Comprehensive knowledge of extreme values is required for designing offshore structures and ocean current turbines. However, data on the return levels of ocean currents are rarely available. This is the case for the Mexican Caribbean, where enormous energy potential in the ocean currents has recently been detected. In this study, long-term numerical data from the Hybrid Coordinate Ocean Model for a depth of 50m was adjusted via linear quantile regression to short-term empirical data for a depth of 49m. The error of the results was estimated using simplified extreme value analysis. Based on the numerical data, a comprehensive extreme value analysis was conducted using the peaks over threshold method and fitting a Generalized Pareto Distribution to the data. This method relies on filtering peaks with a moving time window and an automated threshold selection based on a reparameterised scale parameter of the Generalized Pareto Distribution. The adjusted numerical model is shown to underestimate the empirical data with the error converging to almost 22% for rare events (return period > 10years). The method showed consistent results in the domain, with some anomalies only at the boundaries of the underlying numerical model. The methodology is suitable for estimating the return levels of ocean currents provided by HYCOM, although further research is needed to reduce the error of the numerical model.