Dr. Mohamed Hamdy Eid, Dr. Attila Kovacs and Professor Peter Szűcs

A new research article published in highly ranked Q1 journal by our researchers, Dr. Mohamed Hamdy Eid, Dr. Attila Kovács, and Professor Péter Szűcs, in Q1 journal (Water). The article, titled “Enhancing Karst Spring Discharge Simulation Through a Hybrid XGBoost–BiLSTM Machine Learning Framework,” addresses a critical challenge in sustainable water resource management: the accurate simulation of karst spring discharge.
Karst aquifers are vital global water resources, supplying drinking water to nearly a quarter of the world’s population. However, their complex, heterogeneous, and non-linear hydrodynamics make predicting their behavior exceptionally difficult. This study focuses on the Jósva Spring in Hungary, located within the Aggtelek Karst area. The research team conducted a comprehensive comparative assessment of five state-of-the-art machine learning models to forecast the daily discharge of this complex system.
The study evaluated traditional ensemble methods (Random Forest and Extra Trees), a powerful gradient-boosting algorithm (XGBoost), and advanced deep learning architectures (BiLSTM and a novel Hybrid XGBoost-BiLSTM model). The models were trained using a five-year dataset comprising rainfall, temperature, and upstream discharge data.
The key finding of this research is the superior performance of the novel Hybrid XGBoost-BiLSTM model. This innovative architecture synergistically combines the powerful feature extraction capabilities of XGBoost with the temporal dependency modeling of a BiLSTM network. The hybrid model achieved the highest predictive accuracy on unseen test data, significantly outperforming standalone models. It effectively captured both the rapid, non-linear flashy recharge characteristic of conduit flow and the slow, sequential recession dynamics of the rock matrix.
Furthermore, the study’s feature importance analysis confirmed a distinct 3-day hydraulic travel time from the upstream catchment, providing valuable empirical validation for conceptual modeling of the region. This research offers clear, evidence-based guidance for developing robust operational prediction tools, ultimately contributing to improved water allocation strategies, enhanced flood preparedness, and the sustainable management of vulnerable karst water resources in a changing climate.

Tovább a cikkhez: mdpi.com/journal/water