Supervisors: Adina Moraru, Tor Andre Myrvoll, Andrew Perkis
Figures. Left: aerial photo of Utvik (SW Norway) after the flood in 2017 (norgeibilder.no); right: drone images of a flooded area studied with neural networks (Rahnemoonfar et al., 2018).
Problem description
The aim of this thesis is to use AI-based imaging methods (e.g. CNN, RNN), where aerial photos of rivers are analyzed and the AI algorithm is trained to identify features in the photos (such as the borders of the river channel, bridges, areas of accumulation of sediment, areas full of vegetation, etc.) and extract this information for further use in flood risk analysis.
This list of features can be used as criterion to identify where floods happen, and which areas of the given picture are critical. If these critical locations are known, they could be used to refine the grid where physics-based flood simulations are computed.
Problem motivation
This thesis is held within the World of Wild Waters (WoWW) project and aims at constructing realistic flood scenarios and real-time numerical simulations based on real data, and improving risk assessment and its communication to stakeholders and decision-makers. The gathered knowledge will help develop a methodology that will increase flood simulation efficiency, eventually contributing to the state-of-the-art of real-time flood simulations.
Outline of the thesis
- Aerial photos of several Norwegian rivers will be provided;
- Implementation of Machine Learning for the identification of river features;
- Calibrate and refine the model with field data (drone) or documentation (image/video) of a flood event;
- Hard-code the identified river features into conditions that allow identifying flood-prone areas.
- Assessment of the performance of the model in terms of e.g. computational speed and resulting precision.
Desired background
- Strong programming skills (preferably Python);
- Strong understanding of statistical modeling, probability and inference;
- Inclined to research on machine learning-based imagery methods;
- Basic understanding of machine learning techniques (including neural networks).