A novel approach based on the neural network (NN) ensemble technique is formulated and used for development of a NN stochastic convection parameterization climate numerical weather prediction (NWP) models. This fast built learning from data simulated by cloud-resolving model (CRM) initialized with forced observed meteorological available 4-month boreal winter November 1992 to February 1993. CRM-simulated were averaged processed implicitly define parameterization. learned using an NNs. The...

Abstract The approach to accurate and fast-calculating model physics using neural network emulations was previously developed by the authors for both longwave shortwave radiation parameterizations or full radiation, which is most time-consuming component of physics. It successfully tested a moderate-resolution uncoupled NCAR Community Atmospheric Model (CAM) that driven climatological SST decadal climate simulation mode. In this study, has been further implemented into NCEP coupled Climate...

Abstract. The ability of machine-learning-based (ML-based) model components to generalize the previously unseen inputs and its impact on stability models that use these have been receiving a lot recent attention, especially in context ML-based parameterizations. At same time, emulators existing physically based parameterizations can be stable, accurate, fast when used they were specifically designed for. In this work we show shallow-neural-network-based radiative transfer developed almost...

An approach to calculating model physics using neural network emulations, previously proposed and developed by the authors, has been implemented in this study for both longwave shortwave radiation parameterizations, or full radiation, most time-consuming component of physics. The highly accurate emulations NCAR Community Atmospheric Model (CAM) parameterizations are 150 20 times as fast original/control respectively. was used a decadal climate simulation with CAM. A detailed comparison...

A novel approach based on the neural network (NN) technique is formulated and used for development of a NN ensemble stochastic convection parameterization numerical climate weather prediction models. This fast built data from Cloud Resolving Model (CRM) simulations initialized with TOGA-COARE data. CRM emulated are averaged projected onto General Circulation (GCM) space atmospheric states to implicitly define parameterization. comprised as an networks. The developed NNs trained tested....

Abstract This paper describes a microphysics parameterization based on integral moments of the full drop size distributions (DSDs) as opposed to partial approach (sometimes referred Kessler-type parameterization) integrated separately over cloud and rain portion spectrum. does not assume prescribed form DSD but employs model variables that have clear physical meaning: concentration surface area, water content, precipitation flux, radar reflectivity. These can be directly measured assimilated...

Abstract. The ability of Machine-Learning (ML) based model components to generalize the previously unseen inputs, and resulting stability models that use these components, has been receiving a lot recent attention, especially when it comes ML-based parameterizations. At same time, emulators existing parameterizations can be stable, accurate, fast used in they were specifically designed for. In this work we show shallow-neural-network-based radiative transfer developed almost decade ago for...

In this paper the use of neural network emulation technique, developed earlier by authors, is investigated in application to ensembles general circulation models used for weather prediction and climate simulation. It shown that technique allows us: (1) introduce fast versions model physics (or components physics) can speed up calculations any type ensemble 2 -3 times; (2) conveniently an naturally perturbations a component develop perturbed physics), (3) make computation time entire (in case...

A new application of the NN ensemble approach is presented. It applied to emulations model physics in complex numerical climate models, and aimed at improving accuracy simulations. In particular, this long wave radiation widely used National Center for Atmospheric Research Community Model. shown that practically all individual neural network we have trained process development an optimal LWR emulation can be within simulation. Using results a significant reduction simulation errors, namely:...

Development of neural network (NN) emulations depends significantly on our ability to generate a representative training set. Because high dimensionality the input domain that is in order several hundreds or more, it rather difficult cover entire domain, especially its "far corners" associated with rare events, even when we use model simulated data for NN training. In this situation emulating may be forced extrapolate, which beyond generalization and lead larger errors outputs. A new...