Akademik Veri Yönetim Sistemi
Sezen C., Hartmann A., Chen Z.
TÜBİTAK Projesi, 2219 - Yurt Dışı Doktora Sonrası Araştırma Burs Programı, 2023 - 2024
Karst aquifers are important for freshwater supply but difficult to manage, due to highly variable water levels and spring discharge rates. Numerical models are meaningful tools and can provide useful information for managing karst water resources. In recent years, the increasing frequency of extreme events, such as droughts and floods make the modelling studies more critical. In this regard, forecasting aquifer discharge behaviour under extreme flow conditions is highly important to prevent natural disasters as well as efficient water resources management. Many factors such as heterogeneous aquifer properties, changing climate characteristics, anthropogenic effects and geographical formations can make the modelling process more challenging. In recent years, different modelling approaches, such as conceptual, physically-based and machine learning models have been implemented to improve the modelling performance in karst systems. All these approaches can have limitations or advantages for the modelling. Conceptual models can generally require fewer parameters than physically-based models; however, they can have challenges, especially in the calibration process, in highly non-linear catchments. The physically-based models can need a large dataset including both hydrometeorological variables and aquifer parameters related to catchment characteristics. The machine learning models are based on the procedure, which establishes a relationship between input and output variables without taking into the physical process in the basin consideration for modelling.
The main targets of the research project can be stated under three headings. Firstly, it is aimed that a hybrid modelling approach by combining conceptual hydrological and machine learning models, which is suitable for karst systems with highly variable discharge behaviour, will be developed. Secondly, the hybrid modelling performance will be evaluated by focusing on extreme events (flood and drought), which have been considered rarely in previous research. Finally, the proposed hybrid-modelling approach will be applied for different test sites in Turkey, Slovenia and Switzerland, which are characterized by different hydrological and hydrogeological settings in order to evaluate its transferability. In this respect, it is aimed that the Génie Rural à 6 paramètres Journalier (GR6J) daily lumped conceptual model and VarKarst karst model, will be combined with the wavelet-based machine learning models, such as deep neural network (DNN) and regression tree ensemble (RTE) models for karst discharge forecasting. The development and implementation of the proposed hybrid modelling approach will be done by using open source software. The source code and research results will be published in well-known international peer-review journals. This research, which is based on improving the hybrid modelling approach in the karst systems, will be guiding for the relevant authorities with regard to the determination of discharge capacity of karst aquifers and forecasting of extreme events, such as floods and droughts.