Bilingual
En/Fa
Geographical Research is Published in both Persian and English Full-text.
Volume 38, Issue 2 (2023)
GeoRes 2023, 38(2): 181-190 |
Back to browse issues page
Article Type:
Subject:
History
Received: 2022/03/6 | Accepted: 2023/05/27 | Published: 2023/06/5
Received: 2022/03/6 | Accepted: 2023/05/27 | Published: 2023/06/5
How to cite this article
Habibi Razi A, Azizi Z. Object-based and pixel-based methods in land cover changes detection using Landsat and Worldview imagery (case study: West of Tehran). GeoRes 2023; 38 (2) :181-190
URL: http://georesearch.ir/article-1-1303-en.html
URL: http://georesearch.ir/article-1-1303-en.html
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Rights and permissions
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
1- Department of Remote Sensing and GIS, Faculty of Natural Resource and Environment, Science and Research Branch, Islamic Azad University, Tehran, Iran
Full-Text (HTML) (42 Views)
Introduction
One of the most significant advantages of satellites for Earth observation is the classification of changes and monitoring [Jovanovic et al., 2015]. The use of remote sensing data, due to its provision of up-to-date information, repetitive coverage, and cost-effectiveness, holds a prominent position in the assessment of natural resources. Moreover, change detection is one of the essential requirements in land management and evaluation. Accordingly, the extent of land use and land cover changes, which results from the change detection process, can be estimated using multi-temporal remote sensing imagery [Rezaei Moghaddam et al., 2015; Mirzapour et al., 2023].
The 21st century is an era of continuous and dynamic transformations in land management. Urbanization is rapidly advancing and utilizing unmanaged lands on a large scale. Sustainable management of land, environmental, and human resources requires a comprehensive understanding of land use structures and land use changes, particularly in urban lands under intense human pressure [Szarek-Iwaniuk, 2021]. Today, urban growth is a multidimensional spatial and demographic process in which cities and urban settlements become centers of population concentration due to their specific economic and social characteristics, which are an integral part of the development of human societies [Dadras et al., 2015].
Urban land use planning is a set of purposeful activities that organize the built environment and provide, to the extent possible, the needs and demands of urban communities in land utilization. Based on this, urban lands are generally divided into two major categories: developed and undeveloped lands. Developed lands refer to areas that have undergone construction, whereas undeveloped or vacant lands are those without a specific use [Poormohammadi, 2016]. Ultimately, if land cover changes occur rapidly and without planning, they disrupt environmental balance and impose irreversible consequences on natural resources. As a result, ecosystems face severe challenges, and natural resources are exposed to significant risks.
Among the most important methods for identifying, determining, and detecting land cover boundaries, as well as monitoring changes in recent years, is the use of remote sensing satellite imagery and geographic data. Land use/land cover classification methods are employed to better understand essential programs such as urban and suburban land use planning during city growth [Namdar et al., 2018], agricultural practices and cropping patterns [Alharthi et al., 2020], forest management, natural resource conservation, afforestation development, and deforestation reduction [Azizi et al., 2010; Alawamy et al., 2020], water body management and monitoring [Hamzeh & Torabi, 2021; Mafi et al., 2021; Ali Bakhshi et al., 2020], drought and desertification monitoring [Kamali Maskooni et al., 2019], environmental pollution assessment [Yousefi Kebria et al., 2020; Mokhtari et al., 2017], coastal change monitoring [Abdollahi et al., 2019], as well as soil erosion, salinity, and moisture monitoring [Ozsahin et al., 2020]. These studies contribute to a better understanding of urban and agricultural patterns, forest management, and other processes for the future management of land cover and land use, while also enabling decision-makers to accurately comprehend land cover utilization, urban development, and agricultural characteristics.
In this regard, several studies have been conducted; however, none have employed a combined use of Landsat 7 and WorldView-2 images. The findings of Shaterian et al., who monitored land use changes in Shahrekord County using Landsat imagery, revealed increases in urban, agricultural, and industrial land uses, mainly due to land degradation caused by airport construction and the destruction of rangelands to expand urban and agricultural areas [Shaterian et al., 2019]. Javaheri and Tarahi, in their article, utilized Landsat 5 and 7 imagery to detect land use changes over the past 35 years. Their results indicated a noticeable reduction in forested and dense rangelands across three time intervals, where the area of dense rangelands increased until 2000 but subsequently decreased until 2019, while residential areas, water bodies, and orchards expanded [Javaheri & Tarahi, 2021]. An evaluation of three algorithms, Support Vector Machine, Convolutional Neural Network, and Random Forest, demonstrated the efficiency of all three approaches, with accuracies ranging between 70% and 90%. However, the Convolutional Neural Network algorithm, with an overall accuracy of 90%, outperformed the other two methods in image classification [Xie et al., 2021]. In another study, Jaber Hussein et al. aim to identify four major land use classes, urban areas, barren lands, vegetation, and roads, using QuickBird images and both object-based and maximum likelihood algorithms. The results showed that the object-based approach performed better compared to the maximum likelihood method [Jaber Hussein et al., 2022].
