Predicting Carbonate Chemistry on the Northwest Atlantic Shelf Using Neural Networks

The Northwest Atlantic Shelf (NAS) region has experienced accelerated warming, heatwaves, and is susceptible to ocean acidification, yet also suffers from a paucity of carbonate chemistry observations, particularly at depth. We address this critical data gap by developing three different neural network models to predict dissolved inorganic carbon (DIC) and total alkalinity (TA) in the NAS region from more readily available hydrographic and satellite data. The models predicted DIC with r (2) between 0.913-0.963 and root mean square errors (RMSE) between 15.4-23.7 (mu mol kg(-1)) and TA with r (2) between 0.986-0.983 and RMSE between 9.0-10.4 (mu mol kg(-1)) on an unseen test data set that was not used in training the models. Cross-validation analysis revealed that all models were insensitive to the choice of training data and had good generalization performance on unseen data. Uncertainty in DIC and TA were low (coefficients of variation 0.1%-1%). Compared with other predictive models of carbonate system variables in this region, a larger and more diverse data set with full seasonal coverage and a more sophisticated model architecture resulted in a robust predictive model with higher accuracy and precision across all seasons. We used one of the models to generate a reconstructed seasonal distribution of carbonate chemistry fields based on DIC and TA predictions that shows a clear seasonal progression and large spatial gradients consistent with observations. The distinct models will allow for a range of applications based on the predictor variables available and will be useful to understand and address ocean sustainability challenges. Plain Language Summary The U.S. northeast coast is particularly susceptible to climate change and ocean acidification. However, the lack of observations on seawater carbonate chemistry makes it difficult to assess the impacts of ocean acidification on the region. We address this information gap by developing three different machine learning models to predict carbonate system parameters from more readily available field and satellite data. The models predicted carbonate system parameters with high accuracy and good precision. Compared with other models of carbonate chemistry variables for this region, a larger data set with full seasonal and vertical coverage of the water column and a more complex model architecture resulted in a robust model with low error and uncertainty across all four seasons as well as in surface and subsurface waters. The reconstructed distributions of carbonate chemistry fields on U.S. northeast coast based on one of the models were consistent with observations. We anticipate that the distinct versions of the model will allow for a wide range of different applications based on the predictor variables that are available.