HR: 0800h
AN: IN31A-1120 [Abstracts]
TI: Improving the Forecast Accuracy of an Ocean Observation and Prediction System by Adaptive Control of the Sensor Network
AU: * Talukder, A
EM: Ashit.Talukder@jpl.nasa.gov
AF: Jet Propulsion Laboratory, 4800 Oak Grove Dr., Pasadena, CA 91109, United States
AU: Panangadan, A V
EM: Anand.V.Panangadan@jpl.nasa.gov
AF: Jet Propulsion Laboratory, 4800 Oak Grove Dr., Pasadena, CA 91109, United States
AU: Blumberg, A F
EM: ablumber@stevens.edu
AF: Stevens Institute of Technology, Civil, Environmental and Ocean Engineering
Castle Point on Hudson, Hoboken, NJ 07030, United States
AU: Herrington, T
EM: Thomas.Herrington@stevens.edu
AF: Stevens Institute of Technology, Civil, Environmental and Ocean Engineering
Castle Point on Hudson, Hoboken, NJ 07030, United States
AU: Georgas, N
EM: ngeorgas@stevens.edu
AF: Stevens Institute of Technology, Civil, Environmental and Ocean Engineering
Castle Point on Hudson, Hoboken, NJ 07030, United States
AB:
The New York Harbor Observation and Prediction System (NYHOPS) is a real-time, estuarine and coastal
ocean observing and modeling system for the New York Harbor and surrounding waters. Real-time
measurements from in-situ mobile and stationary sensors in the NYHOPS networks are assimilated into
marine forecasts in order to reduce the discrepancy with ground truth. The forecasts are obtained from the
ECOMSED hydrodynamic model, a shallow water derivative of the Princeton Ocean Model.
Currently, all sensors in the NYHOPS system are operated in a fixed mode with uniform sampling rates. This
technology infusion effort demonstrates the use of Model Predictive Control (MPC) to autonomously adapt
the operation of both mobile and stationary sensors in response to changing events that are -automatically
detected from the ECOMSED forecasts. The controller focuses sensing resources on those regions that are
expected to be impacted by the detected events. The MPC approach involves formulating the problem of
calculating the optimal sensor parameters as a constrained multi-objective optimization problem. We have
developed an objective function that takes into account the spatiotemporal relationship of the in-situ sensor
locations and the locations of events detected by the model.
Experiments in simulation were carried out using data collected during a freshwater flooding event. The
location of the resulting freshwater plume was calculated from the corresponding model forecasts and was
used by the MPC controller to derive control parameters for the sensing assets. The operational parameters
that are controlled include the sampling rates of stationary sensors, paths of unmanned underwater vehicles
(UUVs), and data transfer routes between sensors and the central modeling computer. The simulation
experiments show that MPC-based sensor control reduces the RMS error in the forecast by a factor of 380%
as compared to uniform sampling. The paths of multiple UUVs were simultaneously calculated such that
measurements from on-board sensors would lead to maximal reduction in the forecast error after data
assimilation. The MPC controller also reduces the consumption of system resources such as energy
expended in sampling and wireless communication.
The MPC-based control approach can be generalized to accept data from remote sensing satellites. This will
enable in-situ sensors to be regulated using forecasts generated by assimilating local high resolution in-situ
measurements with wide-area observations from remote sensing satellites.
DE: 0520 Data analysis: algorithms and implementation
DE: 4217 Coastal processes
DE: 4260 Ocean data assimilation and reanalysis (3225)
DE: 4262 Ocean observing systems
DE: 4263 Ocean predictability and prediction (3238)
SC: Earth and Space Informatics [IN]
MN: 2008 Fall Meeting