@InProceedings{MindlinGoyHurTedZil:2022:SiChSe,
author = "Mindlin, Julia and Goyal, Rishav and Hurtado, Santiago Ignacio and
Tedeschi, Renata Gon{\c{c}}alves and Zilli, Marcia",
affiliation = "{Universidad de Buenos Aires} and {University of New South Wales}
and {National University of La Plata} and {Instituto Nacional de
Pesquisas Espaciais (INPE)} and {University of Oxford}",
title = "A Simple Characterization of Sea Surface Temperature Patterns that
Represent the Seasonal Evolution of El Niņo Southern Oscillation
Flavors",
year = "2022",
organization = "AGU Fall Meeting",
publisher = "AGU",
abstract = "In the last decade, much of the attention on El NinoSouthern
Oscillation (ENSO), especially its teleconnections, has focused on
differentiating among the events' {"}flavors{"}. Despite that, the
definition of each flavor, useful when classifying events and
performing composite analysis, is highly variable in the
literature. Furthermore, most of the literature focuses on
preferential seasons when the sea surface temperature (SST)
anomalies are maximized, resulting in deep convection and Rossby
Wave triggering. This work uses k-means clustering algorithms to
distinguish significantly different patterns for SST monthly
variability considering four SST datasets with varying spatial
resolution: Kaplan Extended SST v2, HadlSST, COBE-SST2, ERSSTv5.
SST anomalies are clustered over the entire tropical Pacific Ocean
(140E,15S to 280E,15N) rather than restricting it to arbitrary
regions as in traditional ENSO indices. The analysis also treats
ENSO as an evolving system by considering the entire year,
classifying monthly SST anomalies into a limited number (seven) of
SST patterns (clusters). We first train the clustering algorithm
with satellite-era data (1979-2013), identifying seven patterns
tested against white noise using a classifiability index. The
number of patterns is similar across datasets, attesting to the
robustness of the identified patterns. After that, we classify the
dataset (1900-2020) based on these SST patterns and investigate
the seasonal evolution of transition probabilities of the SST
patterns, providing a picture of the seasonal evolution of the
flavors. Given the robustness and simplicity of the method, it is
easily applicable to classify ENSO flavors in a variety of
datasets, including historical and future projections of SST. It
also allows a simple representation of nonlinearity between
positive and negative ENSO, with a classification method valid for
any month of the year.",
conference-location = "Chicago, IL",
conference-year = "12-16 Dec. 2022",
urlaccessdate = "06 jun. 2024"
}