responsible for unplanned by Abir and Saha (2021), in Pabna municipality, Bangla-
urban expansion in an overpopulated country like Bangla- desh, utilizing winter simulation data. Rashid et al. (2021),
desh, which results in a gradual decline of agricultural land, reported spatiotemporal changes in land surface tempera-
water bodies, and vegetation within city areas (Abir and Saha ture and vegetation cover in Cox’s Bazar city, Bangladesh,
2021). Besides, the increasing growth rates of population focusing on refugee camps and surrounding areas, following
and industrialization act as catalysts for changing land sur- the Rohingya influx from Myanmar. Based on these recent
face temperature (Kumari et al. 2018). Currently, more than studies, population intensity and vegetation were observed
45% of the world’s population lives in urban regions, and it to have a significant correlation with the change in LST over
is expected to exceed 60% by 2030 (Zhang et al. 2013). As short to long timescales.
a result of population dynamics and extreme urbanization, Such studies are scarce for the ancient city of
built-up areas are increasing at an alarming rate, while green Mymensingh, and examination of the seasonal or interannual
and healthy vegetated areas are declining significantly (Abir patterns of NDBI and its relationship with LST also lack
and Saha 2021). These changes negatively impact both the scientific investigation over this study area, although they
urban environment and natural weather patterns and affect are necessary. Bangladesh has four distinct seasons: win-
the quality of life and human health, to the extent that they ter (December to February), pre-monsoon (March to May),
form a major current global concern. monsoon (June to August), and post-monsoon (September
Natural green spaces are an important feature that influ- to November). As a subtropical monsoonal climate system,
ences the distribution of land surface temperature (Yuan excessive cloud cover (75–90%) occurs during the monsoon
et al. 2017). Many researchers have studied the relation- period (Abir and Saha 2021), which obstructed retrieval of
ships between LST and various vegetation indices (VI), data for the current study within this season. Therefore, the
namely, normalized difference vegetation index (NDVI), current study primarily highlighted the relationship between
normalized difference built-up index (NDBI), soil-adjusted LST, NDVI, and NDBI on Mymensingh district over the
vegetation index (SAVI), and greenness vegetation index period from 1988 to 2016, for winter, summer, and autumn
(GVI). However, NDVI is broadly used in monitoring the seasons.
relationships between land surface temperature and vegeta- In the current study, the spatiotemporal and seasonal vari-
tion, while NDBI has been used to determine the level of ation of LST in relation to NDVI and NDBI were investi-
urbanization (Kumar and Shekhar 2015; Guha et al. 2020). gated for the urbanized Mymensingh district, Bangladesh.
Thermal infrared (TIR) bands are used to identify the rela- The main objectives of this study were to (1) assess the
tionships between LST and NDVI through remote sensing spatial and temporal distribution of LST, NDVI, and NDBI

13
36378 Environmental Science and Pollution Research (2022) 29:36376–36391

from 1988 to 2016 and (2) observe the correlation of LST three seasons (winter, summer, and autumn) were used for
with NDVI and NDBI. Multiyear simulation data from 19 the years 1999, 2004, 2008, 2012, and 2016. Unfortunately,
Landsat satellites were used and were further cross-validated cloud-free images of summer, 1988 and winter, 1992 were
with MOD11A1 to confirm the LST results. Subsequently, not found, and hence the summer season was excluded from
a simple linear regression model (SLRM) was applied to analyses.
define the relationship between them. Overall, this study will
enhance understanding of changing urban vegetation and Data analyses
land surface temperature patterns associated with built-up
areas in an urbanized city within a subtropical region. In this study, ENVI 5.3 and ArcGIS 10.8 software were used
for spatial data analyses. Microsoft Excel and R program-
ming language were used for statistical data preparation and
Materials and methods analyses, respectively. The general working procedures,
such as image preprocessing and atmospheric correction,
Study area description and data collection and retrieval processes of LST, NDVI, and NDBI are briefly
described in Figure 2.
Study area description
Image preprocessing and atmospheric correction
Mymensingh is the largest district in Bangladesh. The geo-
graphical location of the study area is between 24°02′31″ to Due to sensor, solar, atmospheric, and topographic effects,
25°25′56″ N and 89°39′00″ to 91°15′35″ E, covering an area Landsat satellite images always contain some distortions
of 4363.48 ­km2. It consists of 94.18% land area, 2.43% riv- (Young et al. 2017). Therefore, before using the Landsat
erine area, and 3.35% forest reserve (Figure 1). Mymensingh images for any quantitative analysis they were preprocessed
district is bound on the north by Meghalaya state of India to minimize distortions with the general aim of achieving a
and Garo Hills, on the south by Gazipur District, on the better result. In the case of Landsat Level-1 products gener-
east by Netrokona and Kishoreganj districts, and on the ally, a geometric correction is not compulsory as images are
west by Sherpur, Jamalpur, and Tangail districts. Currently, registered and orthorectified through a systematic process
this district is urbanized and consists of thirteen upazilas, that involves ground control points and a digital elevation
namely Bhaluka, Dhubaura, Fulbaria, Gaffargaon, Gauripur, model (DEM) (Rashid et al. 2021; Guha and Govil 2021;
Haluaghat, Ishwarganj, Mymensingh Sadar, Muktagachha, Young et al. 2017). In this study, atmospheric correction
Nandail, Phulpur, Tarakanda, and Trishal. The climate of (for all sensor data) and the removal of stripes and gaps of
this area is characterized by subtropical monsoon. The win- Landsat ETM+ data were carried out using the ENVI 5.3
ter season is characterized by dry and infrequent rainfall, software. Other spatial and temporal analysess used ArcGIS
while short-term thunderstorms occur in the monsoon sea- 10.8 software.
son. Heavy rainfall is observed in the southwest monsoon
season when approximately 75% rainfall occurs. The average Normalized difference vegetation index (NDVI)
annual temperature of the area ranges from 18 to 34 °C, and
annual rainfall averages around 2541 mm (District Statistics The NDVI was the first and is the most common vegeta-
2011: Mymensingh). tion index used to explore and detect green vegetation areas
(Mancino et al. 2020). It is a simple numerical index that
Data collection uses the visible and near-infrared bands of the electromag-
netic spectrum to assess whether or not the object being
In this study, a total of nineteen, little (<2%) or cloud-free observed contains live green vegetation (Kumar and Shekhar
Landsat images were freely downloaded from the United 2015). It is used to quantify vegetation greenness and pro-
States Geological Survey’s EarthExplorer (USGS) website vides understanding of vegetation density, which can be used
(https://​earth​explo​rer.​usgs.​gov) for the years 1988, 1992, to monitor changes in plant health. In this study, NDVI was
1999, 2004, 2008, 2012, and 2016, in order to investigate calculated as a ratio between the red (R) and near-infrared
the spatial and temporal changes in LST, NDVI, and NDBI. (NIR) bands of the electromagnetic spectrum because these
Amongst them, eleven were Landsat 5 Thematic Mapper two bands have the highest absorption rate by chlorophyll.
(TM) level-1 images, five were Landsat 7 Enhanced The- The NDVI value varies from −1 to +1, with −1 and +1
matic Mapper Plus (ETM+) images, and the rest were referring to nonexistence and existence, respectively, of
Landsat 8 Operational Land Imager (OLI)/Thermal Infra- green vegetation. Purevdorj et al. (1998) used the following
red Sensor (TIRS) level-1 data (Table 1). To observe the equation to calculate the value of NDVI:
seasonal variations of LST, NDVI, and NDBI, images of

