Evaluating Gaussian processes for matched-field processing localization using minimum mean squared error criterion

Gaussian processes (GPs) can densify and denoise sparsely sampled signals and have been applied in matched-field processing (MFP) localization to improve localization accuracy and robustness. Given a known true field, the minimum mean squared error criterion is proposed to evaluate the performance of GP interpolation and its application in MFP localization. This approach allows for the performance comparison of different kernel functions and likelihood functions, assisting in identifying the optimal hyperparameters, interpolation results, and localization outcomes. It also highlights possible challenges faced by existing GP methods under limited data conditions while establishing a performance upper bound for GPs-MFP.