Journal of Environmental Management 245 (2019) 348–357

Contents lists available at ScienceDirect

Journal of Environmental Management

journal homepage: www.elsevier.com/locate/jenvman

Research article

Analysis and simulation of land cover changes and their impacts on land T
surface temperature in a lower Himalayan region
Siddique Ullah, Khalid Ahmad, Raja Umer Sajjad, Arshad Mehmood Abbasi, Abdul Nazeer,
Adnan Ahmad Tahir∗
Department of Environmental Sciences, COMSATS University Islamabad (CUI), Abbottabad Campus, 22060, Abbottabad, Pakistan

A R T I C LE I N FO A B S T R A C T

Keywords: Rapid urbanization is changing the existing patterns of Land Use Land Cover (LULC) globally which is conse-
Land use land cover quently increasing the Land Surface Temperature (LST) in many regions. Present study was focused on esti-
Land surface temperature mating the current and simulating the future LULC and LST trends in the alpine environment of lower Himalayan
Artificial neural network region of Pakistan. Past patterns of LULC and LST were identified through the Support Vector Machine (SVM)
Cellular automata model
and multi-spectral Landsat satellite images during 1987–2017 data period. The Cellular automata (CA) model
Support vector machine
Urban heat island
and Artificial Neural Network (ANN) were applied to simulate future (years 2032 and 2047) LULC and LST
changes, respectively, using their past patterns. CA model was validated for the simulated and the estimated
LULC for the year 2017 with an overall Kappa (K) value of 0.77 using validation modules in QGIS and IDRISI
software. ANN method was validated by correlating the observed and simulated LST for the year 2017 with
correlation coefficient (R) and Mean Square Error (MSE) values of 0.81 and 0.51, respectively. Results indicated
a change in the LULC and LST for instance the built-up area was increased by 4.43% while agricultural area and
bare soil were reduced by 2.74% and 4.42%, respectively, from 1987 to 2017. The analysis of LST for different
LULC classes indicated that built-up area has highest temperature followed by barren, agriculture and vegetation
surfaces. Simulation of future LULC and LST showed that the built-up area will be increased by 2.27% (in 2032)
and 4.13% (in 2047) which led 42% (in 2032) and 60% (in 2047) of the study area as compared to 26% area (in
2017) to experience LST greater than 27 °C. A strong correlation between built-up area changes and LST was thus
found signifying major challenge to urban planners mitigating the consequent of Urban Heat Island (UHI)
phenomenon. It is suggested that future urban planning should focus on urban plantation to counter UHI
phenomena in the region of lower Himalayas.

1. Introduction
Landscape changes occur due to both natural and man-made rea-
sons. The recent decades experienced significant landscape changes
mainly due to rapid urban growth which has negatively influenced land
surface temperature (LST) (Yuen and Kong, 2009). The urban areas
comprising the 54% of the global population are continuously subjected
to complex interactions of economic development and regional ex-
ploitation (Weng, 2001). This brings change in land use land cover
(LULC) which affects the urban geophysical condition (Herold et al.,
2003) and creates environmental disturbances including changes in
climate (Xiao et al., 2006) and LST change. Land cover indicates the
part of earth covered by natural features while human occupied area is
called the land use but they are often studied together due to their
strong interconnection (Verburg et al., 2009). LST varies both spatially
and temporally over land surface features. For example, a significant
temperature increase in urban areas may be observed compared to the
surrounding areas (Zhang et al., 2017). The urban (built-up) area which
was only 3% in 1950 is expected to reach 66% in 2050 (United Nations,
2018) globally. Due to the rapid expansions in urban environment,
vegetation cover is being replaced with built-up areas which increase
LST by trapping more incoming solar radiation on the land surface. This
is resulting in Urban Heat Island (UHI) creation worldwide (Yang et al.,
2017; Ahmed, 2018; El-Hattab et al., 2018).
Dense vegetation has cooling effect on the urban environment
(Matthews et al., 2015) and impervious surfaces with low vegetation
and low albedo have positive relationship with LST and contribute for
UHI formation. The LULC data are vital for long-term sustainable po-
licies (Long et al., 2008) regarding urban planning. LULC mapping is
the efficient tool for spatial and temporal change detection of different

∗
Corresponding author.
E-mail address: Adnantahir@cuiatd.edu.pk (A.A. Tahir).

