A Bayesian Hierarchical Model for Joint 3D Ocean Properties Reconstructions
Ricardo Lemos1, Bruno Sanso2
1Environmental Research Division, NOAA/NMFS, Pacific Grove, CA, United States; 2Applied Math and Statistics, University of California, Santa Cruz, CA, United States

We present HOMER: a Hierarchical Ocean Model for Extended Reconstructions. The goal is to obtain smooth three dimensional fields of temperature and salinity, as well as long term climatologies, on a monthly time scale. Using a hierarchical Bayesian model we are able to incorporate information from different sources, knowledge about measurement error as well as structural constraints. The latter are imposed in order to make post hoc corrections unnecessary. Inhomogeneities in space are handled using process convolutions based on flexible compactly supported kernels. We consider historical records of up to century long and geographical domains that cover substantial parts of the ocean. This produces massive datasets that require heavy computations. We develop carefully designed Markov chain Monte Carlo algorithms on distributed machines to speed up computations.

Keywords: Hierarchical Models; Bayesian Statistics; Spatio-temporal models

Biography: Bruno Sanso is Professor of Statistics and Chair of the Department of Applied Mathematics and Statistics at the University of California Santa Cruz. He is an elected member of the International Statistical Institute and a Fellow of the American Statistical Association.He is an expert in Bayesian spatio-temporal modeling, environmental and geostatistical applications, modeling of extreme values and statistical assessment of climate variability.