Adaptive bias correction for improved subseasonal forecasting.

Area of Science:

• Meteorology
• Climate Science
• Machine Learning

Background:

• Subseasonal forecasts (2-6 weeks ahead) are vital for water management, wildfire control, and mitigating floods/droughts.
• Current dynamical models have limitations in predicting temperature and precipitation due to errors in atmospheric dynamics and physics.

Purpose of the Study:

• To introduce and evaluate an adaptive bias correction (ABC) method for enhancing subseasonal weather predictions.
• To improve the accuracy of temperature and precipitation forecasts beyond current operational model capabilities.

Main Methods:

• Developed an adaptive bias correction (ABC) method integrating machine learning with observational data.
• Applied the ABC method to the European Centre for Medium-Range Weather Forecasts (ECMWF) subseasonal model.

Main Results:

• ABC improved temperature forecasting skill by 60-90% in the contiguous U.S.
• ABC enhanced precipitation forecasting skill by 40-69% in the contiguous U.S.
• Developed a workflow to explain skill gains and identify optimal forecasting windows.

Conclusions:

• The adaptive bias correction (ABC) method offers a substantial improvement over existing subseasonal forecasting techniques.
• This approach provides a practical framework for leveraging machine learning to enhance climate predictions and inform critical resource management decisions.