Bilingual
En/Fa
Geographical Research is Published in both Persian and English Full-text.
Volume 37, Issue 3 (2022)
GeoRes 2022, 37(3): 351-360 |
Back to browse issues page
Article Type:
Subject:
History
Received: 2022/06/15 | Accepted: 2022/08/21 | Published: 2022/07/1
Received: 2022/06/15 | Accepted: 2022/08/21 | Published: 2022/07/1
How to cite this article
Ebadi L. Optimum Method for Water Body Extraction from Multispectral Satellite Images. GeoRes 2022; 37 (3) :351-360
URL: http://georesearch.ir/article-1-1361-en.html
URL: http://georesearch.ir/article-1-1361-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:
Authors
L. Ebadi * 
“Environmental Hazard Institute” and “Department of Surveying Engineering, Faculty of Engineering”, Golestan University, Aliabad Katoul, Iran
Full-Text (HTML) (100 Views)
Introduction
Water resources are among the most vital components of life and sustainable development. Monitoring large water bodies such as oceans, seas, lakes, and gulfs is essential for the proper management and utilization of water resources. Numerous studies and projects related to natural resources, the environment, agriculture, fisheries, and tourism rely on up-to-date information about water bodies. Due to the large spatial extent and inaccessibility of these aquatic areas, ground-based measurement methods are often time-consuming, costly, and sometimes impractical. Satellite imagery, with its wide spatial coverage and high temporal frequency, provides an appropriate option for mapping water bodies.
The Landsat satellite series is one of the oldest remote-sensing missions, continuously acquiring imagery of the Earth’s surface since 1972. In addition to being freely available, Landsat imagery possesses several advantageous characteristics, including medium spatial resolution (30 m), suitable spectral bands (visible and infrared regions), a 16-day revisit time, and a long historical record (over 50 years), that make it an ideal resource for long-term monitoring of surface phenomena, including water bodies [Saif & Najmi, 2013].
Optical satellite sensors capture surface features across different portions of the electromagnetic spectrum, including visible and infrared wavelengths, and store them in separate spectral bands. Since surface features such as water, soil, and vegetation exhibit distinct spectral behavior across these bands, various spectral techniques can be employed to distinguish land-cover types from one another [Attarchi et al., 2020]. Numerous spectral methods have been utilized to differentiate water from land in satellite imagery, which can be broadly categorized into two groups: image classification methods and spectral indices. Classification methods are divided into supervised and unsupervised approaches. Unsupervised classification is generally used for preliminary interpretation due to its lower accuracy, while supervised classification requires training data, making it unsuitable for fully automated workflows. Furthermore, collecting training samples is often time-consuming and subject to human error. Previous research has frequently utilized classification techniques to generate water-body maps [Sotoudehpour et al., 2020].
Spectral indices are mathematical expressions applied pixel-by-pixel to different spectral bands. Their purpose is to enhance specific surface phenomena, such as vegetation, water, or soil and they have been widely used in previous studies [Faramarzi & Nouri, 2015]. The following section reviews several studies that have employed these indices.
Du et al. [2014] have applied NDWI with three different band combinations of Landsat 8 imagery to the Yangtze and Huai River basins. They have concluded that using green and mid-infrared bands, equivalent to the Modified NDWI (MNDWI), yields the best results. Xie et al. [2016] have applied various combinations of NDWI and the AWEI index on Landsat 8 imagery to map multiple water types, including clear, turbid, and eutrophic water. Their results showed that clear water is more easily and accurately identifiable than other types, and that AWEI, NDWI47, and NDWI37 achieved the highest overall accuracies of 98.55%, 95.50%, and 96.61% for clear, turbid, and eutrophic water, respectively. They used both manual thresholding and Otsu’s method to determine optimal thresholds for water extraction, demonstrating that Otsu’s method can replace manual threshold selection.
Kaplan and Avdan [Kaplan & Avdan, 2017] have employed NDWI combined with object-based classification (OBIA) using Sentinel-2 imagery (10 m resolution) to extract water bodies in urban and mountainous regions. They argued that since NDWI was originally designed for Landsat, a threshold of zero is unsuitable for Sentinel-2 data and used a threshold of 0.1 instead. Their findings confirmed that combining spectral indices with classification is highly effective for distinguishing water from shadows in complex environments.
