The Egyptian Journal of Remote Sensing and Space Sciences 26 (2023) 341–350

Contents lists available at ScienceDirect

The Egyptian Journal of Remote Sensing and Space Sciences

journal homepage: www.sciencedirect.com

Research Paper

Development of a map for land use and land cover classification of the
Northern Border Region using remote sensing and GIS
Abdulbasit A. Darem a,⇑, Asma A. Alhashmi a, Aloyoun M. Almadani a, Ali K. Alanazi a, Geraldine A. Sutantra b
a
Department of Computer Science, Northern Border University, Arar 9280, Saudi Arabia
b
Agricultural Engineering College, Padjadjaran University, Sumedang 45363, Indonesia

a r t i c l e i n f o a b s t r a c t

Article history: The land use and land cover study (LULC) play an essential role in regional socio-economic development
Received 8 February 2023 and natural resource management to develop sustainable development in vegetation changes, water
Revised 16 April 2023 quantity and quality, land resources, and coastal management. This study uses remote sensing data to
Accepted 30 April 2023
investigate LULC in the Northern Border Region (NBR) in the Kingdom of Saudi Arabia. The purpose of this
Available online 12 May 2023
study is to obtain a better understanding of the patterns and drivers of changes in LULC in the NBR over
the past three decades. Remote sensing data from Landsat imagery between 1990 and 2022 were used to
Keywords:
classify LULC types, and a time series analysis was performed using Landsat imagery to detect changes
GIS
LULC
over time. The classification finds four main classes: bare land, built-up area, rocks, and vegetation.
Remote Sensing The results indicate a significant increase in urban development. The outcomes revealed that most urban-
Kingdom of Saudi Arabia ization occurred in the outskirts of the cities, where previously there were bare soil lands. The main dri-
Northern Border Region vers of urbanization were population growth and economic development. These findings have important
implications for city planning, the management of green spaces, and the sustainable development of
cities. Maximum Likelihood classifier was used to perform the classification. The accuracy assessment
demonstrated satisfactory results, with an overall accuracy of 92.6%. The study paves the way for further
monitoring LULC changes in the NBR geographic location. The technique used was adequate to address
the objectives of this study.
Ó 2023 National Authority of Remote Sensing & Space Science. Published by Elsevier B.V. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction veys and other methods such as satellite imagery. The data can
then be compiled and organized into a decision-making hierarchy.
LULC is a source of information for countries to make appropri- The main objective of the classification system is to accurately
ate decisions based on remote sensing (RS) data. This knowledge identify the type of land cover present in a given area. The most
and detection of land change are considered essential factors for widely used method for LULC classification is traditional visual
planning land conservation, management, and development. interpretation. It involves passing a trained expert over the land
Understanding the dynamics of environmental change to ensure area in question and requesting detailed reports on the present
sustainable development and formulate policies for the optimal land cover. The specialist is asked to provide a subjective assess-
future use of these lands. LULC classification is an important tech- ment of the land, such as its vegetation cover or water bodies. Such
nique to assess the relationship between the environment and a visual interpretation helps provide a detailed understanding of a
human activities (Singh et al., 2021). It studies how different land given area but can be unreliable and unsuitable for large-scale
uses are distributed in a particular area. This type of analysis is LULC analyses. An alternative to visual interpretation is object-
widely used in a variety of applications, including conservation based image analysis (OBIA). This method involves segmenting
planning, urban planning, and disaster management. LULC has an image into distinct objects based on their physical characteris-
become increasingly important as the human population grows tics, such as shape and texture, and assigning a LULC class to each
and the planet’s resources are stretched to their limits. LULC clas- object. This approach is helpful for large-scale LULC classification,
sification involves identifying and categorizing different types of as it can process large amounts of image data quickly and accu-
land cover within an area, collecting data through field or RS sur- rately. OBIA is more reliable and requires less user input than tra-
ditional visual interpretation. The use of machine learning and
⇑ Corresponding author. artificial intelligence is gaining attention as a new method for LULC
E-mail address: [email protected] (A.A. Darem). classification. This approach involves training a computer model to

https://doi.org/10.1016/j.ejrs.2023.04.005
1110-9823/Ó 2023 National Authority of Remote Sensing & Space Science. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

identify LULC classes from satellite images by learning from istics. Accurate LULC classification is essential for various land
labelled training data. It can be done using various models, such management, natural resource management, and urban planning
as convolutional neural networks (Balarabe and Jordanov, 2021), applications. Due to the rapid advancement of RS technology,
deep belief networks (Lv et al., 2015; Vignesh and Thyagharajan, high-resolution images are now available with a wide range of
2019), support vector machines (Hamad, 2020; Kadavi and Lee, spectral and spatial resolutions, allowing for accurate and timely
2018; Rezaei Moghaddam et al., 2015; Aslami et al., 2015), and land cover classification. However, due to the complexity of land
others. Machine learning has the advantage of quickly processing features, the classification of LULC remains challenging. A key fac-
massive amounts of data and providing more accurate results than tor affecting the accuracy of LULC classification is the selection of
traditional LULC classification methods. features for classification. There are many classification methods.
This research study is significant because the growing popula- The following sections briefly describe the classification and fea-
tion and the increasing socio-economic had created pressure on ture selection methods.
LULC, which led to unplanned and uncontrolled changes. The The three main methods for classifying LULC objects using RS
Northern Border Region is a critical ecological and economic zone and GIS are satellite imagery, spectral analysis, and digital terrain
that encompasses a variety of land uses and covers, including analysis. First, satellite imagery involves using satellites to acquire
urban areas, agricultural lands that has not been used, and natural images of land surfaces from different angles. These images are
ecosystems. However, the region is undergoing rapid and extensive then processed and analyzed to identify and classify LULC objects.
land-use changes, resulting in significant impacts on local commu- The data collected can include details such as elevation, vegetation,
nities, natural resources, and the environment. Therefore, it is and soil type. Second, spectral analysis is used to identify and clas-
essential to develop accurate and efficient tools to monitor and sify LULC objects using electromagnetic radiation (EMR) emitted
manage land use and land cover changes in the region. Remote from these objects. This technique uses different wavelengths of
sensing and GIS technologies offer a powerful means to quantify EMR, such as infrared, visible light, and thermal imaging, to detect
and analyze LULC changes, but their application requires a specific and characterize different LULC objects. Third, digital terrain anal-
method tailored to the unique characteristics of the study site. Our ysis involves using computer algorithms and GIS software to ana-
study aimed to develop a map for land use and land cover classifi- lyze and classify LULC objects. This technique is used to create
cation of the Northern Border Region using remote sensing and detailed contour maps and 3D models of terrain in order to classify
GIS, which can provide a reliable and efficient tool to monitor and identify LULC objects accurately.
and manage land use and land cover changes in the region. The
study’s results can contribute to the development of effective land
management policies and practices, promote sustainable land use 3. Literature review
and resource conservation, and provide critical information for
urban planning, agricultural management, and biodiversity conser- Various studies have been conducted to classify and update
vation. The KSA plays a crucial role in global efforts to preserve the LULC information in different regions using RS and GIS technology
future of the Earth, as it launched the Green Saudi Initiative to with varying levels of success (Sang et al., 2021; Bhattacharya
renew its belief in a sustainable future for all. From this point of et al., 2021; Dibs et al., 2020; Kaya and Görgün, 2020; Saddique
view, there was a need to study the environment and topography et al., 2020; Ghayour et al., 2021). Abdallah et al. (2019) used three
of the Kingdom from several points of view to delineate the maps cloud-free Landsat MSS, ETM +, and OLI images to classify LULC
of LULC. Information and data are needed to help decision-makers classes and land cover changes in Jizan Dam, Saudi Arabia. The
detect the change in the LULC of the NBR and understand the study successfully identified five classes (Vegetated land, Urban
dynamics of environmental change to ensure sustainable develop- area, Bare area (sands), Bare area (rocks), and water bodies) with
ment and formulate policies for the optimal future use of these an overall accuracy of 86.67%. As part of their classification of Mak-
lands. There are no previous studies covering the LULC of the kah and Al-Taif in Saudi Arabian Desert Cities, Alqurashi and
Northern Border region. The generated and processed maps that Kumar (2014) incorporated maximum likelihood and object-
will result from this study will be significant for the future plan- orientated classification methods. Among the five identified classes
ning of the NBR. In this study, RS techniques will be used to map vegetation, urban landscape, barren land, sand, and rocky terrain.
the land use and cover of NBR, one of the most critical regions of There was an average accuracy of 88.9 percent for maximum like-
the Kingdom in terms of environmental diversity, in addition to lihood classification for 1986, 1990, 2000, and 2013, while 92.5
the presence of many essential minerals. Decision-makers can percent for object-oriented classification for those years. According
use the results of this study to manage natural resources in the to the findings, urban areas grew by around 174% in Makkah and
region, help in socio-economic development, plan the appropriate 113% in Al-Taif. As a result of changing average precipitation in this
use of land (agriculture, construction, and urban expansion), and environment, the distribution of vegetation cover over the study
know the size and type of vegetation in the northern border area. area varies from year to year (Alqurashi and Kumar, 2014).
The maps from this study will help find suitable places for mining Mahmoud and Alazba (2016) estimated the spatial and temporal
essential minerals in the northern border region. distribution of actual evapotranspiration (AET) in KSA’s western
The rest of this paper is structured as follows: Section 2 provides and southern regions from 1992 to 2014 using the SEBAL model
background information on the topic, followed by a review of rele- and field observations. Salih (2018) applied different image pre-
vant literature in Section 3. Section 4 describes the research design, processing techniques and a well-known and widely used classifi-
including the deadest materials and procedures. Section 5 presents cation method (i.e., Maximum Likelihood classifier). Also,
the study findings, including the statistical analyses and data classes. Khwarahm (2021) Maximum Likelihood classifier to model and
Section 6 presents the Challenges and Limitations, followed by the analyze LULCC trends in the Sulaimani region in Iraq. Khwarahm
conclusion to summarize the main findings and their implications. et al.(2021) presents a study on predicting and mapping land
cover/land use changes in Erbil, Iraq using a CA-Markov synergy
model. Landsat-7 data was used in this study to classify and map
2. Background land cover types and attributes in Al-Ahsaa Oasis, Eastern Region,
Saudi Arabia. The Landsat images of AlKhobar, Saudi Arabia, have
LULC classification is a fundamental tool for understanding the been used by Rahman (2016) to quantify land use/cover changes
spatial distribution of land features and their associated character- and urban sprawl over 23 years. It uses ISODATA classification
342
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

