









































Pa
ge

 
1



Pa
ge

 
60

American Journal of  
Geospatial Technology (AJGT)

Urbanization and Its Ecological Impacts in Zanzibar: A Spatial-Temporal Analysis with 
Land Change Modeler (1995–2024)

Mohamed Khalfan Mohamed1*

Volume 4 Issue 1, Year 2025
ISSN: 2833-8006 (Online)

DOI: https://doi.org/10.54536/ajgt.v4i1.4619
https://journals.e-palli.com/home/index.php/ajgt

Article Information ABSTRACT

Received: February 23, 2025

Accepted: March 26, 2025

Published: April 30, 2025

This study examines the urban expansion of  Zanzibar from 1995 to 2024, utilizing Landsat 
satellite imagery to highlight the role of  remote sensing in understanding land-use changes 
and their socio-environmental implications. By employing change detection analysis, land 
cover classification techniques, and spectral separability assessments, the research quanti-
fies the spatial and temporal dynamics of  urban growth. Spectral separability was evaluated 
using Transformed Divergence (TD) and Jeffries-Matusita (J-M) distance metrics to ensure 
reliable differentiation between land cover classes, enhancing classification accuracy. The 
analysis focused particularly on urban Unguja Island and its surroundings, where significant 
urban sprawl has occurred over the past three decades. Results indicate a dramatic increase 
in built-up areas, rising from 2,650.5 ha in 1995 to 11,407.2 ha in 2024, corresponding to an 
overall growth of  330.4%. This urban expansion has come at the expense of  natural vege-
tation, which decreased by 26.3% over the study period. While water bodies have remained 
relatively stable, the transformation of  vegetation into urban land highlights the growing en-
vironmental pressure exerted by rapid urbanization. The classification accuracy of  the study 
improved over time, with overall accuracies of  78.33%, 87.22%, and 93.33% for the years 
1995, 2009, and 2024, respectively. The findings emphasize the importance of  implementing 
sustainable urban planning and policy interventions to mitigate the adverse effects of  ur-
ban sprawl on ecological sustainability. Integrating remote sensing data with socio-economic 
analysis is recommended for developing effective land management strategies in Zanzibar.

Keywords
Zanzibar, Land Use Land Cover 
Change, Spectral Separability, 
Transformed Divergence, Jeffries-
Matusita

1 Abdulrahman Al-Sumait University, Zanzibar, Tanzania. 
* Corresponding author’s e-mail: mohamed.khalfan76@gmail.com

INTRODUCTION
Urban growth is a predominant global trend driven by 
economic opportunity, demographic shifts, and rural-
to-urban migration. The United Nations projects that 
by 2050, nearly 68% of  the global population will reside 
in urban areas, with most of  this growth concentrated 
in Asia and Africa, where urban populations are rising 
rapidly (UN, 2019). Urban expansion is particularly 
intense in megacities like Mumbai, Lagos, and Dhaka, 
where populations are expected to double within the 
next few decades due to high migration rates and 
natural population increase (World Bank, 2021). As 
cities grow, they consume significant land resources; 
in developing regions, urban sprawl is encroaching on 
agricultural and forested areas, threatening food security 
and local ecosystems (Seto et al., 2020). This expansion 
is transforming landscapes worldwide, contributing to 
environmental challenges like habitat fragmentation and 
biodiversity loss, as well as increasing the demand for 
housing, infrastructure, and basic services (Angel et al., 
2021). These trends underscore the complex challenges 
governments face in managing urban growth sustainably, 
ensuring cities remain livable, resilient, and equitable 
spaces (UN-Habitat, 2020).
The environmental impacts of  global urban growth are 
substantial, with urban areas accounting for over 70% of  
global greenhouse gas emissions, despite covering only 
3% of  Earth’s surface (UN-Habitat, 2020). Urbanization 
often leads to the phenomenon of  urban heat islands, 
where built-up areas experience higher temperatures than 

surrounding rural regions due to reduced vegetation and 
increased energy use, particularly in large cities like Tokyo 
and New York (Li et al., 2021). Additionally, expanding 
urban areas place immense pressure on water resources, 
as cities draw on local and regional supplies to meet the 
demands of  growing populations (Zhang et al., 2021). 
Addressing these impacts requires integrated planning that 
includes green infrastructure, energy-efficient buildings, 
and compact city designs to reduce resource consumption 
and mitigate environmental degradation (Satterthwaite 
et al., 2020). Cities that have successfully implemented 
sustainable urban planning, such as Singapore and 
Stockholm, illustrate how innovative policies can manage 
urban growth’s environmental impacts while enhancing 
residents’ quality of  life and fostering economic resilience 
(Sassen, 2021).
Remote sensing, particularly through the Landsat 
mission, has revolutionized the mapping of  urban growth 
globally by providing comprehensive, long-term datasets 
that facilitate the monitoring and analysis of  land-use 
changes. Since its launch in 1972, the Landsat program 
has consistently delivered high-quality satellite imagery, 
capturing detailed information on urban environments 
and their expansion (U.S. Geological Survey, 2021). One 
of  the significant advantages of  Landsat is its moderate 
spatial resolution of  30 meters, which enables the 
identification of  various land cover types, including built-
up areas, vegetation, and water bodies. This capability is 
crucial for urban studies, as it allows researchers to assess 
impervious surface expansion, evaluate urban heat islands, 



Pa
ge

 
61

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

and monitor the effects of  urbanization on surrounding 
ecosystems (Seto et al., 2012). The temporal resolution 
of  Landsat, with images acquired approximately every 
16 days, provides insights into rapid urbanization 
dynamics, particularly in developing regions experiencing 
population influx and infrastructural pressures (Zhang et 
al., 2021). By employing advanced analytical techniques, 
such as change detection and classification algorithms, 
researchers can effectively quantify urban growth and 
its impacts on the environment, informing sustainable 
urban planning and management strategies (Tsurusaki & 
Salem, 2024). Landsat’s extensive historical archive allows 
for long-term studies of  urbanization trends, revealing 
critical insights into how cities evolve over time and the 
associated socio-environmental consequences (Valero-
Jorge et al., 2024). For instance, studies have shown that 
urban expansion often encroaches on agricultural and 
natural landscapes, leading to habitat loss and increased 
vulnerability to climate change (Sánchez-Llull et al., 
2025). This highlights the necessity of  integrating remote 
sensing data into urban development frameworks to 
ensure balanced growth that considers environmental 
sustainability. Furthermore, the versatility of  Landsat 
imagery enables its integration with other remote sensing 
platforms and ground-based data, enhancing the accuracy 
of  urban growth assessments and supporting evidence-
based decision-making for sustainable urban futures (Tew 
et al., 2018; Chander et al., 2009).
Urban growth in Zanzibar, Tanzania, reflects broader 
national trends of  increasing urbanization driven by rapid 
population growth, tourism, and rural-to-urban migration 
(World Bank, 2021). Zanzibar is the most urbanized region 
in Tanzania, and its population is expanding at a rate of  
approximately 4.4% annually, surpassing the mainland 
average and creating significant demand for housing, 
infrastructure, and services (NBS Tanzania, 2020). The 
growth of  the tourism sector has spurred economic 
development and job creation, drawing more residents to 
urban centers like Stone Town; however, it has also led 
to land-use changes that encroach on coastal ecosystems, 
including critical mangrove forests (RGoZ, 2020; 
RGoZ, 2022). With limited space for urban expansion 
on the island, rapid population growth contributes to 
overcrowding, water shortages, and inadequate waste 
management systems, particularly in low-income 
neighborhoods (UN-Habitat, 2021). Moreover, the urban 
infrastructure in Zanzibar is struggling to keep pace with 
demand, leading to challenges such as traffic congestion, 
insufficient public transport, and high levels of  air and 
water pollution (RGoZ, 2022World Bank, 2021). In 
response, the Revolutionary Government of  Zanzibar 
has launched the Zanzibar Urban Development Policy, 
aiming to promote sustainable urban growth through 
policies that improve infrastructure, support economic 
diversification, and enhance environmental resilience 
(RGoZ, 2020; RGoZ, 2019). However, the success of  
these efforts relies heavily on sustainable land use planning 
and greater investment in environmental conservation to 

mitigate the pressures of  urbanization on both natural 
resources and community welfare (RGoZ, 2022; World 
Bank, 2021).
While research on Zanzibar’s urban growth has expanded, 
a critical research gap exists in utilizing spatial time series 
data to comprehensively map and analyze long-term 
urban expansion and its ecological impacts. Previous 
studies have primarily relied on general population 
metrics and localized surveys, which are limited in 
capturing the spatial and temporal patterns of  land-
use change over extended periods (World Bank, 2021). 
However, Landsat missions, with their extensive historical 
record spanning nearly five decades, offer a unique and 
valuable dataset for examining urban growth dynamics 
over time. The application of  Land Changer Modeler in 
TerrSet liberaGIS v20.01 provides a robust framework 
for detecting gradual land conversion and encroachment 
on existing ecosystems and agricultural land through 
advanced analytical techniques such as change detection. 
Existing studies have overlooked the integration of  
spatial-temporal analysis essential for understanding the 
progression of  urbanization over decades. Therefore, this 
study aims to bridge this gap by mapping Zanzibar’s urban 
expansion from 1995 to 2025 using Landsat missions 
and advanced change detection techniques to provide a 
comprehensive understanding of  urban growth patterns 
and their implications for ecological sustainability.

