Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia
Globally, the Sea Level Rise (SLR) has been rising due to global warming affecting many nations. According to National Hydraulic Research Institute of Malaysia (NAHRIM), the projected sea-level rise in 2100 will be in the range of 0.25 m and 0.50 m, which will hit hardest the low-lying areas along t...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/98119/1/FSKTM%202021%206%20-%20IR.pdf http://psasir.upm.edu.my/id/eprint/98119/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| id |
my.upm.eprints.98119 |
|---|---|
| record_format |
eprints |
| institution |
Universiti Putra Malaysia |
| building |
UPM Library |
| collection |
Institutional Repository |
| continent |
Asia |
| country |
Malaysia |
| content_provider |
Universiti Putra Malaysia |
| content_source |
UPM Institutional Repository |
| url_provider |
http://psasir.upm.edu.my/ |
| language |
English |
| topic |
Sea-level - Research Shorelines - Malaysia Coastal zone management - Case studies |
| spellingShingle |
Sea-level - Research Shorelines - Malaysia Coastal zone management - Case studies Abd Manaf, Syaifulnizam Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| description |
Globally, the Sea Level Rise (SLR) has been rising due to global warming affecting many nations. According to National Hydraulic Research Institute of Malaysia (NAHRIM), the projected sea-level rise in 2100 will be in the range of 0.25 m and 0.50 m, which will hit hardest the low-lying areas along the Northeast and West coasts of Peninsular Malaysia. Practically, sea level change can be measured using tide gauge and satellite altimetry while land surface can be measured using ground land survey. However, optical satellite images can be used as compliments to these surveying devices to measure shoreline positions. Therefore, identifying coastal erosion-induced sea-level rise entails the analysis of historical data of shoreline changes that have occurred for more than 30 years. Arguably, extracting shorelines from large regions using traditional ground survey techniques is laborious and time-consuming.
Most of studies focused on either shoreline extraction model only or shoreline change model only. Therefore, in this study, the researcher proposes a spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines of Peninsular Malaysia consisting of four phases, namely image fusion, synthetic hyper spectral generation, shoreline extraction, and shoreline prediction. Based on the proposed phases, the use of the proposed model could help improve the tasks of extracting and predicting shorelines not only faster but also more accurately compared to the existing methods. Three studies areas were selected in this study, namely Langkawi, Pontian, and Tumpat, which are potentially at risk to sea-level rise according to the local authority. The main data of the study were multi-temporal satellite images and tides data, which were complemented with other ancillary data observed and collected over a period of about 40 years.
The main problem in the extraction of satellite images is that most raw images contain noises, such as clouds, shadows, and haze, which leads to low accuracy extraction rate. Thus far, pre-processing tasks can only remove haze, but not the other two. In a worst scenario, in which clouds and shadows are located at the boundary areas of land and water regions, extracting shorelines will be problematic because some parts of an extracted image cover and obscure the water and land regions. To make matters worse, such regions cannot be masked out for analysis because important boundary information had already been removed. As such, the fusion of multispectral and Synthetic Aperture Radar (SAR) images can help extract shorelines with higher accuracy rate without losing important spatial boundary information. In this study, a hybrid of Nearest-neighbor Diffusion and Gram-Schmidt techniques was proposed for fusing multispectral and SAR images to produce a synthetic fusion image that closely represents the original image.
Although, Google Earth is freely available, it lacks some important spectral information that is vital for detailed analyses. As an alternative, the extraction of shorelines can be performed more accurately and efficiently by using the Earth observation remote sensing satellites. In this study, spectral bands derived from satellite images, such as spectral band indices, texture, statistical, and transformation features, were utilized to improve the accuracy of land cover classification. The proposed synthetic hyper spectral approach to determine optimal features for land cover classification consisted of feature extraction, feature subset selection, and feature ranking. The results showed that the combination of several basic spectral bands and derived features of multispectral images and their order could help produce the most optimal features. Then, Artificial Neural Network (ANN) with Error Correcting Output Codes (ECOC) was proposed to identify the most suitable classifier for classifying land cover types.