Sarp and Ozcelik [Sarp & Ozcelik, 2017] used NDWI, MNDWI, AWEI, and Support Vector Machine (SVM) classification on Landsat TM and ETM imagery to extract water bodies, employing manual trial-and-error thresholding for the indices. They found that both SVM and MNDWI delivered the highest extraction accuracies. Samiei et al. [2017] monitores 18-year changes in Lake Maharloo using Landsat imagery and two methods: NDWI and thresholding on the SWIR band, using manual trial and error. They considered a threshold of zero for separating water using NDWI.
Kwang et al. [2018] have used Landsat 8 and Sentinel-2 imagery to extract a portion of the Volta River using unsupervised classification and spectral indices (AWEI, MNDWI, NDWI), concluding that MNDWI performed well for both sensors. Sotoudehpour et al. [2019] have mapped coastal areas of Bushehr using Landsat 8 and Sentinel-2 imagery and found MNDWI to be the most accurate index for both datasets. Ahmadnejad et al. [2021] have used NDWI with a zero threshold to delineate shorelines in parts of the Gulf of Mexico using Aster, Landsat 8, and Sentinel-2 imagery.
Asghari et al. [2020] have analyzed the Gamasiab River in Kermanshah province using several spectral indices. Water extraction in narrow and shallow rivers is more challenging, and they tested NDWI, MNDWI, AWEI, and WRI, determining thresholds through histogram interpretation, comparison with reference maps, and visual analysis. They identified AWEI as the most effective index. Solimani Sardo et al. [2021] have analyzed the temporal dynamics of water areas in the Jazmourian Playa using Landsat 8 time series and NDWI, MNDWI, and AWEI indices. In addition to calculating area changes, they observed a strong correlation between annual precipitation and increased water extent derived from MNDWI. Maleki et al. [2022] have evaluated quantitative changes in surface water resources influenced by the Sarpol-e Zahab earthquake using Landsat 8 optical data and water-related indices, including MNDWI.
A review of the existing literature demonstrates that the MNDWI index is highly effective for identifying water in satellite imagery. It is also simpler than many other indices, as it relies on only two spectral bands. The major advantage of spectral indices is that they require no training data, eliminating user intervention and reducing human-induced error. The primary factor influencing the accuracy of spectral-index outputs is the selection of an appropriate threshold to distinguish water from other land-surface features. Many researchers have simply used a threshold of zero, while others have applied manual trial-and-error approaches; only a limited number have employed formalized thresholding techniques.
The objective of this study is to introduce an optimal method for extracting water bodies from multispectral satellite imagery
Methodology
The present study is applied in nature and is based on a case study approach. Its overall framework relies on quantitative methods as well as spatial and visual analyses, utilizing multispectral satellite imagery from Landsat 8 and Sentinel-2.
Gorgan Bay is a semi-enclosed coastal water body located in the southeastern Caspian Sea, separated from the open sea by the Miankaleh Peninsula. Together with the Miankaleh Wetland, it constitutes one of the world’s biosphere reserves and among the most important protected areas along the southern coast of the Caspian Sea. Due to its favorable ecological conditions for marine organisms, the bay has significant economic and environmental importance, particularly for fisheries and recreational activities, and was registered as the first international wetland complex under the Ramsar Convention in Iran. Its extent has undergone considerable changes over time as a result of Caspian Sea level fluctuations and sedimentation processes, making continuous monitoring essential for environmental management and decision-making.
To avoid cloud contamination, satellite images acquired during the summer season were selected, corresponding to late June 2021. Landsat 8 imagery was used to delineate water bodies, taking advantage of its Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). Sentinel-2 multispectral imagery, provided freely by the European Space Agency, was also employed. Sentinel-2 satellites consist of two identical platforms, each carrying the Multispectral Instrument (MSI) with 13 spectral bands at varying spatial resolutions. Owing to their finer spatial resolution and broader spectral range, Sentinel-2 images were used to validate the results derived from Landsat data.
Water surface area in Gorgan Bay was initially estimated using the Modified Normalized Difference Water Index (MNDWI) derived from Landsat 8 imagery. Subsequently, the study area was classified using Sentinel-2 MSI data and a Support Vector Machine (SVM) classification algorithm. The classified Sentinel-2 map served as the reference for evaluating the accuracy of the Landsat-based extraction and for determining the optimal threshold value for water delineation.
Spectral characteristics of water differ markedly from those of soil and vegetation: water strongly absorbs near-infrared wavelengths and appears dark, whereas vegetation and dry soils exhibit high reflectance in this region. Water also shows relatively high reflectance in the green portion of the spectrum, which forms the basis for water-related spectral indices.