The present study was conducted with the aim of identifying land cover types and evaluating object-based and pixel-based methods using the Maximum Likelihood and Support Vector Machine algorithms in parts of Districts 21 and 22 of the Tehran metropolis.
Methodology
This remote sensing study, based on analytical and classification methods, was conducted in parts of Districts 21 and 22 of Tehran, located in the southern foothills of the Alborz Mountains (35°15′15″ to 35°45′41″ N latitude and 51°10′34″ to 51°15′15″ E longitude). Fieldwork and land use surveys in the study area, due to its extent, required up-to-date and comprehensive information on the status of different locations. This was achieved through a combination of literature review (Tehran Statistical Yearbook, 2020), systematic random field visits, and the use of Google Earth software for the study area.
Image Preprocessing
To identify and extract land use/land cover types, Landsat 7 imagery from 2000 and WorldView-2 imagery from 2020 were used. For Landsat 7 images, preprocessing included radiometric correction, atmospheric correction using the Quick Atmospheric Correction (QUAC) method, and enhancement of spatial resolution using the Gram-Schmidt method. For WorldView-2 images, radiometric and atmospheric corrections were also applied using QUAC. For extracting land cover information from satellite imagery, both pixel-based and object-based classification algorithms were applied, specifically the Maximum Likelihood (ML) and Support Vector Machine (SVM) methods.
Maximum Likelihood Algorithm
Classification methods are generally divided into supervised and unsupervised approaches. Supervised methods require prior knowledge, such as the number of classes, their characteristics, and training samples from each class, while unsupervised methods operate automatically based on pixel values without prior information. The Maximum Likelihood algorithm is one of the most common supervised pixel-based classification methods. It is more accurate than many other pixel-based approaches but requires more computational time. This algorithm can be applied to both multispectral and hyperspectral images. In this method, the algorithm estimates statistical parameters from training data and assigns each pixel to the class with the highest probability.
Support Vector Machine Algorithm
Object-based classification considers groups of pixels as homogeneous objects, extracting spatial and spectral features for classification. It involves segmentation of the image into objects and assigning all pixels within an object to a single class based on both spectral and spatial characteristics, such as shape and size. Object-based classification offers several advantages, including higher accuracy, faster processing, and reduced memory requirements compared to pixel-based methods. The Support Vector Machine algorithm is a supervised object-based classifier that separates classes by identifying an optimal decision boundary based on training data, ensuring the highest accuracy in classification.
Accuracy Assessment
The classification results from both methods were evaluated using reference ground truth points. Overall accuracy, which measures the percentage of correctly classified pixels, and the Kappa coefficient, which evaluates classification accuracy relative to a random classification, were used as metrics. For each land cover type, barren lands, urban areas, vegetation, and water bodies, a set of training points was selected (25, 26, 25, and 10 points, respectively), and error matrices were prepared to assess classification performance
Findings
The number of training samples and training pixels in the WorldView-2 image was higher than in the Landsat 7 image due to its superior spatial resolution. Subsequently, satellite images from different years were classified, and the classification outputs, along with error matrices, overall accuracy, and Kappa coefficients, were extracted.
According to the evaluation of classification results, the WorldView-2 image using the Support Vector Machine (SVM) algorithm demonstrated the highest accuracy. Classification of the WorldView-2 image using the Maximum Likelihood method exhibited the highest level of noise due to the high spatial resolution of the image and the presence of shadows around buildings and vegetation. Additionally, the lower accuracy and Kappa coefficient observed reflect the significant real-world changes in land cover over the study period. Differences in results between the two images and methods are primarily attributable to the long 20-year time interval.
Overall accuracy in 2000 and 2020 for the SVM algorithm using WorldView-2 imagery was 86.05%, whereas for the Maximum Likelihood algorithm using Landsat 7 imagery it was 53.98%. Similarly, the Kappa coefficient for the SVM algorithm with WorldView-2 images was 0.807, while for the Maximum Likelihood algorithm with Landsat 7 images it was 0.396. As observed, the higher spatial resolution of WorldView-2 contributed to classification accuracy sufficient to rely on its results.
