Regression Models for Outlier Identification (Hurricanes and Typhoons) in Wave Hindcast Databases

The development of numerical wave prediction models for hindcast applications allows a detailed description of wave climate in locations where long-term instrumental records are not available. Wave hindcast databases (WHDBs) have become a powerful tool for the design of offshore and coastal structures, offering important advantages for the statistical characterization of wave climate all over the globe (continuous time series, wide spatial coverage, constant time span, homogeneous forcing, and more than 60-yr-long time series). However. WHDBs present several deficiencies reported in the literature. One of these deficiencies is related to typhoons and hurricanes, which are inappropriately reproduced by numerical models. The main reasons are (i) the difficulty of specifying accurate wind fields during these events and (ii) the insufficient spatiotemporal resolution used. These difficulties make the data related to these events appear as outliers when compared with instrumental records. These had data distort results from calibration and/or correction techniques. In this paper, several methods for detecting the presence of typhoons and/or hurricane data are presented, and their automatic outlier identification capabilities are analyzed and compared. All the methods are applied to a global wave hindcast database and results are compared with existing hurricane and buoy databases in the Gulf of Mexico, Caribbean Sea, and North Atlantic Ocean.