Permanent water and flash flood detection using global navigation satellite system reflectometry
2021
In this thesis, research for inland water extent and flash floods remote sensing using
Global Navigation Satellite System Reflectometry (GNSS-R) data of the Cyclone
Global Navigation Satellite System (CYGNSS) is presented.
Firstly, a high-resolution Machine Learning (ML) method for detecting inland water
extent using the CYGNSS data is implemented via the Random Under-Sampling
Boosted (RUSBoost) algorithm. The CYGNSS data of the year 2018 over the Congo
and Amazon basins are gridded into 0.01゚ × 0.01゚ cells. The RUSBoost-based classifier
is trained and tested with the CYGNSS data over the Congo basin. The Amazon
basin data that is unknown to the classifier is then used for further evaluation. Using
only three observables extracted from the CYGNSS Delay-Doppler Maps (DDMs),
the proposed technique is able to detect 95.4% and 93.3% of the water bodies over the
Congo and Amazon basins, respectively. The performance of the RUSBoost-based
classifier is also compared with an image processing based inland water detection
method. For the Congo and Amazon basins, the RUSBoost-based classifier has a
3.9% and 14.2% higher water detection accuracies, respectively.
Secondly, a flash flood detection method using the CYGNSS data is investigated.
Considering Hurricane Harvey and Hurricane Irma as two case studies, six different
Data Preparation Approaches (DPAs) for flood detection based on the CYGNSS
data and the RUSBoost classification algorithm are investigated in this thesis. Taking
flood and land as two classes, flash flood detection is tackled as a binary classification
problem. Eleven observables are extracted from the DDMs for feature selection.
These observables, alongside two features from ancillary data, are considered in feature selection. All the combinations of these observables with and without ancillary
data are fed into the classifier with 5-fold cross-validation one-by-one. Based on the
test results, five observables with the ancillary data are selected as a suitable feature
vector for flood detection here. Using the selected feature vector, six different DPAs
are investigated and compared to find the best one for flash flood detection. Then,
the performance of the proposed method is compared with that of a Support Vector
Machine (SVM) based classifier. For Hurricane Harvey and Hurricane Irma, the
selected method is able to detect 89.00% and 85.00% of flooded points, respectively,
with a resolution of 500m × 500m, and the detection accuracy for non-flooded land
points is 97.20% and 71.00%, respectively.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
0
References
0
Citations
NaN
KQI