The selection of suitable satellite images can provide data with high accuracy of extraction that relies on high-resolution satellite images, which are, however, not applicable at the earlier time. Therefore, the use of medium resolution satellite images cannot be avoided. However, due to the medium resolution constraint, the most efficient approach for shoreline extraction of either pixel-based classification or object-based image analysis need to be carried out. Moreover, for the shoreline extraction technique, coastal research studies mostly used a single spectral band or a number of spectral bands of the original satellite image sensors to extract shorelines Therefore, there is a need for more accurate extraction by utilizing the maximum number of spectral bands possible according to the proposed synthetic hyper spectral approach. Based on knowledge from previous phase on which is the most suitable classifier, the proposed optimised ANN with ECOC was the most effective classifier to classify the types of land cover of ranked features of multispectral images, which ultimately helped to extract accurate shorelines. Then, validation assessment was applied to assess the most effective technique to extract shorelines accurately and the result was directly proportional to the classification accuracy, thus indicating that the proposed technique is the most suitable approach.
Due to a lack of spatio-temporal data available for the study areas, a statistical approach for shoreline prediction was adopted. Specifically, the End Point Rate (EPR) technique was performed to the historical data to predict new shorelines’ positions that could be validated by the most recent extracted shoreline data. The projection of future shorelines’ positions could also be carried out by the same prediction model by extrapolating the positions of shorelines of the historical data. Moreover, the EPR technique could detect the trends of shoreline changes more efficiently in more detail.
In this study, it was observed that the implementation of the proposed spatio-temporal GIS-based system for shoreline changes and projection depended on the quality of data that had been extracted and predicted through a series of processes. Thus, the understanding of relevant data was extremely important in modelling the implementation of such a GIS-based system. Specifically, the software prototype was implementation using the web-based Geographical Information System (GIS) in the server environment, consisting of a number of interactive built-in features and widgets. |
| format |
Thesis |
| author |
Abd Manaf, Syaifulnizam |
| author_facet |
Abd Manaf, Syaifulnizam |
| author_sort |
Abd Manaf, Syaifulnizam |
| title |
Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| title_short |
Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| title_full |
Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| title_fullStr |
Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| title_full_unstemmed |
Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia |
| title_sort |
spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in peninsular malaysia |
| publishDate |
2021 |
| url |
http://psasir.upm.edu.my/id/eprint/98119/1/FSKTM%202021%206%20-%20IR.pdf http://psasir.upm.edu.my/id/eprint/98119/ |
| _version_ |
1739829920430817280 |
| spelling |
my.upm.eprints.981192022-07-20T00:45:50Z http://psasir.upm.edu.my/id/eprint/98119/ Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia Abd Manaf, Syaifulnizam Globally, the Sea Level Rise (SLR) has been rising due to global warming affecting many nations. According to National Hydraulic Research Institute of Malaysia (NAHRIM), the projected sea-level rise in 2100 will be in the range of 0.25 m and 0.50 m, which will hit hardest the low-lying areas along the Northeast and West coasts of Peninsular Malaysia. Practically, sea level change can be measured using tide gauge and satellite altimetry while land surface can be measured using ground land survey. However, optical satellite images can be used as compliments to these surveying devices to measure shoreline positions. Therefore, identifying coastal erosion-induced sea-level rise entails the analysis of historical data of shoreline changes that have occurred for more than 30 years. Arguably, extracting shorelines from large regions using traditional ground survey techniques is laborious and time-consuming. Most of studies focused on either shoreline extraction model only or shoreline change model only. Therefore, in this study, the researcher proposes a spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines of Peninsular Malaysia consisting of four phases, namely image fusion, synthetic hyper spectral generation, shoreline extraction, and shoreline prediction. Based on the proposed phases, the use of the proposed model could help improve the tasks of extracting and predicting shorelines not only faster but also more accurately compared to the existing methods. Three studies areas were selected in this study, namely Langkawi, Pontian, and Tumpat, which are potentially at risk to sea-level rise according to the local authority. The main data of the