Further studies of the optimization of a hurricane track prediction model using the adjoint equations

The method of model fitting, or adjoint method, is applied to a barotropic hurricane track forecast model described by DeMaria and Jones using a large sample of forecast cases. The sample includes ail Atlantic tropical cyclon es that reached hurricane intensity during the 1989-93 hurricane seasons (1 41 72-h forecasts of 17 storms). The cases considered by DeMaria and Jones are a subset of the present sample. Model-fitting calculations using strong , weak, strong followed by weak, or weak followed by strong model constrain ts are discussed for data assimilation period, varying from 6 to 72 h. Gene rally, the best track forecasts occur for shorter assimilation periods and for weak constraints, although only the 12-h assimilation with the weak con straint has less track error than the control forecast without assimilation , and only for the 12-h forecast. The principle reason for this lack of imp rovement is that the fit of the model to the observed track is god at the m iddle of the assimilation period, bur not very good at the end where the fo recast begins. When a future track position at 6 h is included in the assim ilation, in order to improve the track Fit at the synoptic data time, the r esulting track errors average about 10% smaller than the control forecast. The control forecast may also be improved in the same way. In that case, th e best assimilation forecasts have 2.5% smaller track errors than the modif ied control forecasts.