methods to classify Landsat TM, ETM+, and OLI data. The rate of 4.1. Study area
urban development increased from 17% between 1990 and 2001
to 43.51% between 2001 and 2013. Furthermore, vegetation The study area is in Saudi Arabia’s Northern Border Region
increased by 110 percent between 1990 and 2001 and 52 percent (NBR), along the borders with Iraq and Jordan. It consists of several
between 2001 and 2013. Several studies also used RS and GIS for towns, villages, land, and semi-arid areas. The elevation of the area
LULC. Alawamy et al. (2020) used time-series Landsat data from is 536 m above sea level. The GPS coordinates of NBR are 30.975
1985 to 2017 to detect and analyze LULC changes in Al-Jabal Al- and 41.038, and its longitude is 41.016666. 30° 590 0.002400 N
Akhdar, Libya. Wijitkosum (2016) illustrated the impact of land and 41° 00 59.997600 E (https://en.wikipedia.org/wiki/Arar,_Saudi_
use and spatial changes on the risk of desertification in degraded Arabia). Its population is 395,000 (the year 2014 census). It is rec-
areas in Thailand. Research studies have shown that using RS and ognized for its fertile grazing plains. The Saudi Arabian Highway 85
GIS technologies provides efficient and accurate ways to detect passes through NBR, a significant supply stop connecting Iraq and
and classify LULC. Furthermore, RS data-driven classification tech- Saudi Arabia. Fig. 2 shows the location of NBR on the map.
nology is increasingly used for LULC mapping to obtain accurate
and timely data.
4.2. Dataset collection

4. Materials and methods The dataset consists of RS imageries of NBR’s maps from 1990 to
2022. It was extracted using Landsat 5 and Landsat 8 from various
The workflow diagram of this study can be illustrated in Fig. 1. satellite images repository like Google Earth Pro, the United States

Fig. 1. Workflow Diagram.

Fig. 2. Study Area NBP in Saudi Arabia.

343
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Table 1
Sensor and Date/time of the scene acquisition from Landsat 5 Thematic Mapper (TM) and Landsat 8 (OLI).

Path Row Landsat 5 (TM) Landsat 8 (OLI)

168 39 12/18/1990 11/14/1995 12/29/2000 11/25/2005 12/9/2010 12/7/2015 12/20/2020 12/18/2022
168 40 12/18/1990 11/14/1995 12/29/2000 11/25/2005 11/23/2010 12/7/2015 12/20/2020 12/18/2022
168 41 12/18/1990 11/14/1995 12/29/2000 11/25/2005 11/23/2010 12/7/2015 11/18/2020 12/18/2022
169 39 12/9/1990 12/7/1995 11/2/2000 6/25/2005 11/30/2010 11/28/2015 12/11/2020 12/25/2022
169 40 12/9/1990 12/7/1995 12/20/2000 10/15/2005 11/30/2010 11/28/2015 12/11/2020 12/25/2022

