Forecast error estimation in the NRL relocatable ocean ensemble forecast system

A globally relocatable regional ocean nowcast/forecast system has been developed to support rapid implementation of new regional forecast domains. The system is in operational use at the Naval Oceanographic Office for a growing number of regional and coastal implementations. The new system is the basis for an ocean ensemble forecast and adaptive sampling capability.

The forecast system consists of core ocean data analysis and forecast modules, software for domain configuration, surface and boundary condition forcing processing, and job control, and global databases for ocean climatology, bathymetry, tides, and river locations and transports. The analysis component is the Navy Coupled Ocean Data Assimilation (NCODA) system, a 3D multivariate optimum interpolation system that produces simultaneous analyses of temperature, salinity, geopotential, and vector velocity using remotely-sensed SST, SSH, and sea ice concentration, plus in situ observations of temperature, salinity, and currents from ships, buoys, XBTs, CTDs, profiling floats, and autonomous gliders. The forecast component is the Navy Coastal Ocean Model (NCOM). The system supports one-way nesting and multiple model update methods. The ensemble system uses the ensemble transform technique with error variance estimates from the NCODA analysis to represent initial condition error. A perturbed atmospheric surface state (or an atmospheric ensemble) may be used to represent uncertainty in surface boundary conditions, and perturbations of the global model may be used to represent resolved errors in the lateral boundary conditions.

The representation of uncertainty in the ensemble analysis and the growth of uncertainty through the ensemble forecast are controlled by the analysis error estimated by the NCODA analysis. Analysis error estimated through model temporal variability or through forecast error via the analysis increment may over- or under-estimate the forecast error due to the limited availability of subsurface observations. We propose a method to estimate analysis error that includes model variability, recent model forecast error, and historical model or climate uncertainty with specified error growth rates, and accounts for the subsurface sampling. By including the growth back to historical error estimates in the long periods between subsurface sampling, we improve the ensemble spread as measured using both verifying analyses and in situ and remote sensing observations.

We present an overview of the ensemble forecast system and the proposed analysis error estimate, with an assessment of the impact of the analysis error on the ensemble spread in limited-area models.