https://doi.org/10.1016/j.jenvman.2019.05.063
Received 5 February 2019; Received in revised form 24 April 2019; Accepted 17 May 2019
Available online 31 May 2019
0301-4797/ © 2019 Elsevier Ltd. All rights reserved.
S. Ullah, et al.
Fig. 1. (a) Location map of the study area (Abbottabad region, Pakistan) with major mountain ranges Hindukush, Karakoram and western Himalaya, (b) Digital
elevation model (DEM) of the district Abbottabad showing elevation change and superimposed with major roads.
land features. For this purpose, satellite remote sensing (RS) in com-
bination with Geographic Information Systems (GIS) is being used ex-
tensively (Weng, 2001; Selçuk et al., 2003). Multi-spectral Landsat re-
mote sensing data has the longest continuous available record (∼45
years) for LULC estimations globally. Inclusion of thermal band in these
data sets makes it easier to study LST in addition to LULC from the same
image. This data was used by many researchers (e.g. Sobrino et al.,
2004; Hu and Jia, 2010; Keshtkar and Voigt, 2016) to estimate the
LULC and LST changes over long time-period.
While the past patterns of LULC and LST change may be detected
from the available remote sensing data, it is a difficult task to model
such dynamic system (Ahmed, 2011) for prediction of future changes.
The most common methods used for LULC simulation and prediction
are Markov chain (Balzter, 2000), cellular automata (Santé et al., 2010;
Zenil, 2010), and Artificial Neural Network (Civco, 1993) models. Each
tool has its own strengths and weaknesses. Markov Chain is preferred
when the land cover change is known but it lacks spatial dependency
and spatial distribution (Balzter, 2000). The Cellular Automata (CA)
model identifies the state of each cell in an array depending on the
previous state of the cells within a neighborhood, according to a set
transition rules (Santé et al., 2010). CA model is commonly used for
simulation of future LULC by using its past pattern (Ahmed et al., 2013;
Parsa and Salehi, 2016; Hadi et al., 2014; Rahman et al., 2017; Rimal
et al., 2017; Guidigan et al., 2018; Akbar et al., 2019). The logistic
regression is used to model transition potential matrix. One of the main
advantages of this model is that it is based on the past pattern and
driving variables such as distance to roads, slope and elevation. This
model is commonly applied in land utilities research (Ménard and
Marceau, 2005). This is similar to land use activities that are frequently
characterized by the interaction of adjoining landscape patterns. The
model's accuracy is validated by Kappa variation using simulated and
observed LULC map (Keshtkar and Voigt, 2016). This research em-
ployed CA model with logistic regression for LULC simulation.
Artificial Neural Network (ANN) and various LULC indices may be
used to predict LST for a region. Ahmed et al. (2013) used Normalized
Difference Vegetation Index (NDVI) to project urban natural landscape
and future LST in Dhaka Bangladesh. Many studies showed that NDVI is
a weaker predictor of LST than other indices like the Normalized Dif-
ference Built Index (NDBI), vegetation fraction and the percentage
Impervious Surface Area (ISA) (Chen et al., 2006; Li and Liu, 2008;
Deng and Wu, 2013). In another study, Hasanlou and Mostofi (2015)
estimated LST based on a linear function of a combination of indices
which included NDVI, NDBI, Normalized Difference Bareness Index
349
Journal of Environmental Management 245 (2019) 348–357
(NDBaI), Normalized Difference Water Index (NDWI), Soil Adjusted
Vegetation Index (SAVI), Enhanced Built-up and Bareness Index (EBBI),
Urban Index (UI), and Built-Up Index (BUI). However, Ahmed et al.
(2013) reported that when several factors are used in a linear regression
model, accuracy of LST simulation may be compromised due to co-
linearity between the factors. Mushore et al. (2017) tested the potential
of various vegetation and non-vegetation indices to predict LST and
reported that the Urban index (UI) was the best predictor of LST. The
indices based prediction cannot accommodate past pattern that is why
ANN is a preferable option for LST simulation. In addition, ANN is
highly suited for a study area where the underlying processes are not
well known (Haykin, 1999). ANN does not require too much of prior
knowledge of the real world complex network dynamics to reconstruct
the underlying process in order to make predictions (Dhamge et al.,
2012). ANN can comprehend complex data characteristics which pre-
sent serious problems for traditional statistical techniques because of its
nonlinear structure. Therefore, ANN presents the best approach to
model, simulate and predict the time series system like LST without
much provision of external influencing factors such as in the alpine
areas of northern Pakistan.
Northern valley areas of Pakistan are being affected by the large
influx of population from surrounding mountainous areas. Therefore,
the urban areas situated in this region and once known for their alpine
forests and pleasant climate are now turning into built-up areas with
diminishing vegetation cover and increasing temperatures. Absence of
ground LULC and LST observations both on spatial and temporal scale
limited the study of these important factors for future urban planning in
this region, so far. Multi-spectral remote sensing data such as Landsat
and modeling approaches such as CA and ANN may prove useful to
study the past and future patterns of LULC and LST in this area. It may
help the urban planners and policy makers to mitigate the UHI effect
from this alpine area. This present study is first of its kind in this
geographical region designed to:
• Identify patterns of LULC change and LST for the period of
1987–2017 using land surface classification maps and thermal
bands retrieved from Landsat data; and
• Simulate the future LULC changes for the years 2032 and 2047, and
their impacts on LST using past patterns and modeling approaches.
2. Case study
Study area i.e. Abbottabad district (Fig. 1), is situated in the Khyber
S. Ullah, et al.
Pakhtunkhwa, KPK, province of northern Pakistan, at 33°45′ to 34°30′
N Latitude and 73°–73°30′ E Longitude. It comprises an area of 1967-
km2 with a population of 1.3 million. This region received a huge po-
pulation from surrounding less developed areas during the past 30 years
mainly due to its weather, peaceful reputation, educational networks
and the perceived security of a garrison. In addition, there was an influx
of migrants from neighboring Azad Kashmir region after the earth
quake destruction in 2005, and from north-western parts of the country
during military operations in 2009 and 2010. Such large-scale migra-
tion brought significant changes in the existing LULC patterns of this
region. Its population increased from 0.8 to 1.3 million from 1998 to
2017 with urban and rural growth of 4.34% and 2.5%, respectively
(Pakistan bureau of statistics, 2017). Total annual rainfall received in
this region is ∼1262 mm, and mean annual temperature is ∼18 °C,
where June is the hottest month with an average temperature of
∼27.3 °C and January is the coldest month with temperatures aver-
aging ∼7.5 °C (Afzaal et al., 2009).
Situated in the lower western-Himalayas, district Abbottabad is a
popular tourist center as well as a major transit point to all major tourist
regions of Pakistan situated in the upper-western Himalayas and
Karakoram ranges. The Karakoram Highway, KKH (Fig. 1), which traces
one of the paths of the ancient Silk Road, is also passing through this
region.
3. Data sets and methods
3.1. Data sets and treatment process
Multi-spectral Landsat satellite data was acquired from United
States Geological Survey (USGS) for the period of 30 years from 1987 to
2017 to assess LULC change and its effect on LST in the study area. The
images were downloaded for the same month (May) and within five-
days span (19 May – 24 May) for all the data years to avoid the effect of
seasonal change. During the download process, cloud cover was set to
be < 15% to have clear weather background of all three images so that
they are comparable and not affected by the weather conditions such as
significant change in temperature or humidity. Cloud cover on entire
Landsat scene was between 6 and 13% for all three images but it was
near zero over our studied region (district Abbottabad, Fig. 1). Details
(date, sensor, scene ID, cloud cover and Path/Row) of the images ac-
quired from USGS online data portal are presented in Table 1.
