Konold et al. (2026) BiasCast: learning and adjusting real time biases from meteorological forecasts to enhance runoff predictions

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-08-12
• Authors: Oliver Konold, Moritz Feigl, Patrick Podest, Christoph Klingler, Karsten Schulz
• DOI: 10.5194/hess-30-5067-2026

Research Groups

• Institute of Hydrology and Water Management, BOKU University, Vienna, Austria
• baseflow AI solutions, Vienna, Austria
• ELLIS Unit, LIT AI Lab, Institute for Machine Learning, Johannes Kepler University (JKU), Linz, Austria

Short Summary

This study investigates training strategies and Long Short-Term Memory network architectures to mitigate meteorological forecast-induced bias in maximum daily discharge predictions. The results show that domain shifts between reanalysis and forecast data lead to substantial skill loss, with median Nash–Sutcliffe Efficiency decreasing from 0.58 to 0.33.

Objective

• Investigate the effectiveness of different training strategies and LSTM architectures in reducing meteorological forecast-induced bias in maximum daily discharge predictions.
• Evaluate the performance of models under different data availability scenarios and training strategies.

Study Configuration

• Spatial Scale: Central Europe, with a focus on 451 catchments from the Extended LamaH-CE dataset.
• Temporal Scale: Daily time step, with a focus on 24-hour ahead maximum daily discharge predictions.

Methodology and Data

• Models used: Long Short-Term Memory (LSTM) networks, including baseline, Encoder–Decoder LSTM, Sequential Forecast LSTM, and Transfer Learning approaches.
• Data sources: Extended LamaH-CE dataset, which includes reanalysis data from ERA5-Land, ECMWF-HRES, E-OBS, MSWEP, and GLEAM, as well as forecast data from ECMWF-HRES.

Main Results

• The results show that domain shifts between reanalysis and forecast data lead to substantial skill loss, with median Nash–Sutcliffe Efficiency decreasing from 0.58 to 0.33.
• The Sequential Forecast LSTM demonstrated the most stable improvements, achieving a median NSE of 0.63.
• Integrating recent discharge observations further enhanced performance, raising median NSE to 0.71 and surpassing even the reanalysis-driven baseline.

Contributions

• This study provides new insights into the effectiveness of different training strategies and LSTM architectures in reducing meteorological forecast-induced bias in maximum daily discharge predictions.
• The results highlight the importance of integrating recent discharge observations and using sequential processing with near real-time discharge observations to improve model performance.

Funding

• Not specified.

Citation

@article{Konold2026BiasCast,
  author = {Konold, Oliver and Feigl, Moritz and Podest, Patrick and Klingler, Christoph and Schulz, Karsten},
  title = {BiasCast: learning and adjusting real time biases from meteorological forecasts to enhance runoff predictions},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-5067-2026},
  url = {https://doi.org/10.5194/hess-30-5067-2026}
}