Summer Internship Final Report

August 2016

Mentor: Dr. Yi Chao, RSS

Author: Hanna Gratch

Introduction:

Less than five percent of the ocean has been discovered (NOAA).[1] Considering the fact that around 70% of the earth is covered with water, there is much more to uncover about our world. The ocean plays a crucial part in regulating the climate, and is thus important to study as scientists learn more about sustaining the Earth. For example, the carbon cycle, which is largely controlled by the ocean, circulates carbon dioxide (a greenhouse gas). Greenhouse gases directly affect the climate. By sampling the water, scientists can learn more about the ocean.

Autonomous underwater vehicles (AUV’s) can be employed to obtain data such as temperature and salinity. While these vehicles allow scientists to collect more data than manually collecting samples, they are expensive to use. Additionally, due to the vastness of the ocean, it is nearly impossible to cover every area. Therefore, these vehicles must perform efficiently. This efficiency relates to how accurately the vehicles follow an intended path in the ocean.

AUV’s and gliders (Seagliders) are used to carry out experiments. There are various differences between these vehicles. AUV’s need to be sent out and retrieved often (they can stay out for up to one week), whereas Seagliders have more endurance and can last for months in the ocean. However, AUV’s have both horizontal and vertical control and Seagliders have only horizontal control. While AUV’s have thrust power, Seagliders rely on their wings and shifting buoyancy to move in the water. Additionally, AUV’s are more expensive than gliders. These are a few factors that scientists must consider before employing a certain instrument.

The gliders are affected by currents due to their slow speed (they lack thrust power). Therefore, planned paths are not perfectly executed because of the changing currents. By using simulations, we can determine the most efficient path without having to worry about the cost of physically testing an AUV or glider. Employing these vehicles takes a considerable amount of effort with tasks ranging from sending out/retrieving the asset, to tracking the asset and ensuring it is functioning properly. It is important to plan the vehicles with simulations to make the real testing process more effective.

The Regional Ocean Modeling System (ROMS) is a model that predicts the ocean properties including currents, temperature and salinity (1). From the ROMS  model, two types of models are generated for simulations: nature model and planning model.

The nature model is the most accurate model, and planning models are less accurate models. Two models are used because the currents cannot be accounted for exactly, meaning that there will always be a margin of error between the most accurate simulation and the real world model. The models are used to reflect the simulation errors between the most accurate model and the ocean model (real ocean).

Rather than evaluating a single planning model, an ensemble of them can be used. An ensemble consists of a group of planning models with slightly varying initial conditions. With ensemble planning, these planning models are evaluated against each other and ranked based on the most accurate model. In this paper I examine if the use of ensembles can improve the robustness of model performance using Seagliders and if different scoring rules and planning methods can further enhance the performance.

 [1] http://oceanservice.noaa.gov/facts/exploration.html

Full Report: http://remotesensingsolutions.com/wp-content/uploads/2016/10/PlanningAutonomous-Underwater-Vehicles-with-Emsembles.pdf