How to deal with inhomogeneous outputs and high dimensionality of neural network emulations of model physics in numerical climate and weather prediction models

Authors:
Vladimir M. Krasnopolsky;Stephen J. Lord;Shrinivas Moorthi;Todd Spindler
Affiliations:
Environmental Modeling Center, National Centers for Environmental Prediction, Camp Springs, MD and Earth System Science Interdisciplinary Center, University of Maryland, MD;Environmental Modeling Center, National Centers for Environmental Prediction, Camp Springs, MD;Environmental Modeling Center, National Centers for Environmental Prediction, Camp Springs, MD;Environmental Modeling Center, National Centers for Environmental Prediction, Camp Springs, MD
Venue:
IJCNN'09 Proceedings of the 2009 international joint conference on Neural Networks
Year:
2009

Citing 2
Cited 0

On the approximate realization of continuous mappings by neural networks

Neural Networks
2006 Special issue: Complex hybrid models combining deterministic and machine learning components for numerical climate modeling and weather prediction

Neural Networks - 2006 special issue: Earth sciences and environmental applications of computational intelligence

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this paper we discuss our pilot study where the NN emulation technique developed previously for computing model radiation parameterizations was applied to the part of the NCEP GFS model physics, GBPHYS, that is complementary to the radiation parameterization. The results of the study showed that not all outputs of GBPHYS are emulated uniformly well with the original approach. Significant differences between the radiation parameterizations and GBPHYS block and challenges for the NN emulation approach due to these differences are demonstrated and discussed. Several approaches that will allow us to deal with the challenges and that will be used to complement the NN emulation approach for dealing with entire model physics are also introduced.