MATERIALS AND METHODS
Study Area
The study area for this research is the urban region of  
Unguja, the largest and most populous island of  the 
Zanzibar Archipelago, located off  the coast of  mainland 
Tanzania in East Africa (Figure 1). Unguja is situated 
between latitudes approximately 5.7°S and 6.5°S, and 
longitudes 39.0°E and 39.6°E. The island covers an area 
of  around 1,666 square kilometers, with Zanzibar City, 
its capital, located on the central-western coast. The 
city comprises two main parts: Stone Town, a historic 
urban area designated as a UNESCO World Heritage 
Site, and Ng’ambo, a rapidly growing urban area that 
has undergone significant expansion over the past few 
decades. Zanzibar experiences a tropical monsoon 
climate characterized by two distinct rainy seasons: the 
long rains (Masika) from March to May, and the short 
rains (Vuli) from October to December (Mohamed 
et al., 2023a). The average annual temperature ranges 
from 25°C to 30°C, with high humidity levels due to its 
coastal location. Zanzibar City serves as the political, 
economic, and cultural hub of  Zanzibar (Mohamed et 
al., 2023b). It has witnessed rapid population growth 
due to rural-to-urban migration and natural population 
increase. The city’s economy largely relies on tourism, 
fishing, trade, and small-scale industries. The growth 
of  tourism has particularly driven urban development, 
especially in coastal areas and transport infrastructure 
(Mohamed et al., 2024). The urban landscape of  Zanzibar 
City has transformed significantly over the study period, 



Pa
ge

 
62

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

with increasing built-up areas replacing vegetation and 
agricultural land (Mohamed et al., 2024). Key drivers 
of  urbanization include population growth, tourism 

development, and infrastructure expansion (Mohamed et 
al., 2023; Mohamed et al., 2024).

Figure 1: Study area

Data sources and pre-processing
The study utilized satellite imagery to monitor urban 
changes in Zanzibar from 1995 to 2024. Landsat images 
for the years 1995, 2009, and 2024 were acquired from 
the US Geological Survey’s Earth Explorer portal 
(http://earthexplorer.usgs.gov/, last accessed 21 
December 2024) to evaluate temporal shifts in urban 
land cover. Given Zanzibar’s persistent cloud cover 
throughout the year, selecting suitable imagery presented 
a considerable challenge. However, the use of  Landsat 
Thematic Mapper (TM), Enhanced Thematic Mapper 
Plus (ETM+), and Landsat 8 provided higher-resolution 

data critical for accurate analysis. Landsat TM and ETM+ 
sensors each consist of  seven spectral bands with a spatial 
resolution of  30 meters for most bands, while their 
thermal bands (band 6) have resolutions of  120 and 60 
meters, respectively. Landsat 8 offers nine spectral bands 
with a 30-meter resolution for most bands and includes 
a 15-meter panchromatic band, enhancing spatial detail 
for improved analysis. The uneven acquisition intervals 
of  Landsat images resulted from the region’s tropical 
climate, characterized by frequent cloud cover and 
rainfall, which significantly constrained the availability of  
clear-sky imagery.

Table 1: The Landsat data used in the study of  Zanzibar urban growth.
S/
No.

Satellite
Name

Bands
Used

Acquisition WRS 
Path/
Row

Sensor 
Type

Cloud 
Cover 
(%)

LULC 
Name

UTM 
Zone

Spatial
Resolution
(Meters)

Date Time

1 Landsat 
5

1-5, 7 1995-09-
06

02.25.15 166/64 TM 1.2 1995 LULC 30 N 30 X30

2 Landsat 
7

1-5, 7 2009-07-
01

02.38.27 166/64 ETM+ 1.5 2009 LULC 30 N 30 X 30

3 Landsat 
8

01-Jul 2024-04-
29

02.49,14 166/64 OLI 0 2024 LULC 30 N 30 X 30

The Landsat module in TerrSet liberaGIS v20.01 was 
employed to enhance the quality of  the Landsat images 
through radiometric correction, converting raw digital 
number (DN) values into surface reflectance. To minimize 
atmospheric interference, the dark-object subtraction 
algorithm was applied, effectively reducing atmospheric 
haze effects. All images were reprojected to the UTM 
Zone 30 North coordinate system to ensure spatial 

consistency across datasets. Furthermore, the Landsat 
images were resampled to a 15-meter spatial resolution 
using the nearest neighbor resampling technique, as 
recommended by Mota et al. (2022) and Aziz et al. (2015). 
This resampling process enhanced the spatial detail, 
enabling a more accurate analysis of  urban growth 
patterns in Zanzibar.



Pa
ge

 
63

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Acquisition of  field data
Ground control points were collected using an eTrex® 20 
GPS Receiver (Garmin, Olathe, KS, USA) in combination 
with Landsat OLI imagery, topographic maps, and aerial 
photographs from the study site. These ground truth data 
points were acquired in February 2024. To enhance detail 
and improve color differentiation, a false-color Landsat 
OLI composite (RGB—754) was applied. Transect lines 
were strategically established across the study areas to 
capture representative samples of  homogeneous land 
cover classes, including built-up areas, vegetation, and 
water bodies. For each land cover class, 60 coordinates 
were recorded with a precision range of  3 to 5 meters. 
These coordinates were then mapped onto the resampled 
Landsat images and converted into polygons using TerrSet 
Geospatial Monitoring and Modeling System (TGMMS) 
v20.01 by tracing the borders of  the corresponding pixels. 
These polygons served as reference data, effectively 
establishing regions of  interest (ROIs) that accurately 
represented various land cover types. The ground truth 
data collection followed a comprehensive approach that 
integrated aerial imagery and historical records. Aerial 
surveys, including drone imagery, provided extensive 
coverage and a detailed perspective of  urban land cover. 
Additionally, archival records, such as land use maps and 
photographs obtained from the Zanzibar Department 
of  Forestry, were invaluable for reconstructing long-
term urban expansion trends. A thorough review of  
scientific literature and government reports further 
contextualized the findings, establishing a solid baseline 
for understanding urban dynamics over time.

Pre-Processing of  Satellite Images 
Using TerrSet Geospatial Monitoring and Modeling 
System (TGMMS) v20.01 software, satellite images were 
pre-processed for different purposes including geometric 
correction, radiometric calibration, atmospheric 
correction, and image enhancement. Geo-referencing 
images was used to remove geometric distortions in the 
acquired data and position the satellite images into a 
geographical coordinate system. Pre-processing satellite 
images is essential before image classification to monitor 
and predict land use/cover changes (Mishra et al., 2017). 
Ground control points (GCPs) were utilized along with 
reference datasets from high-resolution Google Earth 
images, topographic maps, and aerial data obtained from 
Zanzibar’s Department of  Forestry and Department of  
Land Surveying. The satellite images were rectified to 
eliminate atmospheric distortions caused by variations 
in sensor orientation parameters and noise from the 
acquisition platform (Ma et al., 2020). This correction 
significantly minimized data misinterpretation during 
image classification. Additionally, image enhancement 
techniques were applied to improve the visual quality 
and clarity of  the satellite images for subsequent analysis 
(Foody, 2002). 

Spectral separability
To evaluate the spectral separability of  various land 
cover types, their multispectral response patterns were 
thoroughly analyzed. Spectral separability reflects the 
statistical distance between class signatures, which 
significantly influences overall classification accuracy 
(Jackson & Adam, 2021). This study utilized the 
Transformed Divergence (TD) index and the Jeffries-
Matusita (J-M) distance as key metrics for assessing 
separability. Divergence (D) values were computed using 
the mean and variance-covariance matrices of  the feature 
class data, following the methodology outlined by Liu et 
al. (2020).

