OS11A-2008
Online Detection of Mixed Layer Depth for Autonomous Underwater Vehicles

Monday, 14 December 2015
Poster Hall (Moscone South)
Tara Estlin1, Selina Chu2, Rebecca Castano3, Gail Woodward1, Michelle M Gierach3, Andrew F Thompson4 and Steven Schaffer1, (1)Jet Propulsion Laboratory, Pasadena, CA, United States, (2)NASA Jet Propulsion Laboratory, Machine Learning and Instrument Autonomy, Pasadena, CA, United States, (3)NASA Jet Propulsion Laboratory, Pasadena, CA, United States, (4)California Institute of Technology, Pasadena, CA, United States
Abstract:
The accurate determination of the mixed layer depth (MLD) plays a crucial role in studying ocean dynamics and climate change. Various methods to estimate MLD have been proposed [1, 2]. However there is no current consensus on the best model, which leads to large uncertainty in the estimation. The variability, coupled with the complexity of physical, chemical and biological processes involved and the uncertainty and instabilities of the upper ocean surface, makes estimating MLD a challenging task. MLD varies significantly, even across a small spatial area (< 10km), and this depth is fluctuating, even over a short period of time (< 24 hrs), depending on the season.

This abstract describes our proposed online algorithm for detecting mixed layer depth that would operate onboard an autonomous underwater vehicle (AUV). Using an online method permits a more adaptive approach to estimating MLD. Our proposed algorithm is based on an ensemble approach, which includes data mining techniques for real-time peak and change detection, learned seasonal variability profile, combined with MLD estimation criteria in [1]. In this study, we analyze measurements using glider data collected from the OSMOSIS (Ocean Surface Mixing, Ocean Submesoscale Interaction Study) project, concatenated into a year-long time series [3]. The glider data consists of nine full-depth moorings, which were deployed in a 15 km by 15 km box at the Porcupine Abyssal Plain in the northeast Atlantic, centered at 16.2°W, 48.7°N. Our algorithm utilizes direct measurements of salinity, temperature, depth and time and the design is based on the spatial and temporal variability of MLD learned.

We will present our initial work on tracking the MLD based on real-time simulations using the OSMOSIS glider data and discussed for the case of deploying on a single AUV. Using an online algorithm for estimating MLD in-situ enables the system to rapidly adapt to the variability in a real-world environment and also allows for the intelligent operation of the limited sampling resources available on an AUV. We will discuss the autonomy architecture and algorithm design for implementing this methodology and present results from our initial investigation.

[1] de Boyer Montégut, et al, 2004

[2] Holte and Talley, 2009

[3] https://www2.physics.ox.ac.uk/collaborations/osmosis