Analysis of land cover changes over the 20-year period indicated a decrease of 9.61 km² in barren lands, a reduction of 5.13 km² in vegetation cover, an increase of 13.13 km² in built-up areas, and an increase of approximately 1 km² in water bodies due to the construction of Chitgar Lake.
Discussion
The aim of the present study was to evaluate the efficiency of satellite imagery in monitoring and assessing the spatiotemporal changes of land cover in the Tehran metropolitan area. A precise understanding of land use characteristics and structure is essential for studying their impacts on human life and the environment, and it is a critical requirement for planning and decision-making by authorities. Urban development represents a major form of land extraction and conversion, as it is associated with population growth, accessibility to health services, urban facilities, and economic factors. Analysis of urban growth using spatial data and historical and current features is considered a fundamental requirement of urban geographic studies, future planning, and the formulation of general policies for urban development [Dadras et al., 2015]. Accordingly, monitoring land cover is an integral part of land management. Many stakeholders, including landowners, governments, NGOs, and occasionally international organizations, are interested in tracking urban changes to manage them effectively. This process can be achieved using land cover maps, although most of these maps are produced for a single time point [Masiliunas et al., 2019].
One approach to representing land cover change is the production of multiple single-date maps and the analysis of differences between them. However, if the probability of a pixel belonging to a particular class is similar between two classes, even minor changes in the input data may lead to a different land cover classification. Urban development, as mentioned, is a major form of land conversion due to population increase and the availability of urban services, and analyzing urban growth through spatial data from past and present remains essential for urban geographic studies and policy development [Dadras et al., 2015].
The object-based algorithm proved to be the most suitable method for extracting the four main land use classes, barren land, urban areas, roads, and vegetation and showed higher feature extraction capability for identifying urban areas and buildings [Jabar et al., 2021]. The results of this study indicated that the object-based method achieves high accuracy in detecting urban areas. Roostaei et al. have shown that the Support Vector Machine (SVM) algorithm has higher accuracy than the Maximum Likelihood method [Roostaei et al., 2019]. Urban structures, including buildings, shopping centers, and recreational parks, have been among the primary factors driving land use changes [Ragabi, 2019]. The presence of Chitgar Park and proximity to industrial areas in western Tehran may have contributed to land use changes, ultimately leading to increased physical urban growth, local population expansion, and migration to cities [Rahnama et al., 2022]. Additionally, previous studies have confirmed that human activities are among the most significant drivers of land use changes, particularly in urban lands, in the study area [Mohammadpoor et al., 2022; Khedmatzade et al., 2019].
Barren lands, vegetated areas, and urban lands were sensitive to SVM parameters. Nevertheless, further research is required to confirm the general applicability and transferability of this approach. Similar studies using Landsat images have evaluated the performance of various machine learning algorithms [Li et al., 2018], while analyses with Sentinel-2 imagery have shown that Convolutional Neural Networks outperform traditional machine learning methods [Wang et al., 2019]. The importance of integrating multi-source data and machine learning techniques for accurate and efficient land cover analysis has also been emphasized in studies by Zeo et al. and Zhang et al. [Zeo et al., 2020; Zhang et al., 2021].
Overall, the SVM algorithm is recommended as an efficient and precise approach for classifying satellite imagery to produce land use maps, although parameter optimization requires extensive knowledge and repeated experimentation with different settings. Chen et al. have also confirmed the effectiveness of advanced remote sensing methods combined with learning techniques for detecting land cover changes using multi-temporal images. These methods provide higher accuracy than traditional change detection approaches, especially in complex landscapes with mixed urban land cover [Chen et al., 2023].
In summary, these studies highlight the importance of advanced remote sensing technologies and machine learning techniques for accurate and efficient land cover analysis, which can have significant implications for environmental management and decision-making processes.
Conclusion
Based on overall accuracy and the Kappa coefficient, the Support Vector Machine (SVM) algorithm demonstrated superior performance compared to the Maximum Likelihood algorithm. This is because SVM considers not only spectral reflectance but also the shape, size, and texture across bands, whereas the Maximum Likelihood algorithm exhibits higher levels of interference and misclassification
Acknowledgments: No acknowledgments were declared by the authors.
Ethical Approval: No ethical approval issues were reported by the authors.
Conflict of Interest: No conflicts of interest were reported by the authors.
