Unit:
Department of Geology and GeoenvirommentLibrary of the School of Science
Author:
Kotinas Vasileios
Dissertation committee:
Γκουρνέλος Θεόδωρος, Καθηγητής, Τμήμα Γεωλογιας και Γεωπεριβάλλοντος, Ε.Κ.Π.Α
Ευελπίδου Νίκη-Νικολέττα, Καθηγήτρια, Τμήμα Γεωλογιας και Γεωπεριβάλλοντος, Ε.Κ.Π.Α
Πούλος Σεραφείμ , Καθηγητής, Τμήμα Γεωλογιας και Γεωπεριβάλλοντος, Ε.Κ.Π.Α
Καρύμπαλης Ευθύμιος, Καθηγητής, Τμήμα Γεωγραφίας, ΧΑΡΟΚΟΠΕΙΟ ΠΑΝΕΠΙΣΤΗΜΙΟ
Νάστος Παναγιώτης, Καθηγητής, Τμήμα Γεωλογιας και Γεωπεριβάλλοντος, Ε.Κ.Π.Α
Χαλκιάς Χρίστος, Καθηγητής, Τμήμα Γεωγραφίας, ΧΑΡΟΚΟΠΕΙΟ ΠΑΝΕΠΙΣΤΗΜΙΟ
Βασιλάκης Εμμανουήλ , Επίκουρος Καθηγητής, Τμήμα Γεωλογιας και Γεωπεριβάλλοντος, Ε.Κ.Π.Α
Original Title:
Μελέτη της εδαφικής διάβρωσης και των πλημμυρών στο νησί της Σάμου με την εφαρμογή μεθόδων τεχνητής νοημοσύνης (Νευρωνικά Δίκτυα, Ασαφής Λογική) και Γεωγραφικών Συστημάτων Πληροφοριών.
Translated title:
Study of Soil Erosion and Flash Floods with the use of Artificial Intelligence (Fuzzy Logic, Artificial Neural Networks) and Geographic Information Systems. A study case in the island of Samos, Greece.
Summary:
Our planet is formed through slow geologic processes, but there are some rare occurrences of extreme events which can change the landscape rapidly. These are also known as natural disasters which are increasing during the last century mainly because of climatic change. The most important natural disasters are earthquakes, volcanic eruptions, typhoons and floods, while soil erosion can also be considered as a natural disaster.
The aim of this thesis is to identify erosion and flood risk, by determining the relationships between various factors/processes through the use of artificial intelligence (fuzzy logic and artificial neural networks) and apply this methodology in a study case area (Samos island, Greece). Samos island is selected because of interesting geological and geomorphological characteristics and the occurrence of extended natural hazards.
The primary data that are used include 1: 50,000 topographic maps, digital elevation models with a resolution ranging from 30x30m to 5x5m, 1: 50,000 geological maps, meteorological-climatic data for the last 40 years, satellite images SENTINEL-2 (moderate resolution) and Worldview-2 (high resolution), and land use data (from the European Project CORINE 2018) . Additional data were obtained after extensive fieldwork, which includes mapping and measurement of various topographic, geomorphological and hydrological variables as well as the collection of historical data of flood events.
For the data analysis we use Geographic Information Systems (GIS) as well as MATLAB, R, PYTHON and ANACONDA software - programming languages. By analyzing the primary data, mainly through ArcMap and SAGA GIS, various secondary parameters are calculated (e.g. slope, river network, lithology, NDVI index, topographic indices), including some modern topographic indicators such as HAND and DUNE. All data are collected in a geo-database that is created for the study area. By combining various parameters using different models (either classical or artificial intelligence methods) we can estimate erosion and flood risk for the study area.
The results are obtained through the application of classical methods for estimating erosion and hydrological conditions (RUSLE & SCS-CN method) as well as the implementation of models for determining the risk of erosion and flood risk using artificial intelligence methods. In particular, a large number of different Mamdami-type fuzzy logic and SOM-type neural network systems are implemented for the estimation of erosion in the study area. Flood potential is being investigated in four hydrological basins of the island of Samos, at the sub-basin level through the implementation of fuzzy logic systems and the creation of an automated fuzzy system for predicting the soil thickness of systems.
The comparison between the models, shows that both the number of inputs and the model characteristics greatly affect the results obtained, and an optimal model is selected for each application. Through the proposed artificial intelligence methods we have the ability to accurately identify the spatial distribution of areas with high risk of erosion and / or flooding, using data that is not inherently accurate. These methods can prove valuable at local, national, and European level for better planning and management of the environment.
Main subject category:
Science
Keywords:
Natural Hazards Modelling, Soil Erosion, Flood Risk, Geomorphology, Artificial Intelligence, Fuzzy Logic, Neural Networks
Number of references:
349