Normalized-difference spectral indices are commonly calculated from the difference between two spectral bands divided by their sum, highlighting specific landscape features. Building on this principle, the normalized-difference water index (NDWI) was introduced to enhance water-related features using green and near-infrared wavelengths. However, this index often overestimates water extent due to noise caused by non-water features. To address this issue, a modified version of the index (MNDWI) was developed by replacing the near-infrared band with the shortwave infrared band, reducing noise and improving water delineation accuracy. Positive MNDWI values typically represent water surfaces.
Because spectral indices must be extracted from atmospherically and radiometrically corrected images, Level-2 surface reflectance Landsat 8 products were used. Landsat 8 and Sentinel-2 images were selected to be as close as possible in acquisition date to ensure consistency. Preprocessing steps included normalization of reflectance values and clipping of the study area mosaic. After generating the MNDWI layer, water pixels were initially separated using a zero threshold.
To validate these results, Sentinel-2 imagery at 10-meter resolution was classified using the SVM algorithm, which has demonstrated strong performance in numerous remote sensing applications. Comparisons revealed that shallow and turbid nearshore waters were not accurately identified using the zero threshold applied to the MNDWI layer, necessitating the determination of an optimized threshold.
Image thresholding aims to divide pixels into two brightness-based categories to produce a binary map. Since the histogram of the MNDWI image generally exhibits a bimodal distribution, global thresholding methods are appropriate. Two global thresholding approaches were applied: the mean thresholding method and Otsu’s method. Mean thresholding selects the average pixel intensity of the image as the threshold, offering computational simplicity and speed. Otsu’s method identifies the threshold that maximizes inter-class variance (or equivalently minimizes intra-class variance) to optimally separate the two histogram peaks. Both thresholds were applied to the MNDWI image to extract the water surface
Findings
To calculate the area of the Gorgan Bay, the maps generated from the SVM classification and the thresholding methods applied to the MNDWI index were imported into ArcGIS. After creating polygons representing the boundaries of the bay, the corresponding areas were measured. To assess the accuracy of the SVM classification, reference data were collected from the imagery, and the water class achieved an accuracy of 99.5% based on the error matrix. For estimating the accuracy of the three thresholding methods, the SVM output was considered the reference, and accuracies were computed using the error matrix.
The results indicated that the Otsu thresholding method applied to the MNDWI index produced the highest accuracy (99.2%) in separating water bodies. Visual inspection also revealed that the spatial output of the Otsu method closely corresponded with that of the SVM classification. The mean thresholding method also yielded acceptable accuracy and spatial agreement, whereas the zero-threshold method substantially underestimated the true area of the bay.
Subsequently, the three thresholding methods were used to extract the area of Gorgan Bay from Landsat 8 images over a ten-year period (2013–2022). The extracted area values obtained from the mean and Otsu thresholding methods were highly correlated with one another, while both methods showed considerably lower correlation with the zero-threshold method.
Water resources are among the most vital components of life and sustainable development. Monitoring large water bodies such as oceans, seas, lakes, and gulfs is essential for the proper management and utilization of water resources. Numerous studies and projects related to natural resources, the environment, agriculture, fisheries, and tourism rely on up-to-date information about water bodies. Due to the large spatial extent and inaccessibility of these aquatic areas, ground-based measurement methods are often time-consuming, costly, and sometimes impractical. Satellite imagery, with its wide spatial coverage and high temporal frequency, provides an appropriate option for mapping water bodies.
The Landsat satellite series is one of the oldest remote-sensing missions, continuously acquiring imagery of the Earth’s surface since 1972. In addition to being freely available, Landsat imagery possesses several advantageous characteristics, including medium spatial resolution (30 m), suitable spectral bands (visible and infrared regions), a 16-day revisit time, and a long historical record (over 50 years), that make it an ideal resource for long-term monitoring of surface phenomena, including water bodies [Saif & Najmi, 2013].
Optical satellite sensors capture surface features across different portions of the electromagnetic spectrum, including visible and infrared wavelengths, and store them in separate spectral bands. Since surface features such as water, soil, and vegetation exhibit distinct spectral behavior across these bands, various spectral techniques can be employed to distinguish land-cover types from one another [Attarchi et al., 2020]. Numerous spectral methods have been utilized to differentiate water from land in satellite imagery, which can be broadly categorized into two groups: image classification methods and spectral indices. Classification methods are divided into supervised and unsupervised approaches. Unsupervised classification is generally used for preliminary interpretation due to its lower accuracy, while supervised classification requires training data, making it unsuitable for fully automated workflows. Furthermore, collecting training samples is often time-consuming and subject to human error. Previous research has frequently utilized classification techniques to generate water-body maps [Sotoudehpour et al., 2020].
