Semestral Journal of Centro Argentino de Meteorólogos, which is published  since 1970 and serves on the Core of Argentine Scientific Journals since 2005. Meteorologica publishes original papers in the field of atmospheric sciences and oceanography.

Registration number of intellectual property: 2023-95212445-APN-DNDA#MJ

ISSN 1850-468X


Sofia Ruiz Suarez, Mariela Sued, Luciano Vidal, Paola Salio, Daniela Rodriguez, Stephen Nesbitt y Yanina Garcia Skabar

Servicio Meteorológico Nacional, Buenos Aires, Argentina
Instituto de Cálculo, Facultad de Ciencias Exactas y Naturales - UBA
Centro de Investigaciones del Mar y la Atmósfera- UBA
Departamento de Ciencias de la Atmosfera y los Océanos - UBA
UMI-Instituto Franco Argentino sobre Estudios del Clima y sus Impactos CNRS 3351, Buenos Aires, Argentina
Department of Atmospheric Sciences, University of Illinois at Urbana-Champaign, Urbana-Champaign, USA
Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)

Manuscript recevied on 8th August 2017, in its final form on 2nd January 2018


Data coming from meteorological radars is of the utmost importance for the diagnosis and monitoring of precipitation systems and their possible associated severe phenomena. The echoes caused by objectives that are not meteorological introduce errors in the information. Therefore, it is necessary to detect their presence before using this data. This paper presents four supervised classification techniques based on different models which seek to give an answer to this problem. In addition, as an important part of this work, resampling techniques were implemented on the training set in order to further asses the results. Resampling methods are an indispensable tool in modern statistics. Those techniques provide additional information about the model of interest by repeatedly drawing samples from then data.
Based on data from a C-band Dual-PolarizationDoppler weather radar located in Anguil and from a previous expert’s manual classification, four supervised classification methods with different degrees of flexibility in their structure were implemented: Lineal Model, Quadratic Model, Logistic Model and Bayes Naive Model. Finally, the results of each of them were assessed and compared. Although difficulties were encountered in classifying boundary zones between classes, the results obtained were adequate, showing the best performance in the least flexible model, the linear one. It is considered necessary to keep working in this line of research in order to include more cases in the analysis and allow a better inference on the results.