The Transformed Divergence (TD) metric was employed 
to minimize the influence of  highly separable classes, 
which could otherwise skew the average divergence value 
and diminish the reliability of  the measure (Kavzoglu & 
Mather, 2000).

In this context, tr[·] denotes the trace of  a matrix, which 
is the sum of  its diagonal elements. Σi and Σj represent 
the variance-covariance matrices corresponding to classes 
i and j, respectively, while μi and μj are the associated 
mean vectors. The constant c defines the range of  TD 
values.
The Jeffries-Matusita (J-M) distance measures the 
separability between the distributions of  two classes, ωi 
and ωj, as described by Jia et al. (1999).
JMij = 2(1-e-Bij)             (3)
where Bij represents the Bhattacharyya distance, 
calculated as (Kailath, 1967):

In this context, μi and μj represent the mean reflectance 
values for species i and j, respectively, while Σi and Σj 
denote their corresponding covariance matrices. The 
determinants of  these matrices are given by ǀΣiǀ and ǀΣjǀ. 
The notation ln refers to the natural logarithm, and T 
denotes the matrix transposition operation. According 
to Jensen (2005), class separability is assessed based 
on the resulting value: values above 1.9 indicate good 
separability, values between 1.7 and 1.9 suggest moderate 
separability, and values below 1.7 imply poor or non-
existent separability.

Pairwise feature comparison
A pairwise comparison technique was used to compute 
similarity scores for each pair of  Landsat bands based on 
the method outlined by Li et al. (2016). The bands are 
considered co-referent if  their similarity score exceeds a 
specified threshold. The outcomes of  these comparisons 
are structured into a matrix that represents the ratings 
between the k – th and p - th criteria. This matrix exhibits 



Pa
ge

 
64

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

reciprocity cₚₖ = cₖₚ⁻¹ and has unity along its diagonal 
(Ckp = 1 when k = p).  Table 2 presents a correlation 

matrix that illustrates the relationships between the 
Landsat 8 bands.

Table 2: Spectral separability as calculated by TD index (Equation (2)) and J-M distance (Equation (3)); BA-built areas, 
V-vegetation, W-water.

TM (1995) ETM+ (2009) OLI (2020)
LULC Pairs TD Index J-M Distance TD Index J-M Distance TD Index J-M Distance

BA/V 1.99 1.98 2.00 2.00 2.00 2.00
BA/W 2.00 2.00 2.00 2.00 2.00 2.00
V/W 2.00 2.00 2.00 2.00 2.00 2.00

Table 3: Classification system for land use and land cover (LULC)
S/No. Land Use/Land Cover Types Description
1 Bult-up/urban Regions that encompass residential, industrial, and commercial zones, 

as well as mixed-use structures, roadways, and other transportation 
infrastructure.

2 Vegetation It includes agricultural and horticultural lands, crop and fallow fields, 
forests, shrubs, Coconut trees, and various other types of  plantations.

3 Water Bodies These areas encompass the city's permanent bodies of  water, including 
seawater, streams, ponds, and various reservoirs.

Image classification
Image classification is a widely used and effective technique 
for processing satellite imagery data (Abd El-Hamid et al., 
2020). This method enables the detection, identification, 
and categorization of  various features within an image 
based on the actual land cover classes they represent on 
the Earth’s surface (Huang et al., 2020). The maximum 
likelihood classification (MLC) algorithm was applied 
to classify Landsat images due to its well-established 
theoretical framework and versatility in handling various 
data types, land use, and land cover (LULC) categories 
across different satellite systems (Foody, 2002). This study 
employed a supervised classification approach utilizing 
the Maximum Likelihood Classifier (MLC) to classify the 
acquired satellite imagery. The MLC, commonly known as 
the Bayesian decision rule, is among the most frequently 
applied supervised classification algorithms due to its 
accessibility and straightforward training process (Huang 
et al., 2020). In this technique, pixels within the satellite 
images are classified based on their likelihood of  belonging 
to a particular land use and land cover (LULC) category. 
The method operates under the assumption that all 
classes have equal probabilities and that the input bands 
follow a normal distribution. The study utilized specific 
input bands to generate false-color composite maps. For 

Landsat 5 TM and Landsat 7 ETM+, bands 4, 3, and 2 
were used, while for Landsat 8 OLI, bands 5, 4, and 3 
were selected. The spectral signatures of  individual image 
pixels were compared with the training samples from the 
study area to classify the satellite images into three major 
land use/land cover (LULC) categories. These categories 
included built-up or urban areas, vegetation, and water 
bodies, as outlined in Table 3. As described by (Vali et al., 
2020; Valero-Jorge et al., 2024; World Bank, 2021), the 
algorithm used to calculate the maximum likelihood (Li) 
of  an unknown measurement vector (x) belonging to a 
specific known class (Mc) is based on the application of  
the Bayesian equation, which is presented in Equation (1).
Lᵢ(x) = ln p(a_c) - [0.5 ln(|Cov_c|)] - [0.5(X - M_c)^T 
(Cov_c⁻¹)(X - M_c)]                          (5)
The discriminant function in the maximum likelihood 
algorithm is represented by Li(x), where the class is 
denoted as ac, with i ranging from 1 to M, where M is 
the total number of  classes. x is an n-dimensional vector 
representing the pixel, and n corresponds to the number 
of  bands. p(ac) indicates the probability of  class ac 
at position x for a given pixel. The determinant of  the 
covariance matrix for the data in class ac is represented by 
|Covc|, while Covc denotes the inverse of  the covariance 
matrix, and Mc is the mean vector for the class.

Accuracy assessment
In this research, an accuracy assessment of  the classified 
Landsat images was performed using ArcGIS version 
10.1 to evaluate the reliability and effectiveness of  the 
classification process. The first step in the accuracy 
assessment involved gathering ground truth data, which 
consisted of  field observations or high-resolution 
reference datasets, to serve as the standard for comparison 
with the classified imagery (Cohen, 1960). These reference 
points were carefully selected to represent a variety of  

land cover types within the study area. The confusion 
matrix was generated by comparing the classified image 
with the reference data. This was done by creating a 
point or polygon layer that contained the reference data 
and then using ArcGIS tools such as “Extract Values 
to Points” to compare the ground truth values with the 
classified image results. The confusion matrix, which is 
typically created using the “Tabulate Area” or “Reclassify” 
tool, includes key metrics such as user’s accuracy (UA), 
producer’s accuracy (PA), overall accuracy, and the kappa 



Pa
ge

 
65

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

statistic. The user’s accuracy (UA) and the producer’s 
accuracy (PA) were calculated for each land cover class 
to determine the classification’s performance from both 
the user’s and the producer’s perspectives (Fleiss et al., 
2003). The overall accuracy, representing the percentage 
of  correctly classified pixels, was also derived from the 
confusion matrix. The kappa statistic was calculated to 
measure the agreement between the classified data and 
the reference data, correcting for random chance. The 
kappa coefficient is a statistical measure that ranges 
from -1 to 1, though it typically lies between 0 and 1. 
It is used to evaluate the degree of  agreement between 
observed and predicted classifications. The interpretation 
of  kappa values is generally categorized as follows: (i) 
values between 0.00 and 0.20 indicate slight agreement, 
(ii) values from 0.21 to 0.40 represent fair agreement, (iii) 
values ranging from 0.41 to 0.60 correspond to moderate 
agreement, (iv) values from 0.61 to 0.80 reflect substantial 
agreement, and (v) values between 0.81 and 1.00 signify 
near-perfect or almost perfect agreement (Zanotta et al., 
2018). These metrics were visualized and interpreted 
within ArcGIS to assess the classification quality and 
guide any necessary improvements or adjustments. The 
results from the accuracy assessment were used to validate 
the classification method and ensure the classified image 
accurately represented the land cover changes observed 
in the field (Cohen, 1960). This process provided valuable 
insights into the accuracy of  the LULC maps and the 
effectiveness of  the classification approach employed in 
the research.

where, OA represents the overall accuracy, which refers 
to the proportion of  instances that are accurately 
classified. Pc denotes the count of  positive cases correctly 
identified, while Nc refers to the number of  negative 
cases accurately classified. Fp corresponds to the number 
of  negative cases that are incorrectly labeled as positive, 
and Fn represents the positive cases that are mistakenly 
classified as negative. P(e) is the expected probability of  
agreement by chance, calculated as the ratio of  the sum 
of  the products of  marginal probabilities for each class to 
the total number of  entries across all classes.