Spectral indices are mathematical expressions applied pixel-by-pixel to different spectral bands. Their purpose is to enhance specific surface phenomena, such as vegetation, water, or soil and they have been widely used in previous studies [Faramarzi & Nouri, 2015]. The following section reviews several studies that have employed these indices.
Du et al. [2014] have applied NDWI with three different band combinations of Landsat 8 imagery to the Yangtze and Huai River basins. They have concluded that using green and mid-infrared bands, equivalent to the Modified NDWI (MNDWI), yields the best results. Xie et al. [2016] have applied various combinations of NDWI and the AWEI index on Landsat 8 imagery to map multiple water types, including clear, turbid, and eutrophic water. Their results showed that clear water is more easily and accurately identifiable than other types, and that AWEI, NDWI47, and NDWI37 achieved the highest overall accuracies of 98.55%, 95.50%, and 96.61% for clear, turbid, and eutrophic water, respectively. They used both manual thresholding and Otsu’s method to determine optimal thresholds for water extraction, demonstrating that Otsu’s method can replace manual threshold selection.
Kaplan and Avdan [Kaplan & Avdan, 2017] have employed NDWI combined with object-based classification (OBIA) using Sentinel-2 imagery (10 m resolution) to extract water bodies in urban and mountainous regions. They argued that since NDWI was originally designed for Landsat, a threshold of zero is unsuitable for Sentinel-2 data and used a threshold of 0.1 instead. Their findings confirmed that combining spectral indices with classification is highly effective for distinguishing water from shadows in complex environments.
Sarp and Ozcelik [Sarp & Ozcelik, 2017] used NDWI, MNDWI, AWEI, and Support Vector Machine (SVM) classification on Landsat TM and ETM imagery to extract water bodies, employing manual trial-and-error thresholding for the indices. They found that both SVM and MNDWI delivered the highest extraction accuracies. Samiei et al. [2017] monitores 18-year changes in Lake Maharloo using Landsat imagery and two methods: NDWI and thresholding on the SWIR band, using manual trial and error. They considered a threshold of zero for separating water using NDWI.
Kwang et al. [2018] have used Landsat 8 and Sentinel-2 imagery to extract a portion of the Volta River using unsupervised classification and spectral indices (AWEI, MNDWI, NDWI), concluding that MNDWI performed well for both sensors. Sotoudehpour et al. [2019] have mapped coastal areas of Bushehr using Landsat 8 and Sentinel-2 imagery and found MNDWI to be the most accurate index for both datasets. Ahmadnejad et al. [2021] have used NDWI with a zero threshold to delineate shorelines in parts of the Gulf of Mexico using Aster, Landsat 8, and Sentinel-2 imagery.
Asghari et al. [2020] have analyzed the Gamasiab River in Kermanshah province using several spectral indices. Water extraction in narrow and shallow rivers is more challenging, and they tested NDWI, MNDWI, AWEI, and WRI, determining thresholds through histogram interpretation, comparison with reference maps, and visual analysis. They identified AWEI as the most effective index. Solimani Sardo et al. [2021] have analyzed the temporal dynamics of water areas in the Jazmourian Playa using Landsat 8 time series and NDWI, MNDWI, and AWEI indices. In addition to calculating area changes, they observed a strong correlation between annual precipitation and increased water extent derived from MNDWI. Maleki et al. [2022] have evaluated quantitative changes in surface water resources influenced by the Sarpol-e Zahab earthquake using Landsat 8 optical data and water-related indices, including MNDWI.
A review of the existing literature demonstrates that the MNDWI index is highly effective for identifying water in satellite imagery. It is also simpler than many other indices, as it relies on only two spectral bands. The major advantage of spectral indices is that they require no training data, eliminating user intervention and reducing human-induced error. The primary factor influencing the accuracy of spectral-index outputs is the selection of an appropriate threshold to distinguish water from other land-surface features. Many researchers have simply used a threshold of zero, while others have applied manual trial-and-error approaches; only a limited number have employed formalized thresholding techniques.
