Despite significant advances of the dynamic atmosphere models over the last decades of the 20th century, the empirical atmospheric models have been widely used due mostly to both its general applicability and its little dependence on the computational resources. This study is show comparison of precipitation simulation to Northeast Brazil (NEB) - 1971-2000 from large scale dynamical modeling and regional model (downscaling) and the forecast of empirical modeling (K-nearest-neighbor (k-NN). Were user the general circulation model ECHAM4.5 together two regional models, the Regional Spectral Model (RSM/97) from the National Centers for Atmospheric Prediction-NCEP and the Regional Atmospheric Model System (RAMS) developed at Colorado State University. The regional models were nested in ECHAM4.5, forced with the observed Sea Surface Temperature as a boundary condition, for the period from February to May. The results show that the empirical model presented a smaller absolute error than the dynamic models for the periods February to April (FMA) and March to May (MAM) in isolated areas of the north of Maranhão and Piauí states, west and south of Ceará, center-south of Piauí and west and northeast of Bahia state. Regarding model biases, ECHAM4.5 and the RSM/97 produced, to a large extent, a humid bias over large areas of NEB, however with an average precipitation for the northern sector of NEB (2ºS-12ºS and 45ºW-37ºW) close to the observations. RAMS and the analog method had a dominance of a dry bias over NEB, with precipitation totals below the observed values. The model skills (using the Heidke score) were evaluated for three categories Dry (S), Normal (N), Rainy (C), showed that the analogous method has low skills, between 0,1 and 0,3 in all categories, while dynamic models presented superior skills, with larger values for categories S and C (of the order of 0,4 the 0,5), exceeding 0,6 in some areas of the northern sector of NEB for category C as seen in the models ECHAM4.5 and RSM/97 models.
Analogs; Previsibility; Downscaling