Change detection analysis 
This study, the Land Change Modeler (LCM) in the 
TerrSet Geospatial Monitoring and Modeling System 
(TGMMS), was utilized to perform a change detection 
analysis of  Landsat images from 1995, 2009, and 2024 
to map urban growth in Zanzibar. The primary objective 
was to assess the expansion of  urban areas. The analysis 
focused on three main land cover classes: Build-up, 
Vegetation, and Water. The first step involved classifying 
the Landsat images for each period (1995, 2009, and 2024) 
into the three land cover classes using TerrSet’s Maximum 
Likelihood Classification (MLC) algorithm. This method 
was chosen to handle the spectral variability across 

different land cover types and provide a statistically sound 
classification (Shivakumar et al., 2018). Ground truth data 
was used to train the classification model, ensuring that 
the results accurately represented the land cover types in 
Zanzibar. After classification, the Land Change Modeler 
was used to perform change detection between the three 
time periods. The tool generates a change matrix that 
compares the classified land cover maps from 1995, 2009, 
and 2024, allowing for the identification of  changes in the 
land cover classes over time. This process reveals areas 
where Build-up areas have expanded at the expense of  
Vegetation or Water, and how the spatial distribution of  
these classes has shifted.
The Land Change Modeler’s “Change Detection” 
module quantifies the amount of  change in each land 
cover class by calculating the area of  transition between 
different classes (Pontius & Cheuk, 2006). The analysis 
showed how Vegetation and Water were converted to 
Build-up areas. This allows for a clear understanding 
of  urban expansion patterns, including the rate and 
direction of  growth. The Land Change Modeler was also 
used to predict future changes based on historical trends. 
By analyzing the patterns of  urban growth from 1995 
to 2024, the tool modeled future land cover scenarios 
and estimated how the Build-up area may continue to 
expand, potentially encroaching on Vegetation in the 
next 50 years. The results of  the change detection analysis 
provided a detailed view of  urban growth in Zanzibar, 
highlighting the transformation of  land cover over the 
past few decades. 

RESULTS AND DISCUSSIONS
Spectral separability between land cover classes
The analysis of  average spectral reflectance curves for the 
land cover/use classes in the study area revealed distinct 
patterns, as depicted in Figure 2. These curves, along 
with their standard deviations (SDs), show considerable 
spectral overlaps between the three LULC classes (Built-
up, Vegetation, and Water) across the seven Landsat 
bands, especially within 1 SD. The visible bands (bands 
1 to 4: coastal aerosol, blue, green, and red) exhibited the 
highest level of  overlap. The general reflectance trend 
showed that the visible bands had the lowest values, 
followed by band 7 (Short-wave Infrared 2), band 6 
(Short-wave Infrared 1), and band 5 (Near Infrared). 
Among these, the near-infrared band (band 5) displayed 
the most noticeable reflectance differences between water 
and non-water classes. Outside the range of  1 SD, only 
the Near Infrared (band 5) and Short-wave Infrared 
1 (band 6) bands were effective in distinguishing water 
from other land cover types. These findings highlight 
the significance of  specific spectral bands in accurately 
differentiating water from other land cover classes.
Table 2 presents the spectral separability assessment 
of  four LULC classes—built-up areas, water, and 
vegetation—using Landsat OLI imagery (bands 1–7) and 
evaluated through the Transformed Divergence (TD) 
and Jeffries–Matusita (J-M) distance metrics. The analysis 



Pa
ge

 
66

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Figure 2: Mean spectral response—extracted from the Landsat OLI imagery - of  land cover and land use classes in 
Urban Zanzibar.

showed that for Landsat OLI (2024), ETM+ (2009), 
and TM (1995) imagery, the TD values for all LULC 
classes surpassed the threshold of  1.9, indicating a clear 
distinction between classes. However, the assessment of  
Landsat TM imagery revealed limitations in differentiating 
certain classes: both the TD index and J-M distance 
demonstrated poor separability between water and built-
up areas, as well as between vegetation and built-up areas. 
These results suggest that the older Landsat TM sensor 
has a reduced capacity for distinguishing specific LULC 
classes, highlighting the superior performance of  newer 
sensors (ETM+, OLI) for classification tasks.

Pairwise band comparison
The pairwise correlation analysis between the Landsat 
bands for the urban Zanzibar scene reveals significant 
relationships among the spectral bands, particularly 
within the visible spectrum. As shown in Table 3, 
Bands 1 (Coastal Aerosol), 2 (Blue), 3 (Green), and 4 
(Red) exhibit extremely high correlations, with values 
ranging from 0.979 to 0.999. This strong inter-band 
correlation suggests that these bands capture similar 
spectral information, likely due to their proximity within 

the electromagnetic spectrum and their sensitivity to 
surface reflectance characteristics of  built-up areas 
and vegetation. In contrast, Bands 5 (Near Infrared), 
6 (Short-Wave Infrared 1), and 7 (Short-Wave Infrared 
2) show moderate to high correlations with the visible 
bands but display a noticeable decline in correlation 
values, particularly with Band 5 (NIR) showing lower 
correlations with Bands 1–4 (0.681 to 0.783). The higher 
correlation between Bands 6 and 7 (0.961) reflects their 
shared sensitivity to moisture content and built-up 
surfaces, which makes them valuable for distinguishing 
urban areas from vegetative and water-covered regions. 
Furthermore, the correlation between Bands 5 and 6 is 
notably high (0.963), emphasizing the importance of  the 
NIR and SWIR regions for differentiating water bodies 
from other land cover classes. These findings indicate 
that the near-infrared and Short-Wave Infrared bands 
offer critical information that complements the visible 
bands, enhancing the ability to discriminate between 
urban, vegetation, and water classes. The results highlight 
the need for careful selection of  spectral bands when 
performing classification tasks in urban environments 
like Zanzibar.

Table 3: Pairwise correlations between Landsat bands for the scene covering Urban Zanzibar.
Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 Band 7

Band 1 ----- 0.999 0.980 0.994 0.681 0.790 0.902
Band 2 ----- 0.979 0.993 0.662 0.776 0.893
Band 3 ----- 0.985 0.783 0.868 0.936
Band 4 ----- 0.707 0.820 0.925
Band 5 ----- 0.963 0.869
Band 6 ----- 0.961
Band 7 -----

Image classification 
The analysis of  land cover changes in urban Zanzibar 
from 1995 to 2024 reveals a pronounced trend of  
urban expansion accompanied by a substantial decline 
in vegetative cover (Table 4; Figure 3). This pattern 

reflects rapid urbanization driven by population growth, 
infrastructure development, and the expansion of  
settlements. Notably, built-up areas have significantly 
increased from 2,650.5 hectares in 1995 to 5,862.4 ha in 
2009 and further to 11,407.2 ha in 2024, representing a 



Pa
ge

 
67

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

growth of  approximately 120.9% between 1995 and 2009 
and an additional 94.6% from 2009 to 2024. This rapid 
increase in built-up areas underscores the aggressive 
nature of  urbanization within the study area, highlighting 
the urgent need for sustainable urban planning and 
management. Conversely, vegetation cover has been 
steadily declining throughout the same period, decreasing 
from 33,252.1 ha in 1995 to 29,801.8 ha in 2009 and 
subsequently to 24,496.2 ha in 2024, corresponding 

to an overall reduction of  approximately 26.3% from 
1995 to 2024. This decline in vegetation is likely due 
to the conversion of  natural land to built-up areas and 
agricultural activities to accommodate the growing 
population. In contrast, water bodies have exhibited 
minimal variation over the study period, with coverage 
increasing slightly from 21,773.6 ha in 1995 to 22,012.0 
ha in 2009, before decreasing marginally to 21,772.8 ha 
in 2024. 

Table 4: Land Cover Changes in Urban Zanzibar from 1995 to 2024 (in Hectares)
LULC 1995 2009 2024

Land Cover in Hectares
Built up 2,650.5 5,862.4 11,407.2
Vegetation 33,252.1 29,801.8 24,496.2
Water 21,773.6 22,012.0 21,772.8

Figure 3: Zanzibar urban growth from 1995, 2009, and 2024

The analysis of  land cover changes in urban Zanzibar 
over the periods 1995–2024 and 2009–2024 reveals a 
consistent trend of  rapid urban expansion at the expense 
of  vegetation cover, with minimal changes in water bodies 
(Figure 4). From 2009 to 2024, built-up areas experienced 
significant growth, with approximately 72,691 ha of  land 
converted to urban use, while only around 11,082 ha 
were lost to other land covers. This net positive change 
underscores the aggressive urbanization occurring 
within the region. During the same period, vegetation 
exhibited a substantial net loss of  approximately 69,765 
ha, with only minor gains of  about 10,813 ha, suggesting 

limited reforestation or conservation efforts. Water 
bodies remained relatively stable, with a minor decrease 
of  3,289 ha and a small gain of  632 ha, indicating 
that urbanization has had minimal impact on aquatic 
environments thus far. However, when considering the 
long-term change from 1995 to 2024, the expansion 
of  built-up areas is even more pronounced, with gains 
reaching approximately 90,000 ha. Vegetation loss during 
this period is similarly alarming, with around 70,000 ha 
lost and only minimal recovery. The minimal variation in 
water bodies between these periods suggests a degree of  
resilience or protection, although slight fluctuations could 



Pa
ge

 
68

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

indicate emerging pressures from urban growth. The 
overall pattern highlights the continuous and aggressive 
urbanization of  Zanzibar, primarily driven by population 

growth, infrastructure development, and settlement 
expansion.

