Predicting three-dimensional ocean density structure from remotely sensed surface observations
Date
2010
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Knowing ocean density is essential to understanding ocean dynamics. Many processes are inuenced by ocean density structure, including geostrophic flow, heat transfer, acoustics, light refraction, the timing of phytoplankton blooms, nutrient pumping, water column mixing, and stability. In the ocean, density changes due to heat flux, wind mixing, and fresh water input. Biological responses such as phytoplankton abundance respond quickly to changes in density structure, although in non-conservative ways, because mixing often induces phytoplankton growth. Many of the environmental conditions that either drive (e.g., wind) or are driven by (e.g., phytoplankton) density are observable from remote sensing platforms. Therefore, the relationships between environmental conditions and density structure make it possible to use satellite observations to systematically predict subsurface density structure. In this study, I compiled over 300,000 density profiles from the Argo program, used nonlinear regression techniques to reduce them to five numeric parameters, and matched them with MODIS-Aqua sea surface temperature, normalized water leaving radiance at 412, 443, 488, 531, 551, and 667 nm, surface wind speeds from QuikSCAT, and sea surface height anomalies from Jason-1. I then trained a series of neural networks to develop a set of models that predict vertical density structure down to 2000 m depth based on remote observations. Model RMS errors range from 0.186 kg m..3 to 0.751 kg m..3. The models cover a vast majority of the global ocean (>90%), demonstrating that density profiles can be predicted in virtually every region of the ocean with reasonable errors.