less than spectral peak) of the spectrum based on a constant proportion of the energy transferred out of the mid-
range frequencies. Wave-wave interactions also transfer energy to the high frequency region of the spectrum where
it is assumed that energy is lost due to breaking processes. Wave propagation was handled in the original WAVAD
model by means of a first order upwinding scheme, although this was subsequently altered as discussed in Section 5.
The numerical model grid covered a domain extending from 120 E to 66 W and from 64 S to 56 N with a
resolution of 1.0. The model bathymetry for much of the Pacific Ocean was derived from a global (ETOPO30)
database. A total of twenty-three frequencies were used for the hindcasts in conjunction with a directional
resolution of 22.5 (16 direction bands). The frequencies were spaced at a constant ratio of 1.1 starting at 0.038 Hz.
Ice coverage was included in the model using NCEP/NCAR ice data, and was updated on a 6-hourly basis
throughout the duration of the hindcast.
A forty-year hindcast (1961-2000) was performed. The hindcast model output was archived every two hours at grid
points along the Chilean coastline and at each of the data comparison points. Both the full directional spectra as
well as files containing the integrated summary parameters were stored at each of the archival points. Special
compression algorithms were developed to efficiently handle the directional spectra by only storing those direction-
frequency bins containing energy above a pre-determined minimum threshold.
4.2 The Wind Field
Wind fields derived from the NCEP/NCAR Reanalysis Project data set (Kalanay et al., 1996) were used as the
primary driving mechanism for the wave model. The U (east-west) and V (north-south) wind fields at 10 m
elevation above ground were extracted from the NCEP/NCAR global database for a grid network that covered the
entire Pacific Ocean. The NCEP data are available on a 6-hourly basis for a grid resolution of 1.875 longitude by
an average 1.905 latitude (actual grid is Gaussian), and were interpolated onto the more refined WAVAD input grid
using a four-point bi-linear interpolation scheme. This approach provided a driving wind field for the model at six-
hour intervals. The WAVAD model automatically interpolated the six-hour wind fields to the selected model time
step of two hours.
5.
KEY MODELLING ISSUES
5.1 Systematic Correction of Extratropical Cyclone Winds
It has been recognized (e.g. Cox, 2000) that the NCEP/NCAR winds under-estimate wind speeds in larger storm
events as the limited resolution of the meteorological model cannot fully represent the complexities of a storm event.
In this case, as the focus was on swell events generated by mid-latitude extratropical cyclones, it was important to
develop some means for a systematic correction the winds generated by these cyclones. Traditional approaches to
improving the wind field representation, such as kinematic analysis, are not truly feasible due to the limited
meteorological observation network in the southern Pacific Ocean. For this study, various assessments were
performed, including:
Comparison of the NCEP/NCAR storm wind structure to that provides by scatterometer data, such as
Quikscat. Although initial promising, the data handling and manipulation required for this level of effort
quickly proved beyond the scope of the project.
Use of wind speed estimates derived from satellite altimeters did not provide useful due to the potentially
low saturation limits of these sensors with respect to wind speed estimation.
Ultimately, a relatively simple approach was developed based on comparisons of the hindcast model results to
Topex data. Altimeter estimates of wave heights over the period 1992 to 2000 were derived from two distinct storm
generating regions of the Pacific Ocean: (1) the Northern Mid-latitudes [35-55N] and (2) the Southern Mid-
latitudes [35S-65S]. Trial wave model simulations were then carried out in which various factoring schemes were
applied to the higher wind speeds (>10 m/s). The model results were then interpolated to each and every Topex data
point in the regional boxes. The goodness of fit was evaluated by means of quantile-quantile wave height
Previous Page
