Supervisors: Adina Moraru, Tor Andre Myrvoll, Andrew Perkis
Figures. Left: raw physical model image, center: example of classified image, right: ML model (Son, 2020).
This master’s thesis will be part of the World of Wild Waters (WoWW) project.
WoWW aims to create digital twins based on the real-time visualization of floods in Norway.
The visualized flood scenarios are based on physical and prediction models.
Starting point:
There is a flood physical model with high-resolution both temporally and pixel-wise, which allows customizing the size of the dataset used in this master’s thesis.
The images obtained from the physical model at different time intervals are available for classification and building a Deep Neural Network model that identifies the hotspots with highest flood probability.
The classification can be done using a set of pre-defined conditions that characterize hotspots where flooding occurs in a river. These conditions are provided from previous studies.
Target of the thesis:
Use flood images from the above described physical model together with a set of pre-defined conditions and create a Deep Neural Network that identifies flood hotspots. Prerequisites are:
i) able to use Python and
ii) be familiar with Deep Neural Networks.
Expectations from the student. To:
- Select a suitable Machine Learning method
- Use Machine Learning to classify images from a physical model
- Build a Machine Learning model that can predict the location of flood hotspots
- Analyze the performance of the prediction model