The Use of the Breeding Method for Nested WRF-ARW Simulations

The breeding method, originally developed by Toth and Kalnay (1993), is based on the notion that fast-growing errors (modes) are “bred” in each analysis cycle. Perturbed initial conditions are generated and should capture the uncertainty in the control analysis. Rescaled perturbations are then introduced according to a specific breeding interval, by first subtracting the control analysis from the perturbed forecast. In this manner, the bred vectors offer a good estimate of the growing error fields in the analysis (the “errors of the day”). In an ensemble sense, the mean should result in a better forecast when compared to the control forecast, as long as the initial uncertainty is represented by the ensemble.

We apply the breeding method to high-resolution WRF-ARW forecasts covering Vermont and New Hampshire. The forecast duration is currently 48 hours (after the spin-up period), and the configuration consists of three domains (9/3/1-km horizontal resolution); so it is important to understand the effects of the high-resolution nests on the regional instability. During the spin-up period, we apply 3D variational assimilation using WRFDA and conventional observations to generate a new analysis for each domain every 3 hours. These analyses are used to create the new perturbations at each breeding interval.

The breeding method is the first step in understanding and utilizing ensembles and determining what is feasible from an operational perspective at high resolution. We will present the breeding results obtained thus far, challenges, and future work, which includes the utilization of the Localized Ensemble Transform Kalman Filter (LETKF) for data assimilation.