KNOU-Flood: Advancing global streamflow forecasting with lead-aware probabilistic machine learning
KNOU-Flood extends an existing machine learning recipe for streamflow forecasting by incorporating lead-aware forecasts and a probabilistic output layer. The model uses a single-layer LSTM core but improves performance by using longer input windows, integrating multiple forecast sources like GraphCast, and training with a latent-noise ensemble. This results in significant skill gains over the original method under real operational conditions.