A standard pretreatment process was applied to Landsat images
before deriving LULC maps. This process included atmospheric cor-
rection and radiometric calibration. The data was then analyzed for
LULC change and LST through ArcMap, Envi 5.3, IDRISI and QGIS
softwares. Support Vector Machine (SVM) method from ENVI 5.3
software was applied to derive the LULC classification maps for years
1987, 2002 and 2017. SVM is a non parametric classifier equipped with
a set of related learning algorithms used for regression and classifica-
tion. The theory was suggested by Vapnik and Chervonenkis (1968),
and later detailed by Vapnik (1999). The LST was calculated from the
thermal bands of available Landsat satellite images for the study area.
LULC maps were simulated to project the future variation in year
2032 and 2047 using cellular automata (CA) model. The main purpose
of the selection of 2032 and 2047 for prediction was to assess short (15-
years) and long (30-years) term scenario for the urban planning as
mentioned by Mushore et al. (2017). Past LULC trends were used as
Table 1
Details of downloaded Landsat satellite images. (Source: US Geological Survey).
Date Acquired (M/D/Y) Scene ID
5/24/1987 LT51500361987114ISP00
5/19/2002 LE71500362002139SGS00
5/20/2017 LC81500362017140LGN00
350
Journal of Environmental Management 245 (2019) 348–357
dependent variables and slope, elevation and distance to roads as in-
dependent variables to project future LULC change. Road network
vector data (GIS shape file) for the study area was acquired from the
Survey of Pakistan Department and was used to calculate distances to
roads. Digital Elevation Model (DEM) data having spatial resolution of
30-m was obtained from Shuttle Radar Topography Mission (SRTM) of
the USGS. It was used to estimate the elevation and slope of the study
area. The ANN model was used to simulate LST projections for the years
2032 and 2047, based on the past patterns of LST from 1987 to 2017.
Details of the data treatment process and modeling approach are pre-
sented in sections 3.1 to 3.4.
3.1.1. Derivation of LULC classification maps for 1987‒2017 data period
The acquired Landsat images were classified into five LULC classes
(built-up, agriculture, vegetation, bare soil and water surface) for the
year of 1987, 2002, and 2017, based on SVM method in ENVI 5.3
software. The images were investigated based on their spectral and
areal profiles to develop training sites using supplementary information
and baseline data from various sources (e.g. Ground survey points,
Google images etc.). Approximately, forty training samples for each
LULC class were selected. The accuracy of the classification was mea-
sured using confusion matrix and kappa index which is considered as
the best quantitative measurement of classification accuracy (Foody,
2002).
3.1.2. Estimation of LST for 1987‒2017 data period
The LST was derived from the thermal bands of geometrically and
radiometrically corrected available Landsat satellite images of 1987,
2002 and 2017. Landsat sensors acquire the thermal data and store in
the form of Digital Numbers (DN). These DNs were converted to LST
using the four steps process (Artis and Carnahan, 1982) in this study for
the available data.
• In step 1, Digital Numbers (DNs) were converted to Radiance using
equation (1):
Lλ = LMIN + (LMAX − LMIN) x DN/255 (1)
where: Lλ = Spectral radiance; Following values of LMIN and LMAX
were used for three sensors:
LMIN = 1.238 and LMAX = 15.30 for Landsat 5; LMIN = 1.238 and
LMAX = 15.600 for Landsat 7; and LMIN = 0.10033 and
LMAX = 22.00180 for Landsat 8 (band 10 and 11 both).
• Step 2 involved the conversion of Radiance into brightness tem-
perature (TB) in Kelvin using equation (2):
TB = K2/In ((K1 / Lλ ) + 1) (2)
where K1 = Calibration Constant 1, and K2 = Calibration Constant 2.
Following values of K1 and K2 were used for three sensors:
K1 = 607.76 and K2 = 1260.56 for Landsat 5; K1 = 666.09 and
K2 = 1282.71 for Landsat 7; K1 = 774.89 and K2 = 1321.08 for
Landsat 8 (band 10); K1 = 480.88 and K2 = 1201.14 for Landsat 8
(band 11).
• Step 3 involved conversion of TB from Kelvin to Celsius using
equation (3):
TB (°C) = TB (in Kelvin) − 273.15 (3)
Sensor Cloud cover (%) Path/Row
Landsat 5 TM ∼6 150/36
Landsat 7 ETM+ ∼8 150/36
Landsat 8 OLI ∼13 150/36
S. Ullah, et al.
• Finally, the brightness temperature at sensor was converted to LST
in step 4 by using equation (4):
LST = TB/ [1 + (λ ∗ TB/ρ )∗ ln(ε )] (4)
where: λ = wavelength of emitted radiance, the peak response and the
average of the limiting wavelengths (λ = 11.5 μm); ρ = h∗c/σ
(1.438 × 102 mK).
H = Planck's constant (6.626∗10−34 J s)
σ = Boltzmann constant (1.380 649 ∗10−23 J/K)
c = velocity of light (2.998∗108 m/s)
ε = Surface emissivity
Surface emissivity was calculated from PV (Proportion of vegeta-
tion) which is derived from NDVI of the respective years of 1987, 2002
and 2017. The PV is derived by the following formulae proposed by
Sobrino et al. (2004):
PV = (NDVI − NDVImin / NDVImax − NDVImin)2 (5)
Finally the surface emissivity was calculated by using equation (6)
(Sobrino et al., 2004):
Surface emissivity, ε = 0.004 PV + 0.986 (6)
LST from Landsat 8 was derived from band 10 and 11 and then
averaged using cell statistics to calculate the average based on per cell.
3.1.3. Standardization of LST
Due to topographic and seasonal variations related to mountainous
areas in the thermal images of different Landsat data years, the stan-
dardization of LST was performed for comparability. It is not proper to
directly compare LST variability of the images from different years and
seasons. The purpose of standardization was to bring all the variables in
to proportion of one another. The cloud contaminated pixels were re-
moved through gross techniques and the LST was standardized (Salama
et al., 2012) using equation (7);
LSTs = LST − LSTu / LSTΩ (7)
where: LSTs = Standardized LST, LSTu = Mean of estimated LST from
1987 to 2017, LSTΩ = Standard deviation of LST from 1987 to 2017.
3.2. Classification of temperature zones
To calculate the proportional area lying in different temperature
ranges (Fig. 2), the temperature zones were categorized into six classes
i.e. < 15, 15 to < 21, 21 to < 24, 24 to < 27, 27 to 30 and ≥ 30 °C.
The purpose of categorization is to assess the trends of variations in
different temperature zones during study period. The lowest tempera-
ture zone was set to be lower than 15 °C and the highest temperature
zone as greater than or equal to 30 °C.
3.3. Simulating LULC projections for years 2032 and 2047
Cellular Automata (CA) model was applied to predict future LULC
changes using MOLUSCE tool in QGIS. Land use models such as CA
model cover both static and dynamic aspect of LULC changes. CA model
was used to predict future land cover pattern because of its reputed
accuracy (Santé et al., 2010). The prediction was based on two types of
variables; the dependent variables such as past pattern of LULC changes
estimated from 2002 to 2017 Landsat images and the independent
variables such as distance to roads, elevation and slope. The distance to
roads was calculated using roads vector data of the study area and
Euclidian distance function in ArcMap software. SRTM DEM was used
for elevation and slope estimation for the study area using ArcMap
software. Above mentioned variables were used as input to generate
transition potential matrix. The sampling mode was defined as a
random. To define maximum iteration and neighborhood pixel for the
351
Journal of Environmental Management 245 (2019) 348–357
model, the maximum iteration was 1000 and neighborhood pixel was
9 cells i.e. 3 × 3 cells.
Once transition potential matrix was modeled using logistic re-
gression method, then CA model was applied to simulate the future
LULC maps for the years 2032 and 2047 using MOLUSCE tool in QGIS
software. In order to ensure that the model is reliable in predicting
LULC change for a specific projected year, it must be validated using
existing datasets. Therefore, the CA model was validated for its accu-
racy to simulate LULC of year 2017 which was then compared to esti-
mated LULC (from Landsat data) of the same year for the study area.
The validate module in the IDRISI Taiga software was used for vali-
dating the model by producing several Kappa (K) parameters: K-no, K-
location, and K-standard that are used to identify the accuracy of the
model. Additionally QGIS-MULUSCE validation module was also used
