Forecasting Ocean Mesoscale Eddies in the Northwest Pacific in a Dynamic Ocean Forecast System

Reliable forecasting of ocean mesoscale eddies is essential for applications such as scientific investigation, ecosystem management, and environmental services. However, comprehensive, large-sample evaluations of eddy forecasts from dynamical ocean prediction systems remain largely absent. This study evaluates the performance of the LICOM Forecast System (LFS), a global eddy-resolving ocean forecast system, in predicting mesoscale eddies over the Northwest Pacific. One year of 1–15 day sea level anomaly (SLA) forecasts was compared with observations using the GEM-M eddy identification and tracking algorithm. A novel distance-based matching framework is developed to objectively link forecasted and observed eddies. This framework pairs correctly forecasted eddies between observation and forecast, while the remaining eddies are classified as missing eddies or false eddies.Statistically, the system slightly underestimates eddy number (~8%) and amplitude (~22%), while overestimating eddy radius (~4%) and velocity (~24%). Despite these biases, LFS reproduces the large-scale spatial distribution of mesoscale variability in both eddy-rich and eddy-poor regions. Further, the matching outcomes reveal that LFS successfully forecasts ~63% of observed eddies at a 1-day lead time, while 37% of the observed eddies were missed, and 31% of the forecasted eddies were false. A key finding is that forecast skill is strongly dependent on eddy dynamical characteristics. Eddies with larger amplitudes and slower propagation velocities are more likely to be correctly predicted and exhibit smaller location errors. Quantitative analysis reveals a significant relationship between eddy amplitude and forecast location errors, particularly for weak eddies (amplitude smaller than 1.1 cm), and a robust linear dependence between eddy propagation speed and forecast error. For eddies with amplitudes greater than 1 cm and velocities below 1 km/day, the mean location errors is reduced to ~71 km at a 1-day lead time, compared to ~80 km for the full sample. This provides practical guidance for the forecasting applications: for eddies with larger amplitudes and slower velocity, the forecast system demonstrated greater accuracy in predicting their location. This study establishes a systematic and scalable framework for evaluating mesoscale eddy forecasts and demonstrates that eddy predictability is closely linked to intrinsic dynamical properties. Also, the proposed matching-based validation framework further distinguishes between correct, missing, and false forecast eddies, providing new insight into the structural limitations of dynamical ocean forecasts and offering a diagnostic tool for evaluating forecast system performance.