Repeated Gaussian Processes for Modeling Atmospheric Profiles

Speaker Name: 
Dr. Cliff Bergman
Speaker Organization: 
Lawrence Livermore National Lab
Start Time: 
Monday, April 15, 2019 - 4:00pm
End Time: 
Monday, April 15, 2019 - 5:00pm
Rajarshi Guhaniyogi


Radiosondes collect data on various atmospheric parameters at
multiple irregular heights to provide information about the atmospheric
current profile. Often, binning of the data is used to build a statistical
model of the data. As an alternative, we propose treating each record as a
single realization of a 1D Gaussian Process with a shared mean and
covariance structure. A
deformation correlation structure is used to account for non-stationarity
of the covariance structure. By treating each record as independent,
computation costs are kept relatively minimal allowing for analysis of
large numbers of records. In this talk, discussion of the statistical model
and applications will be presented.