Probabilistic Modeling and Inference for Extreme Rainfall Episodes in Central America

• Activity Number: 32
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Luis Cid-Serrano*
• Companies: Universidad del Bio Bio
• Keywords: rainfall ; Central America ; Extreme value distribution

Abstract

Trends in rainfall in Central America are addressed as part climatic global change. Considering that climate in the area is characterized by heavy rainfall in brief periods of time, we are interested in probabilistic modeling of extremely heavy rain episodes in Central America, using the generalized extreme value distribution (EVD). Locally, we wished to establish if there exist differences in the distribution of these extreme events between the Pacific and the Caribbean oceans gradients. We used 30 years of daily data from 104 weather stations located throughout the continent, to estimate the parameters of the distribution for each station. As extreme values, we considered the 95 percentile of the daily rainfall data of each station. We fitted the EVD to the data and generated 104 vector of parameters (location, shape and scale), one for each weather station. Using multivariate analysis techniques, we classified the stations according to the values of the parameters. We compared the mean values of the parameters for two regions along Central America. We did not find differences in the amount of rain between the regions, but dispersion was significantly higher in the Caribbean.