Could Machine Learning Break the Convection Parameterization Deadlock?

Abstract

Representing unresolved moist convection in coarse‐scale climate models remains one of the main bottlenecks of current climate simulations. Many of the biases present with parameterized convection are strongly reduced when convection is explicitly resolved (i.e., in cloud resolving models at high spatial resolution approximately a kilometer or so). We here present a novel approach to convective parameterization based on machine learning, using an aquaplanet with prescribed sea surface temperatures as a proof of concept. A deep neural network is trained with a superparameterized version of a climate model in which convection is resolved by thousands of embedded 2‐D cloud resolving models. The machine learning representation of convection, which we call the Cloud Brain (CBRAIN), can skillfully predict many of the convective heating, moistening, and radiative features of superparameterization that are most important to climate simulation, although an unintended side effect is to reduce some of the superparameterization's inherent variance. Since as few as three months' high‐frequency global training data prove sufficient to provide this skill, the approach presented here opens up a new possibility for a future class of convection parameterizations in climate models that are built “top‐down,” that is, by learning salient features of convection from unusually explicit simulations.

Using Radar Data to Calibrate a Stochastic Parametrization of Organized Convection

“…By contrast, Rasp et al () and Brenowitz and Bretherton () use machine learning to constrain the convective tendency to high‐resolution simulation data based on a multiscale modeling framework and cloud resolving modeling, respectively. See also the work of Gentine et al () and Schneider et al ().…”

Applying Machine Learning to Improve Simulations of a Chaotic Dynamical System Using Empirical Error Correction

“…Krasnopolsky et al (, ) also found that an ANN could be used to cheaply reproduce the output of the radiation scheme in the National Center for Atmospheric Research Community Atmosphere Model. More recent work has focused on replacing atmospheric models' convection schemes with ANNs, in order not just to reduce the cost of presently used schemes but to allow more expensive, higher‐quality schemes to be used, such as superparameterization (Gentine et al, ; Rasp et al, ) or emulations of convection‐resolving models (Brenowitz & Bretherton, ). O'Gorman and Dwyer () also showed that a model's convection scheme could be replaced by a random forest algorithm, and the model could run stably and reasonably reproduce precipitation extremes.…”

Using Deep Neural Networks as Cost‐Effective Surrogate Models for Super‐Parameterized E3SM Radiative Transfer

“…Deep learning is a subset of machine learning and has demonstrated tremendous increase in accuracy in the areas of image and speech recognition. Deep learning algorithms are also gaining popularity in climate and earth science research (Gentine et al, ; Scher, ). Gentine et al () used deep NN (DNNs) to emulate the effects of unresolved clouds and convection, such as the vertical transport of heat and moisture and the interaction of radiation with cloud and water vapor in an idealized simulation over an aquaplanet using the Super‐Parameterized Community Atmosphere Model.…”

“…Deep learning algorithms are also gaining popularity in climate and earth science research (Gentine et al, ; Scher, ). Gentine et al () used deep NN (DNNs) to emulate the effects of unresolved clouds and convection, such as the vertical transport of heat and moisture and the interaction of radiation with cloud and water vapor in an idealized simulation over an aquaplanet using the Super‐Parameterized Community Atmosphere Model. More recently, Scher () trained deep convolution NNs on a simple GCM and demonstrated that trained convolution NNs can successfully predict the model outputs several time steps ahead.…”