Authors’ Contributions: Habibi Razi A (First Author): Methodologist/Principal Researcher/Discussion Writer/Data Analyst (65%); Azizi Z (Second Author): Assistant Researcher/Introduction writer/Data Analyst (35%).
Funding: This study is derived from the master’s thesis of Alireza Habibi Rezaei, supervised by the second author. All costs were covered by the student, and no institution or organization provided financial support
One of the most significant advantages of satellites for Earth observation is the classification of changes and monitoring [Jovanovic et al., 2015]. The use of remote sensing data, due to its provision of up-to-date information, repetitive coverage, and cost-effectiveness, holds a prominent position in the assessment of natural resources. Moreover, change detection is one of the essential requirements in land management and evaluation. Accordingly, the extent of land use and land cover changes, which results from the change detection process, can be estimated using multi-temporal remote sensing imagery [Rezaei Moghaddam et al., 2015; Mirzapour et al., 2023].
The 21st century is an era of continuous and dynamic transformations in land management. Urbanization is rapidly advancing and utilizing unmanaged lands on a large scale. Sustainable management of land, environmental, and human resources requires a comprehensive understanding of land use structures and land use changes, particularly in urban lands under intense human pressure [Szarek-Iwaniuk, 2021]. Today, urban growth is a multidimensional spatial and demographic process in which cities and urban settlements become centers of population concentration due to their specific economic and social characteristics, which are an integral part of the development of human societies [Dadras et al., 2015].
Urban land use planning is a set of purposeful activities that organize the built environment and provide, to the extent possible, the needs and demands of urban communities in land utilization. Based on this, urban lands are generally divided into two major categories: developed and undeveloped lands. Developed lands refer to areas that have undergone construction, whereas undeveloped or vacant lands are those without a specific use [Poormohammadi, 2016]. Ultimately, if land cover changes occur rapidly and without planning, they disrupt environmental balance and impose irreversible consequences on natural resources. As a result, ecosystems face severe challenges, and natural resources are exposed to significant risks.
Among the most important methods for identifying, determining, and detecting land cover boundaries, as well as monitoring changes in recent years, is the use of remote sensing satellite imagery and geographic data. Land use/land cover classification methods are employed to better understand essential programs such as urban and suburban land use planning during city growth [Namdar et al., 2018], agricultural practices and cropping patterns [Alharthi et al., 2020], forest management, natural resource conservation, afforestation development, and deforestation reduction [Azizi et al., 2010; Alawamy et al., 2020], water body management and monitoring [Hamzeh & Torabi, 2021; Mafi et al., 2021; Ali Bakhshi et al., 2020], drought and desertification monitoring [Kamali Maskooni et al., 2019], environmental pollution assessment [Yousefi Kebria et al., 2020; Mokhtari et al., 2017], coastal change monitoring [Abdollahi et al., 2019], as well as soil erosion, salinity, and moisture monitoring [Ozsahin et al., 2020]. These studies contribute to a better understanding of urban and agricultural patterns, forest management, and other processes for the future management of land cover and land use, while also enabling decision-makers to accurately comprehend land cover utilization, urban development, and agricultural characteristics.
In this regard, several studies have been conducted; however, none have employed a combined use of Landsat 7 and WorldView-2 images. The findings of Shaterian et al., who monitored land use changes in Shahrekord County using Landsat imagery, revealed increases in urban, agricultural, and industrial land uses, mainly due to land degradation caused by airport construction and the destruction of rangelands to expand urban and agricultural areas [Shaterian et al., 2019]. Javaheri and Tarahi, in their article, utilized Landsat 5 and 7 imagery to detect land use changes over the past 35 years. Their results indicated a noticeable reduction in forested and dense rangelands across three time intervals, where the area of dense rangelands increased until 2000 but subsequently decreased until 2019, while residential areas, water bodies, and orchards expanded [Javaheri & Tarahi, 2021]. An evaluation of three algorithms, Support Vector Machine, Convolutional Neural Network, and Random Forest, demonstrated the efficiency of all three approaches, with accuracies ranging between 70% and 90%. However, the Convolutional Neural Network algorithm, with an overall accuracy of 90%, outperformed the other two methods in image classification [Xie et al., 2021]. In another study, Jaber Hussein et al. aim to identify four major land use classes, urban areas, barren lands, vegetation, and roads, using QuickBird images and both object-based and maximum likelihood algorithms. The results showed that the object-based approach performed better compared to the maximum likelihood method [Jaber Hussein et al., 2022].
The present study was conducted with the aim of identifying land cover types and evaluating object-based and pixel-based methods using the Maximum Likelihood and Support Vector Machine algorithms in parts of Districts 21 and 22 of the Tehran metropolis.
