Azusa Takeishi - Can we machine-learn the raindrop formation process? скачать в хорошем качестве

Azusa Takeishi - Can we machine-learn the raindrop formation process?

Millimeter-sized raindrops in warm clouds form by condensation and numerous collisions of micrometer-sized cloud droplets. Due mainly to the complexity of the latter process among numerous droplets in various sizes, the rain formation processes are usually parametrized in climate models. This parameterization, however, is often a source of significant discrepancies among models, in terms of rainfall, cloud microphysics, and potentially resultant dynamical conditions as well. We utilize the machine-learning method to obtain a new parameterization that mimics the rain formation processes in a detailed bin model. Although the high computational cost makes it impractical to directly embed the bin model within climate models, this machine-learned parameterization enables the prediction of rain formation processes similar to the bin model.