Evaluation of ECMWF wind data for wave hindcast in Chabahar zone

(1)

380 Saket, et al.

Evaluation of ECMWF wind data for wave hindcast in Chabahar

zone

Arvin Saket Amir Etemad-Shahidi , Mohammad Hadi Moeini School of Civil Engineering, Iran

University of Science and Technology, 16765-163, Tehran, Iran.

[email protected] [email protected]

Griffith School of Engineering, Griffith University, Queensland 4222, Australia. [email protected]

INTRODUCTION

Wind waves are the most important environmental forces acting on the marine structures. Due to the lack of long-term wave measurements in many regions, wave prediction plays a key role in the design of coastal and offshore infrastructures. The capability of this prediction leads to a more accurate design of structures. Nowadays, numerical wind wave models have become one of the most useful tools for wave hindcast and forecast. Since wind wave models are very sensitive to wind field variations and the quality of numerical wave hindcasts depends to a large extent on the quality and the accuracy of the wind fields (Teixeira et al., 1995, Holthuijsen et al., 1996, Moeini et al., 2012), the selection of appropriate wind source is a vital step in the wave modeling.

There are many sources of wind such as synoptic stations measurements, buoys, satellites measurements and weather prediction models, as an input of the numerical models. Due to the temporal or geographic limitation of wind measurements, weather prediction models are usually used as wind fields for the wave simulating. European Center for Medium Range Weather Forecasts (ECMWF) data is one of the most commonly used sources of wind fields. The ECMWF is a meteorological data assimilation project, in which historical observational data spanning an extended period is implemented through a single consistent analysis in forecast models. It is essential to evaluate

the ECMWF numerical weather prediction data locally and if needed modify it using measured wind data before simulating the wave field. Many investigators have studied the quality of this modeled surface wind for wave simulating in different areas. Signell et al. (2005) evaluated the ECMWF for using it in wave modeling of the Adriatic Sea. SWAN model was used for wave simulating and they revealed that the ECMWF global model T511 underestimates the wind speed. Cavaleri and Scalvo (2006) also assessed the quality of the same wind model in the Mediterranean Sea and showed the same result. Moeini et al. (2010) applied the ECMWF and the measured wind data for wave modeling in the Persian Gulf using the SWAN model. They found that the model overestimates the low wave heights because of the overestimation of low wind speeds and underestimates the higher ones because of the underestimation of high wind speeds by ECMWF in all directions. Vicinanza et al. (2013) studied the wave energy potential of the north-west of Sardinia by an analysis of wave measurements carried out by the Italian Wave Buoys Network and the corresponding hindcast data by the ECMWF. They used the Mike21 NSW for wave simulation and compared the ECMWF dataset with the buoy records. They found that because of the underestimate of ECMWF data, the wave parameters which were simulated by ECMWF lead to underestimate the wave energy in the studied area. In addition to these studies, Cavaleri and Bertotti (2004), Caires and Sterl (2005), and Brenner et al. (2007) showed the underestimation of the ECMWF wind fields.

ABSTRACT

Saket, A., Etemad-Shahidi, A., Moeini, M.H., 2013. Evaluation of ECMWF wind data for wave hindcast in Chabahar zone, Proceedings 12th International Coastal Symposium (Plymouth, England), Journal of Coastal Research, Special Issue No. 65, pp. 380-385, ISSN 0749-0208.

Wind waves are the most important environmental forces acting on the marine structures. Due to the incompleteness of measured wave parameters, wave prediction plays a key role in the design of coastal and offshore structures. Nowadays, numerical wind wave models are widely used for wave hindcast and forecast. Since wind is the most important forcing term in the numerical wind wave model, the selection of appropriate wind source is a vital step in the wave modeling. In the present study; two wind sources i.e. the measured synoptic and the ECMWF (European Center for Medium Range Weather Forecasts) data, were evaluated for wave simulation near the Chabahar zone. To simulate wave parameters the third generation spectral SWAN model was utilized and the results were compared with those of in situ measurements in a depth of about 17 m. The whitecapping dissipation coefficient and bottom friction factor were used for calibration of the model. The sensitivity analysis showed that other physical parameters have no specific effect on the wave characteristics. Calibration of whitecapping dissipation rate led to the overestimation of high waves. Therefore, a combination of whitecapping dissipation and bottom friction factors was used to calibrate the model. It was found that the SWAN model forced by ECMWF wind data predicted the south-west and west waves successfully while underestimated the east, south-east and south waves. This was mainly due to well prediction of south-west and west wind and underestimation of wind from the east to the south by the ECMWF model. In addition, it was revealed that synoptic wind data can be used as an appropriate wind source for wave hindcasting at the studied area.

