Water Level Forecasting in the Gamcheon River, Korea, Using TimeGPT
A real-world forecasting study that tests foundation-model time-series methods against hydrological constraints and evaluation choices.
A study of foundation-model forecasting for river water levels. We compare TimeGPT with classical and linear baselines under rolling-origin evaluation on the Gamcheon River setting.
problem
River forecasting is not only a point-prediction problem: event timing, rainfall context, station sensitivity, and evaluation windows matter for operational usefulness.
key idea
Use TimeGPT as a foundation-model baseline while comparing it against classical/linear approaches under rolling-origin evaluation in a real river setting.
my role
Co-author; contributed to the AI forecasting and evaluation framing.
methods
- • TimeGPT forecasting
- • Rolling-origin evaluation
- • Hydrological time-series preprocessing
evidence / results
- • Published in IJFIS
- • Provides an applied evidence line for domain-specific forecasting systems
why this belongs in the portfolio
- • Connects foundation time-series models to river-water forecasting
- • Highlights evaluation choices that matter outside notebook benchmarks
authors
Jon-Lark Kim, Jae-Hyun Baek, Keon-Hwi Kim, Tae Hyo Baek, Chang-Lae Jang
venue / status
International Journal of Fuzzy Logic and Intelligent Systems
Published with DOI available.
tags