Geological Survey (USGS)1 and (diva)2 website. These images were cies in the image’s brightness values (Kuemmerle et al., 2013).
clear of cloud cover. Table 1. Shows the Sensor and Date/time of Geometric and radiometric corrections are made in this step. The
the scene acquisition from Landsat 5 Thematic Mapper (TM) and accuracy of the product is improved by radiometric and geometric
Landsat 8 (OLI). adjustments, which are both highly helpful in removing impacts
brought on by varying correction orders. To prevent radiometric
4.3. Pre-processing errors or distortions, radiometric calibration and atmospheric cor-
rection recalculation of the pixel’s radiance is carried out. Geomet-
Data pre-processing is an essential part of LULC classification, as ric correction involves employing registration to eliminate
it affects the accuracy of the resulting classification outputs. Data geometric distortion, corrections of local incident angle with
pre-processing is a method used prior to LULC classification to toposheet, and knowledge of the geodetic coordinates of some
optimize the accuracy of the classification results. It helps the data related ground sites (Kuemmerle et al., 2013; Alshari and Gawali,
to be improved to provide better results. Data pre-processing 2021; Verburg et al., 2011; Shi et al., 2020; Alshari and Gawali,
encompasses various stages, such as radiometric corrections, data 2021). v) Image enhancement techniques used for noise removal
enhancement, and spatial filtering, which are essential for obtain- and enhanced features in the images retrieved from Landsat 5
ing more accurate classifications. Radiometric corrections are per- and Landsat 8, making them easier to identify. These operations
formed to compensate for sensor gain and offset effects, which can seek to improve satellite images for improved classification and
adversely affect the LULC classification results. The primary radio- repair the degraded image to depict the original scene (Alqadhi
metric corrections used in LULC classification are signal-to-noise et al., 2021) accurately. Because these procedures are carried out
ratio adjustment and histogram equalization. The signal-to-noise before the data is used for a specific purpose, correcting shortcom-
ratio adjustment minimizes the variation so that the output image ings and eliminating faults from the data are referred to as pre-
has a more even signal distribution. Histogram equalization com- processing (Ramdani et al., 2021). Due to atmospheric effects and
pensates for the non-uniform luminance distribution in regions the sensing system’s limits, the likely range of pixel values may
with varying illumination intensity by redistributing the pixel val- not be fully utilized while acquiring data while creating RS ima-
ues across the entire dataset. gery. As a result, acquired data may be of low quality, such as hav-
It is crucial to pre-process the RS data to correct the radiometric ing little contrast, a dark overlay, or much radiometric noise (Linda
and geometric inaccuracy that occurs during acquisition, scanning, Theres and Selvakumar, 2022). Therefore, it refers to the functions
and transmission. The pre-processing steps include clipping maps, performed regularly to enhance the geometric and radiometric
image registration, geo-referencing, radiometric correction, image quality of the photographs. vi) Composite Band Images are created
enhancement, campsite bands, and layout maps. These steps by combining multiple bands of information from environmental
improve image quality and remove noise sources, providing a more spectrum sensors. Each information band contains data from sev-
reliable dataset. The steps are described as follows: i) Clipping eral spectral regions within the larger electromagnetic spectrum,
maps are used to cut out specific regions of interest from the orig- such as near-infrared, visible, and shortwave infrared. Each band
inal images. It is done by cutting out unneeded background infor- collects information about different aspects of the environment,
mation, allowing images to be analyzed more precisely. ii) Image such as reflectance and vegetation indices. When multiple infor-
registration is a process of aligning images of different areas mation bands are combined in a composite band image, a greater
together, allowing similar features to be compared. It can also understanding of the characteristics of the environment can be
match the coordinate systems of two images taken at the exact appreciated. These composite images are then used to identify
location (Ramdani et al., 2021). The image registration is done and map LULC features. vii) Layout Maps are used to map the
through SOMINFO, which must be installed in the plugin manage- boundaries of the land cover classes, providing an outline of the
ment from the plugin’s menu inside QGIS. iii) Geo-referencing is a land cover types within the image. The layout process maps the
process to assign locations to images, allowing a computer to iden- information obtained from composite band images into limited
tify the location of various land cover objects. In contrast, geomet- patterns corresponding to LULC features. For example, areas of
ric correction entails moving a pixel’s location using coordinates high near-infrared reflectance may be classified as evergreen for-
without altering its luminous reflectance and recalculating its ests, while areas of low visible and near-infrared reflectance may
radian value using resampling. Then, geometric rectification is be classified as agricultural land. Layout maps also allow for
done using ArcMap with the projection WGS 1984 UTM 37 N by
placing the pixels in their accurate planimetric (x, y) map coordi-
nates. iv) The radiometric correction is then used to normalize Table 2
the look of the image, making it easier to identify land cover Classes’ Description.
classes. To increase the accuracy of the brightness value magni- No. Class Description
tudes, radiometric correction of RS photos mainly entails the pro-
1. Bare land Bare soil and land do not contain the population of people
cessing of digital images. The primary objective of radiometric like the desert.
correction is to reduce the impact of inaccuracies or inconsisten- 2. Built-up The built area can be a large or small building.
Area
1
3. Rocks Mountains or bare rocks
https://www.usgs.gov/. 4. Vegetation Space containing sparse grasslands.
2
https://www.diva-gis.org/.

344
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Table 3
LULC Changes in NBR between 1990 and 2022.

Classes Years/Area (km2)

1990 1995 2000 2005 2010 2015 2020 2022
Bare land 114,053 113,272 113,174 113,080 113,325 112,660 112,603 112,977
Built-up Area 21 24 31 57 77 105 122 133
Rocks 3417 4193 4277 4346 4084 4714 4751 4359
Vegetation 1 3 10 9 11 13 16 23
Total 117,492 117,492 117,492 117,492 117,492 117,492 117,492 117,492