Figure 4: Land cover gain, loss, and net change from 1995 to 2024

Accuracy assessment
The accuracy assessment of  the land use and land 
cover (LULC) classification results for urban Zanzibar 
demonstrates a progressive improvement in classification 
accuracy over the study period from 1995 to 2024. The 
overall accuracy and Kappa coefficient values indicate the 
reliability and consistency of  the classification process. 
In 1995, the classification achieved an overall accuracy 
of  78.33% with a Kappa coefficient of  0.76, reflecting 
moderate agreement between the classified map and 

ground reference data. By 2009, the accuracy significantly 
improved to 87.22% with a corresponding Kappa 
coefficient of  0.84, indicating substantial agreement. This 
improvement is attributed to enhanced image quality 
from ETM+ and OLI satellite missions and the adoption 
of  more advanced classification techniques. In 2024, the 
classification accuracy reached its highest level at 93.33%, 
with a Kappa coefficient of  0.88, indicating a near-perfect 
agreement between classification results and ground 
reference data.

Table 5: The overall accuracies and Kappa coefficients for LULC in urban Zanzibar
Year Ground data Classification results Overall accuracy (%) Kappa coefficients
1995 180 141 78.33% 0.76
2009 180 157 87.22% 0.84
2020 180 168 93.33% 0.88

Change detection analysis 
The crosstabulation analysis of  1995 and 2024 land 
cover maps highlights significant land cover transitions 
in urban Zanzibar, revealing notable urban expansion 
and landscape dynamics (Table 6; Figure 5). The most 
substantial transition is from Vegetation to Build-up, 
with a conversion of  approximately 9,236.88 hectares, 
indicating extensive urbanization over the study period. 
This change suggests the rapid transformation of  
vegetated areas into urban landscapes, likely driven by 
population growth, infrastructure development, and 
economic activities. The persistence of  Build-up areas 
(1,994.7 ha) further confirms urban growth’s stability and 

consolidation over nearly three decades. Additionally, the 
conversion from Water to Build-up (175.5 ha) illustrates 
the construction of  houses, hotels, and infrastructure 
development in the coastal areas. A considerable portion 
of  the landscape remains vegetated (23,995.3 ha), 
suggesting that while urban expansion is substantial, 
natural or semi-natural areas still dominate much of  the 
landscape. Findings also reveal that the conversion of  
Build-up areas back to Vegetation (499.0 ha) is relatively 
minor. The minimal transition from Water to Vegetation 
(1.8 ha) is negligible and may reflect natural processes 
such as sediment deposition or slight vegetative growth 
along water bodies.



Pa
ge

 
69

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Table 6: Land cover transition from 1995 to 2024 in urban Zanzibar
No. Land cover Transition Rate of  Change (in ha.)
1 Build up to Build up 1994.7
2 Vegetation to Build up 9236.8
3 Water to Build up 175.5
4 Build up to Vegetation 499.0
5 Vegetation to vegetation 23995.3
6 Water to vegetation 1.8
7 Build up to Water 156.7
8 Vegetation to Water 19.8
9 Water to water 21596.2

Table 7: Exchange between vegetation and build-up 
areas
Land Cover 
Exchange

1995-2009 2009-2024 1995-2024

Land Cover 
Exchange

1995-2009 2009-2024 1995-2024

Vegetation to 
Build Up

4009.4 6251.4 9236.8

Build Up to 
Vegetation

563.8 967.9 499.0

Figure 5: Land cover transitions from 1995 to 2024

The findings of  land cover exchanges in urban Zanzibar 
between 1995 and 2024 reveal a clear pattern of  extensive 
vegetation loss due to urban expansion, with some minor 
regeneration of  built-up areas back to vegetation (Table 7; 
Figure 6). Between 1995 and 2009, 4,009.4 ha of  vegetation 
were converted to built-up areas. This period marked 
the beginning of  significant urbanization when natural 
vegetation was increasingly cleared for infrastructure 
development and urban growth. However, 563.8 ha of  
built-up land was reclaimed by vegetation, which could 
be attributed to greening efforts, abandonment of  urban 

areas, or natural vegetation regrowth. The period from 
2009 to 2024 shows an even more accelerated rate of  
vegetation loss, with 6,251.4 ha being converted to built-
up areas. During this period, 967.9 ha of  built-up land 
was reclaimed by vegetation. The overall analysis from 
1995 to 2024 indicates that a total of  9,236.8 ha of  
vegetation was converted to built-up areas, underscoring 
the significant scale of  urbanization that has taken place 
over the past 29 years. Conversely, only 499.0 ha of  built-
up areas were reclaimed by vegetation throughout this 
period, which is relatively small compared to the vast 
areas lost to urbanization. The findings highlight a clear 
and consistent pattern of  vegetation cover loss to urban 
expansion in urban Zanzibar, with accelerated rates of  
conversion occurring between 2009 and 2024.

Discussion
Spectral separability between land cover classes
The analysis of  average spectral reflectance curves for 
the land cover/use classes in the study area revealed 
distinct patterns. These curves, along with their standard 
deviations (SDs), show considerable spectral overlaps 
between the three LULC classes (Built-up, Vegetation, and 
Water) across the seven Landsat bands, especially within 1 
SD. This observation is consistent with previous research 
that highlights the challenges of  spectral separability 
between these land cover types, particularly when using 
multispectral sensors (Vali et al., 2020). Gavade and 
Rajpurohit (2020) found that the visible bands (bands 1 
to 4: coastal aerosol, blue, green, and red) exhibited the 
highest level of  overlap. These findings align with studies 
that show the visible spectrum generally struggles to 
distinguish between vegetation and built-up areas due to 



Pa
ge

 
70

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Figure 6: Transition from vegetation to build-up areas

similar reflectance characteristics in these bands (Ma et 
al., 2020). The general reflectance trend showed that the 
visible bands had the lowest values, followed by band 7 
(Short-wave Infrared 2), band 6 (Short-wave Infrared 1), 
and band 5 (Near Infrared) (Nalepa et al., 2019). This is 
in agreement with other studies that have reported the 
effectiveness of  infrared bands, particularly bands 5 and 
7, in enhancing classification accuracy for land cover 
mapping (Gavade & Rajpurohit, 2020). Among these, the 
near-infrared band (band 5) displayed the most noticeable 
reflectance differences between water and non-water 
classes. This is consistent with the findings of  a study by 
Vali et al. (2020), who demonstrated that the near-infrared 
region is sensitive to water body characteristics, enabling 
better discrimination between water and vegetation. 
Outside the range of  1 SD, only the Near Infrared 
(band 5) and Short-wave Infrared 1 (band 6) bands were 
effective in distinguishing water from other land cover 
types (Nelapa et al., 2019). Similar results were reported 
by Ma et al. (2020), who emphasized the significance of  
these bands in water body identification, especially when 
the spectral reflectance of  water overlaps with that of  
vegetation in the visible bands. 
The spectral separability assessment of  built-up areas, 
water, and vegetation using Landsat OLI (2024), ETM+ 
(2009), and TM (1995) imagery, evaluated through 
Transformed Divergence (TD) and Jeffries-Matusita (J-
M) distance metrics, demonstrated that newer sensors 
(OLI and ETM+) provide superior separability due to 
their enhanced spatial and spectral resolution, with TD 
values surpassing the 1.9 threshold for all land cover 
classes. This aligns with findings that newer sensors, 

particularly OLI, improve classification accuracy for 
challenging land covers like water and urban areas 
(Harrak et al., 2025). In contrast, Landsat TM showed 
limitations in distinguishing water from built-up areas and 
vegetation, likely due to spectral overlap and the absence 
of  a coastal aerosol band (Lewis et al., 2018; Günen & 
Atasever, 2024). These limitations are consistent with 
previous studies highlighting the reduced capacity of  
TM to separate LULC classes due to its coarser spectral 
resolution and fewer bands (Shao et al., 2023). Recent 
studies also emphasize the improved classification 
performance of  OLI and ETM+ sensors for urban and 
water bodies, which are critical for accurate land cover 
classification and environmental monitoring (Zhang et al., 
2025; Harrak et al., 2025).

