Uncertainty Analysis from the Hydrological Perspective of Meteorological Climate Data and Machine Learning-Based Rainfall-Runoff Model through the Application of Meta-Gaussian Techniques

In recent years, extreme heavy rains caused by climate change have led to an increase in flood damage caused by river flooding and levee breaches. As a result, the importance of the rainfall-runoff model is being highlighted for disaster preparedness and efficient water resource management by predicting flooding caused by extreme rainfall. In the past, rainfall-runoff models based on physical models were mainly used, but recently, research is being actively conducted to apply machine learning models thanks to the development of artificial intelligence technology. Therefore, this study applied time-series machine learning models such as LSTM (Long Short-Term Memory) and Bi-LSTM (Bidirectional-LSTM), which are time-series deep learning models based on RNN (Recurrent Neural Network) that are actively used in the field of hydrology, such as predicting inflow and water level. In the existing case of prior research, the performance of models using machine learning techniques was evaluated using statistical validation indicators. This evaluation method has the limitation that it can only check the overall simulated performance, so this study also used an accuracy verification index from a hydrological perspective. In addition, in real watersheds, rainfall that falls on the watershed is discharged through various processes such as evaporation, transpiration, and infiltration. Since these complex processes are not considered in model development and are developed only on a data basis, there is a possibility that there is uncertainty in the simulation results. Therefore, this study analyzed the uncertainty of the simulated results of the model using the Meta-Gaussian method. It is believed that the analysis results of this study will lay the foundation for evaluating the reliability of the model by supplementing the limitations of data-driven models that cannot consider physical hydrological processes through uncertainty analysis