because the increment of human activity, such as construction and

urban development, will continue to drive growth. The built-up
area in NBR is shown in Fig. 3. Due to the changing average precip-
itation in this environment, the distribution of vegetation cover
over the study area varied slightly from year to year. The vegeta-
tion is limited to sparse grassland and public parks. There are no
patterns to indicate any crops. There is no constant increase or
decrease in the changes to the Rocks and Bare Land classes. It could
have happened because the satellite captured rocks covered by
sand or vice versa. Table 4 shows the LULC change trends of the
study area in 1990 compared to 2022.
Our work on developing a map for land use and land cover clas-
sification of the Northern Border Region using remote sensing and
GIS is an important contribution to the field of environmental
monitoring and natural resource management. Our approach uti-
Main Cities lizes advanced remote sensing and GIS techniques to achieve a
Towns and villages higher accuracy of land use and land cover classification, which
is essential for sustainable land management, natural resource
conservation, and environmental monitoring.
Fig. 3. The built-up area in NBR.
Our approach has several potential applications in different
fields, such as urban planning, agricultural management, and bio-
images to be compared, helping to identify detailed changes in the diversity conservation. In urban planning, accurate land use and
LULC features over time. Figures 13 and 14 show the layout map land cover classification maps can be used to identify areas suitable
for 2012 and 2022, respectively. for development, estimate population density, and monitor
changes in the built environment. In agricultural management,
such maps can be used to identify land use patterns and estimate
5. Results & discussion
crop yields, which can inform decisions on land use planning and
resource allocation. In biodiversity conservation, our approach
To classify the re-travel segmented images. We used maximum
can be used to identify areas with high ecological value and poten-
likelihood classification (MLC) to classify images based on specific
tial threats to biodiversity, which can inform conservation
features. Eq. (1) is used as a discriminant function for each pixel in
strategies.
the image. The classified images are used to analyze and predict
This work also contributes to the ongoing efforts to monitor and
changes over time.
understand the land use and land cover changes in the Northern
X X
1 Border Region. Such information is essential for policymakers, land
g i ðxÞ ¼ 1n pðwi Þ  1=2 1n j j  1=2ðx  mi ÞT ðx  mi Þ ð1Þ managers, and researchers to develop effective strategies to pro-
i i
mote sustainable land use practices and natural resource
Where: x = n-dimensional data (where n is the number of bands); conservation.
i = class; p(xi) = probability that class xi occurs in the image and is
assumed the same for all classes; |Ri| = determinant of the covariance
matrix of the data in class xi; R-1 i = its inverse matrix; mi = mean 6. Accuracy assessment
vector.
This study recognized four classes: bare land, Built-up area, The confusion matrix of ground truth data and the Kappa coef-
rocks, and vegetation. The classes’ details are described in Table 2, ficient were used to assess accuracy. Ground-truth data were gath-
Table 3, and Fig. 4. ered from fieldwork carried out in NBR. A confusion matrix is a
All LULC classes in the northern border region from year to year popular tool for evaluating LULC classification techniques’ accuracy
have changed. The built-up area increased from 1990 with 21 km2 (Rizvon and Jayakumar, 2022; Tan et al., 2021). It can be used alone
wide to 2022 with 132 km2 wide area. Vegetation also increased or with other accuracy assessment techniques to assess the quality
from 1990 at 1 km2 wide to 2022 at 23 km2 wide. Changes that of the classification. A confusion matrix is a powerful tool for
happened to Rocks and Bare Land are not increased or decreased assessing accuracy in LULC classification (Alshari et al., 2022). It
constantly. The possibility that made it happen is because the is used to express the performance of the classification and calcu-
satellite captured rocks covered by sand or vice versa. late false positives, false negatives, true positives, and true
The built-up area class experienced the most significant change negatives.
in NBR. The results showed that urban areas increased by approx- The accuracy assessment with the confusion matrix often offers
imately 628% between 1990 and 2022. It is a predictable situation more profound insights into the results of the classification pro-
345
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Fig. 4. LULC Changes in Northern Border Regine from 1990 to 2022.

346
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Table 4
LULC change trends of the study area.

Classes 1990 2022 LULC change (1990–2022) %

2
Area (km2) % Area (km ) %
Bare Land 114,053 97.07 112,977 96.16 0.92
Builtup Area 21 0.02 133 0.11 0.10
Rocks 3417 2.91 4359 3.71 0.80
Vegetation 1 0.00 23 0.02 0.02
Total 117,492 100 117,492 100

Table 5
Confusion matrix in 1990.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 18 0 2 0 20 90
2 Built-up Area 0 17 3 0 20 85
3 Rocks 0 0 20 0 20 100
4 Vegetation 0 0 0 20 20 100
Total 18 17 25 20 Overall 94
Producer Acc. (%) 100 100 80 100 Acc. (%)

Table 6
Confusion matrix in 1995.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 19 0 1 0 20 95
2 Built-up Area 0 16 1 3 20 80
3 Rocks 1 0 17 2 20 85
4 Vegetation 0 2 2 16 20 80
Total 20 18 21 21 Overall 85
Producer Acc. (%) 95 89 81 76 Acc. (%)

Table 7
Confusion Matrix in 2000.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 20 0 0 0 20 100
2 Built-up Area 1 17 0 2 20 85
3 Rocks 0 0 20 0 20 100
4 Vegetation 0 0 0 20 20 100
Total 21 17 20 22 Overall 96
Producer Acc. (%) 95 100 100 91 Acc. (%)