ADDITIONAL INDEX WORDS:Wind waves, Wind fields, SWAN model, Chabahar

www.JCRonline.org

____________________

DOI: 10.2112/SI65-065.1 received 07 December 2012; accepted 06 March 2013.

© Coastal Education & Research Foundation 2013

(2)

Evaluation of ECMWF wind data for wave hindcast in Chabahar zone 381

The present study investigates the quality of two sources of surface winds, i.e. the measured synoptic data and the ECMWF numerical weather prediction data for the wave modeling near the Chabahar zone in the Gulf of Oman. For simulating wave parameters the third generation spectral SWAN model (Booij et al., 1999) was utilized, and the model outputs were calibrated and verified with recorded wave data. Finally, the accuracy of both measured and ECMWF wind data were evaluated for wave simulating in the studied area.

STUDY AREA AND FIELD DATA

The Chabahar zone in Sistan va Baluchestan province is located in the south-east of Iran and north-east of the Gulf of Oman (Figure 1). The Chabahar port, located at 25.29°N and 60.63°E, is officially designed as a free trade and industrial zone in the Gulf of Oman and is the closest Iranian port to the Indian Ocean. Therefore, this area has become an important transit route among countries situated in the northern part of the Indian Ocean and Central Asia. In addition, according to the great importance of this port, huge investment has been allocated for marine and hinterland constructions in the area. Taking into account these reasons, an accurate prediction of the wave climate is essential in the Chababar zone.

The recorded wave data used in this study were collected by a Datawell Directional Waverider buoy, located near the Chabahar port at 25.267°N and 60.65°E and deployed by Islamic Republic of Iran Meteorological Organization (IRIMO) (Figure 1). The depth of water at the measurement station was about 17m and the buoy was programmed to collect the data set at intervals of three hours from May, 1998 to September, 2000 with some gaps.

As previously mentioned, there is a direct correlation between

the quality of wind data as an input of numerical models and the results of numerical wave hindcast. Therefore, selecting an appropriate wind field is crucial for wave modeling and hindcasting. In the present study, two wind sources i.e. the ECMWF numerical weather prediction data and the measured synoptic data, were investigated for wave simulation at the study area using the SWAN model. The operational ECMWF wind data was applied as a wind field because of its appropriate spatial and temporal resolutions which are 0.5 degree and 6 hours, respectively. In addition, this modeled surface wind is widely used for wave modeling in Iranian seas and the accuracy of this source of wind is crucial in the studied area. The synoptic data was selected as a measured wind data, according to its proximity to the Gulf of Oman and the wave recording location. This station is located on a flat land; therefore, the local topography has negligible effect on wind characteristics at the synoptic station. The measured wind data has been recorded 3-hourly by the Chabahar meteorological station at the location of 25.28°N and 60.62°E (Figure 1). The elevation of the anemometer in this station was 8 m above the ground level. Since the SWAN model uses wind speed at a 10-meter elevation, the measured wind speeds require to be adjusted prior to using in the numerical model. In order to modifying the wind velocities, following equation was applied (CEM, 2003):

7 1 10 10 z U U z (1)

where U10 and Uz are wind speeds at 10 and z meters over the ground level, respectively. It should be state that the ECMWF wind data is predicted at the height of 10 m and does not need to be modified before using in the SWAN model.

(3)

382 Saket, et al.

NUMERICAL MODEL

SWAN MODEL

In order to model the wave characteristics in this study, the spectral wave model SWAN (Simulating WAves Nearshore) cycle III version 40.72AB (The SWAN team, 2009) was employed. SWAN is a third-generation spectral wind wave model, developed for estimation of wave parameters in coastal regions (Booij et al., 1999; Ris et al., 1999). The model solves the transport equation with no particular shape which is assumed for the wave spectrum (Komen et al., 1984; Bolaños-Sanchez et al., 2007). Action density spectrum is considered in SWAN model rather than energy density spectrum because, unlike energy density, action density is conserved in the presence of currents. Action density is defined as: / ) , ( ) , ( E N (2)

where is the energy density with the relative frequency (i.e. the radian frequency in a frame of reference moving with current velocity) and the propagation direction (i.e. the direction normal to the wave crest of each spectral component).