to calculate overall kappa coefficients and %-correctness between
classified and predicted LULC map of 2017. Thus, the accuracy of CA
model was assessed by two approaches mentioned above before using it
for the future projections.
3.4. Simulating LST projection for years 2032 and 2047
The increasing LST in urban areas in current global warming sce-
nario is the cause of concern for urban planners (Dhar et al., 2019). A
Multi-layer Feed forward back Propagation Artificial Neural Network
method in MATLAB software (Maduako et al., 2016) was applied to
model, simulate and predict future LST pattern using past trends in the
study area. The Multi-Layer Perceptron (MLP) neural network makes
automatic decisions about the network parameters and how they should
be changed to better model the network. MLP algorithm is based on the
principle of error correction learning. In MLP, when the network re-
ceives a pattern, at first, it processes and produces a possibly low ac-
curate random output. It then does a self-computed error function by
calculating the difference between this random output and the expected
(target) output. Using the “Leveraging back-propagation” algorithm,
correction weights are computed between the output layer and the
hidden layers and also between the hidden layers and the input layers.
This back and forth iterative process goes on until an acceptable error
between the network output and the desired output is reached (Silva
and Clarke, 2002).
The simulation of LST in this study was based on the LST data trends
from 1987 to 2017. The study area was sub-divided into spatial grids of
500-m x 500-m to generate sample point size of spatial units using QGIS
software. This grid size was chosen considering the minimum space
range at which characteristics of one point can significantly affect a
change in LST (Lee and Jung, 2014). The sample data was used to
develop Neural Network in MATLAB for LST prediction. In addition, we
added longitude and latitude of the defined sample spatial units to in-
crease the model efficiency; the more the input parameters, the higher
the efficiency of the network model. The LST prediction involved the
network creation, network training, performance evaluation of the
network and prediction. The confidence of the network was assessed by
Mean Square Error (MSE) and correlation coefficient (R) values. The
regression analysis provides measures about how well the variations in
the output results were described by the target data set. If the value is 1
then there is a perfect correlation between output and target data set.
However, it is rare to get the value of 1. The Graphic User Interface
(GUI) was developed to check the performance indicator before
adopting the network. Once the performance indicators of the network
were satisfactory, then these were saved for prediction.
The values of 0.8 and 0.5 for R and MSE, respectively, were
achieved during assessment. Call back function was used to simulate
and predict LST for 2032 and 2047. The number of hidden layers was
selected on the basis of many trials with respect to MSE and R values.
The hidden layers are important because these affect results by enabling
the network to exhibit non-linear behavior. For the current study, three
hidden layers were selected. The initial learning rate (μ) during this
S. Ullah, et al.
Fig. 2. Variations in the pattern of temperature zones in district Abbottabad from 1987 to 2017.
study was set to 0.1 and the decay rate (β) was utilized to control it. The
standard decay rate varies from 0 to 1 (0 < β < 1). The decay rate of
0.9 was utilized to update the learning rate such as if the error function
between the present and the previous iteration is on the increase, the β
upgrades the learning rate μ by division while it multiplies it when the
error function is on the decrease in order to recede the μ. The de-
scription of the method is given in detail by Maduako et al. (2016).
4. Results and discussion
After applying the methodology detailed in section 3, we derived
the past pattern maps of LULC and LST distribution and simulated fu-
ture distribution maps of both the variables for the study area. Details
are presented in the following sections.
4.1. Past pattern of LULC class variations
Past patterns (1987‒2017) of variations in LULC classes derived
from Landsat data and SVM method are presented in Fig. 3. The overall
classification accuracy of SVM was 82.44%, 89.91% and 94.68% for the
year of 1987, 2002 and 2017, respectively (Table 2).
Two trends were clearly distinct (Fig. 3); first, built-up area in-
creased gradually and second, agriculture and bare soil area declined
over the study period. The results indicated the conversion of agri-
cultural and bare soil area into built-up and vegetation area which was
increased by 4.43% and 2.90%, respectively, in the past 30 years
(Table 3). Several factors contributed for this increase in built-up area
in the studied region since 2006 such as urban population growth and
population influx due to earth quake in 2005 and unrest situation in the
allied eastern and western areas. The results of this current study sup-
ported the findings of Weng (2001) who reported that both the eco-
nomical and geopolitical factors contributed to urban growth.
The total vegetation content in the study area showed increasing
trend especially at higher altitude (Figs. 1 and 3), which is attributed to
the recovery of forest area and the “Billion Tree Tsunami Afforestation”
Project, conducted by KPK Provincial government (IUCN, 2017). Raza
et al. (2012) reported that vegetation in district Abbottabad increased
due to massive plantation by the forest department and by pushing out
the people who were practicing agriculture activity at forest land. Ac-
cording to Ali et al. (2015), massive reforestation has been carried out
in the villages located in the vicinity of district Abbottabad. But there is
352
Journal of Environmental Management 245 (2019) 348–357
a significant decrease in vegetation inside and outskirts of the main city
area (Fig. 3). This is because of population influx to the city and sur-
roundings which converted the vegetation cover into built-up areas. In
addition, the plantation project mentioned by Ali et al. (2015) was
carried out in the forested area located at high altitude and away from
the urban area.
4.2. Past pattern of LST changes
Past patterns (1987–2017) of LST changes derived from Landsat
thermal bands using different equations (explained in section 3) are
presented in Fig. 4. The maximum temperature value was increased
from 32 °C (in 1987) to 39 °C (in 2017) during the study period (Fig. 4).
The area with LST > 27 °C was ∼0.03% in 1987, which increased up to
∼26% in 2017 (Figs. 2 and 4). The area with temperature range be-
tween 24 and 27 °C was increased from ∼5% (in 1987) to ∼29% (in
2017), while the area of temperature range between 21 and 24 °C de-
creased (Fig. 2) because most part of this area was converted to higher
temperature zone (i.e. 24‒27 °C). Reasons for these variations could be
the result of both climate change and urbanization. Rapid urbanization
leads to great deal of construction which enhances LST, which is further
amplified by climate change. It seems that area was affected by both
climate change and urban warming.
The annual mean temperature has augmented by approximately
0.5 °C along with an overall increasing trend in the precipitation during
previous 50 years in Pakistan while the number of annual heat wave
days has an increase of nearly five times greater than what was 30 years
before (ADB, 2017). Although heat release from the surface of the Hi-
malayan Plateau is the major source of rainfall as temperature differ-
ence is created between the air at high altitude of Himalayan Plateau
and the air at the nearby lower elevation ground surface but the effect
of this phenomena is minimum at the southern and lower elevation
portions (Boos and Kuang, 2010) of the Himalaya such as our study
area. Another factor which affects climate of Himalayan region is the
insulation of warm, moist air over continental India from the cold dry
extra tropics, the area between the subtropics and Polar Regions. A
recent theory emphases that local warming affecting air temperature of
the Himalayan region is caused by the LULC change (rather than by
climate change) thus altering temperature and humidity of the
boundary layer air (Boos and Kuang, 2010; Cao et al., 2019).
S. Ullah, et al. Journal of Environmental Management 245 (2019) 348–357