Methodology
This remote sensing study, based on analytical and classification methods, was conducted in parts of Districts 21 and 22 of Tehran, located in the southern foothills of the Alborz Mountains (35°15′15″ to 35°45′41″ N latitude and 51°10′34″ to 51°15′15″ E longitude). Fieldwork and land use surveys in the study area, due to its extent, required up-to-date and comprehensive information on the status of different locations. This was achieved through a combination of literature review (Tehran Statistical Yearbook, 2020), systematic random field visits, and the use of Google Earth software for the study area.
Image Preprocessing
To identify and extract land use/land cover types, Landsat 7 imagery from 2000 and WorldView-2 imagery from 2020 were used. For Landsat 7 images, preprocessing included radiometric correction, atmospheric correction using the Quick Atmospheric Correction (QUAC) method, and enhancement of spatial resolution using the Gram-Schmidt method. For WorldView-2 images, radiometric and atmospheric corrections were also applied using QUAC. For extracting land cover information from satellite imagery, both pixel-based and object-based classification algorithms were applied, specifically the Maximum Likelihood (ML) and Support Vector Machine (SVM) methods.
Maximum Likelihood Algorithm
Classification methods are generally divided into supervised and unsupervised approaches. Supervised methods require prior knowledge, such as the number of classes, their characteristics, and training samples from each class, while unsupervised methods operate automatically based on pixel values without prior information. The Maximum Likelihood algorithm is one of the most common supervised pixel-based classification methods. It is more accurate than many other pixel-based approaches but requires more computational time. This algorithm can be applied to both multispectral and hyperspectral images. In this method, the algorithm estimates statistical parameters from training data and assigns each pixel to the class with the highest probability.
Support Vector Machine Algorithm
Object-based classification considers groups of pixels as homogeneous objects, extracting spatial and spectral features for classification. It involves segmentation of the image into objects and assigning all pixels within an object to a single class based on both spectral and spatial characteristics, such as shape and size. Object-based classification offers several advantages, including higher accuracy, faster processing, and reduced memory requirements compared to pixel-based methods. The Support Vector Machine algorithm is a supervised object-based classifier that separates classes by identifying an optimal decision boundary based on training data, ensuring the highest accuracy in classification.
Accuracy Assessment
The classification results from both methods were evaluated using reference ground truth points. Overall accuracy, which measures the percentage of correctly classified pixels, and the Kappa coefficient, which evaluates classification accuracy relative to a random classification, were used as metrics. For each land cover type, barren lands, urban areas, vegetation, and water bodies, a set of training points was selected (25, 26, 25, and 10 points, respectively), and error matrices were prepared to assess classification performance
Findings
The number of training samples and training pixels in the WorldView-2 image was higher than in the Landsat 7 image due to its superior spatial resolution. Subsequently, satellite images from different years were classified, and the classification outputs, along with error matrices, overall accuracy, and Kappa coefficients, were extracted.
According to the evaluation of classification results, the WorldView-2 image using the Support Vector Machine (SVM) algorithm demonstrated the highest accuracy. Classification of the WorldView-2 image using the Maximum Likelihood method exhibited the highest level of noise due to the high spatial resolution of the image and the presence of shadows around buildings and vegetation. Additionally, the lower accuracy and Kappa coefficient observed reflect the significant real-world changes in land cover over the study period. Differences in results between the two images and methods are primarily attributable to the long 20-year time interval.
Overall accuracy in 2000 and 2020 for the SVM algorithm using WorldView-2 imagery was 86.05%, whereas for the Maximum Likelihood algorithm using Landsat 7 imagery it was 53.98%. Similarly, the Kappa coefficient for the SVM algorithm with WorldView-2 images was 0.807, while for the Maximum Likelihood algorithm with Landsat 7 images it was 0.396. As observed, the higher spatial resolution of WorldView-2 contributed to classification accuracy sufficient to rely on its results.
Analysis of land cover changes over the 20-year period indicated a decrease of 9.61 km² in barren lands, a reduction of 5.13 km² in vegetation cover, an increase of 13.13 km² in built-up areas, and an increase of approximately 1 km² in water bodies due to the construction of Chitgar Lake.
