Michael Pritchard, co-author, UCI assistant professor of Earth system science said that “Clouds play a key role in maintaining Earth’s climate by a cyclic process that involves transportation of heat and moisture, reflecting and absorbing the sun’s rays and by trapping infrared heat rays by producing precipitation
Based on Standard climate prediction models we can estimate cloud physics using simple numerical algorithms that can help to produce simulations extending out as much as a century even though there are some imperfections limiting their usefulness.
The researcher’s team wanted to explore deep machine learning to provide an efficient, objective and data-driven alternative that could rapidly implement mainstream climate predictions where it is based on computer algorithms to execute learning abilities and performance improvement. They started a training on a deep neural network to predict thousands of tiny, two-dimensional, cloud-resolving models that interact with the planetary-scale weather.
Stephan Rasp, an LMU doctoral student in meteorology mentioned that the neural network learned to represent the fundamental physical constraints where the clouds move heat and vapor towards earth that executes processing power towards the cloud-modeling approach.
The researcher’s hopes to conduct further follow-on studies to extend their methodology to set up advanced models that include realistic geography to understand the limitations of machine learning for interpolation versus extrapolation. This study makes sense to apply some of the new principles to climate science which is established on larger data sets especially in these days where new global models are beginning to resolve actual clouds.