In addition, SWAN model solves the spectral action balance equation for the Cartesian coordinates (Ris et al., 1999):

S N c N c N c y N c x N t x y (3)

The first term on the left-hand side shows the local rate of change of action density in time, the next two terms represent the propagation of action in geographical space where cx and cy are the propagation velocity in the x and y-spaces, respectively. The fourth term represents the shifting of the relative frequency due to variations in depths and currents in which, c is the propagation velocity in the -space. The last term on the left-hand side demonstrates depth-induced and current-induced refraction and propagation in directional space in which, c is the propagation

velocity in the -space (Booij et al., 1999). The term

represents sources or sinks of wave energy. This term shows the effects of generation, dissipation (by whitecapping, bottom friction and depth-induced wave breaking) and nonlinear wave-wave interactions. Further details can be found in Booij et al., 1999, Ris et al., 1999 and Holthuijsen et al., 1989.

MODEL SETUP

For simulating the wave parameters, the SWAN model was executed in the third generation and two-dimensional non-stationary mode. The model was implemented on a nautical coordinates covering the Gulf of Oman. The geographical domain was divided into 92 x 82 uniform cells with 0.0333° resolutions in x and y directions, respectively. The bathymetry data was obtained from the Iranian National Center for Oceanography (INCO). The spectral space was resolved in 18 directions (20° angular bandwidth) and 25 logarithmically spaced frequencies, between 0.04 Hz and 1 Hz. Since it has been found that using the Komen et al. (Komen et al., 1984) theory for wind input parameterizations result in a more accurate prediction of significant wave height (Moeini and Etemad-Shahidi, 2007), this expression was considered for the exponential growth of wind input. Quadruplet nonlinear interactions, depth-induced wave breaking and bottom friction coefficients were changed to obtain best results for both the wave height. The model execution time step was 10 minutes.

RESULTS AND DISCUSSION

For calibrating the SWAN model, recorded wave data from 16/02/2000 to 06/03/2000 with a wide range of wave conditions was selected. The calibration of the model is a process which the model parameters are tuned to obtain best results against the measurements. To calibrate the model, the traditional trial and error method was used.

The sensitivity analysis showed that the whitecapping

0.0 0.5 1.0 1.5 2.0 Measured

Modeled (Only whitecapping)

Modeled (Combination of whitecapping and bottom friction)

(a)

0.0 0.5 1.0 1.5 2.0 16/02/2000 19/02/2000 22/02/2000 25/02/2000 28/02/2000 2/03/2000 5/03/2000 8/03/2000 Time

(b)

Figure 2. Comparison of the calibrated model with the whitecapping dissipation and a combination of whitecapping dissipation and bottom friction factors, against the measurements in the calibration period (a) ECMWF wind forced data (b) Synoptic wind forced data.

(4)

dissipation coefficient and bottom friction factor were the tunable parameter for calibrating the model and other physical parameters have negligible effect on the wave characteristics. Therefore, the default values were selected for these tunable parameters in the mentioned sink terms. The default tunable coefficient value of whitecapping (cds2 = 0.7e-5) in the Komen et al. theory for the exponential growth of wind input, led to the underestimation of the results. In addition, considering the location of the buoy, the depth of water had shallow water conditions for high waves and calibration of the whitecapping dissipation rate with the default tunable factor of bottom friction (cfjon=0.067), led to the overestimation of high waves. Therefore, a combination of whitecapping dissipation and bottom friction factors was used for calibrating the model and the value of these tunable parameters of whitecapping dissipation and bottom friction were increased to 0.9e-5 and 1.3 respectively, to obtain the best simulation. A qualitative view of the calibration result, in terms of significant wave height is illustrated in Figure 2(a). This figure represents the comparison of the calibrated model results with the whitecapping dissipation and a combination of whitecapping dissipation and bottom friction coefficients, forced by ECMWF wind data, against the buoy measurements in this period. In addition, the model was calibrated by forcing the synoptic wind data, in the same process (Figure 2 b).

For the quantitative comparison of the results, the following statistics were used:

n i i i y x n RMSE 1 2 ) ( 1 (4) i i x y Bias (5)

where RMSE is the root mean square error. xi represents the measured and yi represents the modeled wave heights.

x

and

y

are the mean value of measured and modeled wave heights,

respectively and n is the total number of data. The summary of statistical analysis for the calibration period is given in Table 1. As seen, there is a very good agreement between the model results and measurements.