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36379

Fig. 1  Location of the study

area showing the land elevation.
The map was generated using
ArcGIS 10.8 software

Table 1  Characteristics of the Scene ID Satellite Date (D-M-Y) Time (BST) Path/row Season
collected multi-date Landsat
satellite data LT51370431988050BKT00 Landsat 5 TM 19.02.1988 09:54:05 137/043 Winter
LT51370431988290BKT00 Landsat 5 TM 16.10.1988 09:54:59 137/043 Autumn
LT51370431992077BKT00 Landsat 5 TM 17.03.1992 09:48:37 137/043 Summer
LT51370431992317ISP00 Landsat 5 TM 12.11.1992 09:45:28 137/043 Autumn
LT51370431999032BKT00 Landsat 5 TM 01.02.1999 10:03:30 137/043 Winter
LT51370431999080BKT00 Landsat 5 TM 21.03.1999 10:03:20 137/043 Summer
LT51370431999320BKT00 Landsat 5 TM 16.11.1999 10:00:04 137/043 Autumn
LT51370432004350BKT01 Landsat 5 TM 15.12.2004 10:10:07 137/043 Winter
LT51370432004062BKT00 Landsat 5 TM 02.03.2004 10:03:25 137/043 Summer
LT51370432004318BKT01 Landsat 5 TM 13.11.2004 10:09:27 137/043 Autumn
LE71370432008001SGS00 Landsat 7 ETM+ 01.01.2008 10:08:29 137/043 Winter
LT51370432008105BKT01 Landsat 5 TM 14.04.2008 10:13:45 137/043 Summer
LE71370432008289PFS00 Landsat 7 ETM+ 15.10.2008 10:13:05 137/043 Autumn
LE71370432012028PFS00 Landsat 7 ETM+ 28.01.2012 10:18:26 137/043 Winter
LE71370432012076PFS00 Landsat 7 ETM+ 16.03.2012 10:18:32 137/043 Summer
LE71370432012300PFS02 Landsat 7 ETM+ 26.10.2012 10:20:15 137/043 Autumn
LC81370432016047LGN01 Landsat 8OLI/TIRS 16.02.2016 10:24:20 137/043 Winter
LC81370432016063LGN01 Landsat 8 OLI/TIRS 03.03.2016 10:24:18 137/043 Summer
LC81370432016335LGN01 Landsat 8 OLI/TIRS 30.11.2016 10:24:45 137/043 Autumn

D-M-Y, day-month-year; BST, Bangladesh Standard Time

NIR − R while for Landsat 8 products, the calculation of NDVI =

NDVI =
NIR + R (Band 5 − Band 4)/(Band 5 + Band 4).
In the case of Landsat 4–7 products, this results in a cal-
Normalized difference built‑up index (NDBI) The NDBI is a
culation of NDVI = (Band 4 − Band 3)/(Band 4 + Band 3),
useful proxy of the built-up area as well as of barren soil. It
is calculated as a ratio between the middle infrared (MIR)

13
36380 Environmental Science and Pollution Research (2022) 29:36376–36391

band and the near infrared (NIR) band (Chen et al. 2013). Conversion of digital number (DN) to spectral radiance
Like NDVI, the value of NDBI varies between −1 and +1, for Landsat 5 and Landsat 7
where higher values indicate built-up area and barren land,
and lower values represent vegetation and water bodies. The thermal data in satellite imagery of Landsat sensors are
NDBI can be calculated as the following equation: stored in Digital Numbers (DNs) as a way of representing
pixels which have not yet been calibrated into meaningful
MIR − NIR
NDBI = units. DNs were converted to radiance to produce a mean-
MIR + NIR
ingful unit after acquiring the satellite images (Isaya Ndossi
In the case of Landsat 4–7 products, NDBI = (Band 5 and Avdan 2016). The method described in the Landsat 7
− Band 4)/(Band 5 + Band 4), while for Landsat 8 products, data users handbook (USGS 2019a, b) was followed for
NDBI = (Band 6 − Band 5)/(Band 6 + Band 5). converting the DN values to spectral radiance units at the
sensor’s aperture using the thermal band (band 6), for both
Retrieval of LST of the Landsat 5 and Landsat 7 datasets. The conversion
formula is as follows:
In this study, the RTE method was used for LST retrieval ( )
from the Landsat images (Oguz 2013). L𝜆 = Grescale ∗ Qcal + Brescale

Fig. 2  Work flowchart describ-

ing the general methodology,
data collection, and processing

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36381

which is also expressed as: can also be converted from spectral radiance (as described
( ) above) to brightness temperature, which is the effective tem-
Lmax𝜆 − Lmin𝜆
L𝜆 = .(Qcal − Qcalmin) + Lmin𝜆 perature viewed by the satellite under an assumption of unity
Qcalmax − Qcalmin emissivity (USGS 2019a, b). The conversion formula is as
follows:
where L λ represents spectral radiance at the sensor’s
aperture in Watts/(m2*sr*µm); Grescale is rescaled gain in K2
TB =
Watts/(m2*sr*µm)/DN; Brescale is rescaled bias in Watts/ ln( K1 + 1)
(m2*sr*µm); Qcal is quantized calibrated pixel value in DN L𝜆

