Ensemble prediction is implemented for Northeast Brazil using RAMS mesoscale model, initialized with data from the CPTEC atmospheric general circulation model (AGCM). Methods for the ensemble generation consist of using different physical parametrizations and nudging timescales (Newtonian relaxation) for each model run (ENSFI), or perturbing the initial conditions (ENSCI and ENSCI-FRONT). These methods are evaluated against the AGCM analyses and data of automatic meteorological stations located in Ceará State. Perturbations in the initial conditions for ENSCI and ENSCI-FRONT are based on the method "lagged-average forecasting", which takes previous model executions valid for the time and region of interest, and are included in the horizontal components of the wind, imposing a standard deviation of 5 m.s-1. ENSCI-FRONT also includes perturbations in the regional model lateral boundary conditions to deal with a rapid decrease of the ensemble dispersion with the forecast range found when perturbations are only included in the initial conditions. The mean of the perturbed forecasts obtained with ENSCI-FRONT or ENSFI give better results than the unperturbed forecast in most cases.
Ensemble forecasts; Numerical models; Verification
