Associate Professor, Center for Water Cycle, Marine Environment and Disaster Management
The primary goals of this research consist of three key aspects: employing advanced deep learning approaches to address hydrological problems, evaluating the suitability of such deep learning techniques, and simultaneously developing new deep learning architectures specifically designed for hydrological challenges. Deep learning has emerged as a prevalent approach across a wide range of fields, and its popularity within hydrology has been steadily increasing. However, due to the extensive array of deep learning methods and their continuous improvement and development, it is essential to integrate these state-of-the-art techniques into hydrological studies. Furthermore, while the use of deep learning techniques in hydrology continues to expand, there is a noticeable lack of research exploring their suitability. Most investigations focus on introducing novel methods or analyzing their accuracy. It is crucial, however, to examine the nature of relationships between variables that deep learning algorithms can identify. This research endeavor will address these issues and subsequently attempt to establish new methods based on the knowledge gained from the investigations mentioned earlier.
The current study employs deep learning architectures, such as Recurrent Neural Networks (RNN) and Convolutional Neural Networks (CNN). RNNs are particularly effective for processing time series data, which is prevalent in hydrological settings. Consequently, RNNs will be applied to address relevant hydrological problems. Simultaneously, CNNs are proficient at deriving information from two-dimensional or multi-dimensional datasets, including atmospheric data, precipitation field data, ocean data, and more. CNNs will be deployed for these applications. In addition, new deep learning architectures, such as graph neural networks and transformers, will also be employed. These architectures, along with others, will be used to tackle a variety of hydrological issues. For example, the research aims to improve the accuracy of precipitation or flood forecasting and enhance the reliability of future projections of hydrological variables through the application of deep learning techniques. As a result, this study holds the substantial potential to inform and support disaster prevention and mitigation efforts.