A Mixed distribution with EV1 and GEV components for analyzing heterogeneous samples

Abstract

FLOOD CHARACTERISTICS ARE REQUIRED TO SOLVE SEVERAL WATER-ENGINEERING PROBLEMS. TRADITIONAL FLOOD FREQUENCY ANALYSIS INVOLVES THE ASSUMPTION OF HOMOGENEITY OF THE FLOOD DISTRIBUTION. HOW EVER, FLOODS ARE OF TEN GENERATED BY DISTRIBUTIONS COMPOSED OF A MIXTURE OF TWO OR MORE POPULATIONS. DIFFERENCES BETWEEN THE POPULATIONS MAY BE THE RESULT, FOR INSTANCE, OF THE ENSO PHENOMENON. IF THESE PHYSICAL PROCESSES ARE NOT CONSIDERED IN CONVENTIONAL FLOOD FREQUENCY ANALYSIS, THE T-YEAR FLOOD ESTIMATE CAN BE IN EFFICIENT FOR DESIGN PURPOSES. IN ORDER TO MODEL HETEROGENEOUS SAMPLES, A MIXED DISTRIBUTION WITH EXTREME VALUE TYPE I (EV1 OR GUMBEL) AND GENERAL EXTREME VALUE (GEV) COMPONENTS IS PROPOSED. A REGION IN NORTH WESTERN MEXICO WITH 35 GAUGING STATIONS HAS BEEN SELECTED TO APPLY THE MODEL AND AT-SITE QUANTILES WERE ESTIMATED BASED ON THE MAXIMUM LIKE LIHOOD PROCEDURE. RESULTS PRODUCED BY FIT TING THE EV1-GEV DISTRIBUTION WERE COMPARED THROUGH THE USE OF A GOODNESS-OF-FIT TEST WITH THOSE OBTAINED BY THE MIXED GUMBEL AND MIXED GEV DISTRIBUTIONS. THE EV1-GEV DISTRIBUTION WAS THE BEST OPTION FOR THE 40% OF ANALYZED SAMPLES AND THUS IT IS SUGGESTED ITS APPLICATION WHEN MODELING HETEROGENEOUS SERIES IN FLOOD FREQUENCY ANALYSIS.