After the above testing, the nested grid system - comprised of the inner and outer grids - was used for forecasting purposes. For this purpose, a protocol to automat-
ically download data by interconnecting two different FTP servers was developed. The system is run on a PC with a 3.40 GHz Pentium 4 processor and 2 GB of RAM, and uses a MATLAB code which is automated to run on a daily basis. The salient features of the system are shown in Fig. 77. The system is initialized at 00z and the input forcing functions are downloaded. The first 12 hours (denoted as “MODEL SIM TIME” in Fig. 77) include the time lag associated with the WW3 output (∼5.5 hours) and the WRF output (∼2.5 hours), and also the model computational time (∼ 2.5 hours). Overall about 12 hours of real time are lost in the modeling effort, and hence, even though each simulation is made for 48 h, the forecasts are provided at 12z for the next 36 h. Upon completion of the simulation, contour plots of SWH, Tp,
Dp, and wave-induced velocities are generated and transferred to the AOOS FTP
server. During the forecast mode, the sea-state at 24 h is also saved in order to initialize the next 00z run (the following day). The system obviously results in a 12 h overlap in the forecasts (between the last 12 hours of the previous days’ forecast and the first 12 hours of the current forecast). Consequently, for the recent 12 h, there will be an effect from two wind-fields which may or may not improve the wave forecast in some cases.
6.5 Discussion
Wave forecasts in coastal regions are needed for engineering and other applica- tions. In this section, a high-resolution coastal wave forecasting system for PWS was developed using multiple nested grids. The scheme consisted of a three-way coupling between NOAA’s global model for wave boundary conditions, UAF’s local WRF model for wind-fields, and two nested grids for obtaining wave-fields in PWS.
Fig. 77. PWS forecasting system protocol.
The system will provide daily forecasts (starting at 12z) for 36 hours on a continuous basis.
Although the system, as it stands, provides reasonable results for the hindcast studies described earlier, various errors that can accrue in the forecasts must be recognized. The assumption of JONSWAP spectral form to compute directional spectra at the open boundaries does not account for swell energy and may create errors, even though the total energy is conserved (this is briefly examined in the next section). In addition, errors can accrue from inaccuracies in the input forcing functions as obtained from other models (winds, SWHs at the open boundary, etc.)
and in the bathymetric representation. As regards the wind-fields, it was found in Section 4.4 that the random errors of the order of 10% have a relatively small effect on the predictions (see also Panchang et al. 2008). In any case, as stated in Section 1.3.1, if the model is to be run in the forecast mode, one has no choice but to accept that the flaws in the system exist and evaluate its reliability.
The reliability assessment of the wave forecasts is performed in the next section. Unfortunately in CI, there are currently no active buoys. Also, the historic wave data spans only a few months and that too with long gaps. Hence, the estimation of forecast uncertainty is only performed for PWS.
7. FORECAST SYSTEM RELIABILITY IN PRINCE WILLIAM SOUND, ALASKA
7.1 Introduction
After the hindcasting and testing work summarized in Section 6, operational forecasting was initiated in the PWS domain. In this section1, the issue of forecast
reliability is addressed in context of wave predictions in PWS. The motivation for this study stems from a need to provide reliable wave forecasts in order to meet various critical requirements associated with shipping, commercial and sport fishing vessel safety, and oil spill response. These needs mainly originate from lessons learned during the infamous ”Exxon Valdez oil spill” in PWS in 1989 that destroyed most of region’s ecosystem (Piatt et al., 1990). Mostly, the lack of suitable information on ocean climate affected the subsequent recovery and clean-up efforts. As a result, the PWS Oil Spill Recovery Institute (OSRI) has placed much emphasis on accurate environmental prediction.
For this purpose, wave forecasts are analyzed for roughly a year (June 2007 to May 2008) using the measured data. As stated earlier, this time period was selected, in part, due to the deployment of two wave gauges inside PWS - one near the outpost island (410), and the other at the entrance of Valdez arm (411) (Fig. 63). The data from the gauges were provided to us by PWS OSRI. Apart from the two gauges, measured data from four NDBC buoy locations (46076, 46061, 46060, and 46081) and a suite of satellite tracks were utilized in our statistical analysis. In the context of system reliability, these datasets were used to obtain reliability measures (in addition to the usual statistics) which can be used by managers of engineering operations,
1Parts of this section were published by Singhal et al. (2010). The author of this dissertation is
who are ultimately interested in knowing the likelihood of a predicted condition to actually occur and the associated error bounds.