Menu Close

Statistical Post-Processing Approaches

Ana­log Ensem­ble prob­a­bilis­tic per­for­mance against logis­tic regres­sion for 6‑hourly pre­cip­i­ta­tion data at rain gauges over Ger­many


Post-pro­cess­ing in weath­er and cli­mate is aimed at enhanc­ing the out­put of metero­log­i­cal mod­els by apply­ing a sta­tis­ti­cal mod­el­ing frame­work using the sim­u­lat­ed esti­mates togeth­er with obser­va­tions in a train­ing data set. These meth­ods are often used to enhance weath­er fore­casts by account­ing for the sys­tem­at­ic errors in the mod­el out­put. Post-pro­cess­ing approach­es can also be applied to cli­ma­to­log­i­cal mod­el data, i.e., reanaly­ses, to bring the esti­mates clos­er to the truth or to increase the tar­get­ed res­o­lu­tion such as in sta­tis­ti­cal down­scal­ing approach­es.

One aspect we focus on is the devel­op­ment of prob­a­bilis­tic post-pro­cess­ing of reanaly­sis data. Such a method can­not only pro­vide improved esti­mates of rel­e­vant mete­o­ro­log­i­cal para­me­ters like wind gusts or pre­cip­i­ta­tion, but also enable the assess­ment of the uncer­tain­ty which is of great impor­tance for risk assess­ment, e.g., in the field of renew­able ener­gy.

Exam­ples for cur­rent devel­op­ments are

  • Down­scal­ing post-pro­cess­ing scheme of region­al reanaly­ses using the ana­log ensem­ble approach
  • Mul­ti-method post-pro­cess­ing of weath­er fore­casts aimed at radar data with clas­si­cal com­pared to neur­al net­work (AI) approach­es
  • Prob­a­bilis­tic post-pro­cess­ing for the spa­tial dis­tri­b­u­tion of wind gusts (e.g. gusts at dif­fer­ent ver­ti­cal heights or over extend­ed spa­tial areas).