Table 8
Confusion matrix in 2005.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 20 0 0 0 20 100
2 Built-up Area 0 19 1 0 20 95
3 Rocks 1 0 17 2 20 85
4 Vegetation 0 0 0 20 20 100
Total 21 19 18 22 Overall 95
Producer Acc. (%) 95 100 94 91 Acc. (%)

Pn Pn
cess, making it a valuable tool for LULC classification tasks. Addi- N i¼1 mi;i  ðGi C i Þ
k¼ Pn i¼1 ð2Þ
tionally, it can be used to compare the performance of different 2
N  i¼1 ðGi C i Þ
techniques or assess the accuracy of a particular technique con-
cerning a reference classifier. The confusion matrix results are Where: N is the total number of classified values compared to
described in Tables 5–12. The kappa coefficient is calculated as truth values; i is the class number; mi,i is the number of values
follows: belonging to the truth class i that have also been classified as class

347
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Table 9
Confusion Matrix in 2010.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 20 0 0 0 20 100
2 Built-up Area 1 17 0 2 20 85
3 Rocks 2 0 18 0 20 90
4 Vegetation 0 1 0 19 20 95
Total 23 18 18 21 Overall 93
Producer Acc. (%) 87 94 100 90 Acc. (%)

Table 10
Confusion matrix in 2015.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 20 0 0 0 20 100
2 Built-up Area 0 19 1 0 20 95
3 Rocks 1 0 18 1 20 90
4 Vegetation 2 0 1 17 20 85
Total 23 19 20 18 Overall 93
Producer Acc. (%) 87 100 90 94 Acc. (%)

Table 11
Confusion Matrix in 2020.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 20 0 0 0 20 100
2 Built-up Area 0 17 0 3 20 85
3 Rocks 4 0 16 0 20 80
4 Vegetation 0 0 0 20 20 100
Total 24 17 16 23 Overall 91
Producer Acc. (%) 83 100 100 87 Acc. (%)

Table 12
Confusion matrix in 2022.

ID Classes Classification results Total User Acc. (%)

1 2 3 4
1 Bare land 19 0 0 1 20 95
2 Built-up Area 0 19 1 0 20 95
3 Rocks 2 0 17 1 20 85
4 Vegetation 0 0 0 20 20 100
Total 21 19 18 22 Overall 94
Producer Acc. (%) 90 100 94 91 Acc. (%)

Table 13 7. Challenges and limitations

Overall accuracy and Kappa coefficient.

Year Accuracy Kappa Coefficient While the current developments in RS and GIS LULC classifica-
1990 94% 0.92% tion are promising, there are still limitations and challenges. First,
1995 85% 0.80% the accuracy of collected data can vary depending on the type of
2000 96% 0.95% equipment used and the environment in which the data was col-
2005 95% 0.93%
lected. Due to the nature of the classification process and the chal-
2010 93% 0.90%
2015 93% 0.90%
lenges of detecting small-scale land-use changes, there is a need to
2020 91% 0.88% improve the accuracy of the classification process. It can be
2022 94% 0.92% achieved by deploying more sophisticated RS platforms and algo-
Overall accuracy 92.6% 90% rithms. For example, one future direction is combining many
machine learning and deep learning algorithms to improve the
i; Gi is the total number of truth values belonging to class i; Ci is the accuracy of LULC classification. Furthermore, multimodal data
total number of predicted values belonging to class i. fusion can be applied to improve the quality of classification algo-
Table 13 shows the overall accuracy and Kappa coefficient. The rithms, and pixel-based classification techniques can be used to
Kappa coefficient values in each year are in substantial and perfect analyze subtle changes in LULC at a fine scale. Second, data acqui-
agreement, with a range of 80% to 95%. The overall precision is sition and processing costs can be high, making it difficult for smal-
92.6%, indicating that the classification has a reasonable accuracy ler organizations or researchers to obtain the resources needed to
rate. carry out the analysis. Finally, the resolution of the imagery can

