Towards a Cost-Effective Monitoring System of Coastal Ocean Acidification in the US North East

PI: Themistoklis Sapsis, MIT, Harold Hemond, MIT, John Leonard, Massachusetts Institute of Technology, Changsheng Chen, UMass - Dartmouth, George Karniadakis, MIT, Michael DeFilippo, MIT, Robert C Beardsley, WHOI, Carolina Bastidas, MIT, Paris Perdikaris, MIT Sea Grant, Thomas Consi, MIT Sea Grant

Project Number:2017-R/RCM-49

Start Date:2017-02-01End Date:2019-01-31

Proposal Summary

Objectives: Accurate and repeatable data collection within the coastal marine environment acquired over an extended duration of time enables us to analyze and understand the effects of climate change, particularly ocean acidification, to ocean dynamics and marine life. It is important to increase monitoring of coastal acidification (COA) within the coastal environment, where the effect of anthropogenic input on vital natural resources is maximized. However, instrumenting the entire Boston Harbor, Mass Bay and GOM with pH sensors is not practical and, in fact, it is very expensive! On the other hand, computer simulation can help increase the accuracy of the forecasting and alleviate some of the cost but at the present time there are no validated biochemical simulation models for the Northeast, i.e. no simulated outputs for pH, pCO2, TA or DIC are available in our region. We propose an integrated methodology based on theory, measurements and simulation that will work synergistically to design a cost-effective and robust COA monitoring platform in the Northeast. In particular, our proposal includes the use of all available information from monitoring stations in Massachusetts Bay, Boston Harbor and the rivers that input to this system. The approach that we propose is based on deep Gaussian Processes (GP) that will enable the use of fusion of information from diverse sources at variable fidelity, including high fidelity laboratory analysis for DIC and TA, high accuracy in-situ pH and pCO2 measurements, computer simulations, and other measurements of these and other parameters (such as T, Sal and nutrients) using more readily affordable equipment. This vision of the new paradigm is a multi-fidelity Bayesian framework, where all uncertainties are accounted for and propagated in the final prediction that is based on information provided at all the levels of fidelity. The total uncertainty is also part of the prediction and can be used for designing future stations and new monitoring campaigns – a form of “active learning” in our multi-fidelity paradigm inspired by recent developments in machine learning.

More specifically, we propose to conduct integrated theoretical, simulation and experimental work in order to optimize research and outcomes of ocean acidification in the coastal environment of New England. The long-term goal is to develop a cost-effective COA/OA monitoring infrastructure for Gulf of Maine (GOM), but we will test this idea first in the Boston/Harbor/Mass Bay region for which we have some in situ measurements available thanks to MWRA, see Table 1. We have already validated the FVCOM simulation model in this region but we need to incorporate and validate a new biogeochemical model (see technical details in section 2.2). We will leverage existing stations by MWRA and conduct new measurements but in addition we will start a new experimental campaign in instrumenting the coastal stations and rivers and Mass. Bay to obtain multi-fidelity measurements (see sections 2.3 & 2.4). In the first stage, the theoretical work will involve GP regression analysis of existing data from Massachusets Water Resources Authority (MWRA) measurements (temperature salinity, and pH among other variables), satellite data (SST, SSS and Chl), and FVCOM simulation outputs (temperature, salinity, Chl-a 3D fields but no pH). This analysis will establish (for the first time) pH stochastic response surfaces and empirical correlations in the Boston Harbor that can be used to generate initial conditions and to calibrate parameters in the new biogeochemical model of FVCOM (i.e. FVCOM-ERSEM). The corresponding uncertainty quantification (UQ) analysis that will be conducted will be used to determine where we will require new critical measurements, and how to best plan for such an experimental campaign given limited budgets, including opportunistic cruise data collection. Subsequently, we will design the multi-fidelity Bayesian framework using both linear GPs and deep GPs so that we best exploit the synergy of low and high fidelity measurements as well as the simulation outputs to obtain a robust predictive capability endowed with uncertainty quantification. We will train and validate the multi-fidelity framework and its predictive capability at the end of the first and also at the second year when we will have a lot of new measurements.

Methodology: We propose an integrated and synergistic theoretical-experimental-computational approach that will result in an effective new framework of data-driven prediction of COA in the Northeast.

