Modeling Yearly Cycle of the Ocean's Mixed Layer Depth by Process Convolution

• Activity Number: 53
• Type: Contributed
• Date/Time: Sunday, August 3, 2003 : 4:00 PM to 5:50 PM
• Sponsor: Section on Bayesian Stat. Sciences
• Author(s): Ana Grohovac-Rappold*+ and Michael L. Lavine and Susan Lozier
• Companies: Duke University and Duke University and Duke University
• Keywords: annual cycle ; Bayesian model ; process convolution ; spatio-temporal model

Abstract

Oceanographers are interested in the vertical temperature profiles of the ocean, more specifically, in determining changes with time in the depth M of a wind-mixed surface layer. Temperature profiles, as well as M, vary with time and space. We define M as the depth at which we observe the fastest change in temperature, and model its variability through the continuous space-time Gaussian random field. The random field is constructed by convolving low dimensional white noise process with a time-space specific kernel as in Higdon (1998). We propose to model the annual cyclic pattern of M by restricting the domain in time to a circle whose circumference corresponds to the length of a yearly cycle. Two dimensional spatial kernel is estimated from the spatial dependence reflected in surface heat-wind forcing fields. Such construction has computationally attractive features for large datasets as the dimension of the model is the dimension of the white noise process and not that of the observations.