348
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

be limited, making it difficult to detect small LULC objects Alqadhi, S., Mallick, J., Balha, A., Bindajam, A., Singh, C.K., Hoa, P.V., 2021. Spatial and
decadal prediction of land use/land cover using multi-layer perceptron-neural
accurately.
network (MLP-NN) algorithm for a semi-arid region of Asir, Saudi Arabia. Earth
Sci. Inf. 14 (3), 1547–1562.
8. Conclusions Alqurashi, A.F., Kumar, L., 2014. Land Use and Land Cover Change Detection in the
Saudi Arabian Desert Cities of Makkah and Al-Taif Using Satellite Data. Adv.
Remote Sensing 03 (03), 106–119.
This study aimed to investigate LULC changes in NBR over the Alshari, E.A., Gawali, B.W., 2021. Development of a classification system for LULC
past three decades to understand the patterns and drivers of LULC. using remote sensing and GIS. Global Trans. Proc. 2 (1), 8–17.
Alshari, E.A., Gawali, B.W., 2021. Developing a science of land change: challenges
The study utilized RS data from Landsat imagery between 1990 and methodological issues. Proc. Natl. Acad. Sci. 2 (1), 73–79.
and 2022 and conducted a time-series analysis to detect changes Alshari, E.A., Gawali, B.W., Li, Y., 2022. Analysis of Machine Learning Techniques for
over time. Four classes were identified: bare land, built-up area, Sentinel-2A Satellite Images. J. Electr. Computer Eng. 2022, 1–16.
Aslami, F., Ghorbani, A., Sobhani, B., Panahandeh, M., 2015. Comparing artificial
rocks, and vegetation. The results showed that there had been a
neural network, support vector machine, and object-based methods in
significant increase in urbanization and a decrease in bare soil preparation land use/cover maps using landSat-8 images. J. RS and GIS
and green spaces during this period. Most of the urbanization Natural Resour. 6 (3), 1–14.
Balarabe, A.T., Jordanov, I., 2021.. LULC Image Classification with Convolutional
occurred in the outskirts of the cities, where previously there
Neural Network. IEEE, pp. 5985–5988.
was bare soil. The main driver of urbanization was population Bhattacharya, R.K., Das Chatterjee, N., Das, K., 2021. Land use and land cover change
growth and economic development. The finding showed how and its resultant erosion susceptible level: An appraisal using RUSLE and
human factors and processes have greatly affected the shapes Logistic Regression in a tropical plateau basin of West Bengal, India. Environ.
Dev. Sustain. 23 (2), 1411–1446.
and disappearance of many forms of the Earth’s surface between Dibs, H., Hasab, H.A., Al-Rifaie, J.K., Al-Ansari, N., 2020. An optimal approach for
1990 and 2022. The study provides a reliable baseline for monitor- land-use/land-cover mapping by integration and fusion of multispectral
ing changes in the LULC of the study area and can be used as a Landsat OLI images: a case study in Baghdad, Iraq. Water Air Soil Pollut. 231
(9), 1–15.
starting point in other studies that involve a change in land cover Ghayour, L., Neshat, A., Paryani, S., Shahabi, H., Shirzadi, A., Chen, W., Al-Ansari, N.,
in the NBR. The outcomes show that the classification scheme Geertsema, M., Pourmehdi Amiri, M., Gholamnia, M., Dou, J., Ahmad, A., 2021.
accurately mapped the NBR’s LULC classes, with an overall accu- Performance evaluation of sentinel-2 and Landsat 8 OLI data for land cover/use
classification using a comparison between machine learning algorithms.
racy of 92.6% and a Kappa coefficient of 0.90. This study demon- Remote Sens. (Basel) 13 (7), 1349.
strated the usefulness of remote sensing and GIS techniques for Hamad, R., 2020. An Assessment of Artificial Neural Networks, Support Vector
mapping and monitoring LULC change dynamics in the Northern Machines, and Decision Trees for Land Cover Classification Using Sentinel-2A
Data. Sciences 8 (6), 459–464.
Border Region over an extended period of time. However, the lim-
https://en.wikipedia.org/wiki/Arar,_Saudi_Arabia.
itations and challenges identified in this study also highlight the Kadavi, P.R., Lee, C.W., 2018. Land cover classification analysis of volcanic island in
need for more advanced techniques to predict LULC change Aleutian Arc using an artificial neural network (ANN) and a support vector
machine (SVM) from Landsat imagery. Geosci. J. 22 (4), 653–665.
dynamics in the future. Therefore, it is recommended that future _
Kaya, I.A., Görgün, E.K., 2020. Land use and land cover change monitoring in
studies in the region should explore more advanced machine Bandırma (Turkey) using remote sensing and geographic information systems.
learning and artificial intelligence techniques, such as deep learn- Environ. Monit. Assess. 192 (7), 1–18.
ing, to improve the accuracy and efficiency of LULC classification Khwarahm, N.R., 2021. Spatial modeling of land use and land cover change in
Sulaimani, Iraq, using multitemporal satellite data. Environ. Monit. Assess. 193
and prediction. In addition, incorporating ancillary data sources, (3), 148.
such as socioeconomic and environmental variables, could provide Khwarahm, N.R., Qader, S., Ararat, K., Fadhil Al-Quraishi, A.M., 2021. Predicting and
a more comprehensive understanding of the driving factors of mapping land cover/land use changes in Erbil/Iraq using CA-Markov synergy
model. Earth Sci. Inf. 14 (1), 393–406.
LULC change in the region. Such studies would not only advance Kuemmerle, T., Erb, K., Meyfroidt, P., Müller, D., Verburg, P.H., Estel, S., Haberl, H.,
our knowledge of LULC dynamics in the region but also inform Hostert, P., Jepsen, M.R., Kastner, T., Levers, C., Lindner, M., Plutzar, C., Verkerk, P.
effective land management and planning strategies for sustainable J., van der Zanden, E.H., Reenberg, A., 2013. Challenges and opportunities in
mapping land use intensity globally. Curr. Opin. Environ. Sustain. 5 (5), 484–493.
development. Linda Theres, B., Selvakumar, R., 2022. SelvakumarComparison of land
use/landcover classifier for monitoring urban dynamics using spatially
Declaration of Competing Interest enhanced Landsat dataset. Environ. Earth Sci. 81 (5), 1–8.
Qi Lv, Yong Dou, Xin Niu, Jiaqing Xu, Jinbo Xu, Fei Xia, ‘‘Urban Land Use and Land
Cover Classification Using Remotely Sensed SAR Data through Deep Belief
The authors declare that they have no known competing finan- Networks,” Journal of Sensors, vol. 2015, Article ID 538063, 10 pages, 2015.
cial interests or personal relationships that could have appeared Mahmoud, S.H., Alazba, A.A., 2016. A coupled remote sensing and the Surface
Energy Balance based algorithms to estimate actual evapotranspiration over the
to influence the work reported in this paper.
western and southern regions of Saudi Arabia. J. Asian Earth Sci. 124, 269–283.
Rahman, M.T., 2016. Detection of land use/land cover changes and urban sprawl in
Acknowledgements Al-Khobar, Saudi Arabia: an analysis of multi-temporal remote sensing data. Int
J Geo-Inf 5, 15–32.
Ramdani, F., Setiawan, B., Rusydi, A., Furqon, M. (2021). An Artificial Neural Network
The authors gratefully acknowledge the approval and the sup- Approach to Predict the Future Land Use Land Cover of Great Malang Region,
port of this research study by grant no. SCIA-2022-11-1647 from Indonesia.
the Deanship of Scientific Research at Northern Border University, Rezaei Moghaddam, M., Valizadeh Kamran, K., Andaryani, S., Almaspoor, F., 2015.
Comparison of ANN and SVM methods in extraction Land Use/Land Cover maps
Arar, K.S.A. We also express sincere thanks to the USGS and (diva) from Landsat 8 satellite image (Case Study: Sufi Chay Basin). Geography
for making the Landsat data available for public use through their Planning 19 (52), 163–183.
portals. Rizvon, S.S., Jayakumar, K., 2022. Strength prediction models for recycled aggregate
concrete using Random Forests, ANN, and LASSO. J. Building Pathology
Rehabilitation 7 (1), 1–10.
Saddique, N., Mahmood, T., Bernhofer, C., 2020. Quantifying the impacts of land
use/land cover change on the water balance in the afforested River Basin,
References Pakistan. Environ. Earth Sci. 79 (19), 1–13.
Salih, A., 2018. Classification and mapping of land cover types and attributes in Al-
Ahsaa Oasis, Eastern Region, Saudi Arabia using Landsat-7 data. J. Remote Sens.
Abdallah, S., Abd elmohemen, M., Hemdan, S., Ibrahem, K., 2019. Assessment of land
GIS 7 (1), 228–234.
use/land cover changes induced by Jizan Dam, Saudi Arabia, and their effect on
Sang, X., Guo, Q., Wu, X., Xie, T., He, C., Zang, J., Qiao, Y., Wu, H., Li, Y., 2021. The
soil organic carbon. Arab. J. Geosci. 12 (11). https://doi.org/10.1007/s12517-
Effect of DEM on the Land Use/Cover Classification Accuracy of Landsat OLI
019-4474-1.
Images. J. Indian Soc. Remote Sens. 49 (7), 1507–1518.
Alawamy, J.S., Balasundram, S.K., Mohd. Hanif, A.H., Boon Sung, C.T., 2020. Detecting
Shi, W., Zhang, M., Zhang, R., Chen, S., Zhan, Z., 2020. Change detection based on
and analyzing land use and land cover changes in the region of Al-Jabal Al-
artificial intelligence: State-of-the-art and challenges. Remote Sens. (Basel) 12
Akhdar, Libya using time-series Landsat data from 1985 to 2017. Sustainability
(10), 1688.
12 (11), 4490.