Fig. 3. LULC maps of the study area estimated from Landsat data for years 1987, 2002 and 2017.

Table 2 the LST was found to be increased for all LULC classes, even for vege-
Accuracy assessment of Support Vector Machine in terms of confusion matrix to tation during the study periods (Fig. 5). It was clear from Fig. 2 that
classify the LULC. temperature increased beyond the built-up area which was due to
Year User Accuracy Producer Overall Kappa change in the surface characteristics and global warming. Terando et al.
(%) Accuracy (%) Accuracy (%) coefficient (2014) also noted that temperature increased due to urban growth were
superimposed on rising global temperatures which led to increased LST
1987 83.57 80.43 82.44 0.76
even for vegetative area. A similar urban warming effect was reported
2002 89.11 88.80 89.91 0.86
2017 95.05 91.61 94.68 0.92
by Ren et al. (2008) both in rural and urban areas, in China. However,
the rate of temperature increase in our study area was comparatively
higher probably because of the rapid urban growth experienced in the
Table 3 present study area. Other reason of this kind of difference in tempera-
Distribution area of LULC classes from 1987 to 2017 (in hectares) and net ture increase rate may be the different season of the images; however,
change (in %). the results of this study strongly indicated the presence of urban
Class name 1987 2002 2017 Net change (%)
warming effect in the studied region. Most of the higher temperature
were observed in built-up area (Fig. 5), followed by the barren land
Built-up area 10100 11400 17500 +4.43 while comparatively lower temperatures were found on vegetation and
Agriculture 44000 42700 39400 −2.74 water surfaces. These outcomes supported the findings of Adegoke et al.
Bare Soil 43200 42200 35800 −4.42
(2003), who reported higher LST for the built-up area and bare soil
Vegetation 68200 68900 73000 +2.90
Water Body 1600 1500 1300 −0.18 while a lower LST for other categories such as vegetation and water
bodies.

