The MASCOT project is using and developing methods in statistics and robotics to conduct effective exploration of ocean variables. We have a particular focus on frontal systems; zones where rivers meet the fjord and the freshwater mixes with the more saline fjord waters, and zones where fjords meet the ocean. Such fronts have shown to be important in biological oceanography. By effectively mapping these zones, we can contribute with decision support for fisheries and environmental monitoring. We have so far conducted tests in the Trondheim fjord and off the coast of Porto, Portugal.
In the MASCOT project, we rely on information from satellites or numerical ocean models to construct prior models for the variables of interest such as temperature, salinity, intensity of plankton, etc. Such a prior model is a statistical representation that captures the important trends, uncertainties and correlations in space and time. Critically, the model must scale well and be possible to update quickly when new information becomes available, as this facilitate real-time operation by an underwater vehicle. We are using such models in conjunction with algorithms from artificial intelligence to plan informative sampling paths for the underwater vehicle, so that it explores targets at the right time and right place, which are expected to have a substantial effect on the model.
Website: https://wiki.math.ntnu.no/mascot
The project aims to build the scientific foundations of statistical sampling for oceanographic applications by formulating novel algorithmic methods in statistics and blending it with ocean model
predictions, to embed and test on autonomous vehicles. By sampling, we refer to the design of
experiments in spatio-temporal domains, enabling autonomous platforms to decide on an optimal
strategy of where and when to gather data, in a cost-effective manner.
Renowned oceanographer Walter Munk called the 20th century, the century
of undersampling, something particularly
relevant in ocean-facing Norway with its complex fjord systems
intermixed with coastal skerries.
To improve the state of sampling modern tools and methods, including
the use of autonomous platforms, oceanographic models and satellite
remote sensing at various spatio-temporal scales are
critical. However, without adequate understanding of the theoretical
underpinnings of how, when and where to sample, these
tools and methods are insufficient in our vast and harsh oceans.
The focus of this proposal is in designing, implementing and
testing algorithms for efficient spatio-temporal sampling of the coastal oceans,
with a broader impact to commingling methods in spatial and computational statistics,
with oceanography, with novel methods in automatic control including
artificial intelligence for adaptive sampling. Deliverables
include testing of the new algorithms in field experiments in
Norwegian waters with existing robotic assets.