Discussion
The aim of the present study was to evaluate the efficiency of satellite imagery in monitoring and assessing the spatiotemporal changes of land cover in the Tehran metropolitan area. A precise understanding of land use characteristics and structure is essential for studying their impacts on human life and the environment, and it is a critical requirement for planning and decision-making by authorities. Urban development represents a major form of land extraction and conversion, as it is associated with population growth, accessibility to health services, urban facilities, and economic factors. Analysis of urban growth using spatial data and historical and current features is considered a fundamental requirement of urban geographic studies, future planning, and the formulation of general policies for urban development [Dadras et al., 2015]. Accordingly, monitoring land cover is an integral part of land management. Many stakeholders, including landowners, governments, NGOs, and occasionally international organizations, are interested in tracking urban changes to manage them effectively. This process can be achieved using land cover maps, although most of these maps are produced for a single time point [Masiliunas et al., 2019].
One approach to representing land cover change is the production of multiple single-date maps and the analysis of differences between them. However, if the probability of a pixel belonging to a particular class is similar between two classes, even minor changes in the input data may lead to a different land cover classification. Urban development, as mentioned, is a major form of land conversion due to population increase and the availability of urban services, and analyzing urban growth through spatial data from past and present remains essential for urban geographic studies and policy development [Dadras et al., 2015].
The object-based algorithm proved to be the most suitable method for extracting the four main land use classes, barren land, urban areas, roads, and vegetation and showed higher feature extraction capability for identifying urban areas and buildings [Jabar et al., 2021]. The results of this study indicated that the object-based method achieves high accuracy in detecting urban areas. Roostaei et al. have shown that the Support Vector Machine (SVM) algorithm has higher accuracy than the Maximum Likelihood method [Roostaei et al., 2019]. Urban structures, including buildings, shopping centers, and recreational parks, have been among the primary factors driving land use changes [Ragabi, 2019]. The presence of Chitgar Park and proximity to industrial areas in western Tehran may have contributed to land use changes, ultimately leading to increased physical urban growth, local population expansion, and migration to cities [Rahnama et al., 2022]. Additionally, previous studies have confirmed that human activities are among the most significant drivers of land use changes, particularly in urban lands, in the study area [Mohammadpoor et al., 2022; Khedmatzade et al., 2019].
Barren lands, vegetated areas, and urban lands were sensitive to SVM parameters. Nevertheless, further research is required to confirm the general applicability and transferability of this approach. Similar studies using Landsat images have evaluated the performance of various machine learning algorithms [Li et al., 2018], while analyses with Sentinel-2 imagery have shown that Convolutional Neural Networks outperform traditional machine learning methods [Wang et al., 2019]. The importance of integrating multi-source data and machine learning techniques for accurate and efficient land cover analysis has also been emphasized in studies by Zeo et al. and Zhang et al. [Zeo et al., 2020; Zhang et al., 2021].
Overall, the SVM algorithm is recommended as an efficient and precise approach for classifying satellite imagery to produce land use maps, although parameter optimization requires extensive knowledge and repeated experimentation with different settings. Chen et al. have also confirmed the effectiveness of advanced remote sensing methods combined with learning techniques for detecting land cover changes using multi-temporal images. These methods provide higher accuracy than traditional change detection approaches, especially in complex landscapes with mixed urban land cover [Chen et al., 2023].
In summary, these studies highlight the importance of advanced remote sensing technologies and machine learning techniques for accurate and efficient land cover analysis, which can have significant implications for environmental management and decision-making processes.
Conclusion
Based on overall accuracy and the Kappa coefficient, the Support Vector Machine (SVM) algorithm demonstrated superior performance compared to the Maximum Likelihood algorithm. This is because SVM considers not only spectral reflectance but also the shape, size, and texture across bands, whereas the Maximum Likelihood algorithm exhibits higher levels of interference and misclassification
Acknowledgments: No acknowledgments were declared by the authors.
Ethical Approval: No ethical approval issues were reported by the authors.
Conflict of Interest: No conflicts of interest were reported by the authors.
Authors’ Contributions: Habibi Razi A (First Author): Methodologist/Principal Researcher/Discussion Writer/Data Analyst (65%); Azizi Z (Second Author): Assistant Researcher/Introduction writer/Data Analyst (35%).