Pairwise band comparison
The pairwise correlation analysis of  Landsat bands for 
the urban Zanzibar scene revealed high correlations 
among visible bands (1–4), with values ranging from 
0.979 to 0.999, indicating they capture similar spectral 
information. This is consistent with previous studies 
showing visible bands’ effectiveness in distinguishing 
built-up areas and vegetation (Cottrell et al., 2024). Bands 
5 (NIR), 6 (SWIR 1), and 7 (SWIR 2) showed moderate 
correlations with visible bands, with Band 5 having 
lower correlations (0.681 to 0.783). The high correlation 
between Bands 6 and 7 (0.961) highlights their shared 
sensitivity to moisture content, aiding in distinguishing 
urban areas from vegetation and water (Gokool et al., 
2024). The strong correlation between Bands 5 and 6 
(0.963) emphasizes the importance of  NIR and SWIR 



Pa
ge

 
71

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

bands for differentiating water bodies (Liu et al., 2020), 
reinforcing the need for careful spectral band selection in 
urban environments like Zanzibar.

Image classification 
The analysis of  land cover change in urban Zanzibar 
from 1995 to 2024 highlights significant urban expansion, 
with built-up areas increasing by 120.9% from 1995 
to 2009 and a further 94.6% from 2009 to 2024. This 
rapid urbanization, largely driven by population growth 
and infrastructure development, mirrors similar trends 
observed in coastal cities worldwide (Tsurusaki & Salem, 
2024). For instance, studies in coastal cities like Lagos, 
Nigeria, and Jakarta, Indonesia, also report significant 
urban sprawl at the expense of  natural ecosystems, 
particularly mangrove forests, and wetlands, due to 
high population growth and economic development 
(Kukkonen et al., 2017; Seto et al., 2012). Zanzibar’s 26.3% 
loss in vegetation cover from 1995 to 2024 is consistent 
with the global pattern of  urban encroachment into 
green spaces. For example, a study on urbanization in the 
Philippines found a comparable decline in forest cover as 
urban areas expanded rapidly, threatening biodiversity and 
ecosystem services (Tossoukpe et al., 2025). Additionally, 
the reduction in vegetation in Zanzibar, likely driven by 
land conversion for urban settlements and agriculture, 
mirrors findings from other African coastal cities, 
such as Mombasa, Kenya, where land-use changes for 
urbanization have led to similar declines in coastal forest 
and mangrove habitats (Haldar et al., 2024). On the other 
hand, the minimal variation observed in Zanzibar’s water 
bodies over the study period stands in contrast to trends 
in cities like Dhaka, Bangladesh, where urbanization has 
significantly altered hydrological patterns, often leading to 
the reduction of  freshwater bodies (Tessema & Abebe, 
2023). This relative stability in Zanzibar could indicate 
effective conservation efforts or natural limitations in the 
expansion of  urban development into water-rich areas, 
but it also suggests that Zanzibar may still face risks from 
future urbanization if  water resource management is not 
carefully integrated into development plans (Barman 
et al., 2024). The findings from Zanzibar thus resonate 
with broader global challenges faced by coastal cities, 
reinforcing the need for sustainable urban planning that 
prioritizes ecological conservation, integrated ecosystem 
management, and climate resilience to mitigate the 
adverse effects of  rapid urban growth (Puplampu & 
Boafo, 2021; Blakime et al., 2024).

Accuracy assessment
The accuracy assessment of  land use and land cover 
(LULC) classification for urban Zanzibar shows a 
consistent improvement over the study period from 
1995 to 2024, reflecting advancements in remote sensing 
technology and classification methodologies (Sánchez-
Llull et al., 2025). The overall accuracy increased from 
78.33% in 1995, with a Kappa coefficient of  0.76 
indicating moderate agreement, to 87.22% in 2009, 

with a Kappa coefficient of  0.84 indicating substantial 
agreement. The significant improvement observed 
in 2009 is mainly attributed to the use of  the Landsat 
ETM+ sensor, which offered better spectral resolution 
and data quality than the older Landsat TM sensor (Mota 
et al., 2022). Additionally, improvements in classification 
algorithms and the availability of  higher-quality training 
data contributed to this enhanced accuracy (Valero-Jorge 
et al., 2024). Similar findings have been reported by other 
studies demonstrating the positive impact of  improved 
sensor capabilities and methodological advancements on 
LULC classification accuracy (Foody, 2020; Sánchez-Llull 
et al., 2025). The classification accuracy reached its highest 
level in 2024, achieving an overall accuracy of  93.33% 
with a Kappa coefficient of  0.88, indicating near-perfect 
agreement between the classified maps and reference 
data. This high accuracy is largely due to the superior 
spatial, spectral, and radiometric resolution of  the 
Landsat OLI sensor, complemented by the application 
of  advanced classification techniques such as machine 
learning, which effectively handles complex landscapes 
and mixed pixels (Foody, 2020; Kolić et al., 2025). The 
progressive improvement in classification accuracy aligns 
with previous studies demonstrating that enhanced 
image quality and sophisticated algorithms significantly 
improve LULC mapping accuracy (Nikolakopoulos & 
Petropoulos, 2025). The high accuracy achieved in 2024 
highlights the reliability of  the classification results and 
strengthens their potential use for analyzing urban change 
and ecological impacts in Zanzibar.

Change detection analysis 
The crosstabulation analysis of  1995 and 2024 land cover 
maps for urban Zanzibar reveals substantial land cover 
transitions, with the most prominent change being the 
conversion of  vegetation to built-up areas, amounting to 
approximately 9,236.88 hectares. This finding indicates 
extensive urbanization over the nearly three-decade study 
period, driven by rapid population growth, infrastructure 
development, and increased economic activities. Similar 
studies worldwide have reported comparable trends, 
particularly in developing countries experiencing rapid 
urban expansion. For instance, a study by Seto et al. 
(2012) highlights that urban areas in Asia and Africa are 
expanding at unprecedented rates, resulting in significant 
loss of  vegetation cover and natural habitats. Likewise, 
Cao et al. (2020) reported substantial conversion of  
forested and vegetated areas to built-up land in Zhoushan 
Island, East China due to urban growth, often at the 
expense of  ecological sustainability.
The persistence of  built-up areas (1,994.7 ha) underscores 
the consolidation and stability of  urban growth in 
Zanzibar, reflecting patterns also observed in rapidly 
urbanizing cities globally. For example, studies conducted 
in cities across Pacific urban villages have demonstrated 
similar trends where urban areas continue to expand while 
remaining relatively stable once established (Jones, 2016). 
Additionally, the conversion of  water to built-up areas 



Pa
ge

 
72

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

(175.5 ha) highlights coastal development, including the 
construction of  houses, hotels, and other infrastructure, 
a trend also noted in coastal cities worldwide where 
tourism-driven urbanization is prevalent (Tew et al., 
2018). However, the relatively low transition of  built-up 
areas back to vegetation (499.0 ha) indicates limited urban 
greening efforts, a phenomenon consistent with findings 
in cities across Africa and Asia where urbanization is 
typically associated with irreversible land cover changes 
(Zhou et al., 2018). Furthermore, the negligible transition 
from water to vegetation (1.8 ha) likely reflects minor 
natural processes, such as sediment deposition or 
vegetation encroachment along water bodies. Collectively, 
these findings align with broader global patterns of  urban 
expansion and landscape change, underscoring the need 
for sustainable urban planning to balance development 
with environmental conservation (Petrisor, et al., 2020).

CONCLUSION
The findings of  this study reveal a rapid and 
extensive urban expansion in Zanzibar from 1995 to 
2024, predominantly driven by population growth, 
infrastructure development, and economic activities. 
This urbanization has resulted in a significant decline in 
vegetation cover, with approximately 26.3% of  vegetation 
lost over the study period. Despite the relative stability of  
water bodies, the conversion of  vegetated areas to built-
up land highlights the adverse impacts of  urban growth 
on ecological sustainability. The application of  advanced 
remote sensing techniques, including change detection 
analysis using Landsat imagery, has proven effective in 
providing a comprehensive understanding of  Zanzibar’s 
urbanization patterns. The results emphasize the urgent 
need for sustainable urban planning and conservation 
strategies to mitigate the negative environmental 
consequences of  urban sprawl. Future studies should 
focus on integrating high-resolution satellite data and 
socio-economic factors to enhance the monitoring and 
management of  urban growth in Zanzibar.

