Menu Close

Statistical Downscaling of a High-Resolution Precipitation Reanalysis Using the Analog Ensemble Method

Keller, J.D., Delle Monache, L., Alessan­dri­ni, S. (2017) in J. Appl. Mete­or. Cli­ma­tol., 56, 2081–2095

Link to arti­cle

This study explores the first appli­ca­tion of an ana­log-based method to down­scale pre­cip­i­ta­tion esti­mates from a region­al reanaly­sis. The uti­lized ana­log ensem­ble (AnEn) approach defines a met­ric with which a set of analogs (i.e., the ensem­ble) can be sam­pled from the obser­va­tions in the train­ing peri­od. From the deter­mined AnEn esti­mates, the uncer­tain­ty of the gen­er­at­ed pre­cip­i­ta­tion time series also can eas­i­ly be assessed. The study inves­ti­gates tun­ing para­me­ters of AnEn, such as the choice of pre­dic­tors or the ensem­ble size, to opti­mize the per­for­mance. The approach is imple­ment­ed and tuned on the basis of a set of over 700 rain gauges with 6‑hourly mea­sure­ments for Ger­many and a 6.2‑km region­al reanaly­sis for Europe, which pro­vides the pre­dic­tors. The obtained AnEn esti­mates are eval­u­at­ed against the obser­va­tions over a 4‑yr ver­i­fi­ca­tion peri­od. With respect to deter­min­is­tic qual­i­ty, the results show that AnEn is able to out­per­form the reanaly­sis itself depend­ing on loca­tion and pre­cip­i­ta­tion inten­si­ty. Fur­ther, AnEn pro­duces supe­ri­or results in prob­a­bilis­tic mea­sures against a ran­dom-ensem­ble approach as well as a logis­tic regres­sion. As a proof of con­cept, the described imple­men­ta­tion allows for the esti­ma­tion of syn­thet­ic prob­a­bilis­tic obser­va­tion time series for peri­ods for which mea­sure­ments are not avail­able.