Funding: This study is derived from the master’s thesis of Alireza Habibi Rezaei, supervised by the second author. All costs were covered by the student, and no institution or organization provided financial support
References
1. Abdollahi A, Pourkhosravani M, Afsari F, Madadizadeh F (2018). Monitoring coastal changes using remote sensing (case study: Kish coastal). Geography and Environmental Planning. 30(75):1-16. [Persian] [Link]
2. Alharthi A, El-Sheikh MA, Elhag M, Alatar AA, Abbadi GA, Abdel-Salam EM, et al. (2020). Remote sensing of 10 years changes in the vegetation cover of the northwestern coastal land of Red Sea, Saudi Arabia. Saudi Journal of. Biological Sciences. 27(11):3169-3179. [Link] [DOI:10.1016/j.sjbs.2020.07.021]
3. Ali Bakhshi T, Azizi Z, Vafaeinejad A, Aghamohammadi Zanjirabadi H (2020). Survey of area changes in water basins of Shahid Abbaspour Dam caused by 2019 floods using Google Earth engine. Iranian Journal of Ecohydrology. 7(2):345-357. [Persian] [Link]
4. Arrechea-Castillo DA, Solano-Correa YT, Muñoz-Ordóñez JF, Pencue-Fierro EL Figueroa-Casas A (2023). Multiclass land use and land cover classification of Andean Sub-Basins in Colombia with Sentinel-2 and Deep Learning. Remote Sensing. 15(10):2521. [Link] [DOI:10.3390/rs15102521]
5. Azizi Z, Najafi A, Fatehi P, Pirbavaghar M (2010). Forest stand volume estimation using satellite IRS_P6 (LISS_IV) data (case study: Lirehsar, Tonekabon). Iranian Journal of Forest and Poplar Research. 18(1):143-151. [Persain] [Link]
6. Dadras M, Helmi ZM, Shafri NA, Biswajeet P, Safarpour S (2015). Spatio-temporal analysis of urban growth from remote sensing data in Bandar Abbas city, Iran. The Egyptian Journal of Remote Sensing and Space Science. 18(1):35-52. [Link] [DOI:10.1016/j.ejrs.2015.03.005]
7. Emami H, Hasani M, Mirzaei F (2019). Training of Envi software functions in remote sensing and spatial information system engineering. Tabriz: Asr Zendigi [Persian] [Link]
8. Faizizadeh B (2016). Comparison of pixel-based and object-oriented methods in the preparation of land use maps (case study: Eastern plain of Lake Urmia) [dissertation]. Tabriz: University of Tabriz. [Persian] [Link]
9. Fatemi SB, Rezaei Y (2016). Principles of remote sensing. Tehran: Azadeh publication. [Persian] [Link]
10. Ghayour L, Neshat A, Paryani S, Shahabi H, Shirzadi A, Chen W, et al. (2021). Performance evaluation of Sentinel-2 and Landsat 8 OLI data for land cover/use classification using a comparison between machine learning algorithms. Remote Sensing. 13(7):Unknown Pages. [Link] [DOI:10.3390/rs13071349]
11. Hamzeh S, Torabi O (2021). Investigation the changes in the water area of Gorgan Bay and its relationship with precipitation and water level of the Caspian Sea by using remote sensing data. Iranian Journal of Ecohydrology. 8(2):475-484. [Persian] [Link]
12. Jaber Hussein S, Shareef MA, Merzah ZF (2022). Object-based approaches for land use-land cover classification using high resolution quick bird satellite imagery (a case study: Kerbela, Iraq). Geodesy and Cartography. 48(2):85-91. [Link] [DOI:10.3846/gac.2022.14453]
13. Javaheri S, Tarahi A (2021). Detection land use changes using satellite images during the period 1984-2019 (case study of Kamiyaran city). Journal of GIS and RS Application in Planning. 12(1):41-56. [Persian] [Link]
14. Jovanovic D, Govedarica M, Sabo F, Bugarinović Ž, Novović O, Beker T, et al. (2015). Land cover change detection by using remote sensing-A case study of Zlatibor (Serbia). Geographica Pannonica. 19(4):162-173. [Link] [DOI:10.5937/GeoPan1504162J]
15. Kamali Maskooni E, Kamali MA, Khanamani A (2019). Investigation and preparation of desertification map based on the Iranian model of desertification potential (IMDPA) with an emphasis on two criteria of soil and vegetation (case study: Faryab-Kerman province). Journal of Environmental Science and Technology. 22(12):163-178. [Persian] [Link]
16. Khedmatzadeh A, Feizizade B (2022). Assessing urban quality of life using Remote Sensing and GIS (case study: Urmia urban region). Journal of Applied Researches in Geographical Sciences. 22(67):259-279. [Persian] [Link] [DOI:10.52547/jgs.22.67.259]
17. Mafi M, Azizi Z, Karimi P, Alemi Safaval P (2021). Investigating the trend of water level changes in Allahabad wetland by using temporal images. Iranian Journal of Ecohydrology. 8(2):321-329. [Link]