study were multi-temporal satellite images and tides data, which were complemented with other ancillary data observed and collected over a period of about 40 years. The main problem in the extraction of satellite images is that most raw images contain noises, such as clouds, shadows, and haze, which leads to low accuracy extraction rate. Thus far, pre-processing tasks can only remove haze, but not the other two. In a worst scenario, in which clouds and shadows are located at the boundary areas of land and water regions, extracting shorelines will be problematic because some parts of an extracted image cover and obscure the water and land regions. To make matters worse, such regions cannot be masked out for analysis because important boundary information had already been removed. As such, the fusion of multispectral and Synthetic Aperture Radar (SAR) images can help extract shorelines with higher accuracy rate without losing important spatial boundary information. In this study, a hybrid of Nearest-neighbor Diffusion and Gram-Schmidt techniques was proposed for fusing multispectral and SAR images to produce a synthetic fusion image that closely represents the original image. Although, Google Earth is freely available, it lacks some important spectral information that is vital for detailed analyses. As an alternative, the extraction of shorelines can be performed more accurately and efficiently by using the Earth observation remote sensing satellites. In this study, spectral bands derived from satellite images, such as spectral band indices, texture, statistical, and transformation features, were utilized to improve the accuracy of land cover classification. The proposed synthetic hyper spectral approach to determine optimal features for land cover classification consisted of feature extraction, feature subset selection, and feature ranking. The results showed that the combination of several basic spectral bands and derived features of multispectral images and their order could help produce the most optimal features. Then, Artificial Neural Network (ANN) with Error Correcting Output Codes (ECOC) was proposed to identify the most suitable classifier for classifying land cover types. The selection of suitable satellite images can provide data with high accuracy of extraction that relies on high-resolution satellite images, which are, however, not applicable at the earlier time. Therefore, the use of medium resolution satellite images cannot be avoided. However, due to the medium resolution constraint, the most efficient approach for shoreline extraction of either pixel-based classification or object-based image analysis need to be carried out. Moreover, for the shoreline extraction technique, coastal research studies mostly used a single spectral band or a number of spectral bands of the original satellite image sensors to extract shorelines Therefore, there is a need for more accurate extraction by utilizing the maximum number of spectral bands possible according to the proposed synthetic hyper spectral approach. Based on knowledge from previous phase on which is the most suitable classifier, the proposed optimised ANN with ECOC was the most effective classifier to classify the types of land cover of ranked features of multispectral images, which ultimately helped to extract accurate shorelines. Then, validation assessment was applied to assess the most effective technique to extract shorelines accurately and the result was directly proportional to the classification accuracy, thus indicating that the proposed technique is the most suitable approach. Due to a lack of spatio-temporal data available for the study areas, a statistical approach for shoreline prediction was adopted. Specifically, the End Point Rate (EPR) technique was performed to the historical data to predict new shorelines’ positions that could be validated by the most recent extracted shoreline data. The projection of future shorelines’ positions could also be carried out by the same prediction model by extrapolating the positions of shorelines of the historical data. Moreover, the EPR technique could detect the trends of shoreline changes more efficiently in more detail. In this study, it was observed that the implementation of the proposed spatio-temporal GIS-based system for shoreline changes and projection depended on the quality of data that had been extracted and predicted through a series of processes. Thus, the understanding of relevant data was extremely important in modelling the implementation of such a GIS-based system. Specifically, the software prototype was implementation using the web-based Geographical Information System (GIS) in the server environment, consisting of a number of interactive built-in features and widgets. 2021-08 Thesis NonPeerReviewed text en http://psasir.upm.edu.my/id/eprint/98119/1/FSKTM%202021%206%20-%20IR.pdf Abd Manaf, Syaifulnizam (2021) Spatio-temporal synthetic hyper spectral model for extracting and predicting shorelines in Peninsular Malaysia. Doctoral thesis, Universiti Putra Malaysia. Sea-level - Research Shorelines - Malaysia Coastal zone management - Case studies |
| score |
13.211869 |