Adaptive Sampling of Dynamic Processes in Coastal Areas and Fjords using AUVs

The coastal ocean and the upper water column are domains where the need for more sophisticated robotic sampling approaches is critical. Spatial and temporal variability, in addition to episodic events, makes traditional observation practices prone to undersampling.

We describe the use of a data-driven sampling approach developed for adaptive detection and mapping of frontal zones in fjords using autonomous underwater vehicles (AUVs). Fjord systems are challenging to study as they exhibit a number of time-varying structures and gradients on the surface as well as internally in the water column. Determining what resources to deploy and when and where to sample, is therefore an essential question facing scientists in this domain.

Borrowing ideas from statistical quality control and change detection, we use cumulative sums to evaluate changes in temperature and salinity across the fjord channel, aiming to locate and map the most interesting region of a fjord, i.e. regions with strong gradients. The method is adaptive and based on the premise that the AUV will start at the fjord mouth and traverse inwards in discrete transects until a feature of interest is detected. The AUV will then autonomously allocate sampling efforts to sample the feature with higher resolution.

We will present results from a successful test in Korsfjorden (Norway) November 2017 using a Kongsberg Hugin 1000 AUV, where a subsurface frontal feature was detected and mapped autonomously without human intervention. The observations from the experiment emphasize the importance of autonomy for sampling dynamic phenomena such as spatially-evolving gradients. The inclusion of embedded autonomy such as the approach demonstrated here forms a basis for enabling larger observation networks and multi-coordination of robotic arrays.