corresponding to the thermal bands of the sensors, i.e., where TB represents effective at-satellite temperature in Kel-
Landsat TM/ETM + Band 6; Qcalmin is the minimum quan- vin; K1 and K2 are band-specific thermal conversion constant
tized calibrated pixel value in DN; Qcalmax is the maximum from the metadata and these values may vary depending on
quantized calibrated pixel value in DN; Lminλ is spectral radi- the sensor and the wavelengths by which the thermal bands
ance scaled to Qcalmin in Watts/(m2*sr*µm); Lmaxλ is spec- operate (Table S1); ln is natural algorithm and Lλ is spectral
tral radiance scaled to Qcalmax in Watts/(m2*sr*µm). These radiance in Watts/(m2*sr*µm).
values are found in the metadata file (MTL file) provided in This conversion formula was used for all sensor data, and
satellite data (USGS 2019a, b). after getting the at-sensor brightness temperature in Kelvin,
the acquired value was converted from Kelvin into degree
Conversion of top of atmospheric (TOA) spectral radiance Celsius by using the following formula:
for Landsat 8
TB (◦ C) = TB (K) − 273.15
The Thermal Infrared Sensor (TIRS) measures land surface
temperature in two thermal bands (band 10 and band 11). In
Calculation of land surface emissivity (LSE)
this study, band 10 was utilized for LST calculation as band
11 has calibration uncertainty due to stray light contamina-
LSE is a proportional factor that scales blackbody radi-
tion (Yu et al. 2014; Maithani et al. 2020). According to
ance to predict emitted radiance. In this study, the tempera-
the Landsat 8 data users handbook (USGS 2019a, b), using
ture values were calculated in reference to blackbody (as
32-bit floating-point calculation, the Landsat 8 imageries are
described above). A material cannot satisfy the properties
processed in absolute radiance unit subsequently converted
of a perfect blackbody as the concept of blackbody does
into 16-bit integer values in the final level 1 products, and
not exist practically (Isaya Ndossi and Avdan 2016), so the
these are converted to spectral radiance with the help of radi-
knowledge of land surface emissivity (LSE) is necessary to
ance scaling factors using the following equation:
apply the above methods to a Landsat image. For estimating
L𝜆 = ML ∗ Qcal + AL the real LST, it is important to relate and make corrections
of the at-sensor brightness temperature and LST, using emis-
where Lλ represents the top of atmospheric spectral radiance sivity values of the land surfaces (Krtalić et al. 2020). In this
in W/(m2 * sr * μm); ML is radiance multiplicative scaling study, LSE was calculated using the NDVI method using the
factor for the band; AL is radiance additive scaling factor following formula (Kumari et al. 2018):
for the band; Qcal is quantized and calibrated pixel value in
digital number (DN) corresponding to the thermal bands of 𝜀 = 0.004PV + 0.896
the sensors, i.e., Landsat OLI/TIRS band 10.
where ε is land surface emissivity and Pv is the proportion
of vegetation.
Conversion from spectral radiance to at‑sensor brightness
temperature PV =
(( ) ( ))2
NDVI − NDVI min ∕ NDVI max − NDVI MIN (IsayaNdossiandAvdan2016)

Brightness temperature is the temperature that is needed Though the value of emissivity of a surface can vary between
by a blackbody in order for it to be able to emit the same 0 and 1, most of the objects show spectral emissivity very
amount of radiation per unit of surface area as the body close to 1 (Stathopoulou et al. 2007).
being observed (Isaya Ndossi and Avdan 2016). Apparent
brightness temperature or at-sensor brightness temperature Estimation of land surface temperature (LST)
(TB) is usually computed by means of Planck’s law inver-
sion. In the case of the Landsat series, the thermal band data The next and final step after the calculation of LSE was the
estimation and correction of real land surface temperature.
The emissivity corrected, land surface temperature (LST)

13
36382 Environmental Science and Pollution Research (2022) 29:36376–36391

was computed by the following equation developed by Artis −0.21, −0.11, and −0.16, the higher values were 0.68,
and Carnahan (Tariq et al. 2020; Alademomi et al. 2020). 0.67, 0.71, 0.65, 0.66, 0.58, and 0.64, and the mean values
were 0.34, 0.34, 0.36, 0.32, 0.43, 0.36, and 0.34 (Table 2
TB
Ts = ( ) and Figure S1b). In our study, the maximum and minimum
1 + 𝜆.
TB
. ln𝜀 NDVI values were observed in 1999 and 1988, respectively
𝜌
(Figures S4). The mean NDVI were shown to be stable from
where Ts represents the land surface temperature (°C); TB is 1988 to 1992 (0.34), then increased to 0.02 in 1999, and
the brightness temperature; λ is wavelength of emitted radi- again decreased to 0.04 in 2004. The maximum NDVI was
ance (Table S2); ε is land surface emissivity; ρ = h × c/σ = accounted for in 2008 (0.43), subsequently decreasing 0.07
1.4388 × ­10−2 mK or 14388 µmK (σ = Boltzmann constant and 0.02 onward.
= 1.38 × 1­ 0−23 ­JK−1, h = Planck’s constant = 6.626 × 1­ 0−34 NDBI lower values were −0.73, −0.54, −0.63, −0.61,
Js, c = velocity of light = 2.998 × 108 ­ms−1). −0.54, −0.54, and −0.44, and the higher values were 0.46,
0.37, 0.46, 0.47, 0.42, 0.30, and 0.25, with mean values of
0.03, 0.04, 0.03, 0.05, −0.06, −0.17, and −0.15 (Table 2
Results and Figure S1c). In this study, the maximum and minimum
values were observed in 2004 (0.47) and 1988 (−0.73)
LST, NDVI, and NDBI distribution in the study area (Figure S5). The results showed that NDBI increased (0.01)
from 1988 to 1992, decreased (0.01) in 1999, and again
In this study, the lower values of LST were 20.6 °C, 20.8 increased (0.02) in 2004. The mean value of NDBI continu-
°C, 20.0 °C, 19.7 °C, 17.5 °C, 17.7 °C, and 19.0 °C, and ously decreased from 2004 until 2012, then again showed an
higher values were 33.2 °C, 31.4 °C, 33.5 °C, 30.8 °C, 30.2 increase (0.02) from 2012 to 2016 (Table 2).
°C, 34.5 °C, and 31.4 °C with the mean values of 24.9 °C,
23.7 °C, 24.3 °C, 23.2 °C, 22.0 °C, 22.2 °C, and 22.7 °C Spatial and temporal patterns of LST, NDVI,
for the year 1988, 1992, 1999, 2004, 2008, 2012, and 2016, and NDBI
respectively (Table 2 and Figure S1a). The value for the
regions above mean LST (pink color) and the regions below In the current study, relatively higher LST were observed
mean LST (black color) is given in Figure S2. Our results in the southwestern, central, and northern regions. Consist-
showed that the maximum LST was in 2012, followed by ently lower NDVI as well as higher NDBI was associated
1999 and 1988, while the minimum value of LST was in with spatial LST distribution (Figures S3–S5). From 2008 to
2008, followed by 2012. The LST values showed a fluctua- 2016, LST-NDVI showed a consistent relationship whereby
tion within a decadal period such as 1988 to 1992 showed decreasing vegetation led to increasing LST. However,
a 1.2 °C decrease; from 1992 to 1999, it showed a 0.6 °C inconsistency was also found for 1999 to 2004 (an oppo-
increase, from 1999 to 2012, it showed a 2.1 °C decrease and site trend). From 1988 to 1992, LST showed a decrease, but
again increased around 0.5 °C within 4 years (2012 to 2016) NDVI was unchanged during the same period (Figures S3
(Table 2 and Figure S3). and S4). To investigate the nonuniform pattern of LST dis-
The current study found fluctuating land surface tempera- tribution that might have arisen from changing the built-up
ture over the study period within the study area, which could area (increasing population intensity/or human migration),
be associated with vegetation, built-up features, and other LST-NDBI correlation was used (Figures S5 and S7).
factors. In addressing the hypotheses, NDVI and NDBI of Our study found remarkable seasonal variations in LST,
the study area were investigated. Results showed that the NDVI, and NDBI distributions from 1988 to 2016. A graphi-
lower values of NDVI were -0.39, −0.18, −0.33, −0.28, cal comparison was used to show the dynamic change of

