WHOI PO

Title: Data-DrivenModeling of 4D Ocean and Coastal Acidification in the Massachusetts and CapeCod Bays from Surface Measurements

Abstract: A significantportion of atmospheric CO2 emissions is absorbed by the ocean, leading toacidified seawater and altered carbonate composition that is harmful to marinelife. Despite these harmful effects, assessing the severity of ocean and coastalacidification (OCA) remains challenging due to the scarcity of in-situmeasurements and the high costs of computational modeling. We develop adata-driven framework to model OCA indicators and test it in Massachusetts Bayand Stellwagen Bank, a region where OCA impacts fishing and tourism industries.First, we train a neural network to predict vertical temperature and salinityprofiles using surface quantities from satellites and in-situ measurements. Therelationship between 2D surface and 3D subsurface properties is captured usingprincipal component analysis applied to a high-resolution historical reanalysisdataset. Next, we use Bayesian regression to estimate region-specificrelationships for total alkalinity, dissolved inorganic carbon, and aragonitesaturation state as functions of in-depth temperature, salinity, and surfacechlorophyll-a concentration. Finally, we generate daily 4D field predictionsfrom surface measurements using the neural network followed by regressionmodels. The framework is evaluated against withheld measurements across depths,locations, and seasons. This model provides valuable insights into theevolution of OCA, and the uncertainty quantification can inform future planningand optimal sensor placement.