Forecasting Wind-driven Ocean Waves
277
14.3.2 Operational applications
The WAM model has been implemented operationally in many forecasting centres. The European Centre for Medium-range Weather Forecasts (ECMWF) runs a
global version of the model with a half-degree resolution to make 10-day forecasts.
KNMI runs a North Sea version ofthe model. An example ofthe output ofthe global model is given in Fig. 14.5, which shows isolines of significant wave height, as
computed for one particular time. Quasi-synchronous wave height observations by
ERS-2 are also plotted. The performance ofthe model has been tested against buoy
and satellite observations. As an example, Fig. 14.6 shows the bias as computed for
every month from early 1995 until March 1997. The comparison is made both for
ali observations and for separate regions. Fig. 14.7 shows, for the northem hemisphere, the anomaly correlations between the forecast and the verifying analysis for
the 3-,5-, 7- and 10-day forecast (January 1996 though March 1997). These correlations are a measure of the forecast skill. More details on model validation can be
found in Komen et al. (1994) and in Janssen, Hanssen and Bidlot (1996).
Wave forecasting is only one application of a wave model. Another important
application is the use of a wave model to study the wave climate (Guenther et al.,
1998). Interest in wave climate studies has recently increased when it was realised
that the wave climate may exhibit a considerable amount of decadal variability
(Carter and Draper, 1988; Bouws et al., 1996; WASA, 1998). This variability is
associated with the North Atlantic Oscillation (NAO), which is characterized by
the mean pressure difference between Iceland and the Acores. The WAM model
has also been used to estimate the effect of C02 doubling on the wave climate of
the North Atlantic (Rider et al., 1996) using surf ace winds from global change simulations with coupled atmosphere/ocean general circulation models.
14.4 Outlook
To make useful wave fore- Of hindcasts one needs good quality input winds. Fortunately, atmospheric models have considerable skill, which allows us to make usefuI wave forecasts. A further improvement of wind analysis and forecast over the
ocean will be crucial to increase the quality ofwave computations.
Wave models can be further improved. There are three broad areas for improvement: numerical resolution (both spatial and spectral), numerics (propagation
scheme and the integration ofthe non-linear transfer integral) and physics (i.e. the
representation ofwave growth and dissipation in the source terms). One should not
forget to test "improvements". After all, what really counts is the quality of the
wave predictions, when compared with observations. The WAM model has been
tested in numerous hindcasts and in many years of operational application. Repeating such tests is a formidable task. Maybe inverse modelling, using an adjoint
(Hersbach, 1998) can be useful here.
Wave forecasts can also improve by the assimilation of wave observations.
Present operational methods are still based on optimum interpolation (see, for
277
14.3.2 Operational applications
The WAM model has been implemented operationally in many forecasting centres. The European Centre for Medium-range Weather Forecasts (ECMWF) runs a
global version of the model with a half-degree resolution to make 10-day forecasts.
KNMI runs a North Sea version ofthe model. An example ofthe output ofthe global model is given in Fig. 14.5, which shows isolines of significant wave height, as
computed for one particular time. Quasi-synchronous wave height observations by
ERS-2 are also plotted. The performance ofthe model has been tested against buoy
and satellite observations. As an example, Fig. 14.6 shows the bias as computed for
every month from early 1995 until March 1997. The comparison is made both for
ali observations and for separate regions. Fig. 14.7 shows, for the northem hemisphere, the anomaly correlations between the forecast and the verifying analysis for
the 3-,5-, 7- and 10-day forecast (January 1996 though March 1997). These correlations are a measure of the forecast skill. More details on model validation can be
found in Komen et al. (1994) and in Janssen, Hanssen and Bidlot (1996).
Wave forecasting is only one application of a wave model. Another important
application is the use of a wave model to study the wave climate (Guenther et al.,
1998). Interest in wave climate studies has recently increased when it was realised
that the wave climate may exhibit a considerable amount of decadal variability
(Carter and Draper, 1988; Bouws et al., 1996; WASA, 1998). This variability is
associated with the North Atlantic Oscillation (NAO), which is characterized by
the mean pressure difference between Iceland and the Acores. The WAM model
has also been used to estimate the effect of C02 doubling on the wave climate of
the North Atlantic (Rider et al., 1996) using surf ace winds from global change simulations with coupled atmosphere/ocean general circulation models.
14.4 Outlook
To make useful wave fore- Of hindcasts one needs good quality input winds. Fortunately, atmospheric models have considerable skill, which allows us to make usefuI wave forecasts. A further improvement of wind analysis and forecast over the
ocean will be crucial to increase the quality ofwave computations.
Wave models can be further improved. There are three broad areas for improvement: numerical resolution (both spatial and spectral), numerics (propagation
scheme and the integration ofthe non-linear transfer integral) and physics (i.e. the
representation ofwave growth and dissipation in the source terms). One should not
forget to test "improvements". After all, what really counts is the quality of the
wave predictions, when compared with observations. The WAM model has been
tested in numerous hindcasts and in many years of operational application. Repeating such tests is a formidable task. Maybe inverse modelling, using an adjoint
(Hersbach, 1998) can be useful here.
Wave forecasts can also improve by the assimilation of wave observations.
Present operational methods are still based on optimum interpolation (see, for