Table 2  LST (°C), NDVI, and Year (mean) LST NDVI NDBI
NDBI variation in different
years during the study period Max. Min. Mean SD Max. Min. Mean SD Max. Min. Mean SD
(1988–2016)
1988 33.2 20.6 24.9 1.00 0.68 −0.39 0.34 0.12 0.46 −0.73 0.03 0.12
1992 31.4 20.8 23.7 1.16 0.67 −0.18 0.34 0.08 0.37 −0.54 0.04 0.10
1999 33.5 20.0 24.3 1.69 0.71 −0.33 0.36 0.08 0.46 −0.63 0.03 0.12
2004 30.8 19.7 23.2 0.78 0.65 −0.28 0.32 0.09 0.47 −0.61 0.05 0.10
2008 30.2 17.5 22.0 0.52 0.66 −0.21 0.43 0.09 0.42 −0.54 −0.06 0.08
2012 34.5 17.7 22.2 1.05 0.58 −0.11 0.36 0.06 0.30 −0.54 −0.17 0.09
2016 31.4 19.0 22.7 0.68 0.64 −0.16 0.34 0.08 0.25 −0.44 −0.15 0.06

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36383

LST (Figure 3), NDVI (Figure 4), and NDBI distributions equator from north to south at 10:00 a.m. +/− 15 min mean
(Figure 5). Nineteen Landsat satellite images were collected local time (USGS website, https://​www.​usgs.​gov/), while
to examine the seasonal variation of LST, NDVI, and NDBI MODIS Terra sensor passes the equator from north to south
over the study area. The seasonal mean LST ranged from 18.7 at roughly 10:30 a.m. local time (NSIDC website, https://​
to 29.2 °C, with a substantially higher LST found in summer nsidc.​org/​data/​modis/​index.​html). Therefore, MOD11A1
season in 1999 (42.9 °C), followed by autumn in 1988 (41.1 data sets with a spatial resolution of 1000 m were used for
°C), and summer in 1992 (40.9 °C) (Figures 3 and S6a). In comparing the retrieved LST for specific dates (Table 4).
terms of NDVI, the mean value varied from 0.15 to 0.55. The MOD11A1 data was freely downloaded from the USGS
Higher NDVI was found (0.89) in the summer season in 1992, website (https://e​ arthe​ xplor​ er.u​ sgs.g​ ov). Using ArcGIS 10.8
followed by autumn in 1988 (0.81), and in 2016 (0.81), while software, approximately 200 random point values were taken
lower means were found in winter in 2012 (0.53), followed by for each Landsat and MODIS image to evaluate the correla-
winter in 2004 (0.61), and 2008 (0.61) (Figures 4 and S6b). tion between them. Using MOD11A1 and Landsat data, the
The mean NDBI varied from −0.24 to 0.13 (Figure S6c). results reveal a significant correlation (in all cases with cor-
The highest NDBI was observed in autumn, 2008 (0.61), fol- relation coefficient value, r > 0.40) within two data sets of
lowed by summer in 1992 (0.55), and autumn in 1999 (0.53), derived LST values (Table 4). The dynamic changes of LST
whereas the lowest value was found in autumn 2008 (−0.96), retrieved from MOD11A1 showed nearly similar results to
followed by autumn in 1988 and 2012 (both −0.94) (Figure 5 those retrieved from the Landsat data (Figure 6), and thus,
and Table 3). Overall, the seasonal minimum LST values were this consistency indicated a significant level of accuracy of
observed mainly in winter (except for 1999 and 2016, the low- the method applied in the current study.
est were in autumn) with the lowest LST value recorded as
8.7 °C in 2008 (Table 3 and Figure S6a). It has been observed Correlation of LST with NDVI and NDBI
that the degree of LST change is significantly correlated with
changing yearly NDVI and NDBI (Figure S7), indicating that Pearson correlation was applied to show the relationships
seasonal variation of LST might be driven by a similar extent of LST with NDVI and NDBI (Figures S8 and S9). We took
of NDVI and NDBI variations within the area. Taken together, 30 random points values for each image to evaluate the cor-
the results indicated a progressively increasing trend in land relations using ArcGIS 10.8 software. In these analyses, the
surface temperature, with prominent variations in LST distri- correlation coefficient (r) value −1 to +1 was an indicator of
bution found with respect to different seasons and locations strength. The value +1 indicated strong positive, and −1 indi-
of the urbanized area. cated strong negative correlation. In case of LST-NDVI, the
strongest significant negative correlation was found in 2016
Comparison of the retrieved LST result with MODIS satellite (r = −0.923), followed by the years 2008 (r = −0.919), 2012
data (r = −0.917), 2004 (r = −0.906), 1992 (r = −0.899), 1988
(r = −0.891), and 1999 (r = −0.536) (Table S3). In terms
A compatible validation, by in situ measured data, or by of LST-NDBI relationship, a strong significant positive cor-
other possible ways of measurements, is crucially impor- relation was found in 1999 (r = 0.900), followed by the years
tant for any satellite-driven LST result (Guha et al. 2020; 1992 (r = 0.862), 2012 (r = 0.794), 1988 (r = 0.715), 2008 (r
Guha and Govil 2021). But it was almost impossible for = 0.699), 2016 (r = 0.693), and 2004 (r = 0.676) (Table S3).
this study to validate the satellite-retrieved LST results with Using a simple linear regression model, the results
in situ data, especially for extended time-series data. Many showed a significant negative trend of LST with respect to
researchers have proposed using MODIS data along with changes in NDVI for all years during the study period. The
Landsat-derived LST for accuracy assessment purposes model showed significantly strong relationships between
(Bindhu et al. 2013; Zhang and He 2013; Chan et al. 2018; LST and NDVI (Figure 7). The coefficient of determina-
Guha et al. 2018; Guha et al. 2020; How Jin Aik et al. 2020). tion (R2) revealed the predictability of the model, and it was
Therefore, in this study, MODIS databases have been used to accounted as 0.79, 0.81, 0.29, 0.82, 0.85, 0.84, and 0.85
get a tentative idea of the accuracy of the LST results driven for years 1988, 1992, 1999, 2004, 2008, 2012, and 2016,
by Landsat imageries. It is important to note that the MODIS respectively (Figure 7 and Table S4), which indicated that
Terra satellite sensor was launched in 1999 (December) and the change of LST significantly depended on NDVI vari-
began collecting information from 2000 (February). There- ation within the study area. On the other hand, the coeffi-
fore, the present study failed to compare the Landsat images cient of determination for LST-NDBI was 0.51, 0.74, 0.81,
with MODIS data that were acquired before that date (e.g., 0.46, 0.49, 0.63, and 0.48 for the years 1988, 1992, 1999,
1988, 1992, and 1999). In addition, there is about an ~30- 2004, 2008, 2012, and 2016, respectively, which indicated
min difference between Landsat and MODIS Terra satellite significant and upward trends of LST and NDBI correlation
data acquisition time, because Landsat sensor crosses the (Figures 7 and Table S5).