The objective of this study is to introduce an optimal method for extracting water bodies from multispectral satellite imagery
Methodology
The present study is applied in nature and is based on a case study approach. Its overall framework relies on quantitative methods as well as spatial and visual analyses, utilizing multispectral satellite imagery from Landsat 8 and Sentinel-2.
Gorgan Bay is a semi-enclosed coastal water body located in the southeastern Caspian Sea, separated from the open sea by the Miankaleh Peninsula. Together with the Miankaleh Wetland, it constitutes one of the world’s biosphere reserves and among the most important protected areas along the southern coast of the Caspian Sea. Due to its favorable ecological conditions for marine organisms, the bay has significant economic and environmental importance, particularly for fisheries and recreational activities, and was registered as the first international wetland complex under the Ramsar Convention in Iran. Its extent has undergone considerable changes over time as a result of Caspian Sea level fluctuations and sedimentation processes, making continuous monitoring essential for environmental management and decision-making.
To avoid cloud contamination, satellite images acquired during the summer season were selected, corresponding to late June 2021. Landsat 8 imagery was used to delineate water bodies, taking advantage of its Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS). Sentinel-2 multispectral imagery, provided freely by the European Space Agency, was also employed. Sentinel-2 satellites consist of two identical platforms, each carrying the Multispectral Instrument (MSI) with 13 spectral bands at varying spatial resolutions. Owing to their finer spatial resolution and broader spectral range, Sentinel-2 images were used to validate the results derived from Landsat data.
Water surface area in Gorgan Bay was initially estimated using the Modified Normalized Difference Water Index (MNDWI) derived from Landsat 8 imagery. Subsequently, the study area was classified using Sentinel-2 MSI data and a Support Vector Machine (SVM) classification algorithm. The classified Sentinel-2 map served as the reference for evaluating the accuracy of the Landsat-based extraction and for determining the optimal threshold value for water delineation.
Spectral characteristics of water differ markedly from those of soil and vegetation: water strongly absorbs near-infrared wavelengths and appears dark, whereas vegetation and dry soils exhibit high reflectance in this region. Water also shows relatively high reflectance in the green portion of the spectrum, which forms the basis for water-related spectral indices.
Normalized-difference spectral indices are commonly calculated from the difference between two spectral bands divided by their sum, highlighting specific landscape features. Building on this principle, the normalized-difference water index (NDWI) was introduced to enhance water-related features using green and near-infrared wavelengths. However, this index often overestimates water extent due to noise caused by non-water features. To address this issue, a modified version of the index (MNDWI) was developed by replacing the near-infrared band with the shortwave infrared band, reducing noise and improving water delineation accuracy. Positive MNDWI values typically represent water surfaces.
Because spectral indices must be extracted from atmospherically and radiometrically corrected images, Level-2 surface reflectance Landsat 8 products were used. Landsat 8 and Sentinel-2 images were selected to be as close as possible in acquisition date to ensure consistency. Preprocessing steps included normalization of reflectance values and clipping of the study area mosaic. After generating the MNDWI layer, water pixels were initially separated using a zero threshold.
To validate these results, Sentinel-2 imagery at 10-meter resolution was classified using the SVM algorithm, which has demonstrated strong performance in numerous remote sensing applications. Comparisons revealed that shallow and turbid nearshore waters were not accurately identified using the zero threshold applied to the MNDWI layer, necessitating the determination of an optimized threshold.
Image thresholding aims to divide pixels into two brightness-based categories to produce a binary map. Since the histogram of the MNDWI image generally exhibits a bimodal distribution, global thresholding methods are appropriate. Two global thresholding approaches were applied: the mean thresholding method and Otsu’s method. Mean thresholding selects the average pixel intensity of the image as the threshold, offering computational simplicity and speed. Otsu’s method identifies the threshold that maximizes inter-class variance (or equivalently minimizes intra-class variance) to optimally separate the two histogram peaks. Both thresholds were applied to the MNDWI image to extract the water surface
Findings
To calculate the area of the Gorgan Bay, the maps generated from the SVM classification and the thresholding methods applied to the MNDWI index were imported into ArcGIS. After creating polygons representing the boundaries of the bay, the corresponding areas were measured. To assess the accuracy of the SVM classification, reference data were collected from the imagery, and the water class achieved an accuracy of 99.5% based on the error matrix. For estimating the accuracy of the three thresholding methods, the SVM output was considered the reference, and accuracies were computed using the error matrix.
