• English 
  • Spanish 

Statistical downscaling techniques for different forecast ranges

Research on statistical downscaling techniques for different forecast ranges

logo_aemetprj.PNG Project type: National project
Program: AEMET projects
Period: February 2002 - October 2003
Status: Finished
Web: http://www.boe.es/boe/dias/2002/05/04/pdfs/A16390-16391.pdf

Downscaling techniques consist on adapting numerical model predictions to a local/regional scale. The downscaling method considered in this project is an implementation of an analog-based clustering technique which allows to efficiently combining different ensemble forecasts from different models into a single probabilistic forecast with an estimation of the associated predictability (based on the dispersion of the ensembles).

Ensemble prediction models have shown certain skill in the seasonal forecast of synoptic scale patterns (SST, etc.) during some "El Niño" episodes. In this project we use the reanalysis data provided by the multi-model ensemble project DEMETER (http://www.ecmwf.int/research/demeter/) to validate the skill of seasonal regional forecasts in different regions of Peru. We use a downscaling method which extends the analog-based clustering technique using self-organizing maps to arrange the obtained clusters in a 2D lattice. This allows to efficiently combining different ensemble forecasts from different models into a single probabilistic forecast obtained from the distribution of the ensemble forecasts in the 2D lattice. This also gives an estimation of the associated predictability (based on the dispersion of the ensembles). As a result, we obtained skilful regional predictions for strong "El Niño" episodes in two nearby stations of Peru, see figure and documentation below.


2002_Ancell_AEMETprj4_0.png

Keywords: