Alexei A. Belochitski

Decadal Climate Simulations Using Accurate and Fast Neural Network Emulation of Full, Longwave and Shortwave, Radiation*

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 and shortwave radiation parameterizations, or to the full model radiation, the most time-consuming component of model physics. The developed highly accurate neural network emulations of the NCAR Community Atmospheric Model (CAM) longwave and shortwave radiation parameterizations are 150 and 20 times as fast as the original/ control longwave and shortwave radiation parameterizations, respectively. The full neural network model radiation was used for a decadal climate model simulation with the NCAR CAM. A detailed comparison of parallel decadal climate simulations performed with the original NCAR model radiation parameterizations and with their neural network emulations is presented. Almost identical results have been obtained for the parallel decadal simulations. This opens the opportunity of using efficient neural network emulations for the full model radiation for decadal and longer climate simulations as well as for weather prediction.

Accurate and fast neural network emulations of model radiation for the NCEP coupled Climate Forecast System: climate simulations and seasonal predictions.

Abstract The approach to accurate and fast-calculating model physics using neural network emulations was previously developed by the authors for both longwave and shortwave radiation parameterizations or the full model radiation, which is the most time-consuming component of model physics. It was successfully tested for a moderate-resolution uncoupled NCAR Community Atmospheric Model (CAM) that is driven by climatological SST for a decadal climate simulation mode. In this study, the approach has been further developed and implemented into the NCEP coupled Climate Forecast System (CFS) with significantly higher resolution and time-dependent CO2. The higher complexity of NCEP CFS required further adjustments to the neural network emulation methodology. Validation of the approach for the NCEP CFS has been performed through a decadal climate simulation and seasonal predictions. The developed highly-accurate neural network emulations of longwave and shortwave radiation parameterizations are, on average, 16 and...

Ensemble of Neural Network Emulations for Climate Model Physics: The Impact on Climate Simulations

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

Tree approximation of the long wave radiation parameterization in the NCAR CAM global climate model

The computation of Global Climate Models (GCMs) presents significant numerical challenges. This paper presents new algorithms based on sparse occupancy trees for learning and emulating the long wave radiation parameterization in the NCAR CAM climate model. This emulation occupies by far the most significant portion of the computational time in the implementation of the model. From the mathematical point of view this parameterization can be considered as a mapping R^2^2^0->R^3^3 which is to be learned from scattered data samples (x^i,y^i), i=1,...,N. Hence, the problem represents a typical application of high-dimensional statistical learning. The goal is to develop learning schemes that are not only accurate and reliable but also computationally efficient and capable of adapting to time-varying environmental states. The algorithms developed in this paper are compared with other approaches such as neural networks, nearest neighbor methods, and regression trees as to how these various goals are met.

Development of neural network convection parameterizations for numerical climate and weather prediction models using cloud resolving model simulations

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

Using neural network emulations of model physics in numerical model ensembles

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