13
36384 Environmental Science and Pollution Research (2022) 29:36376–36391

Fig. 3  Seasonal and spatiotem-

poral distributions of LST (°C)
in the year 1988 (winter vs.
autumn), 1992 (summer vs.
autumn), 1999, 2004, 2008,
2012, and 2016 (winter vs.
summer, winter vs. autumn, and
summer vs. autumn)

Validation of Pearson correlation results by Spearman rank values showed slight differences from each other for all
correlation method years, but coefficients were still statistically significant
(Table S3).
In this study, Pearson’s method was used to identify the
relationship of LST with NDVI and NDBI (Figures S8
and S9). To ensure the accuracy of the correlation, Discussions
results were further validated by Spearman’s rank cor-
relation method. A comparative scenario between Pear- With respect to rapid urbanization in developing coun-
son’s and Spearman’s correlation coefficients for LST- tries, the monitoring of NDVI time series can evidence the
NDVI indicated the closest results for both methods area has increased or decreased in vegetation. However,
except for the year 1999 (Table S3). In the case of LST- a combined study using NDVI and NDBI with LST time
NDBI, Pearson’s and Spearman’s correlation coefficient series allowed the determination of the effect of seasonal

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36385

Fig. 4  Seasonal and spatiotem-

poral distributions of NDVI
in the year 1988 (winter vs.
autumn), 1992 (summer vs.
autumn), 1999, 2004, 2008,
2012, and 2016 (winter vs.
summer, winter vs. autumn, and
summer vs. autumn)

or interannual behavior of the land cover-associated micro- Distributional variations in LST, NDVI, and NDBI
climatic condition. The RTE model is the easiest, and it is in the study area
effective for retrieving LST from all types of Landsat satel-
lite sensors. In fact, there is no universal method for retriev- In this study, we found nonuniform distribution patterns of
ing accurate LST values from different satellite sensors data LST, NDVI, and NDBI. For example, LST was decreased
as all other methods have been proposed for use in different by 1.2 °C over the first 4 years of the study (1988 to 1992),
circumstances (Kumari et al. 2018; Kumari et al. 2020). In again increasing 0.6 °C within a further 7-year period, again
this study using the RTE model to estimate the LST time decreasing 2.3 °C within 9 further years, and then finally
series, a significant relationship was observed between showed an increase of 0.7 °C within the last 8 years of obser-
NDVI, NDBI, and LST in an urban city, Bangladesh. vation. Consistently, similar fluctuating behavior was also
observed for NDVI and NDBI in this urbanized city. Using a
single satellite image, both increasing and decreasing LST in
different places have been observed, for example in a study

13
36386 Environmental Science and Pollution Research (2022) 29:36376–36391

Fig. 5  Seasonal and spatiotem-

poral distributions of NDBI
in the year 1988 (winter vs.
autumn), 1992 (summer vs.
autumn), 1999, 2004, 2008,
2012, and 2016 (winter vs.
summer, winter vs. autumn, and
summer vs. autumn)

by Chaudhuri and Mishra (2016), wherein they reported a arisen from the methodological and geographical factors
1.83 °C decrease, whereas Abir and Saha (2021) observed a that bias the results, e.g., data acquisition time, annual or
0.63 °C increase in LST. The differences of LST variations seasonal intervals, and so on. In our study, results showed
in different places from decadal time series analysis have changing NDVI and NDBI were simultaneously responsible

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36387

Table 3  Seasonal distribution Year Season LST NDVI NDBI

of LST (°C), NDVI, and NDBI
during the study period Max. Min. Mean SD Max. Min. Mean SD Max. Min. Mean SD

1988 Winter 29.6 17.9 22.2 1.40 0.72 −0.33 0.24 0.11 0.44 −0.72 0.13 0.16
Autumn 41.1 19.3 27.6 1.16 0.81 −0.79 0.45 0.23 0.50 −0.94 −0.06 0.14
1992 Summer 40.9 19.7 25.1 2.01 0.89 −0.27 0.26 0.09 0.55 −0.50 0.12 0.12
Autumn 29.6 20.2 22.4 0.71 0.73 −0.33 0.42 0.13 0.42 −0.77 −0.05 0.11
1999 Winter 29.6 17 21.2 1.41 0.71 −0.76 0.24 0.12 0.51 −0.80 0.06 0.17
Summer 42.9 21.9 29.2 3.85 0.71 −0.76 0.41 0.16 0.43 −0.65 0.07 0.16
Autumn 30 16.6 22.4 0.66 0.79 −0.62 0.42 0.17 0.53 −0.87 −0.03 0.13
2004 Winter 25.8 15.6 20.8 0.79 0.61 −0.33 0.22 0.10 0.48 −0.69 0.13 0.15
Summer 35.7 20.6 25.1 1.78 0.72 −0.26 0.33 0.10 0.46 −0.63 0.05 0.12
Autumn 33 19.3 23.8 0.76 0.74 −0.44 0.39 0.16 0.49 −0.85 −0.02 0.13
2008 Winter 26.4 8.7 19.7 0.92 0.61 −0.28 0.22 0.10 0.34 −0.74 −0.05 0.13
Summer 30.5 17.7 21.2 1.08 0.73 −0.10 0.52 0.10 0.46 −0.60 −0.09 0.12
Autumn 35 17.6 25 0.74 0.80 −0.53 0.55 0.16 0.61 −0.96 −0.03 −0.09
2012 Winter 29.8 9.3 18.7 0.99 0.53 −0.25 0.15 0.08 0.32 −0.84 −0.13 0.14
Summer 38.2 17.6 23.9 2.19 0.70 −0.13 0.42 0.11 0.31 −0.57 −0.14 0.11
Autumn 35.5 21.8 24 0.82 0.76 −0.28 0.51 0.12 0.42 −0.94 −0.24 0.08
2016 Winter 32.5 18.1 22.4 0.93 0.64 −0.15 0.23 0.08 0.27 −0.53 −0.17 0.09
Summer 32.1 16.4 22.9 1.09 0.67 −0.05 0.37 0.08 0.31 −0.54 −0.18 0.08
Autumn 30.2 16.3 22.9 0.71 0.81 −0.54 0.42 0.15 0.32 −0.44 −0.09 0.07

