Abstract's details
Water level estimation in the Mekong River Basin based on a classification of CryoSat-2 SAR data
CoAuthors
Event: 2017 Ocean Surface Topography Science Team Meeting
Session: Science IV: 25 years of satellite altimetry for Cryosphere and Hydrology: from experimental to emerging operational applications
Presentation type: Type Poster
Contribution: not provided
Abstract:
Compared to conventional LRM altimetry CryoSat-2 SAR data is especially useful for measuring small inland waters such as rivers due to the improved along track resolution. Though, a reliable extraction of water levels is still challenging for river with only very few consecutive water measurements. Land-water-masks are often too unreliable for a precise identification of small rivers. We propose a classification approach which overcomes this problem by detecting the water measurements from the altimetry measurements themselves.
The CryoSat-2 SAR data are available with the full stack data which contains all single-looks for one point on the Earth’s surface. Each of the single-looks is a returning waveform from a different looking angle. The SAR (multi-look) waveform is the averaged waveform of the stack. The assemblage of the mean power of each single-look is named RIP waveform. In this study, we use both quantities, i.e., the full stack data available for CryoSat-2 SAR, to classify the measurements and to identify the water returns. The classification is performed with a k-means clustering algorithm on features extracted from the waveform and the RIP waveform as well as the waveform and RIP waveform itself.
The classification approach is applied and tested over the Mekong River Basin. We found better results in the classification after dividing the data in two regions, one for small upstream rivers surrounded by hills and mountains, and one for the wider rivers with flatter surrounding terrain.
Out of 20 clusters provided by the k-means algorithm, those forming the water classes are identified in a region with known river locations and based on the mean waveform and RIP for each class. These classes are used in a next step for classifying all measurements in the whole river basin in water and non-water measurements. The classified water measurements are used in a last step to generate water levels for each CryoSat-2 crossing with a river branch in the Mekong River basin.
The long repeat orbit of CryoSat-2 hinders the setup of water level time series at fixed virtual stations, which can be used for validation against in situ gauge data. Under the assumption of a stable seasonal signal, which holds true for the Mekong River, an internal validation of the data is possible. To this end, we compare, first, the water levels of consecutive passes (time difference 369 days) with each other and, second, water levels that are both close in location and close in season. The validation proves the success of the approach for the main river as well as for smaller tributaries.
The CryoSat-2 SAR data are available with the full stack data which contains all single-looks for one point on the Earth’s surface. Each of the single-looks is a returning waveform from a different looking angle. The SAR (multi-look) waveform is the averaged waveform of the stack. The assemblage of the mean power of each single-look is named RIP waveform. In this study, we use both quantities, i.e., the full stack data available for CryoSat-2 SAR, to classify the measurements and to identify the water returns. The classification is performed with a k-means clustering algorithm on features extracted from the waveform and the RIP waveform as well as the waveform and RIP waveform itself.
The classification approach is applied and tested over the Mekong River Basin. We found better results in the classification after dividing the data in two regions, one for small upstream rivers surrounded by hills and mountains, and one for the wider rivers with flatter surrounding terrain.
Out of 20 clusters provided by the k-means algorithm, those forming the water classes are identified in a region with known river locations and based on the mean waveform and RIP for each class. These classes are used in a next step for classifying all measurements in the whole river basin in water and non-water measurements. The classified water measurements are used in a last step to generate water levels for each CryoSat-2 crossing with a river branch in the Mekong River basin.
The long repeat orbit of CryoSat-2 hinders the setup of water level time series at fixed virtual stations, which can be used for validation against in situ gauge data. Under the assumption of a stable seasonal signal, which holds true for the Mekong River, an internal validation of the data is possible. To this end, we compare, first, the water levels of consecutive passes (time difference 369 days) with each other and, second, water levels that are both close in location and close in season. The validation proves the success of the approach for the main river as well as for smaller tributaries.