Neural Network Applications at NCEP
OMB has been working on different neural network (NN) applications
since 1994. During this time a valuable level of expertise has been developed
in areas of oceanographic, meteorological, and remote sensing applications
of neural networks. A comprehensive software package based on IDL and FORTRAN
has been written for developing neural network applications. This package
includes an interactive system for neural network training and software
for producing a FORTRAN source code for completed applications. Because
this NN software was developed in-house, a much greater level of understanding
and flexibility in the use of NNs has been achieved.
Developed NN software and expertise can be used for:
-
development of more accurate empirical retrieval
algorithms (3 NN
algorithms for SSM/I has been developed)
-
development of retrieval algorithms through NN
inversion of existing forward models
-
development of fast forward models for direct
assimilation of satellite data in NWP and ocean models, NN forward models
work hundreds times faster than physically based forward models (SSM/I
forward model has been developed)
-
acceleration of existing physically based forward
models
-
acceleration of subblocks of numerical models
(a
prototype of fast
NN algorithm for calculating nonlinear energy transfer in wind wave models
has been developed)
-
fast empirical parametrization (and inverse parametrization)
of physical processes in NWP and ocean models (e.g.,
NN
equation of state of sea water and its NN inversion for salinity has
been developed)
-
downscaling interpolation of model output and
developing MOS- like empirical relationships.
NNs can also be used in many other meteorological and oceanographic application
such as nonlinear principal component analysis, variational data assimilation,
forecasting some atmospheric and oceanic parameters (e.g., damaging winds,
precipitations, road temperature, SST and ENSO), time series forecasting
etc.
Vladimir
Krasnopolsky Page
Return to Neural Networks
Page
Return to AMS:
Committee on Artificial Intelligence
Please send comments and questions to Vladimir Krasnopolsky Vladimir.Krasnopolsky@noaa.gov
Last changed June 2000