for LST variations of the study area. A recent study by Spatial and temporal variations in LST are related
Rashid et al. (2021) found a significant negative correlation to NDVI and NDBI of the study area
between land surface temperature and vegetation cover in
Cox’s Bazar city, Bangladesh, after Rohingya influx migra- We observed relatively higher LST and NDBI, but lower
tion from Myanmar with corresponding settlement devel- NDVI, in the southwestern part of the city, followed by cen-
opment. Therefore, increasing built-up area and decreasing tral and northern regions, indicating spatial patterns of LST,
vegetation area can control LST distribution patterns. Previ- NDBI, and NDVI distributions. The patterns might be better
ous studies stated that human activities related to decreas- explained by rapid population growth and increased devel-
ing green vegetation could significantly drive spatial LST opment activities for human settlement or industrial devel-
distribution (Ahmed et al. 2013; Kumar and Shekhar 2015; opment. Recent studies suggest that development activities
Marzban et al. 2018; Zhang et al. 2018; Guha et al. 2020; have a significant negative influence on LST (Guha et al.
Kumari et al. 2020). Consistently, the positive LST-NDBI 2020; Kumari et al. 2020). Consistently, the positive correla-
relationship and negative LST-NDVI relationship suggest tion between LST and NDBI supporting those developmen-
that land cover and vegetation can drive land surface tem- tal activities (e.g., built-up area) is one of the responsible
perature distributional patterns in an urbanized city. factors for the distribution of LST and decreasing NDVI
across time and space. Furthermore, the current study has
illustrated a seasonal variation of LST regarding changes of
NDVI and NDBI. Previous studies stated that the interan-
nual variation in LST related to NDVI variation might be
Table 4  Correlation between Landsat and MODIS satellite sensors driven by air temperature and rainfall variation (Abir and
derived LST results (n = 200)
Saha 2021; Khatun et al. 2016). This may be because the
Acquisition Acquisi- MODIS acquired LST Pearson’s monsoon is characterized by fluctuating air temperature and
date of Land- tion date of (°C) correlation several rain events, which play important roles in chang-
sat data MODIS data coefficient
Max. Min. Mean ing land surface temperature in the northern Bay of Bengal
region (Khatun et al. 2016). In Bangladesh, approximately
01.01.2008 01.01.2008 25.85 21.43 23.73 0.58
8% of annual rainfall occurs in the summer to autumn while
15.10.2008 15.10.2008 37.03 28.93 31.17 0.44
the winter experiences very dry conditions and low rain-
16.03.2012 16.03.2012 35.23 24.43 28.20 0.63
fall (~2%) (Khatun et al. 2016). Summer is characterized
26.10.2012 26.10.2012 33.49 27.29 28.44 0.62
by high air temperature, which causes heatwave conditions

13
36388 Environmental Science and Pollution Research (2022) 29:36376–36391

Fig. 6  Distribution of LST

using MODIS data. a-b, 2008;
c-d, 2012

over the country which might be one of the influencing fac- studies found that rain events, along with other factors such
tors for higher LST within this season. In addition, rela- as soil moisture, latitude, and land topography, also have a
tively higher NDVI is observed in the autumn, which may significant temporal and spatial influence on LST distribu-
be associated with rainfall events and related to increased tion (Tariq et al. 2020; Nse et al. 2020; Mukherjee and Singh
vegetation from farmed crops or other agriculture. Recent 2020). The current study highlighted significant insights into

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36389

Fig. 7  Regression analysis showing the significant relationships of LST with NDVI and NDBI. a and b 1988, c and d 1992, e and f 1999, g and
h 2004, i and j 2008, k and l 2012, and m and n 2016

spatiotemporal NDVI and NDBI variations related to LST Alademomi et al. 2020). Consistently, the negative relation-
distributional patterns in the selected urbanized city. ship of LST-NDVI suggested the importance of urban green
vegetation. On the other hand, the regression analysis results
Validation of different models applied in this study showed a significant upwards trend for all years, which
meant that there was a positive relationship between LST
To assess the question of the validity of the use of Land- and NDBI. Previous studies stated that positive relationships
sat data on LST time series, the cloud-free temporal series between LST-NDBI (e.g., built-up area) were indicative of
were calculated and compared with MOD11A1 data. Using rapid urbanization (Ogashawara and Bastos 2012; Chen et al.
matching data of Landsat and MOD11A1 data sets, results 2013; Guha et al. 2018).
showed a good level of approximation of LST variation
within the study area (Figure 7). Additionally, Pearson and Limitations of the current study
Spearman’s correlations between LST-NDVI and LST-
NDBI showed good consistency, indicating that the appli- Our study showed significant insights into the LST-NDVI
cation of the models was adequate in this study (Figures S8 and LST-NDBI relationships within an urbanized area in
and S9). Generally, correlation-based analysis can show the Bangladesh. In this study, we used 4-year interval data in a
strength and direction of relationships within two variables, small geographical area (regional scale), whereas annual or
while regression analysis additionally predicts the value of monthly data for a large geographical area, i.e., multiple cit-
a dependent variable based on the changes in an independ- ies, would be preferable to obtain a better resolution of LST
ent variable with functions. The regression analyses showed patterns. Importantly, LST-NDVI and LST-NDBI were used
a good fit to the model and a significant downswing trend to observe land surface temperature patterns. However, other
line for all years, which meant that there was a negative indices would also be used in the monitoring program, espe-
correlation between LST and NDVI, so that the higher the cially SAVI, and greenness vegetation index (GVI), because
NDVI value or green vegetation, the lower will be the LST LST of a given region is controlled by numerous physical
value. Having a significant cooling capacity, green vegeta- and climatic parameters, namely land use/land cover, topog-
tion reflects heat through the evapotranspiration process and raphy, altitude, precipitation, soil moisture, vegetation cover-
decreases LST (Ghobadi et al. 2015; Khandelwal et al. 2018; age, the density of population (Mallick and Rahman 2012;

13
36390 Environmental Science and Pollution Research (2022) 29:36376–36391

Khandelwal et al. 2018; Kumari et al. 2018; Kumari et al. Consent to participate Not applicable.
2020).
Consent for publication Not applicable.

