An optimal signal detection approach to the problem of detecting and estimating the greenhouse signal in geophysical noise from acoustic travel time data is presented. The approach incorporates data from a physical ocean model into an optimal signal processing algorithm within the framework of Bayesian Statistical Estimation theory. The problem of estimating the magnitude of warming signal parameters in the presence of mesoscale and additive noise is addressed. The Geophysical Fluid Dynamics Laboratory Modular Ocean Model (GFDL MOM) is used to study both the ocean noise and later the warming signal.;The Optimal Uncertain Field Processor (OUFP) is presented and its performance is evaluated for different models of travel time and a priori information about the travel time uncertainty due to mesoscale variability. Due to lack of data, ocean mesoscale is initially simulated using weighted sums of the Empirical Orthogonal Functions (EOFs) obtained from MOM. The warming signal is assumed to be a range independent exponential function. Accuracy of the travel time model is shown to be critical in processor performance. The OUFP performance is investigated in detail for the special case where the travel time variability due to mesoscale variations is assumed to be Gaussian with known statistics. This processor is called the Optimal Matched Environment Processor (OMEP). The mesoscale EOFs are assumed to have a triangular correlation function with correlation lengths of a 100 km each. Cramer Rao Lower Bounds (CRLBs) are computed and the predicted performance is compared with that achieved by the OMEP. Data obtained from the ocean model suggests that the first five EOFs are correlated up to longer distances than the 100km initially assumed. The presence of energy at distances of up to 1Mm for the first EOF indicates the presence of a basin scale component in the data. The CRLB obtained using the model based EOF correlation functions is substantially higher than the CRLB obtained using the triangular correlation function.;The greenhouse warming signal is modeled as a decrease in outgoing longwave radiation from the ocean boundary. MOM output suggests that the warming signal is not exponential, and that it is range dependent. Only a range independent warming signal is investigated. OMEP performance evaluation results indicate that a priori statistical information about the warming signal is necessary in order to estimate the magnitude of this warming signal, especially at low signal to noise ratios.;Finally, realistic ocean background noise is simulated by incorporating real wind-stress anomaly data from the COADS data set into the ocean model boundary conditions. No assumption is made about the travel time statistics and the performance of the OUFP is re-evaluated by performing Monte Carlo Integration over the environmental uncertainty. The processor performance improves with range but it is found to be insensitive to source depth for the limited number of depths investigated. The OUFP performance initially improves with increasing SNR, but levels off at high SNR due to environmental limitations.
展开▼
