Publication Detail

Observing System Simulation Experiments (OSSEs) with Ensemble Kalman Filters for Massachusetts Coastal Waters

Pengfei Xue
166 pp.
MITSG 12-16
$55.00 DOM / $75.00 INT

Observing System Simulation Experiments (OSSEs) were performed to help design an optimal observing network for Massachusetts coastal waters. Nantucket Sound (Part 1) and Massachusetts Bay (Part 2) were selected as two pilot sites and experiments were carried out using Ensemble Kalman Filter (EnKF) data assimilation method.

Part 1. OSSEs in Nantucket Sound
As a typical shallow water coastal region, Nantucket Sound is a tidal-dominated, well-mixed, “flow-through” system with three openings. Experiments in Nantucket Sound were focused on testing the capability of EnKF in determining the optimal monitoring sites for physical variables. Twin experiments were performed using the Northeast Coastal Ocean Forecast System (NECOFS) hindcast fields. Experiments were carried out using EnKF for data assimilation with Ensemble Transform Kalman Filter (EnTKF) and Proper Orthogonal Decomposition (POD) for selecting the optimal monitoring sites. The Singular Evolutive Interpolated Kalman filter (SEIK) was compared with EnKF for the data assimilation efficiency.

Running the unstructured-grid Finite-Volume Community Ocean Model (FVCOM) with perturbed initial fields of currents, water temperature and salinity show that in this shallow coastal system, the velocity and surface elevation are able to restore themselves back to the true state over an inertial time scale after perturbation without data assimilation, while the water temperature and salinity are not. This suggests that in this vertically well-mixed region with strong tidal influence, monitoring should be targeted at water properties rather than velocities. By placing measurement sites at an entrance or exit or a location with the maximum signal variance (EnTKF) or at extrema of the dominant EOF spatial modes (POD), we evaluated the capability of EnTKF and POD in designing the optimal monitoring site for the forecast model system in this region. The results suggest that understanding the multi-scale dynamical nature of the system is essential in designing an optimal monitoring network since “optimal” sites suggested by an assimilation method may only represent a local scale feature that has little influence on a region-wide system. Comparing EnKF and SEIK simulations shows that SEIK can significantly improve the data assimilation efficiency by reducing the ensemble number and increasing the convergence rate.

Part 2. OSSEs in Massachusetts Bay
The OSSEs in Massachusetts Bay (Mass Bay) are focused on evaluating the observation data sampling strategies for water quality variables with a focus on dissolved oxygen (DO). Long-term (1992-present) water quality monitoring records revealed that the DO concentration in Mass Bay exhibited a seasonal cycle but not a significant interannual variability. A multi-domain nested coupled physical-biogeochemical model was developed with an aim at identifying the key mechanisms controlling the temporal/spatial variability of DO in Mass Bay. Built on good agreement between model results and observations, an EOF analysis indicates that DO is dominated by seasonal and spatial varying modes: highest in March-April and lowest in October and varying more significant in the southern bay than in the northern bay. Although biogeochemical, advection and mixing processes vary year to year, the DO concentration in Mass Bay is controlled by a “self-regulation” process through which the resulting net contribution remains little changed interannually. In the northern Mass Bay, horizontal advection, which is connected to the upstream Western Maine Coastal Current (WMCC), plays an important role in the DO variation; in the southern bay, particularly within Cape Cod Bay, a well-defined local retention mechanism results in a longer residence time, and the influence of local biogeochemical processes on the DO variation therefore increase.

Running the water quality model with perturbed initial field of DO but “true” boundary forcing conditions, the system shows an ability to restore DO back to the true state without data assimilation over a recovery time scale of about one month. Since DO in Mass Bay has bay-wide correlation scale, placing a monitoring site of DO for data assimilation in the Bay can reduce the restoring time scale to a week. Running the model with perturbed boundary forcing, the errors propagate into Mass Bay as a result of the flow advection: entering the Bay from the northern boundary and then spreading southward to Cape Cod Bay with a time scale of about one month. Placing a DO monitoring site near the northern entrance for data assimilation can efficiently control the error spreading and restore the model field back to the true state in the interior of the bay. The results suggest that understanding the upstream boundary-control nature of this system is critical for optimal design of sampling strategies in this region.

type: Full theses / dissertations

Parent Project

Project No.: 2010-R/RC-116
Title: Development and Validation of the Water Quality Model System for Massachusetts Coastal Waters

Other publications by this author (as Lead)