Conflict of interest The authors declare no competing interests.

Conclusion

In this study, the RTE model was used to derive LST. From
NDVI, NDBI, and LST time series analyses, the present
References
study results showed that the northern, central, and south- Abir FA, Saha R (2021) Assessment of land surface temperature and
western regions of the study area experienced relatively land cover variability during winter: a spatio-temporal analysis of
higher land surface temperatures. Higher LST was observed Pabna municipality in Bangladesh. Environ Challenges 4:100167
in summer, while the winter was characterized by lower Ahmed B, Kamruzzaman MD, Zhu X, Rahman M, Choi K (2013) Sim-
ulating land cover changes and their impacts on land surface tem-
LST. Relatively higher NDVI was observed in autumn for all perature in Dhaka, Bangladesh. Remote Sensing 5(11):5969–5998
years, indicating autumn as being healthier and greener than Alademomi AS, Okolie CJ, Daramola OE, Agboola RO, Salami TJ
other seasons. LST-NDBI showed a significant upward trend (2020) Assessing the relationship of LST, NDVI and EVI with
indicating that the built-up area of urbanization had a nega- land cover changes in the Lagos lagoon environment. Quaest Geo-
graphicae 39(3):87–109
tive influence on LST variation. Applying a simple linear Bindhu VM, Narasimhan B, Sudheer KP (2013) Development and veri-
regression model, the results revealed a good fit to the model fication of a non-linear disaggregation method (NL-DisTrad) to
and showed a strong significant negative correlation between downscale MODIS land surface temperature to the spatial scale
LST and NDVI, while NDBI showed a significant positive of Landsat thermal data to estimate evapotranspiration. Remote
Sens Environ 135:118–129
relationship with LST. Based on the findings, it was con- Chaudhuri G, Mishra NB (2016) Spatio-temporal dynamics of land
cluded that urban land surface temperature increased with cover and land surface temperature in Ganges-Brahmaputra delta:
decreasing NDVI. A positive relationship between NDBI a comparative analysis between India and Bangladesh. Appl
and LST was observed. Thus, the findings of this study Geogr 68:68–83
Chan HP, Chang CP, Dao PD (2018) Geothermal anomaly mapping
would be significant for policy makers in developing sustain- using Landsat ETM+ data in Ilan Plain, Northeastern Taiwan.
able and environment-friendly city planning. Overall, our Pure and Applied Geophysics 175(1):303–323
study would like to recommend some suggestive approaches Chen LY, Li MY, Huang F, Xu SL (2013) Relationships of LST to
such as sustainable urban development, tree plantation pro- NDBI and NDVI in Wuhan City based on Landsat ETM+ image.
6th International Congress on Image and Signal Processing
grams in the urban regions, and control of excessive human (CISP) 2:840–845
migration to the city. For example, building multistoried District Statistics (2011) Mymensingh. Available online: (http://​203.​
buildings may save more open space. Secondly, encourag- 112.​2 18.​6 5:​8 008/​WebTe​stApp​l icat​i on/​u serf ​i les/​I mage/​D istr​
ing and promoting farming, whether intensive or extensive, ict%​20Sta​tisti​cs/​Mymen​singh.​pdf). Last visited/retrieved on
2020.10.30
will be a good practice for increasing seasonal vegetation Ghobadi Y, Pradhan B, Shafri HZM, Kabiri K (2015) Assessment of
and supporting sustainable local agriculture. Although our spatial relationship between land surface temperature and landuse/
baseline study showed significant LST variation regarding cover retrieval from multi-temporal remote sensing data in South
observed changes of NDVI and NDBI, a longer duration Karkheh Sub-basin, Iran. Arab J Geosci 8(1):525–537
Guha S, Govil H, Dey A, Gill N (2018) Analytical study of land sur-
study with monthly satellite data and in situ field observa- face temperature with NDVI and NDBI using Landsat 8 OLI and
tional data should be considered to justify this conclusion. TIRS data in Florence and Naples City, Italy. Eur J Remote Sens
51(1):667–678
Supplementary Information The online version contains supplemen- Guha S, Govil H (2021) An assessment on the relationship between
tary material available at https://d​ oi.o​ rg/1​ 0.1​ 007/s​ 11356-0​ 21-1​ 7480-9. land surface temperature and normalized difference vegetation
index. Environ Dev Sustain 23:1944–1963
Author contribution MAA and MMH: conceived the idea and devel- Guha S, Govil H, Diwan P (2020) Monitoring LST-NDVI relationship
oped the methodology. MMH and MLH: collected data and wrote the using Premonsoon Landsat datasets. Adv Meteorol 4539684:15.
first draft. MAA and MMH: validation and visualization. AHMK, https://​doi.​org/​10.​1155/​2020/​45396​84
MHA, and MHI: critically reviewed, corrected, and revised the manu- He J, Zhao W, Li A, Wen F, Yu D (2019) The impact of the terrain
script. All authors approved the final version of the manuscript and effect on land surface temperature variation based on Land-
agreed to submit it. sat-8 observations in mountainous areas. Int J Remote Sens
40(5–6):1808–1827
How JinAik D, Ismail MH, Muharam FM (2020) Land use/land cover
Data availability United States Geological Survey (USGS) website changes and the relationship with land surface temperature using
(https://​earth​explo​rer.​usgs.​gov). Landsat and MODIS imageries in Cameron Highlands, Malaysia.
Land 9(10):372
Declarations IsayaNdossi M, Avdan U (2016) Application of open-source coding
technologies in the production of land surface temperature (LST)
Ethics approval Not applicable.

