Quantifying Anthropogenic CO₂ Influence on the Inorganic Carbon Cycle in Cape Cod Bay via Multi-Linear Regression and Machine Learning Models

Coastal acidification—the near-shore manifestation of ocean acidification—remains far less constrained than open-ocean chemistry, especially in semi-enclosed systems such as Cape Cod Bay (CCB). As anthropogenic CO₂ emissions continue to rise, resolving local variability in carbonate chemistry has become increasingly important. Compared with the neighboring Gulf of Maine, CCB displays much greater fluctuations in parameters like in situ pHₜ, total alkalinity (TA), and dissolved inorganic carbon (DIC), driven by freshwater inputs, submarine groundwater discharge (SGD), and stronger biological and human influences. In this work, we construct predictive models for TA, DIC, and pH at 20 °C using multilinear regression and machine learning algorithms trained on in situ observations. These predictions are then used in CO2SYS to estimate pHₜ with very high accuracy (mean residual < 0.0001; standard deviation < 0.1). Using the modeled carbonate fields, we generate basin-wide maps of calcite and aragonite saturation states for the entire bay through spatial interpolation. Our results indicate that the inorganic carbon system of Cape Cod Bay is governed by the combined effects of seasonal air–sea CO₂ fluxes, terrestrial inputs, and water-mass retention, making it an acidification hotspot that differs markedly from the Gulf of Maine. Locally trained predictive models outperformed ESPER’s regional products, underscoring the importance of site-specific approaches. Together, these findings provide new perspectives on coastal carbon dynamics and establish a scalable framework for forecasting carbonate chemistry in other nearshore environments.