4.3. Temperature variations in different land cover classes

4.4. Simulation of the future LULC
The built-up area had the highest mean temperature followed by
LULC classes significantly changed during the study period
barren land, agriculture and vegetation (Fig. 5). These findings were in
(1987‒2017) in the area (Table 3 & Fig. 3). It was therefore important
agreement with those reported by Weng (2001) who assess the LULC
to simulate the future LULC dynamics because if the past trends persist,
impact on the LST in the Zhujiang Delta, China, to analyse the re-
LST will be changed which may alter both biodiversity and micro-cli-
lationship between LULC classes and thermal signatures. The present
mate of the study area. The future LULC prediction was also important
findings indicated that built-up area increased LST by altering the ve-
as it provides framework for sustainable urban planning. For this
getation cover and replacing it with non-vegetated areas. Additionally,
reason, the validated cellular automata (CA) model was used to

353
S. Ullah, et al. Journal of Environmental Management 245 (2019) 348–357

Fig. 4. LST variation in the study area estimated from Landsat thermal bands for years 1987, 2002 and 2017.

simulate LULC for 2032 and 2047 for the study area. If a constant rate model validation results showed that overall kappa index and %-cor-
in LULC change is assumed then this model is dependable in forecasting rectness values were 0.77 and 84.16, respectively (Table 4) which is an
future change in the LULC. The model was first validated over the year acceptable accuracy level for LULC modeling.
2017 using the observed and simulated LULC in the study area. The Simulation results showed that approximately 12.27% and 14.13%

Fig. 5. Variations in mean LST over different land cover classes in the study area during study period.