After calibration, the model was verified using the buoy recorded data from 26/01/1999 to 07/02/1999. Time series of the modeled significant wave heights and wave direction forced by the ECMWF and synoptic wind data against the buoy measurements during the verification period, are shown in Figure 3. For quantitative comparison of the results during the verification period, the error measures are shown in Table 1. As seen, similar to the calibration period the SWAN model forced by synoptic wind data performs well in simulating significant wave height in all directions. Although, the model forced by ECMWF wind data indicates a good trend of simulation, modeled wave heights are lower than the measured ones. As can be seen, the model predicted the south-west and west waves successfully while underestimated the waves from the east to the south. Considering the underestimation of ECMWF data, the wind speed and direction was evaluated in the verification period.

The qualitative comparison of ECMWF and synoptic in terms 0.0

0.5 1.0 1.5

2.0 Measured Modeled: Synoptic Modeled: ECMWF

(a)

0 60 120 180 240 300 360 26/01/1999 28/01/1999 30/01/1999 01/02/1999 03/02/1999 05/02/1999 07/02/1999 09/02/1999 Time

(b)

Figure 3. Comparison of the modeled parameters, forced by ECMWF and synoptic wind data against the measurements in the verification period (a) significant wave height (b) wave direction.

Table 1. The summary of statistical analysis of the calibration and verification periods.

HS (m) Modeled by ECMWF Modeled by Synoptic Calibration period Bias (m) -0.003 0.002 RMSE (m) 0.197 0.170 Verification period Bias (m) -0.168 0.077 RMSE (m) 0.293 0.197

(5)

384 Saket, et al.

of wind speed and wind direction is illustrated in Figure 4. The quantitative comparison is also shown in Table 2. As seen, the ECMWF model predicted the south-west and west wind well and underestimated the wind speed from the east to the south.

In order to generalize the results, the long-term of ECMWF and synoptic wind data (wind speed and its direction) were assessed. For this purpose, ECMWF wind data was extracted at the location of synoptic station and the wind roses of ECMWF and synoptic station (from 1992 to 2003), were evaluated (Figure 5). As can be seen, the comparison of these wind roses confirms the results which presented in the verification period. Therefore, using ECMWF wind data for simulating the wave parameters in the Chabahar zone lead to underestimate of the waves that are coming from E, SE and S directions. Hence, ECMWF wind speeds should be modified in these directions using the measured wind sources in this area. Moreover, validation of the wave model showed that synoptic wind data can be applied as an appropriate wind source for wave hindcasting at the study area, especially in the Chabahar port.

SUMMARY AND CONCLUSIONS

In this study, two wind sources i.e. the measured synoptic and the ECMWF data, were evaluated for wave simulation near the Chabahar zone located in Sistan va Baluchestan province in the south-east of Iran and north-east of the Gulf of Oman. The wave model SWAN was implemented to simulate wave parameters and the results were compared with the buoy measurements. The whitecapping dissipation coefficient and bottom friction factors were the tunable parameter used for calibration of the SWAN model. The sensitivity analysis showed that other physical parameters have no specific effect on the wave characteristics. Using only whitecapping dissipation rate for calibrating the model led to the overestimation of high waves. Therefore, a combination of whitecapping dissipation and bottom friction factors was used to calibrate the wave model forced by both ECMWF and synoptic data.

Having successfully calibrated the wave model, outputs of the wave model were verified with buoy measurements. The results in this period showed that the model forced by ECMWF wind data 0.0 2.0 4.0 6.0 8.0 10.0 12.0 Synoptic ECMWF

(a)

0 60 120 180 240 300 360 26/01/1999 28/01/1999 30/01/1999 01/02/1999 03/02/1999 05/02/1999 07/02/1999 09/02/1999 Time

(b)

Figure 4. Comparison of the synoptic and ECMWF wind data in the verification period (a) wind speed (b) wind direction.

Table 2. Statistics of synoptic and ECMWF wind data from the east to the west in the verification period.

Wind source Direction

E SE S SW W

Synoptic wind data _{Standard deviation (m/s)}Average (m/s) 6.08 _1.83 7.18 _1.21 6.08 _1.11 4.96 _0.89 5.31 _1.72

ECMWF wind data Average (m/s) 2.79 3.34 3.65 4.17 4.26

(6)

predicted the south-west and west waves successfully while underestimated the waves which are coming from the east, south-east and south. After evaluating the wind roses of ECMWF and synoptic station during 1992 to 2003, it was found that the ECMWF model well predicted the south-west and west windand underestimated wind speeds from the east to the south. Therefore, ECMWF wind speeds should be modified in these directions using the measured wind data for simulating wave parameters in the Chabahar zone. Finally, it was revealed that synoptic wind data can be used as an appropriate wind source for wave hindcasting at the studied area, in particular around the Chabahar port.

