Abstract:
Neural network (NN) is a versatile and generic artificial intelligence
(statistical learning) nonlinear tool. In EMC data/model rich
environment, we could take full advantage of such a flexible tool if
applied it more courageously. Many NN applications have been and are
being developed at EMC. Brief review of these NN applications for: (1)
model initialization, (2) model physics, and (3) post-processing model
outputs is presented. Three of developed NN applications are
discussed in more details: NN observation operator to assimilate
surface parameters (SSH anomaly); fast long and short wave NN radiation
for CGS and GFS; NN nonlinear multi-model ensemble for calculating
precipitation rate over ConUS. NN methodologies developed, working on
these applications, are generic and can be used and are used currently
to solve a variety of problems in post-processing of model outputs,
data assimilation, etc.