Parametric study of the analog ensembles algorithm with clustering methods for hindcasting with multistations

Araújo, Leonardo Oliveira; Balsa, Carlos; Rodrigues, Carlos Veiga; Rufino, José

http://hdl.handle.net/10198/24627

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
Artigo_AnEn_WorldCist21.pdf		822.79 KB	Adobe PDF	Download

Send Feedback

Authors

Araújo, Leonardo Oliveira

Balsa, Carlos

Rodrigues, Carlos Veiga

Rufino, José

Abstract(s)

Weather prediction for locations without or scarce meteorological data available can be attempted by taking meteorological datasets from nearby stations. This hindcasting problem can be successfully solved using the Analog Ensemble (AnEn) method. This paper presents a parametric analysis of the AnEn method, and two variations (based on K-means and fuzzy C-means clustering methods), when used to search for analog ensembles in a historical dataset. The study allowed to identify the parameter combinations that yield the best prediction accuracy, improving 13% on the systematic error and 5% on the random error of the previous results obtained with the same dataset. In addition, important performance gains were achieved at the computational level.

Keywords

Analog ensembles Clustering Time series Meteorological data Hindcasting

URI

http://hdl.handle.net/10198/24627

Citation

Araújo, Leonardo; Balsa, Carlos; Rodrigues, C. Veiga; Rufino, José (2021). Parametric study of the analog ensembles algorithm with clustering methods for hindcasting with multistations. In World Conference on Information Systems and Technologies, WorldCIST 2021. p. 544-559. ISBN 978-3-030-72650-8