Ensemble Forecasting in the Gulf of Mexico: LETKF vs. ET

Patrick J Hogan, US Naval Research Laboratory, Ocean Sciences Division, Stennis Space Center, MS, United States, Clark David Rowley, Naval Research Laboratory, Oceanography, Stennis Space Center, MS, United States, Seth Camp, Naval Research Laboratory, Stennis Space Center, United States, Ole Martin Smedstad, QinetiQ North America, Stennis Space Center, MS, United States and Prasad Thoppil, Naval Research Laboratory, Oceanography Division, Stennis Space Center, MS, United States; John C. Stennis Space Center, Stennis Space Center, MS, United States
Abstract:
Two ensemble forecast systems have been implemented for mesoscale prediction in the Gulf of Mexico. One uses an Ensemble Transform approach, which uses a 3-D variational analysis and forecast error variance to generate perturbed initial conditions (i.e. it’s based on a 3DVar analysis). The other uses a Localized Ensemble Transform Kalman Filter (LETKF) approach, which uses ensemble covariances to generate an analysis and the subsequent perturbed initial conditions (i.e. it’s based on an ensemble analysis). Both are also forced by atmospheric perturbations (both winds and fluxes), although the perturbation generation mechanisms are slightly different, and both assimilate all available ocean observations for the cycling data assimilation used in the analyses. Both systems are configured with forward ocean models with ~3 km horizontal resolution, although the ET based system uses the Navy Coastal Ocean Model and the LETKF system used the Hybrid Coordinate Ocean Model.

A series of 14 day forecasts were generated with each system, and the ensemble mean and spread (forecast uncertainty) between the two are compared and discussed. Various metrics have been calculated to assess the performance of the ensemble, including anomaly correlations, spread-skill relationships, and rank histograms. Both systems have the highest ensemble spread near the edge of the energetic Loop Current Eddy (as expected), although both are also relatively under-dispersive when compared to unassimilated observations.