Work

ERA5 reanaliz veri seti ve uzun kisa süreli bellek modelleri ile nehir akimlarinin yeniden oluşturulmasi

Year: 2024
Type: dissertation
Abstract: Although long-term and continuous river flow data is necessary for water resources management, flood and drought analysis, design and management of hydraulic infrastructures, climate change research, and predicting river flows, there are limited river flow data around the world. Climate Reanalysis datasets provide the most complete picture currently possible of past weather and climate. Recently, deep learning has shown great success in various fields including hydrology and climate research. Within the above framework, the scope of this study is to develop a methodology to reconstruct historical daily and monthly river flows by a Long Short-Term Memory (LSTM) model, utilizing the ERA5 Reanalysis climate dataset as inputs. As a demonstration, the proposed methodology is used at selected stations in the Seyhan and Euphrates Watersheds in Türkiye. During the training period (2000-2011), correlation coefficient (R) is within 0.96-0.99 for monthly flows and 0.92-0.98 for daily flows, Nash-Sutcliffe Model Efficiency (NSE) value is within 0.87-0.99 for monthly flows and 0.80-0.96 for daily flows, and Kling-Gupta Efficiency (KGE) value is within 0.81-0.95 for monthly flows and 0.83-0.92 for daily flows. On the other hand, during the test period (2012-2015), R ranges within 0.92-0.97 for monthly flows and 0.87-0.91 for daily flows, NSE value ranges within 0.84-0.92 for monthly flows, and 0.73-0.82 for daily flows, and KGE value ranges within 0.82-0.90 for monthly flows, and 0.72-0.86 for daily flows. To conclude, the proposed river flow reconstruction methodology can successfully reconstruct river flows at selected locations if sufficiently long training and test datasets are available.
Language: Turkish
Open Access status: green
OpenAlex ID: W7139158092
Cites: 84
Cited by:
Related to: 10