The results indicated that the Otsu thresholding method applied to the MNDWI index produced the highest accuracy (99.2%) in separating water bodies. Visual inspection also revealed that the spatial output of the Otsu method closely corresponded with that of the SVM classification. The mean thresholding method also yielded acceptable accuracy and spatial agreement, whereas the zero-threshold method substantially underestimated the true area of the bay.
Subsequently, the three thresholding methods were used to extract the area of Gorgan Bay from Landsat 8 images over a ten-year period (2013–2022). The extracted area values obtained from the mean and Otsu thresholding methods were highly correlated with one another, while both methods showed considerably lower correlation with the zero-threshold method.
Discussion
The aim of this study was to introduce an optimized method for extracting water bodies from multispectral satellite imagery. Water can be identified in multispectral images due to its unique spectral reflectance characteristics using spectral separation techniques, including classification and spectral indices. Supervised classification methods are less efficient because they require training data, are time-consuming, involve user intervention, and cannot be fully automated. In contrast, spectral indices do not require user-collected training data, are free from human error, and allow for automated processing. Numerous studies have reported that the MNDWI index effectively identifies water in multispectral satellite images [Du et al., 2014; Sarp & Ozcelik, 2017; Kwang et al., 2018; Asghari et al., 2020; Solimani Sardo et al., 2021; Maleki et al., 2022].
The main challenge when using MNDWI was selecting the appropriate threshold to separate water from land. Many researchers simply use a zero threshold, considering positive MNDWI values as water, while some studies, including the original developers of MNDWI, achieved better results by manually adjusting the threshold [Xu, 2006; Sarp & Ozcelik, 2017; Asghari et al., 2020]. Manual thresholding was time-consuming and dependent on the skill and judgment of the user. For long-term monitoring of water bodies, processing large numbers of images made manual thresholding inefficient and non-automatable. In contrast, global thresholding methods, which utilize the statistical properties of the image, can be effectively applied. In one study, fifteen global thresholding methods were tested for water extraction using spectral indices, all producing high accuracies above 92% [Sekertekin, 2021]. Another study combines spectral indices with both manual and Otsu thresholding to map multiple water bodies, concluding that clear water could be detected more easily and accurately than other water types, and that Otsu thresholding is a suitable alternative to manual thresholding [Xie et al., 2016].
The findings of the present study are applicable for large water bodies. For smaller water bodies, particularly rivers influenced by shadows, further research is needed. Given that the combination of MNDWI and Otsu thresholding does not require manual user intervention and allows for automated processing, it is recommended that future studies develop programs for fully automated extraction of water bodies from multispectral satellite imagery.
Conclusion
The accuracy of extracting water bodies from multispectral satellite images using spectral indices depends on selecting an appropriate threshold. The Otsu thresholding method not only improves accuracy but also utilizes the statistical information of the image without requiring user intervention, making it suitable for automating the water detection process.
Acknowledgments: Not applicable.
Ethical Approval: Not applicable.
Conflict of Interest: Not applicable.
Authors’ Contributions: Ladan Ebadi is the sole author of this article (100%).
Funding: Not applicable.