REFERENCES
Abd El-Hamid, H. T., El-Alfy, M. A., & Elnaggar, A. A. 

(2020). Prediction of  future situation of  land use/
cover change and modeling sensitivity to pollution 
in Edku Lake, Egypt based on geospatial analyses. 
GeoJournal.

African Union. (2021). Agenda 2063: The Africa we want.
Ahmad, F., Goparaju, L., & Qayum, A. (2017). LULC 

analysis of  urban spaces using Markov chain 
predictive model at Ranchi in India. Spatial Information 
Research, 25, 351–359. 

Angel, S., Parent, J., & Civco, D. (2021). Urban expansion 
in the world’s cities: 2000-2030. Cambridge University 
Press.

Aziz, F., Kusratmoko, E., & Mandini, M. D. (2020). 
Estimation of  changes in the lake water level and area 
using remote sensing techniques: Case study: Lake 
Toba, North Sumatra. IOP Conference Series: Earth and 

Environmental Science, 561, 012022. 
Blakime, T. H., Komi, K., Adjonou, K., Hlovor, A. K. D., 

Gbafa, K. S., Oyedele, P. B., Polorigni, B., & Kokou, 
K. (2024). Dynamics of  built-up areas and challenges 
of  planning and development of  urban zone of  
Greater Lomé in Togo, West Africa. Land, 13, 84. 

Cao, W., Li, R., Chi, X., Chen, N., Chen, J., Zhang, H., & 
Zhang, F. (2017). Island urbanization and its ecological 
consequences: A case study in the Zhoushan Island, 
East China. Ecological Indicators, 76, 1–14. 

Clarke, K. C. (2018). Cellular automata and agent-based 
models. In M. Fischer & P. Nijkamp (Eds.), Handbook 
of  regional science. Springer.

Cohen, J. (1960). A coefficient of  agreement for nominal 
scales. Educational and Psychological Measurement, 20(1), 
37-46. 

Congalton, R. G. (1991). A review of  assessing the 
accuracy of  classifications of  remotely sensed data. 
Remote Sensing of  Environment, 37(1), 35-46. 

Cottrell, B., Kalacska, M., Arroyo-Mora, J.-P., Lucanus, 
O., Inamdar, D., Løke, T., & Soffer, R. J. (2024). 
Limitations of  a multispectral UAV sensor for satellite 
validation and mapping complex vegetation. Remote 
Sensing, 16(13), 2463. 

Eastman, J. R. (2016). TerrSet geospatial monitoring and 
modeling system manual. Clark Labs, Clark University.

Jones, P. (2016). The emergence of  Pacific urban villages: 
Urbanization trends in the Pacific Islands.

Jones, P. (2016). The emergence of  Pacific urban villages: 
Urbanization trends in the Pacific Islands. Asian 
Development Bank.

Kolić, J., Pernar, R., Seletković, A., & Ančić, M. (2025). 
Determining the accuracy of  structural parameters 
measured from LiDAR images in lowland oak forests. 
Forests, 16(2), 340. 

Kukkonen, M., Muhammad, M., Käyhkö, N., & Luoto, M. 
(2017). Urban expansion in Zanzibar City, Tanzania: 
Analyzing quantity, spatial patterns, and effects of  
alternative planning approaches. Land Use Policy, 71. 

Fleiss, J. L., Levin, B., & Paik, M. C. (2003). The 
measurement of  interrater agreement. In W. A. 
Shewart & S. S. Wilks (Eds.), Statistical methods for rates 
and proportions (pp. 598–626). John Wiley & Sons Inc.

Foody, G. M. (2020). Explaining the unsuitability of  the 
kappa coefficient in the assessment and comparison 
of  the accuracy of  thematic maps obtained by image 
classification. Remote Sensing of  Environment, 239. 

Foody, G. M. (2002). Status of  land cover classification 
accuracy assessment. Remote Sensing of  Environment, 80, 
185–201. 

Gavade, A. B., & Rajpurohit, V. S. (2020). Sparse-FCM and 
deep learning for effective classification of  land area in multi-
spectral satellite images. Evolutionary Intelligence. 

Gidey, E., Dikinya, O., Sebego, R., Segosebe, E., & 
Zenebe, A. (2017). Cellular automata and Markov 
Chain (CA_Markov) model-based predictions of  
future land use and land cover scenarios (2015–2033) 
in Raya, northern Ethiopia. Modeling Earth Systems and 



Pa
ge

 
73

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Environment, 3, 1245–1262. 
Gokool, S., Mahomed, M., Brewer, K., Naiken, V., 

Clulow, A., Sibanda, M., & Mabhaudhi, T. (2024). 
Crop mapping in smallholder farms using unmanned 
aerial vehicle imagery and geospatial cloud computing 
infrastructure. Heliyon, 10, e26913. 

Günen, M. A., & Atasever, U. H. (2024). Remote sensing 
and monitoring of  water resources: A comparative 
study of  different indices and thresholding methods. 
Science of  the Total Environment, 926, 172117. 

Haldar, S., Chatterjee, U., Bhattacharya, S., Paul, S., 
Bindajam, A. A., Mallick, J., & Abdo, H. G. (2024). 
Peri-urban dynamics: Assessing expansion patterns 
and influencing factors. Ecological Processes, 13, 1. 

Harrak, Y., Rachid, A., & Aguejdad, R. (2025). Evaluation 
of  spectral indices and global thresholding methods 
for the automatic extraction of  built-up areas: An 
application to a semi-arid climate using Landsat 8 
imagery. Urban Science, 9(3), 78. 

Huang, Y., Yang, B., Wang, M., Liu, B., & Yang, X. (2020). 
Analysis of  the future land cover change in Beijing 
using CA–Markov chain model. Environmental Earth 
Sciences, 79, 60. 

Jackson, C., & Adam, E. (2021). Machine learning 
classification of  endangered tree species in a tropical 
submontane forest using WorldView-2 multispectral 
satellite imagery and imbalanced dataset. Remote 
Sensing, 13(24). 

Jensen, J. (2005). Introductory digital image processing: A remote 
sensing perspective (4th ed.). Prentice Hall.

Jia, X., & Richards, J. (1999). Segmented principal 
components transformation for efficient hyperspectral 
remote-sensing image display and classification. 
IEEE Transactions on Geoscience and Remote Sensing, 
37(1), 25–27. 

Karimi, H., Jafarnezhad, J., Khaledi, J., & Ahmadi, P. 
(2018). Monitoring and prediction of  land use/
land cover changes using CA-Markov model: A case 
study of  Ravansar County in Iran. Arabian Journal of  
Geosciences, 11, 592. 

Kavzoglu, T., & Mather, P. (2000). Using feature selection 
techniques to produce smaller neural networks with 
better generalization capabilities. Proceedings of  the 
International Geoscience and Remote Sensing Symposium 
(IGARSS), 7, 23–31.

Lewis, A., Lacey, J., Mecklenburg, S., Ross, J., Siqueira, 
A., Killough, B., Szantoi, Z., Tadono, T., Rosenqvist, 
A., Goryl, P., et al. (2018). CEOS analysis-ready 
data for land (CARD4L) overview. Proceedings of  the 
2018 IEEE International Geoscience and Remote Sensing 
Symposium (IGARSS), Valencia, Spain, 7407–7410.

Li, X., Zhou, Y., Asrar, G. R., & Zhu, Z. (2021). 
The impacts of  urban expansion on climate and 
environment: A global review. Environmental Research 
Letters, 16(1), 1–20. 

Li, S. H., Jin, B. X., Wei, X. Y., Jiang, Y. Y., & Wang, 
J. L. (2015). Using CA-Markov model to model 
the spatiotemporal change of  land use/cover in 

Fuxian Lake for decision support. ISPRS Annals of  
Photogrammetry, Remote Sensing and Spatial Information 
Sciences, II-4/W2, 163–168. 

Liu, H., Zhang, F., Zhang, L., Lin, Y., Wang, S., & Xie, 
Y. (2020). UNVI-based time series for vegetation 
discrimination using separability analysis and random 
forest classification. Remote Sensing, 12(3), 529. 

Ma, X., Hong, Y., & Song, Y. (2020). Super resolution 
land cover mapping of  hyperspectral images using the 
deep image prior-based approach. International Journal 
of  Remote Sensing, 41, 2818–2834. 

Mishra, S., Shrivastava, P., & Dhurvey, P. (2017). Change 
detection techniques in remote sensing: A review. 
International Journal of  Wireless and Mobile Communication 
and Industrial Systems, 4, 1–8.

