Modelos lineales generalizados para la predicción de precipitaciones en el Valle Central de Costa Rica, América Central, usando ENOS: una propuesta metodológica

Abstract

The ocean/atmosphere interaction is commonly studied through processes or
phenomena such as El Niño / Southern Oscillation (ENSO). Many of the statistical
studies on the subject, focus on the use models for continuous time series,
adjusting models in the time domain, as in the case of univariate or multivariate
ARIMA models. However, not always the primary concern is to determine exactly
the magnitude of climatic anomalies such as the amount of rainfall in a period
but, categorizing the magnitude of precipitation, to determine the probability
of occurrence of class, conditioning on different categorizations (i.e. terciles)
of ENSO. The objective of this study is to find and / or develop methodological
statistical strategies to estimate these probabilities. In particular, we are interested
in assessing and modeling the relationship between the occurrence of El Niño (La
Niña) and the variability of precipitation in the Central Region of Costa Rica, in
May-June-July season. Data are time series of sea surface temperature (SST) in the
area of Niño 3 and the Southern Oscillation Index (SOI) for atmospheric pressure.
As the response we used rainfall (R) recorded in Juan Santa Maria airport, Costa
Rica. The series were categorized into terciles to build two-way contingency
tables. The tables were analyzed using log-linear and proportional odds models,
to determine the conditional and joint probabilities of rainfall events. We also
estimated a measure of ordinal association.