GLOBAL WAVE CLIMATE TREND AND VARIABILITY ANALYSIS Sofia Caires Meteorological Service of Canada and Royal Netherlands Meteorological Institute P.O. Box 201, NL-3730 AE De Bilt, Netherlands email: caires@knmi.nl Val Swail Meteorological Service of Canada, Toronto, Ontario, Canada 1. INTRODUCTION In a series of works, Wang and Swail (2001, 2002, 2004) have provided the scientific and the ocean and offshore engineering communities with seasonal trends and patterns of variability of significant wave height (H S ) in the Northern Hemisphere during the last 40 years in terms of means, 90 th and 99 th percentiles, and return value estimates. Their studies were based on two 6-hourly reanalyses data sets, covering the 1958-1997 period. The first data set, the Cox and Swail (2001) reanalysis, produced wave fields on a global 1.25º x 2.5º latitude/longitude grid, and was obtained by using the reanalysis winds from American National Center for Environmental Prediction and the National Centers for Atmospheric Research (NCEP/NCAR) (Kalnay et al., 1996) to force the second generation ODGP2 spectral ocean wave model (see Cox and Swail, 2001). The second, the Swail and Cox (2001) reanalysis, produced wave fields on a 0.625º x 0.833º latitude/longitude grid covering the North Atlantic. It was motivated by deficiencies in the NCEP/NCAR reanalysis winds, which led the authors to carry out an intensive kinematic reanalysis of the NCEP/NCAR surface wind fields and use the resulting improved winds to force the OWI 3-G wave model (see appendix of Wang and Swail (2002) and references therein). More recently, another wave reanalysis data set on a global 1.5º x 1.5º latitude/longitude grid covering the period of 1957 to 2001 has been made available - the ERA-40 dataset. This reanalysis was carried out by the European Centre for Medium-Range Weather Forecasts (ECMWF), using its Integrated Forecasting System, a coupled atmosphere-wave model with variational data assimilation. A distinguishing feature of ECMWF's model is its coupling, through the wave height dependent Charnock parameter (see Janssen et al., 2002), to a third generation wave model, the well- known WAM (Komen et al., 1994), which makes wave data a natural output of ERA-40. A large subset of the complete ERA-40 data set, including H S , can be freely downloaded and used for scientific purposes from the website http://data.ecmwf.int/data/ . The results of ERA-40 have been extensively validated against observations (Caires and Sterl, 2005) and other reanalysis data sets (Caires et al., 2004a). These studies concluded that the ERA-40 data set, although severely underestimating high sea states, compares better with the observations in terms of root mean square error and scatter index than the Cox and Swail (2001) data set, and that the Swail and Cox (2000) data set is the best in describing synoptic wave data in the North Atlantic. In terms of long-term variability, the different reanalysis wave data sets differ mainly in the Tropics and, as regards the period before 1981, in the Southern Hemisphere. Besides the underestimation of high percentiles, the ERA-40 data set has another limitation that seriously discourages its use in direct studies of climate variability and trends: the existence of inhomogeneities in time due to the assimilation of different altimeter H S data sets in the ERA-40 computations (see top panel of Figure 1). These two limitations in the ERA-40 H S data set motivated their correction by Caires and Sterl (2005). These authors corrected the data using a nonparametric regression method, the main idea of which was to estimate the expected error between ERA-40 H S and “true” H S conditional on past (up to 12 hours) and present values of the former, using data from locations at which both ERA-40 and Topex measurements were simultaneously available, and then to use this conditional expected value to correct the whole ERA- 40 data. The result was a new 45-year global 6-hourly dataset - the C-ERA40 dataset. Comparisons of the C-