13
Environmental Science and Pollution Research (2022) 29:36376–36391 36391

maps from Landsat: a PyQGIS plugin. Remote Sens 8(5):413. Rashid KJ, Hoque MA, Esha TA, Rahman MA, Paul A (2021) Spa-
https://​doi.​org/​10.​3390/​rs805​0413 tiotemporal changes of vegetation and land surface temperature
Khandelwal S, Goyal R, Kaul N, Mathew A (2018) Assessment of in the refugee camps and its surrounding areas of Bangladesh
land surface temperature variation due to change in elevation of after the Rohingya influx from Myanmar. Environ Dev Sustain
area surrounding Jaipur, India. Egypt J Remote Sens Space Sci 23(3):3562–3577
21(1):87–94 Stathopoulou M, Cartalis C, Petrakis M (2007) Integrating Corine
Khatun MA, Rashid MB, Hygen HO (2016) Climate of Bangla- Land Cover data and Landsat TM for surface emissivity defini-
desh. Bangladesh Meteorological Department Technical Report tion: application to the urban area of Athens, Greece. Int J Remote
08/2016, pp 1–158 Sens 28(15):3291–3304
Krtalić A, Divjak AK, Čmrlec K (2020) Satellite-driven assessment of Tariq A, Riaz I, Ahmad Z, Yang B, Amin M, Kausar R, Rafiq M (2020)
surface urban heat islands in the city of Zagreb, Croatia. ISPRS Land surface temperature relation with normalized satellite indi-
Ann Photogramm Remote Sens Spat Inf Sci 3:757–764 ces for the estimation of spatio-temporal trends in temperature
Kumar D, Shekhar S (2015) Statistical analysis of land surface tem- among various land use land cover classes of an arid Potohar
perature–vegetation indexes relationship through thermal remote region using Landsat data. Environ Earth Sci 79(1):1–15
sensing. Ecotoxicol Environ Saf 121:39–44 USGS (2019a) Landsat 7 (L7) Data users handbook. Available online:
Kumari B, Tayyab M, Mallick J, Khan MF, Rahman A (2018) Satellite- https://​www.​usgs.​gov/​media/​files/​lands​at-7-​data-​users​handb​ook.
driven land surface temperature (LST) using Landsat 5, 7 (TM/ Last visited/retrieved on 2020.12.20
ETM+ SLC) and Landsat 8 (OLI/TIRS) data and its association USGS (2019b) Landsat 8 (L8) Data users handbook. Available online:
with built-up and green cover over urban Delhi, India. Remote https://​www.​usgs.​gov/​media/​files/​lands​at-8-​data-​users​handb​ook.
Sens Earth Syst Sci 1(3):63–78 Last visited/retrieved on 2020.12.20
Kumari B, Tayyab M, Ahmed IA, Baig MRI, Khan MF, Rahman A Vázquez DP, Reyes FO, Arboledas LA (1997) A comparative study of
(2020) Longitudinal study of land surface temperature (LST) algorithms for estimating land surface temperature from AVHRR
using mono-and split-window algorithms and its relationship data. Remote Sens Environ 62(3):215–222
with NDVI and NDBI over selected metro cities of India. Arab J Vlassova L, Pérez-Cabello F, Mimbrero MR, Llovería RM, García-
Geosci 13(19):1–19 Martín A (2014) Analysis of the relationship between land surface
Macarof P, Groza S, Statescu F (2018) Investigating correlation LST temperature and wildfire severity in a series of Landsat images.
and vegetation indices using Landsat images for the warmest Remote Sens 6(7):6136–6162
month: a case study of Iasi County. Annals of Valahia University Young NE, Anderson RS, Chignell SM, Vorster AG, Lawrence R,
of Targoviste. Geographical Series 18(1):33–40 Evangelista PH (2017) A survival guide to Landsat preprocess-
Maithani S, Nautiyal G, Sharma A (2020) Investigating the effect of ing. Ecology 98(4):920–932
lockdown during COVID-19 on land surface temperature: study of Yuan X, Wang W, Cui J, Meng F, Kurban A, De Maeyer P (2017)
Dehradun City, India. J Indian Soc Remote Sens 48(9):1297–1311 Vegetation changes and land surface feedbacks drive shifts in local
Mallick J, Rahman A (2012) Impact of population density on the temperatures over Central Asia. Sci Rep 7(1):1–8
surface temperature and micro-climate of Delhi. Curr Sci Yu X, Guo X, Wu Z (2014) Land surface temperature retrieval from
102(12):1708–1713 Landsat 8 TIRS—comparison between radiative transfer equation-
Mancino G, Ferrara A, Padula A, Nolè A (2020) Cross-comparison based method, split window algorithm and single channel method.
between Landsat 8 (OLI) and Landsat 7 (ETM+) derived veg- Remote Sens 6(10):9829–9852
etation indices in a Mediterranean environment. Remote Sens Zhang Z, He G (2013) Generation of Landsat surface temperature prod-
12(2):291. https://​doi.​org/​10.​3390/​rs120​20291 uct for China, 2000–2010. Int J Remote Sens 34(20):7369–7375
Marzban F, Sodoudi S, Preusker R (2018) The influence of land-cover Zhang Y, Odeh IO, Ramadan E (2013) Assessment of land surface
type on the relationship between NDVI–LST and LST-T air. Int J temperature in relation to landscape metrics and fractional vegeta-
Remote Sens 39(5):1377–1398 tion cover in an urban/peri-urban region using Landsat data. Int J
Mukherjee F, Singh D (2020) Assessing land use–land cover change Remote Sens 34(1):168–189
and its impact on land surface temperature using LANDSAT data: Zhang F, Kung H, Johnson VC, LaGrone BI, Wang J (2018) Change
a comparison of two urban areas in India. Earth Syst Environ detection of land surface temperature (LST) and some related
4(2):385–407 parameters using Landsat image: a case study of the Ebinur Lake
Nse OU, Okolie CJ, Nse VO (2020) Dynamics of land cover, land watershed, Xinjiang, China. Wetlands 38(1):65–80
surface temperature and NDVI in Uyo City, Nigeria. Sci Afr
10:e00599. https://​doi.​org/​10.​1016/j.​sciaf.​2020.​e00599 Publisher's note Springer Nature remains neutral with regard to
Ogashawara I, Bastos VDSB (2012) A quantitative approach for ana- jurisdictional claims in published maps and institutional affiliations.
lyzing the relationship between urban heat islands and land cover.
Remote Sens 4(11):3596–3618
Oguz H (2013) LST calculator: A program for retrieving land sur-
face temperature from Landsat TM/ETM+ imagery. Environ Eng
Manag J 12(3):549–555
Purevdorj TS, Tateishi R, Ishiyama T, Honda Y (1998) Relationships
between percent vegetation cover and vegetation indices. Int J
Remote Sens 19(18):3519–3535

13