SIParCS 2020 - Dallas Foster
Using Probabilistic Machine Learning to Estimate Ocean Mixed Layer Depth from Satellite and In-Situ Data Sources
The ocean mixed layer is an important component of the global climate system because it serves as the intermediary between the atmosphere and deep ocean. The increasing coverage of in situ Argo ocean profile data allows for greater analysis of the ocean mixed layer depth (MLD) variability on subseasonal timescales; however, the aggregated impacts of subseasonal atmospheric and oceanic processes on subseasonal MLD variability are not well known on a global scale. Furthermore, fine resolution gridded MLD estimates require optimal interpolation, a process that often ignores important sea surface variables. Using satellite observations for sea surface temperatures, salinity, and sea height as inputs, we construct multiple machine learning architectures to produce weekly ½ degree gridded MLD anomaly fields (relative to a monthly climatology) with calibrated uncertainty estimates. We test a variety of traditional and probabilistic machine learning techniques to compare both accuracy and probabilistic calibration of estimates. We find that incorporating sea surface data through a machine learning model improves the performance of MLD estimation over traditional optimal interpolation routines in terms of both mean prediction error and uncertainty calibration. These preliminary results provide a promising first step to greater understanding of subseasonal MLD phenomena and the relationship between the MLD and sea surface variables.
Mentors: David John Gagne, Daniel Whitt