prikaz prve stranice dokumenta Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning
  Download
PDF 18.28 MB
master's thesis
Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning
2024. urn:nbn:hr:166:541101
Metličić, Nikola
University of Split
Faculty of Science
Department of Physics

Cite this document

Metličić, N. (2024). Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning (Master's thesis). Split: University of Split, Faculty of Science. Retrieved from https://urn.nsk.hr/urn:nbn:hr:166:541101

Metličić, Nikola. "Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning." Master's thesis, University of Split, Faculty of Science, 2024. https://urn.nsk.hr/urn:nbn:hr:166:541101

Metličić, Nikola. "Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning." Master's thesis, University of Split, Faculty of Science, 2024. https://urn.nsk.hr/urn:nbn:hr:166:541101

Metličić, N. (2024). 'Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning', Master's thesis, University of Split, Faculty of Science, accessed 27 December 2024, https://urn.nsk.hr/urn:nbn:hr:166:541101

Metličić N. Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning [Master's thesis]. Split: University of Split, Faculty of Science; 2024 [cited 2024 December 27] Available at: https://urn.nsk.hr/urn:nbn:hr:166:541101

N. Metličić, "Forecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning", Master's thesis, University of Split, Faculty of Science, Split, 2024. Available at: https://urn.nsk.hr/urn:nbn:hr:166:541101

Please login to the repository to save this object to your list.
Metadata
TitleForecasting Height of High-Frequency Sea Level Oscillations in Bakar Bay Using Deep Learning
Title (croatian)Prognoziranje visine visoko-frekventnih oscilacija razine mora u Bakarskom zaljevu primjenom dubokog učenja
AuthorNikola Metličić
MentorJadranka Šepić (mentor)
MentorMatjaž Ličer (komentor)
Committee memberJadranka Šepić (predsjednik povjerenstva)
Committee memberMatjaž Ličer (član povjerenstva)
Committee memberKrešimir Ruić (član povjerenstva)
GranterUniversity of Split
Faculty of Science
(Department of Physics)
Split
Defense date and country2024-09-06, Croatia
Scientific / art field,
discipline and subdiscipline		NATURAL SCIENCES
Physics
Abstract
Significant sea level oscillations can pose severe threats to coastal ecosystems and communities. The high-frequency component of these oscillations, occurring over minutes to hours and driven by atmospheric forces, contributes notably to extreme sea levels. This thesis explores the use of deep learning for forecasting the heights of high-frequency (T < 1 h) sea level oscillations in Bakar Bay. Utilizing tide gauge records and ERA5 reanalysis data, a convolutional neural network was developed to forecast the daily maxima of these oscillations up to 72 hours in advance. The model demonstrated high accuracy for non-extreme events but showed decreased performance for extreme events, particularly with longer forecast lead times. Forecasting extreme events accurately remains a challenge. The thesis also examines the importance of model components through ablation studies and alternative training configurations. This work contributes to the development of more accurate and reliable forecasting models, essential for mitigating coastal flooding risks and supporting the resilience of vulnerable communities. Future improvements could involve integrating finer-resolution atmospheric data and extending the input data time range to enhance the model’s performance for rare extreme events.
Abstract (croatian)
Značajne oscilacije razine mora mogu predstavljati ozbiljnu prijetnju obalnim ekosustavima i zajednicama. Visokofrekventna komponenta ovih oscilacija, koja se javlja na periodima od nekoliko minuta do nekoliko sati i koja je vođena atmosferskim forsiranjima, značajno doprinosi ekstremnim razinama mora. Ovaj rad istražuje primjenu dubokog učenja na prognoziranje visina visokofrekventnih oscilacija razine mora (T < 1 h) u Bakarskom zaljevu. Koristeći podatke mareografa i ERA5 reanalize razvijena je konvolucijska neuronska mreža za prognoziranje dnevnih maksimuma ovih oscilacija do 72 sata unaprijed. Model je pokazao visoku točnost za neekstremne događaje, ali smanjenu učinkovitost za ekstremne događaje, posebno za duže intervale prognoza. Točno prognoziranje ekstremnih događaja ostaje izazov. U radu se ispituje i važnost komponenti modela kroz studije ablacije i alternativnih konfiguracija treniranja. Ovaj rad doprinosi razvoju točnijih i pouzdanijih modela prognoziranja, koji su ključni za smanjenje rizika od obalnih poplava i povećanje otpornosti ugroženih zajednica. Buduća poboljšanja mogla bi uključivati integraciju atmosferskih podataka finije rezolucije i proširenje vremenskog raspona ulaznih podataka kako bi se poboljšala učinkovitost modela za prognoziranje rijetkih ekstremnih događaja.
Keywords
Keywords (croatian)
Languageenglish
URN:NBNurn:nbn:hr:166:541101
Study programmeTitle: Physics; specializations in: specialization in Education, specialization in Astrophysics and Elementary Particle Physics, specialization in Biophysics, specialization in Environmental Physics, specialization in Computational Physics
Course: specialization in Environmental Physics
Study programme type: university
Study level: graduate
Academic / professional title: sveučilišni/a magistar/magistra fizike (sveučilišni/a magistar/magistra fizike)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2024-09-10 12:09:40