1. We will develop the theoretical framework that relies on multi-fidelity and fusion of diverse information sources, including both measurements and simulation outputs.

2. We will establish a marine biogeochemistry and ecosystem model system (FVCOM-ERSEM) for application in monitoring and eventually predicting COA in the Northeast.

3. We will augment existing sampling campaigns in Massachusetts Bay and Boston Harbor by including measurements of dissolved inorganic carbon (DIC) and alkalinity to the standard set of measurements taken by the MWRA. In addition we will instrument the research vessel AUK to take additional readings in transects running from Boston Harbor to Stellwagen Bank.

4. We will develop a system to directly measure the discharge of the Charles and Mystic Rivers into Boston Inner Harbor. Discharge data from a USGS gaging station on the Neponset River will give us the total river discharge into Boston Harbor. We will augment MWRA river samples with samples taken for DIC and alkalinity analysis.

At the end of the two-year project we will deliver and make available to interested parties and stakeholders via MIT Sea Grant the validated multi-fidelity software, the validated biogeochemical FVCOM model for Boston Harbor/Mass Bay, new insights on the effect of rivers connected to Boston Harbor, seasonal trends of acidification in Mass Bay, and stochastic response surfaces of TA, DIC, pH and pCO2, i.e., all main quantities for interpreting COA.

Rationale: Our project is relevant to the MIT Sea Grant College Program strategic plan. These goals include Healthy Coastal Ecosystems and Sustainable Fisheries and Aquaculture, which is one of the main focuses of MIT Sea Grant College Program. Ocean acidification is the focus of many environmental groups both in the public and private sector. The effects of ocean acidification will impact coastal communities not only in the United States, but around the world. Coastal communities rely heavily on their local waters for healthy economies. The economic impacts of ocean acidification are only starting to be understood. From a coastal community prospective, we need to understand the impacts and effects of the increase in ocean pH levels.

The increased development of coastal communities has brought more runoff, sedimentation, nutrients and contaminants, and degradation of habitats in the coastal areas. MIT Sea Grant has a long history of supporting improved water quality, particularly in urban areas such as Boston Harbor. Constant and consistent monitoring of pollutants over a large area, such as Massachusetts Bay is extremely expensive as it requires significant man power and other associated costs. However, we can accomplish this cost-effectively by utilizing data from existing monitoring efforts (e.g. MWRA), augmenting these efforts with additional sampling and analysis, equipping vessels of opportunity with autonomous underway sensors, instrumenting the rivers, and employing autonomous surface vessels to increase sampling in targeted areas where current data is uncertain. Integrating this set of multi-fidelity data with the biogeochemical-augmented coastal model (FVCOM-ERSEM) within our multi-fidelity framework will result in highly accurate predictions of COA that will enable improved and more cost-effective management of coastal resources, environmental preservation, and remediation efforts.

Researchers and managers, particularly those working in coastal resources and processes can greatly benefit from the results of this project. For example, a better depiction of acidification in coastal waters, with higher spatial and temporal resolution for carbonate variables, will benefit the forecasts of commercially important species made with bioeconomic models. Results from this research can also allow others to formulate adaptive measures for aquaculture businesses (such as oyster growers, hatcheries) along the New England coast. Our velocity-index system for continuous measurement of river discharge into Boston Harbor provided critical and often unavailable data for modeling complex harbors and bays with multiple freshwater inputs. A better understanding of river impact on COA will enable more comprehensive pollution control policy that explicitly recognizes the intimate connection of the coast and the coastal watershed. Our measurements for COA parameters (DIC, TA, pH, pCO¬2) will be available for all researchers. These data sets will be the most comprehensive set of parameters to determine carbonate chemistry than any data currently available in Massachusetts Bay and Boston Harbor. One of the most important immediate outcomes of this project is to establish a publicly accessible, user-friendly biogeochemical and ecosystem model system for regional stakeholders and ocean research communities for the assessment and prediction of COA in the northeast coastal region. The model system will provide an essential tool for understanding the complex biogeochemical and ecosystem dynamics in this region and will have a wide variety of environmental prediction and management applications. Outreach also includes publication of results on the NECOFS Web Map Server (WMS) and the NERACOOS website. The multi-fidelity framework that we will develop is unique and can be employed in different contexts in ocean modeling but also in other climate modeling projects.


none yet available


none yet available