18. Management and Planning Organization of Tehran (2021). Statistical yearbook of Tehran, 2019. Tehran: Planning and Budget Organization Publishing. [Persian] [Link]
19. Masiliunas D, Tsendbazar NE, Herold M, Lesiv M, Verbesselt J (2019). Change detection in satellite image time series for continuous land cover map updating. Proceedings of ESA Living Planet Symposium; 2019 May 13; Italy: Milan. [Link]
20. Mirzapour S, Kheirkhah Zarkash MM, Azizi Z (2023). Investigation and forecasting of the drought situation in the Zagros region based on remote sensing indicators and general circulation models of the 20-year period. Iranian Journal of Forest and Range Protection Research. 20(2):317-335. [Link]
21. Mokhtari A, Azizi Z, Rabiaee Fradonbeh S (2017). Epidemiological study and spatial modeling of peste des petits ruminants (PPR) in central area of Iran. Revista MVZ Córdoba. 22(2):5899-5909. [Link] [DOI:10.21897/rmvz.1026]
22. Namdar F, Mahmoudi S, Esmaliouri A, Pazira E (2018). Monitoring of thirty-year land cover changes using remote sensing and GIS (case study: Qaresu watershed, Ardabil). Journal of Environmental Science and Technology. 22(12)):179-190. [Persian] [Link]
23. Ozsahin E, Duru O, Eroglu I (2018). Land use and land cover changes (LULCC), a key to understand soil erosion intensities in the Maritsa Basin. Water. 10(3):Unknown Pages. [Link] [DOI:10.3390/w10030335]
24. Poormohammadi MR (2016). Urban land use planning. Tehran: SAMT Publications. [Persian] [Link]
25. Rahnama MR, Bolori Z (2022). Analysis of the growth effects of Amol City on The surrounding environment using Remote Sensing and Geographic Information System. Political Organizing of Space. 4(4):288-302. [Persian] [Link]
26. Rajabi A (2019). Effects of dam construction on physical development and upstream and downstream dam change using remote sensing (case Study: 15 Khordad dam, Qom). Journal of GIS & RS Application in Planning. 10(4):27-46. [Persian] [Link]
27. Rezaei Moghaddam MR, Valizadeh Kamran Kh, Andaryani S, Almaspoor F (2015). Comparison of ANN and SVM methods in extraction land use/land cover maps from Landsat 8 satellite images (case study: Sufi Chay Basin). Journal of Geography and Planning. 19(52):163-183. [Persian] [Link]
28. Richards JA (1995). Remote sensing digital image analysis, an introduction. 2nd Edition. Heidelberg: Springer. [Link]
29. Roostaei S, Mokhtari D, Valizadeh Kamran K, Khodaei Geshlag L (2019). Comparison of pixel-based algorithm (maximum likelihood) and object-based methods (Support Vector Machine) in classification of land use (Ahar-Varzeghan area). Quantitative Geomorphological Research. 8(1):118-129. [Persian] [Link]
30. Shaterian M, Mousavi SH, Momenbeik Z (2019). Application of remote sensing data in detection of urban land use changes (case study: Shahr-e Kord). Scientific-Research Quarterly of Geographical Data (SEPEHR). 28(111):235-250. [Persian] [Link]
31. Szarek-Iwaniuk P (2021). A comparative analysis of spatial data and land use/land cover classification in urbanized areas and areas subjected to anthropogenic pressure for the example of Poland. Sustainability. 13(6):Unknown Pages. [Link] [DOI:10.3390/su13063070]
32. Theodoridis S, Koutroumbas K (2003). Pattern recognition. 2nd Edition. Massachusettes: Elsevier Academic press. [Link]
33. Xie G, Niculescu S (2021). Mapping and monitoring of land cover/land use (LCLU) changes in the Crozon Peninsula (Brittany, France) from 2007 to 2018 by machine learning algorithms (Support Vector Machine, random forest, and Convolutional Neural Network) and by Post-Classification Comparison (PCC). Remote Sensing. 13(19):Unknown Pages. [Link] [DOI:10.3390/rs13193899]
34. Yang C, Rottensteiner F, Heipke C, (2018). Classification of land cover and land use based on convolutional neural networks. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 4(3):251-258. [Link] [DOI:10.5194/isprs-annals-IV-3-251-2018]
35. Yousefi Kebria D, Abaskhanian G, Ghanbari Parmehr E (2020). Detection of oil spill hotspots in the Caspian Sea using remote sensing (case study: Baku oil extraction facility). Environmental Sciences. 18(3):152-166. [Persian] [Link] [DOI:10.29252/envs.18.3.152]