Mohamed, M. K., Adam, E., & Jackson, C. M. (2023). 
Policy review and regulatory challenges and strategies 
for the sustainable mangrove management in 
Zanzibar. Sustainability, 15, 1557. 

Mohamed, M. K., Adam, E., & Jackson, C. M. (2023). 
The spatial and temporal distribution of  mangrove 
forest cover from 1973 to 2020 in Chwaka Bay and 
Menai Bay, Zanzibar. Applied Sciences, 13, 7962. 

Mohamed, M. K., Adam, E., & Jackson, C. M. (2024). 
Assessing the perception and contribution of  
mangrove ecosystem services to the well-being of  
coastal communities of  Chwaka and Menai Bays, 
Zanzibar. Resources, 13(7). 

Mota, A., Curcino, N., Gioppo, F., Andrade, L., & 
Assunção, C. F. (2022). Digital image classification: 
A comparison of  classic methods for land cover and 
land use mapping. Anuário do Instituto de Geociências, 45, 
1–10.

Nalepa, J., Myller, M., Imai, Y., Honda, K., Takeda, T., 
& Antoniak, M. (2019). Unsupervised segmentation of  
hyperspectral images using 3D convolutional autoencoders. 
arXiv preprint, arXiv:1907.08870.

National Bureau of  Statistics Tanzania. (2022). 2022 
Population and Housing Census: Population distribution 
by administrative units. National Bureau of  Statistics. 
https ://www.nbs.g o. tz/uploads/sta t i s t ics/
documents/sw-1720088450-2022%20PHC%20
Initial%20Results%20-%20English.pdf

Nikolakopoulos, I. A., & Petropoulos, G. P. (2025). 
Obtaining a land use/cover cartography in a typical 
Mediterranean agricultural field combining unmanned 
aerial vehicle data with supervised classifiers. Land, 
14(3), 643. 

Petrisor, A.-I., Hamma, W., Nguyen, H. D., Randazzo, G., 
Muzirafuti, A., Stan, M.-I., Tran, V. T., Așțefănoaiei, 
R., Bui, Q.-T., Vintilă, D.-F., et al. (2020). Degradation 
of  coastlines under the pressure of  urbanization and 
tourism: Evidence on the change of  land systems 
from Europe, Asia, and Africa. Land, 9, 275. 

Pontius, R. G., & Cheuk, M. L. (2006). A generalized 
cross-tabulation matrix to compare soft-classified 
maps at multiple resolutions. International Journal of  
Geographical Information Science, 20(1), 1–30. 



Pa
ge

 
74

https://journals.e-palli.com/home/index.php/ajgt

Am. J. Geo Spat. Technol. 4(1) 60-74, 2025

Puplampu, D. A., & Boafo, Y. A. (2021). Exploring 
the impacts of  urban expansion on green spaces 
availability and delivery of  ecosystem services in the 
Accra Metropolis. Journal of  Environmental Challenges, 
5, 100283. 

Revolutionary Government of  Zanzibar. (2019). Zanzibar 
Strategic Urban Plan: Integrated planning for sustainable 
development. Ministry of  Lands, Housing, Water, and 
Energy, Zanzibar.

Revolutionary Government of  Zanzibar. (2020). 
Zanzibar Urban Development Policy: Promoting sustainable 
urban growth through infrastructure improvement, economic 
diversification, and environmental resilience. Ministry of  
Lands, Housing, Water, and Energy, Zanzibar.

Revolutionary Government of  Zanzibar. (2021). Zanzibar 
Environmental Policy: Enhancing environmental management 
and resilience. Ministry of  Agriculture, Natural 
Resources, Livestock, and Fisheries, Zanzibar.

Revolutionary Government of  Zanzibar. (2022). 
Zanzibar Vision 2050: Towards inclusive and sustainable 
development. President’s Office – Finance, Economy, 
and Development Planning, Zanzibar.

Roy, D. P., Wulder, M. A., Loveland, T. R., et al. (2014). 
Landsat-8: Science and product vision for terrestrial 
global change research. Remote Sensing of  Environment, 
145, 154–172. 

Sánchez-Llull, M., Castellanos Torres, L., Muñoz 
Caravaca, A., Martín Morales, G., Sauvage, S., 
Zulueta-Véliz, Y., Olalde Chang, E. J., León Cabrera, 
J., Vasallo-Rodríguez, L., Ouillon, S., & Sánchez-
Pérez, J.-M. (2025). Assessing classification accuracy 
as a criterion for evaluating the performance of  seven 
topographic correction algorithms in the Trinidad 
Mountains, Cuba. Remote Sensing, 17(6), 1032. 

Sassen, S. (2021). Cities in a world economy (5th ed.). Sage 
Publications.

Satterthwaite, D., Archer, D., & Dodman, D. (2020). 
Addressing urban inequalities: Policies and 
implications. Current Urban Studies, 8(4), 403–418. 

Seto, K. C., Güneralp, B., & Hutyra, L. R. (2020). Global 
urban expansion: A multidimensional approach. 
Annual Review of  Environment and Resources, 45, 79–107. 

Seto, K., Reenberg, A., Boone, C., & Fragkias, M. (2012). 
Urban land teleconnections and sustainability. 
Proceedings of  the National Academy of  Sciences, 109(29), 
13916–13921. 

Shao, Z., Cheng, T., Fu, H., Li, D., & Huang, X. (2023). 
Emerging issues in mapping urban impervious 
surfaces using high-resolution remote sensing images. 
Remote Sensing, 15, 2562. 

Tessema, W. M., & Abebe, G. B. (2023). Quantification 
of  land use/land cover dynamics and urban growth 
in rapidly urbanized countries: The case of  Hawassa 
City, Ethiopia. Urban Planning and Transport Research, 

11, 1. 
Tew, Y. L., Tan, M. L., Samat, N., & Yang, X. (2019). Urban 

expansion analysis using Landsat images in Penang, 
Malaysia. Sains Malaysiana, 48(11), 2307–2315. 

Tossoukpe, A. Y., Dukiya, J., Folega, F., Thiel, M., & 
Okhimamhe, A. A. (2025). Dynamics of  changes in 
spatial patterns of  built-up areas in two metropolitan 
areas of  Grand Lomé and Greater Accra (West 
Africa). Urban Science, 9(3), 84. 

Tsurusaki, N., & Salem, M. (2024). Impacts of  rapid 
urban expansion on peri-urban landscapes in the 
global south: Insights from landscape metrics in 
Greater Cairo. Sustainability, 16, 2316. https://doi.
org/10.xxxx/sus16

UN-Habitat. (2020). The world’s cities in 2020: Data 
booklet. United Nations.

United Nations. (2019). World urbanization prospects: The 
2018 revision. United Nations.

U.S. Geological Survey. (2021). Landsat collection 2 
level-1 and level-2 science products. U.S. Department 
of  the Interior. Retrieved September 23, 2024, from 
https://www.usgs.gov/landsat-missions/landsat-
collection-2

Vali, A., Comai, S., & Matteucci, M. (2020). Deep learning 
for land use and land cover classification based on 
hyperspectral and multispectral Earth observation 
data: A review. Remote Sensing, 12(15), 2495. 

Valero-Jorge, A., González-De Zayas, R., Matos-Pupo, F., 
Becerra-González, A. L., Álvarez-Taboada, F. (2024). 
Mapping and monitoring of  the invasive species 
Dichrostachys cinerea (Marabú) in central Cuba using 
Landsat imagery and machine learning (1994–2022). 
Remote Sensing, 16, 20. 

World Bank. (2021). Transforming Tanzania’s cities: Harnessing 
urbanization for competitiveness, resilience, and livability. The 
World Bank.

Zanotta, D. C., Zortea, M., & Ferreira, M. P. (2018). A 
supervised approach for simultaneous segmentation 
and classification of  remote sensing images. ISPRS 
Journal of  Photogrammetry and Remote Sensing, 142, 162–
173. 

Zhang, C., Wang, Q., & Atkinson, P. M. (2025). 
Unsupervised object-based spectral unmixing for 
subpixel mapping. Remote Sensing of  Environment, 318, 
114514. 

Zhou, Y., Li, X., Asrar, G. R., Smith, S. J., & Imhoff, M. 
(2018). A global record of  annual urban dynamics 
(1992–2013) from nighttime lights. Remote Sensing of  
Environment, 219, 206–220. 

Wang, M., Cai, L., Xu, H., & Zhao, S. (2019). Predicting 
land use changes in northern China using logistic 
regression, cellular automata, and a Markov model. 
Arabian Journal of  Geosciences, 12, 790. 


