Development and Verification of an AI-based Long-Term Daily Runoff Model using Meteorological and Topographic Data

This study aims to develop an AI-based long-term daily runoff model that can jointly consider meteorological data and the topographic data of a watershed, and to verify its applicability and performance. To this end, AI models such as LSTM (Long Short-Term Memory) and Bi-LSTM (Bidirectional-LSTM), which can effectively learn the temporal dependency of time-series data, were applied. In selecting the target watershed for model application, several factors had to be considered: the watershed should have representativeness with a certain minimum size, and at least 10 years of meteorological data as well as daily runoff data should be available. Considering these factors comprehensively, the Yongdam Dam watershed, located in the Geum River basin in South Korea and having more than 10 years of long-term data, was selected as the target watershed in this study. As input data, meteorological data that directly affect runoff response, such as daily precipitation and daily temperature, were used. In addition, to reflect the climatic and topographic characteristics of the watershed, the CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) dataset was processed and used for AI model training. Through this, this study was not limited to building a model for a single watershed, but enabled the AI model to learn from data from multiple watersheds, and constructed a long-term daily runoff model that integratively learns meteorological data and topographic data. To evaluate the performance of the trained model, the simulation results of the AI model developed in this study were numerically and visually compared and verified using various hydrological performance indicators and visualizations, including NSE (Nash-Sutcliffe Efficiency) and RMSE (Root Mean Squared Error). This study considered not only meteorological data but also topographic characteristics in AI-based long-term daily runoff simulation, and it is expected that the model can be applied to future mid- to long-term water resources and disaster management.

Acknowledgements

This research was supported by the Specialized university program for confluence analysis of Weather and Climate Data of the Korea Meteorological Institute (KMI) funded by the Korean government (KMA).