Underwater localization faces many constrains and long-term persistent global localization for autonomous underwater vehicles (AUVs) is very difficult. In this paper, we propose a novel AUV localization method taking advantage of the recent progress in ocean general circulation models (OGCMs). During navigation, the AUV performs intermittent local background flow velocity measurements or estimates using on-board sensors. A series of preloaded flow velocity forecast maps generated by OGCMs are referred by a particle filter in updating particle weights based on resemblance between forecasts and local estimation. A rigorous derivation of the problem in probability theory is presented to reveal the recursive structure of the target distribution function. Simulations in a simple double-gyre velocity field exhibit satisfactory converging localization error. Further simulations in a flow field with local flow fluctuations that are not resolved by OGCMs show similar convergent localization error with a slower converging rate. As a first step towards a new set of underwater localization methods, this work presents promising results and reveals the possibility of realizing converging global underwater localization through partial utilization of the background flow information that is easily accessible.