Comparing Physically Based and Data-Driven Approaches for Seasonal Streamflow Forecasting in California’s Upper Feather River Basin
2026 World Environmental and Water Resources Congress, Alabama, Amerika Birleşik Devletleri, 26 - 29 Nisan 2026, ss.13-22, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1061/9780784486931.002
- Basıldığı Şehir: Alabama
- Basıldığı Ülke: Amerika Birleşik Devletleri
- Sayfa Sayıları: ss.13-22
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
Seasonal streamflow forecasts are essential for reservoir operations, flood management, and irrigation planning, particularly in snow-dominated basins where climate-driven extremes challenge stationarity-based methods. This study evaluates two complementary forecasting paradigms in the Upper Feather River Basin (UFRB), a critical source for California’s State Water Project upstream of Oroville Dam. The first approach is a physically based hybrid framework that couples dynamical downscaling from the Weather Research and Forecasting (WRF) model with the WEHY-HCM snow-hydrology-hydro-climate model. Forecast ensembles are post-processed using a lead time-dependent exponential-smoothing filter, trained on recent hindcast errors, to reduce bias and quantify predictive uncertainty. This method has demonstrated skillful and reliable confidence intervals for operational planning. In parallel, we develop a purely data-driven framework using Long Short-Term Memory (LSTM) networks. Separate models are trained for each initialization month using historical streamflow records, with extended experiments incorporating precipitation and snow water equivalent (SWE) as additional predictors. Preliminary results suggest that physically based forecasts better capture the timing of seasonal peaks, whereas LSTM forecasts provide smoother, lower-bias estimates across months. By contrasting these methods, we identify the relative strengths and weaknesses of physically based and data-driven forecasts, providing guidance on their applicability to snow-fed watersheds facing increasing hydroclimatic variability.