ACKNOWLEDGEMENT

The authors would like to thank Islamic Republic of Iran Meteorological Organization (IRIMO) for providing the buoy and wind data. We would also thank the SWAN group at the Delft University of Technology (Department of Fluid Mechanics) for providing the freely available model.

LITERATURE CITED

Bolaños-Sanchez, R., A. Sanchez-Arcilla and Cateura, J., 2007. Evaluation of two atmospheric models for wind wave modelling in the NW Mediterranean. Journal of Marine Systems, 65, 336-353.

Booij, N., Ris, R.C., Holthuijsen, L.H., 1999. A third generation wave model for coastal regions. 1. Model description and validation. Journal of Geophysical Research, 104, 7649-7666. Brenner, S., I. Gertman and Murashkovsky, A., 2007.

Preoperational ocean forecasting in the southeastern Mediterranean Sea: Implementation and evaluation of the models and selection of the atmospheric forcing. Journal of Marine Systems, 65, 268-287.

Caires, S., Sterl, A., 2005. A new nonparametric method to correct model data: Application to significant wave height from the ERA-40 re-analysis. Journal of Atmospheric and Oceanic Technology, 22(4), 443-459.

Cavaleri, L., Bertotti, L., 2004. Accuracy of the modelled wind and waves in enclosed seas. Tellus, 56(A), 167-175.

Cavaleri, L. and Sclavo, M., 2006. The calibration of wind and wave model data in the Mediterranean Sea. coastal Engineering, 53(7), 613-627.

Coastal Engineering Manual (CEM). Chapter II-2, Meteorology and Wave Climate, Engineer Manual 1110-2-1100. U.S. Army Corps of Engineers. Washington, DC.

Holthuijsen, L. H., Booji, N. and Bertotti, L., 1996. The propagation of wind errors through ocean wave hindcasts. Journal of Offshore Mechanics and Arctic Engineering, 118, 184-189.

Holthuijsen, L. H., N. Booij and Herbers, T. H. C., 1989. A prediction model for stationary, short-crested waves in shallow water with ambient currents. Coastal Engineering, 13, 23-54. Komen, G.J., Hasselmann, S. and Hasselmann, K., 1984. On the

existence of a fully developed wind sea spectrum. Journal of Physical Oceanography, 14, 1271-1285.

Moeini, M. H. and Etemad-Shahidi, A., 2007. Application of two numerical models for wave hindcasting in Lake Erie. Applied Ocean Research, 29 (3), 137-145.

Moeini, M. H., Etemad-Shahidi, A., and Chegini, V., 2010. Wave modeling and extreme value analysis off the northern coast of the Persian Gulf. Applied Ocean Research, 32 (2), 209-218. Moeini, M. H., Etemad-Shahidi, A., Chegini, V. and Rahmani, I,

2012. Wave data assimilation using a hybrid approach in the Persian Gulf. Ocean Dynamics, 62 (5), 785-797.

Ris, R.C., Holthuijsen, L.H. and Booij, N., 1999. A third-generation wave model for coastal regions 2. Verification. Journal of Geophysical Research, 104, 7667-7681.

Saket, A. and Etemad-Shahidi, A., 2012. Wave energy potential along the northern coasts of the Gulf of Oman, Iran. Renewable Energy, 40 (1), 90-97.

Signell, R. P., Carniel, S., Cavaleri, L., Chiggiato, J., Doyle, J. D., Pullen, J. and Sclavo, M., 2005. Assessment of wind quality for oceanographic modelling in semi-enclosed basins. Journal of Marine Systems, 53, 217-233.

Teixeira, J. C., Abreu, M. P. and Guedes Soares, C., 1995. Uncertainty of ocean wave hindcasts due to wind modelling. Journal of Offshore Mechanics and Arctic Engineering, 117, 294-297.

The SWAN team, 2009. SWAN User Manual (Cycle III version 40.72AB), Delft University of Technology, Delft.

Vicinanza, D., Contestabile, P. and Ferrante, V., 2013. Wave energy potential in the north-west of Sardinia (Italy). Renewable Energy, 50, 506-521.

Figure 5. (a) ECMWF and (b) synoptic station wind roses from 1992 to 2003.