354
S. Ullah, et al. Journal of Environmental Management 245 (2019) 348–357

Table 4
Validation of CA and ANN models by analyzing the observed and simulated LULC and LST maps, respectively, over year 2017.
Simulation year Validation of CA model for LULC simulation through two modules

IDRISI Taiga module (Kappa K parameters) QGIS-MULUSCE module

K-no K-location K-location Strata K-standard %-correctness Overall Kappa

2017 0.81 0.79 0.79 0.77 84.16 0.77

Validation of ANN model for LST simulation using MATLAB software
Simulation year No of hidden layer Mean Square Error (MSE) Correlation coefficient (R)
2017 3 0.51 0.81

Fig. 6. Simulated LULC maps of the study area for years 2032 and 2047.

of the area (as compared to base year 2017) will be converted into
urban or built-up area in 2032 and 2047, respectively (Fig. 6). Pre-
dicted LULC change in future could adversely affect the environment
altering both climate and biodiversity of the area.
4.5. Simulation of the future LST
Similar to the LULC classes, LST during the study period
(1987‒2017) indicated a significant change (Fig. 4) in the study area.
Therefore, the LST was also simulated for future period (2032 and
2047). The LST trends estimated from the past data were used in ANN
model to predict its future trends in the area. The simulation accuracy
was validated using correlation coefficient (R) and Mean square error
(MSE) between predicted and observed LST for year 2017 (Table 4).
The comparison of the predicted and satellite derived LST showed a
good agreement which proved the accuracy of ANN prediction. The R
and MSE were 0.81 and 0.51, respectively, which indicated strong
correlation between observed and predicted LST (Table 4). Fig. 7
showed the increasing trend in LST which may affect the air tempera-
ture due to their strong correlation with each other. The results pre-
dicted that approximately 42% and 60% of the study area will fall in the
high temperature zone (i.e. > 27 °C) in 2032 and 2047 (Fig. 7), re-
spectively. This area was 26%, 9.94% and 0.51% in 2017, 2002 and
1987, respectively. Findings in this study showed that temperature was
increasing even beyond the built-up areas for simulation periods. This
was due to the fact that the simulation was based on the past pattern
from 1987 to 2017 where the similar LST expansion beyond built-up
area was indicated. Another reason for expanding LST out of built-up
areas may be the change in surface characteristics, greenhouse effect
and global warming which will affect all areas even in the absence of
increased urban growth (Dereczynski et al., 2013; Argüeso et al., 2014).
This concurs with Terando et al. (2014) who indicated that temperature
rise due to urban growth is superimposing on rising global tempera-
tures. The incoming long wave radiation towards the lower atmosphere
increases with the increase in greenhouse gas concentration. However,
although low-medium density built-up areas and vegetated areas are
expected to warm, extent of their warming is smaller than high density
built-up areas. The increase in LST will affect the thermal capacity of
land cover which lead to Urban heat island (UHI) that emerges from the
modification of land surface which favors heat trapping and anthro-
pogenic heat release (Oke, 1982). It will elevate the urban area tem-
perature as compared to rural one (Stewart and Oke, 2012). This is one
of the major environmental issues harmful for both humans and bio-
diversity as well the environment (Grimmond, 2007). Fifth assessment
report of IPCC et al. (2014) indicated that warming in South Asian
region is likely to be more than the global average. Vegetation cover
around the buildings mitigates the impact of global warming (Amiri
et al., 2009; Smith and Roebber, 2011). Hence, urban plantation is a

