Quantifying Acidification Levels in the US North East Coast via Machine Learning Tools

Objectives: Our research objectives are: (1) to analyze and forecast various quantities of interest and their uncertainty directly related to OA, using a flexible Bayesian data fusion framework that leaves no data behind; (2) to build 3D volumetric maps of these quantities in the Massachusetts Bay area, which can also be used everywhere in the Gulf of Maine. The framework also serves as a paradigm shift for effective monitoring strategies for OA in other coastal waters. Our efforts will lead to continuous surface and volumetric space time prediction and uncertainty maps of OA in Massachusetts Bay and Stellwagen Bank.

Methodology: We will use data from all available sources: multi-resolution satellite images, in-situ measurements (e.g.buoys, drifters) and opportunistic measurements such as cruises. We will also make use of simulation data from physics-driven regional ocean models such as FVCOM and HOPS or reduced-order (lumped control-volume-based) models that we can develop ourselves which can be encoded in informative priors in our Gaussian process regression approach. We will collaborate with Prof. Yue at MIT in guiding the deployment of “smart-buoys” in Stellwagen bank, where there is a dearth of regular in-situ measurements, via uncertainty maps, and further incorporating those measurements in our model.

Rationale: There is a lack of high quality OA data for the Mass Bay and especially in the Stellwagen Bank due to the high cost of collecting the samples (cost of buoys, maintenance, etc.). These regions are critical for the fishery industry, tourism, for a wide range of environmental studies, and the blue economy in general. The ultimate goal is to develop a robust data-driven regional forecasting system of aragonite and other quantities of interest (OoI) – an acidification learning system – that will be continuously updated and improved in time and can be used by diverse stakeholders.