References
1. Ahmadnejad M, Akbarzade Shirsavar M, Farhadipoor S, Mohammadi H (2021). Coastlines detection with satellite images processing technique in part of the gulf of Mexico. Hydrophysics. 6(2):113-126. [Link]
2. Asghari S, Jalilyan R, Pirozineghad N, Madadi A, Yadeghari M (2020). Evaluation of water extraction indices using landsat satellite images (Case study: Gamasiab river of Kermanshah). Journal of Applied Researches in Geographical Sciences. 20(58):53-70. [Persian] [Link] [DOI:10.29252/jgs.20.58.53]
3. Attarchi S, Poorrahimi M, Isazade V (2020). Comparison of spectral indices and object-based classification for built-up area extraction in different urban areas. Geographical Urban Planning Research. 8(1):23-43. [Persian] [Link]
4. Du Z, Li W, Zhou D, Tian L, Ling F, Wang H, et al (2014). Analysis of landsat-8 OLI imagery for land surface water mapping. Remote Sensing Letters. 5(7):672-681. [Link] [DOI:10.1080/2150704X.2014.960606]
5. Faramarzi M, Nouri H (2015). Evaluation of vegetation and thermal indices efficiency and feasibility of Landsat images to agricultural drought analysis (Case study: Malayer). Geographical Researches. 30(3):139-152. [Persian] [Link]
6. Feyisa G, L, Meilby H, Fensholt R, Proud S R (2014). Automated water extraction index: A new technique for surface water mapping using landsat imagery. Remote Sensing of Environment. 140:23-35. [Link] [DOI:10.1016/j.rse.2013.08.029]
7. Kaplan G, Avdan U (2017). Object-based water body extraction model using Sentinel-2 satellite imagery. European Journal of Remote Sensing. 50(1):137-143. [Link] [DOI:10.1080/22797254.2017.1297540]
8. Khoshravan H (2020). Spatial and temporal variations of coastal habitats in Gorgan bay resulting from fluctuations in the Caspian sea. Geographical Data. 29(115):127-138. [Persian] [Link]
9. Kwang C, Osei Jnr E M, Amoah A S (2018). Comparing of landsat 8 and sentinel 2a using water extraction indexes over Volta river. Journal of Geography and Geology. 10(1):1-7. [Link] [DOI:10.5539/jgg.v10n1p1]
10. Maleki A, Khazai A, Abdolmaleki A (2022). Evaluation of quantitative changes in surface water resources affected by the earthquake of Sarpol-e-Zahab using satellite data. Journal of RS and GIS for Natural Resources. 13(1):39-61. [Persian] [Link]
11. McFeeters S K (1996). The use of the normalized difference water index (NDWI) in the delineation of open water features. International Journal of Remote Sensing. 17(7):1425-1432. [Link] [DOI:10.1080/01431169608948714]
12. Otsu N (1979). A threshold selection method from gray-level histograms. IEEE Transactions on Systems Man and Cybernetics. 9(1):62-66. [Link] [DOI:10.1109/TSMC.1979.4310076]
13. Saif A, Najmi N (2013). Changes detection of the Karun's meanders by IRS and landsat images satellite. Geographical Researches. 28(3):213-228. [Persian] [Link]
14. Samiei M, Ghazavi R, Pakparvar M, Vali A A (2017). The effect of climate change on Maharlo lake level change using satellite image processing. Journal of RS and GIS for Natural Resources. 8(1):1-18. [Link]
15. Sarp G, Ozcelik M (2017). Water body extraction and change detection using time series: A case study of lake Burdur, Turkey. Journal of Taibah University for Science. 11(3):381-391. [Link] [DOI:10.1016/j.jtusci.2016.04.005]
16. Sekertekin A A (2021). Survey on Global Thresholding methods for mapping openwater body using sentinel-2 satellite imagery and normalized difference water index. Archives of Computational Methods in Engineering. 28:1335--1347. [Link] [DOI:10.1007/s11831-020-09416-2]
17. Sharbaty S (2019). Modeling the effects of water level decreases on the optimum distance of fish pens in Gorgan bay. Iranian Scientific Fisheries Journal. 28(1) 49-58. [Persian] [Link]
18. Sharbaty S, Ghanghermeh A (2016). Forecasting the effect of decreasing long time trend of Caspian sea water level on the life of Gorgan bay. Journal of Environmental Science and Technology. 17(4):45-59. [Persian] [Link]
19. Shen L, Li C. 18th international conference on geoinformatics (2010). Water body extraction from Landsat ETM+ imagery using adaboost algorithm. Beijing, China. 18-20 June 2010.Canada: IEEE Transactions on Systems Man and Cybernetics. [Link] [DOI:10.1109/GEOINFORMATICS.2010.5567762]
20. Solimani Sardo M, Ebrahimi Z, Zarei M (2021). Extraction of surface water zones of seasonal lake Jazmourian using remote sensing indicators. Physical Geography Quarterly. 14(53):119-134. [Persian] [Link]
21. Sotoudehpour A, Madadi A, Asghari Saraskanrood S (2020). Geomorphological monitoring of Bushehr port coastline. Geographical Researches. 35(2):177-187. [Persian] [Link]
22. Xie H, Luo X, Xu X, Pan H, Tong X (2016). Evaluation of landsat OLI 8 imagery for unsupervised inland water extraction. International Journal of Remote Sensing. 37(8):1826-1844. [Link] [DOI:10.1080/01431161.2016.1168948]
23. Xu H (2006). Modification of normalised difference water index (NDWI) to enhance open water features in remotely sensed imagery. International Journal of Remote Sensing. 27(14):3025-3033. [Link] [DOI:10.1080/01431160600589179]