355
S. Ullah, et al.
Fig. 7. Simulated LST maps of the study area for years 2032 and 2047.
significant mitigation measure against global warming, even when
greenhouse gas emissions and ozone depletion continue to affect the
lower atmosphere.
5. Conclusion
The study was designed to assess LULC and LST during the period
from 1987 to 2017 and to predict future scenario (2032 and 2047) in
the lower Himalayan region of Northern Pakistan. Following can be
concluded from the results:
• A significant change in the LULC classes was indicated such as the
built-up area was increased by 4.43% while agriculture area and
bare soil reduced by 2.74% and 4.42%, respectively, from 1987 to
2017.
• Analysis of LST for different LULC classes showed that built-up area
had the highest temperature followed by the barren land, agri-
culture and vegetation area. It was also noted that mean LST in-
creased for all land cover types, which indicated that the study area
is affected by both the climate change and the urban warming.
• Simulation of future LULC indicated that the built-up area will be
increased by 2.27% and 4.13% for the year 2032 and 2047, re-
spectively, as compared to the base year 2017.
• Simulation of future LST indicated that 42% and 60% of the area
will likely to experience a temperature of greater than 27 °C in 2032
and 2047, respectively, as compared to 26% area in the base year
2017.
Expansion of urban areas is pushing the additional land cover to-
wards the zones of highest temperature, and the entire region may be
converted to UHI if the current LST trends continue in the future. A
possible approach in order to decelerate the formation of large-scale
UHI effect in the future may be a compact-town like decentralization of
urban areas and urban plantation. The development of mountainous
region like Abbottabad should be sustainable and eco-friendly. Future
studies are recommended to examine the consequences and problems
356
Journal of Environmental Management 245 (2019) 348–357
being faced by the population of study area due to urban expansions
and LST increase.
Acknowledgment
Authors thank the United States Geological Survey for providing the
Landsat and SRTM DEM satellite data free of cost for this study. We are
grateful to the anonymous reviewers whose comments and suggestions
significantly improved the manuscript.
References
Adegoke, J.O., Pielke Sr, R.A., Eastman, J., Mahmood, R., Hubbard, K.G., 2003. Impact of
irrigation on midsummer surface fluxes and temperature under dry synoptic condi-
tions: a regional atmospheric model study of the US High Plains. Mon. Weather Rev.
131, 556–564.
Afzaal, M., Haroon, M.A., Zaman, Q., 2009. Interdecadal oscillations and the warming
trend in the area-weighted annual mean temperature of Pakistan. Pakistan J.
Meteorol. 6, 13–19.
Ahmed, B., 2011. Urban Land Cover Change Detection Analysis and Modeling Spatio-
Temporal Growth Dynamics Using Remote Sensing and GIS Techniques: A Case Study
of Dhaka. (Bangladesh).
Ahmed, B., Kamruzzaman, M., Zhu, X., Rahman, M.S., Choi, K., 2013. Simulating land
cover changes and their impacts on land surface temperature in Dhaka, Bangladesh.
Rem. Sens. 5, 5969–5998.
Ahmed, S., 2018. Assessment of urban heat islands and impact of climate change on
socioeconomic over Suez Governorate using remote sensing and GIS techniques.
Egypt. J. Remote Sens. Sp. Sci. 21, 15–25.
Akbar, T.A., Hassan, Q.K., Ishaq, S., Batool, M., Butt, H.J., Jabbar, H., 2019. Investigative
spatial distribution and modelling of existing and future urban land changes and its
impact on urbanization and economy. Rem. Sens. 11. https://doi.org/10.3390/
rs11020105.
Ali, S., Rahman, Z., Khan, S., Khan, T., Ali, W., Iqbal, A., 2015. Assessment of land use
pattern in district Abbottabad through geographic information system and remote
sensing. Biol. 63, 87–96.
Amiri, R., Weng, Q., Alimohammadi, A., Alavipanah, S.K., 2009. Spatial–temporal dy-
namics of land surface temperature in relation to fractional vegetation cover and land
use/cover in the Tabriz urban area, Iran. Remote Sens. Environ. 113, 2606–2617.
Argüeso, D., Evans, J.P., Fita, L., Bormann, K.J., 2014. Temperature response to future
urbanization and climate change. Clim. Dyn. 42, 2183–2199.
Artis, D.A., Carnahan, W.H., 1982. Survey of emissivity variability in thermography of
urban areas. Remote Sens. Environ. 12, 313–329.
Asian Development Bank, 2017. Climate Change Profile of Pakistan. https://doi.org/10.
S. Ullah, et al.
22617/TCS178761.
Balzter, H., 2000. Markov chain models for vegetation dynamics. Ecol. Model. 126,
139–154.
Boos, W.R., Kuang, Z., 2010. Dominant control of the South Asian monsoon by orographic
insulation versus plateau heating. Nature 463, 218.
Cao, W., Huang, L., Liu, L., Zhai, J., Wu, D., 2019. Overestimating impacts of urbanization
on regional temperatures in developing megacity: Beijing as an example. Adv.
Meteorol 2019.
Civco, D.L., 1993. Artificial neural networks for land-cover classification and mapping.
Int. J. Geogr. Inf. Sci. 7, 173–186.
Chen, X.L., Zhao, H.M., Li, P.X., Yin, Z.Y., 2006. Remote sensing image-based analysis of
the relationship between urban heat island and land use cover changes. Remote Sens.
Environ. 104, 133–146.
Deng, C., Wu, C., 2013. Examining the impacts of urban biophysical compositions on
surface urban heat island: a spectral unmixing and thermal mixing approach. Remote
Sens. Environ. 131, 262–274.
Dereczynski, C., Silva, W.L., Marengo, J., 2013. Detection and projections of climate
change in Rio de Janeiro, Brazil. Am. J. Clim. Change 2, 25–33.
Dhamge, N.R., Atmapoojya, S.L., Kadu, M.S., 2012. Genetic algorithm driven ANN model
for runoff estimation. Procedia Technol 6, 501–508.
Dhar, R.B., Chakraborty, S., Chattopadhyay, R., Sikdar, P.K., 2019. Impact of land-use/
land-cover change on land surface temperature using satellite data: a case study of
rajarhat block, north 24-parganas district, West Bengal. J. Indian Soc. Remote Sens.
47, 331–348.
El-Hattab, M., S.M., A.,G.E.,L., 2018. Monitoring and assessment of urban heat islands
over the Southern region of Cairo Governorate, Egypt. Egypt. J. Remote Sens. Sp. Sci.
21, 311–323.
Foody, G.M., 2002. Status of land cover classification accuracy assessment. Remote Sens.
Environ. 80, 185–201.
Grimmond, S.U.E., 2007. Urbanization and global environmental change: local effects of
urban warming. Geogr. J. 173, 83–88.
Guidigan, M.L.G., Sanou, C.L., Ragatoa, D.S., Fafa, C.O., Mishra, V.N., 2018. Assessing
land use/land cover dynamic and its impact in Benin republic using land change
model and CCI-LC products. Earth Syst. Environ. 1–11.
Hadi, S.J., Shafri, H.Z.M., Mahir, M.D., 2014. Modelling LULC for the period 2010-2030
using GIS and Remote sensing: a case study of Tikrit, Iraq. In: IOP Conference Series:
Earth and Environmental Science. IOP Publishing, pp. 12053.
Hasanlou, M., Mostofi, N., 2015. Investigating urban heat island estimation and relation
between various land cover indices in Tehran city using Landsat 8 imagery. In:
Proceedings of the 1st International Electronic Conference on Remote Sensing, pp.
1–11 Basel, Switzerland.
Haykin, S., 1999. Self-organizing maps. In: Neural Networks. A Compr. Found, second ed.
Prentice-Hall.
Herold, M., Goldstein, N.C., Clarke, K.C., 2003. The spatiotemporal form of urban growth:
measurement, analysis and modeling. Remote Sens. Environ. 86, 286–302.
Hu, Y., Jia, G., 2010. Influence of land use change on urban heat island derived from
multi‐sensor data. Int. J. Climatol. 30, 1382–1395.
IPCC, 2014. In: Pachauri, R.K., Meyer, L.A. (Eds.), Climate Change 2014: Synthesis
Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of
the Intergovernmental Panel on Climate Change [Core Writing Team, (Geneva,
Switzerland).
IUCN, 2017. Pakistan's Billion Tree Tsunami Restores 350,000 Hectares of Forests and
Degraded Land to Surpass Bonn Challenge Commitment.
Keshtkar, H., Voigt, W., 2016. A spatiotemporal analysis of landscape change using an
integrated Markov chain and cellular automata models. Model. Earth Syst. Environ.
2, 10.
Lee, W.-S., Jung, S.-G., 2014. The application of a prediction model on land surface
temperature using Artificial Neural Network and Scenario. J. Korea Plann. Assoc. 49,
263. https://doi.org/10.17208/jkpa.2014.02.49.1.263.
Li, H., Liu, Q., 2008. Comparison of NDBI and NDVI as indicators of surface urban heat
island effect in MODIS imagery. In: Proc.SPIE.
Long, H., Wu, X., Wang, W., Dong, G., 2008. Analysis of urban-rural land-use change
during 1995-2006 and its policy dimensional driving forces in Chongqing, China.
Sensors 8, 681–699.
Maduako, I.D., Yun, Z., Patrick, B., 2016. Simulation and prediction of land surface
temperature (LST) dynamics within Ikom City in Nigeria using artificial neural net-
work (ANN). Int. J. Remote Sens. 5, 1–7.
Matthews, T., Lo, A.Y., Byrne, J.A., 2015. Reconceptualizing green infrastructure for
climate change adaptation: barriers to adoption and drivers for uptake by spatial
planners. Landsc. Urban Plan. 138, 155–163.
Ménard, A., Marceau, D.J., 2005. Exploration of spatial scale sensitivity in geographic
357
Journal of Environmental Management 245 (2019) 348–357
cellular automata. Environ. Plan B Urban Anal City Sci. 32, 693–714.
Mushore, T.D., Odindi, J., Dube, T., Mutanga, O., 2017. Prediction of future urban surface
temperatures using medium resolution satellite data in Harare metropolitan city,
Zimbabwe. Build. Environ. 122, 397–410.
Oke, T.R., 1982. The energetic basis of the urban heat island. Q. J. R. Meteorol. Soc. 108,
1–24.
Pakistan bureau of statistics, 2017. District Wise Census Results. [WWW Document]. URL
1.20.19. http://www.pbscensus.gov.pk/sites/default/files/bwpsr/kp/
ABBOTTABAD_SUMMARY.pdf.
Parsa, V.A., Salehi, E., 2016. Spatio-temporal analysis and simulation pattern of land use/
cover changes, case study: Naghadeh, Iran. J. Urban Manag. 5, 43–51.
Rahman, M.T., Aldosary, A.S., Mortoja, M.G., 2017. Modeling future land cover changes
and their effects on the land surface temperatures in the Saudi Arabian eastern
coastal city of Dammam. Land 6, 36.
Raza, A., Raja, I.A., Raza, S., 2012. Land-use change analysis of district Abbottabad,
Pakistan: taking advantage of GIS and remote sensing analysis. Sci. Vis. 18, 43–49.
Ren, G., Zhou, Y., Chu, Z., Zhou, J., Zhang, A., Guo, J., Liu, X., 2008. Urbanization effects
on observed surface air temperature trends in North China. J. Clim. 21, 1333–1348.
Rimal, B., Zhang, L., Keshtkar, H., Wang, N., Lin, Y., 2017. Monitoring and modeling of
spatiotemporal urban expansion and land-use/land-cover change using integrated
Markov chain cellular automata model. ISPRS Int. J. Geo-Inf. 6, 288.
Salama, M.S., Van der Velde, R., Zhong, L., Ma, Y., Ofwono, M., Su, Z., 2012. Decadal
variations of land surface temperature anomalies observed over the Tibetan Plateau
by the Special Sensor Microwave Imager (SSM/I) from 1987 to 2008. Clim. Change
114, 769–781.
Santé, I., García, A.M., Miranda, D., Crecente, R., 2010. Cellular automata models for the
simulation of real-world urban processes: a review and analysis. Landsc. Urban Plan.
96, 108–122.
Selçuk, R., Nisanci, R., Uzun, B., Yalcin, A., Inan, H., Yomralioglu, T., 2003. Monitoring
land-use changes by GIS and remote sensing techniques: case study of Trabzon. In:
Proceedings of 2nd FIG Regional Conference, pp. 2–5 Marrakech, Morocco.
Silva, E.A., Clarke, K.C., 2002. Calibration of the SLEUTH urban growth model for Lisbon
and Porto. Portugal. Comput. Environ. Urban Syst. 26, 525–552. https://doi.org/10.
1016/S0198-9715 (01)00014-X.
Smith, K.R., Roebber, P.J., 2011. Green roof mitigation potential for a proxy future cli-
mate scenario in Chicago, Illinois. J. Appl. Meteorol. Climatol. 50, 507–522. https://
doi.org/10.1175/2010JAMC2337.1.
Sobrino, J.A., Jiménez-Muñoz, J.C., Paolini, L., 2004. Land surface temperature retrieval
from Landsat TM 5. Int. J. Appl. Sci. Technol. 90, 434–440.
Stewart, I.D., Oke, T.R., 2012. Local climate zones for urban temperature studies. Bull.
Am. Meteorol. Soc. 93, 1879–1900.
Terando, A.J., Costanza, J., Belyea, C., Dunn, R.R., McKerrow, A., Collazo, J.A., 2014. The
southern megalopolis: using the past to predict the future of urban sprawl in the
Southeast US. PLoS One 9, e102261.
United Nations, Department of Economic, Social Affairs, P.D., 2018. World Urbanization
Prospects: the 2018 Revision.
Vapnik, V.N., 1999. An overview of statistical learning theory. IEEE Trans. Neural Netw.
10, 988–999.
Vapnik, V.N., Chervonenkis, A.Y., 1968. The uniform convergence of frequencies of the
appearance of events to their probabilities. In: Doklady Akademii Nauk. Russian
Academy of Sciences, pp. 781–783.
Verburg, P.H., Van De Steeg, J., Veldkamp, A., Willemen, L., 2009. From land cover
change to land function dynamics: a major challenge to improve land characteriza-
tion. J. Environ. Manag. 90, 1327–1335.
Weng, Q., 2001. A remote sensing? GIS evaluation of urban expansion and its impact on
surface temperature in the Zhujiang Delta, China. Int. J. Remote Sens. 22,
1999–2014.
Xiao, J., Shen, Y., Ge, J., Tateishi, R., Tang, C., Liang, Y., Huang, Z., 2006. Evaluating
urban expansion and land use change in Shijiazhuang, China, by using GIS and re-
mote sensing. Landsc. Urban Plan. 75, 69–80.
Yang, C., He, X., Yan, F., Yu, L., Bu, K., Yang, J., Chang, L., Zhang, S., 2017. Mapping the
influence of land use/land cover changes on the urban heat island effect—a case
study of Changchun, China. Sustainability 9, 312.
Yuen, B., Kong, L., 2009. Climate change and urban planning in Southeast Asia. SAPIENS
2, 2–3.
Zenil, H., 2010. Compression-based investigation of the dynamical properties of cellular
automata and other systems. J. Complex Syst. 19, 1–28.
Zhang, Y., Balzter, H., Liu, B., Chen, Y., 2017. Analyzing the impacts of urbanization and
seasonal variation on land surface temperature based on subpixel fractional covers
using Landsat images. Appl. Earth Obs. Remote Sens. 10, 1344–1356.