Using Ensemble of Neural Networks to Learn Stochastic Convection Parameterizations for Climate and Numerical Weather Prediction Models from Data Simulated by a Cloud Resolving Model

A novel approach based on the neural network (NN) ensemble technique is formulated and used for development of a NN stochastic convection parameterization for climate and numerical weather prediction (NWP) models. This fast parameterization is built based on learning from data simulated by a cloud-resolving model (CRM) initialized with and forced by the observed meteorological data available for 4-month boreal winter from November 1992 to February 1993. CRM-simulated data were averaged and processed to implicitly define a stochastic convection parameterization. This parameterization is learned from the data using an ensemble of NNs. The NN ensemble members are trained and tested. The inherent uncertainty of the stochastic convection parameterization derived following this approach is estimated. The newly developed NN convection parameterization has been tested in National Center of Atmospheric Research (NCAR) Community Atmospheric Model (CAM). It produced reasonable and promising decadal climate simulations for a large tropical Pacific region. The extent of the adaptive ability of the developed NN parameterization to the changes in the model environment is briefly discussed. This paper is devoted to a proof of concept and discusses methodology, initial results, and the major challenges of using the NN technique for developing convection parameterizations for climate and NWP models. 1. Introduction Clouds and convection are among the most important and complex phenomena of the Earth’s physical climate system. In spite of intense studies for centuries, clouds still provide an intellectual and computational challenge. Because of the vast range of time and space scales involved, researchers and models that they use typically focus on a particular component of a cloud system, with a narrow range of time and space scales, and prescribe features of the cloud that operate outside of that range. For example, microphysical models describing drop scale motions (e.g., drop coagulation) deal with the fine spatial and temporal scales (of order of millimeters (drop size) and seconds). For more detailed discussion of atmospheric moisture physics, see [1–4]. At the other end of the spectrum of representations of clouds is their representation in large-scale models, for example, in general circulation or global climate models (GCMs), which resolve atmospheric features with spatial scales of the order of 100？km, and temporal scales of the order of 10 minutes. Numerical atmospheric and coupled atmospheric-oceanic-land models, or GCMs, used for climate and numerical weather