Study on statistical post2processing of GFS ensemble precipitation forecasts

Coupling ensemble pr ecipitatio n predictio n w ith hydro lo gical mo dels is an impo rtant develo pment direction o f hydro2 log ical fo recasting. Howev er, due to the uncertaint y of the initial atmo spheric conditions and model physics, numerical pr ecipita2 tion for ecasts inevitably hav e errors. In this study , based on GFS ensemble precipitation r efor ecasts with a 1282day lead time, w e analyzed five statistical po st2pr ocessing models that wer e developed based on extended log istic reg ressio n ( ELR) and heterosce2 dastic ext ended lo gistic r egr essio n ( H ELR) alg orithms, and compared t heir co rr ect ion effects on the GFS ensemble precipitatio n forecasts in Xix ian sub2basin of H uai river basin. T he r esults indicated that these five models all made significant impro vements to the GFS raw for ecast; but w ith the ex tensio n o f t he lead time, their correction effects tended to attenuate. In g ener al, the tw o HELR2based models had better perfo rmance compar ed to the other thr ee ELR2based models.

Project Supported:

Clc Number: