University of Thessaly (UTH)
Διερεύνηση κινδύνου πλημμύρας σε λεκάνες απορροής ποταμών με εφαρμογή αλγορίθμου μηχανικής μάθησης, υδραυλικού μοντέλου προσομοίωσης και Γεωγραφικών Συστημάτων Πληροφοριών
Abstract
dc:descriptionIn this thesis, an integrated methodology for flood risk investigation in river basins, with an emphasis on the rapid and reliable estimation of flood extent, was developed. The proposed approach combines Geographic Information Systems, hydrological and two-dimensional hydraulic modelling, and deep learning techniques. The main objective of the thesis is to develop a pre-trained machine learning model capable of predicting flood extents with high accuracy, using limited and easily accessible geospatial input data. The methodology first involves the preprocessing and analysis of geospatial data in a GIS environment, with the aim of producing the necessary thematic layers for hydrological, hydraulic, and computational analysis. In this context, the “HydroTools” toolbox was developed to support procedures such as watershed delineation, the estimation of hydrological parameters, and the assignment of coefficients, including Manning’s roughness coefficient and the Curve Number. Hydrological analysis was then carried out to generate design hydrographs, which were used as input data for the two-dimensional hydraulic simulations. The results of the hydraulic modelling, particularly the final flood extents, were used as reference data for training the deep learning model. The main innovation of the thesis lies in the development and evaluation of a two-dimensional convolutional neural network based on the U-Net architecture for the direct prediction of flood extents. The model uses multichannel raster data as input, including spatial variables related to terrain morphology, flow characteristics, and watershed properties. Its output is also raster-based and can be directly used in a GIS environment, enabling the production of flood extent maps in significantly less time than conventional numerical simulations. The methodology was applied to areas within the Thessaly River Basin District, with the Enipeas River basin serving as the main training area. Additional validation and testing areas were selected in different river reaches and streams. The evaluation of the pre-trained model showed highly satisfactory performance, with a strong correlation between the predicted flood extents and the corresponding reference flood extents. Specifically, the coefficients of determination were approximately R² = 0.997 for the validation areas and R² = 0.999 for the testing areas, while in most examined cases the deviations remained below 10%. In parallel, several complementary applications were developed within the framework of the thesis, including the Flood Event Reporting and Recording System of Greece and a network for monitoring climatic parameters through meteorological stations.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Xafoulis, Nikolaos
- Ξαφούλης, Νικόλαος
Subjects
dc:subject × 19- Πλημμυρικός κίνδυνος
- Λεκάνες απορροής
- Μηχανική μάθηση
- Υδραυλική προσομοίωση
- Γεωγραφικά Συστήματα Πληροφοριών (ΓΣΠ)
- Flood risk
- River basin
- Machine learning
- Hydraulic simulation
- Geographical Information Systems (GIS)
- Περιβαλλοντική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Επιστήμη και Τεχνολογία υδάτων
- Επιστήμη Πολιτικού Μηχανικού
- Πολιτική μηχανική
- Environmental Engineering
- Engineering and Technology
- Water Science and Technology
- Civil Engineering
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/61711