349
A.A. Darem, A.A. Alhashmi, A.M. Almadani et al. Egypt. J. Remote Sensing Space Sci. 26 (2023) 341–350

Singh, R.K., Sinha, V.S.P., Joshi, P.K., Kumar, M., 2021. A multinomial logistic model- Verburg, P.H., Neumann, K., Nol, L., 2011. Challenges in using land use and land
based land use and land cover classification for the South Asian Association for cover data for global change studies. Glob. Chang. Biol. 17 (2), 974–989.
Regional Cooperation nations using Moderate Resolution Imaging Vignesh, T., Thyagharajan, K.K., 2019. Land Use and Land Cover Classification Using
Spectroradiometer product. Environ. Dev. Sustain. 23 (4), 6106–6127. Deep Belief Network for LISS-III Multispectral Satellite Images. Int. J. Innovative
Tan, J., Zuo, J., Xie, X., Ding, M., Xu, Z., Zhou, F., 2021. MLAs Land cover mapping Technol. Exploring Engineering (IJITEE) 9, 94–98.
performance across varying geomorphology with Landsat OLI-8 and minimum Wijitkosum, S., 2016. The impact of land use and spatial changes on desertification
human intervention. Eco. Inform. 61, 101227. risk in degraded areas in Thailand. Sustainable Environ. Res. 26 (2), 84–92.

350