Kunyu: A High-Performing Global Weather Model Beyond Regression Losses
This work addresses the utility gap between numerical and data-driven weather prediction for meteorologists and climate scientists, representing an incremental improvement by combining existing loss types.
The paper tackles the problem of blurring in data-driven global weather forecasting by introducing Kunyu, a model that integrates regression and adversarial losses, resulting in enhanced clarity and realism while outperforming ECMWF HRES in aspects like anomaly extremes and remaining competitive on metrics like RMSE and ACC.
Over the past year, data-driven global weather forecasting has emerged as a new alternative to traditional numerical weather prediction. This innovative approach yields forecasts of comparable accuracy at a tiny fraction of computational costs. Regrettably, as far as I know, existing models exclusively rely on regression losses, producing forecasts with substantial blurring. Such blurring, although compromises practicality, enjoys an unfair advantage on evaluation metrics. In this paper, I present Kunyu, a global data-driven weather forecasting model which delivers accurate predictions across a comprehensive array of atmospheric variables at 0.35° resolution. With both regression and adversarial losses integrated in its training framework, Kunyu generates forecasts with enhanced clarity and realism. Its performance outpaces even ECMWF HRES in some aspects such as the estimation of anomaly extremes, while remaining competitive with ECMWF HRES on evaluation metrics such as RMSE and ACC. Kunyu is an important step forward in closing the utility gap between numerical and data-driven weather prediction.