An operational wave forecasting system for the Portuguese continental coastal area

Transcription

1 Journal of Operational Oceanography ISSN: X (Print) (Online) Journal homepage: An operational wave forecasting system for the Portuguese continental coastal area C Guedes Soares, L Rusu, M Bernardino & P Pilar To cite this article: C Guedes Soares, L Rusu, M Bernardino & P Pilar (2011) An operational wave forecasting system for the Portuguese continental coastal area, Journal of Operational Oceanography, 4:2, 17-27, DOI: / X To link to this article: Published online: 01 Dec Submit your article to this journal Article views: 167 View related articles Citing articles: 15 View citing articles Full Terms & Conditions of access and use can be found at

2 An operational wave forecasting system for the Portuguese continental coastal area C Guedes Soares, FIMarEST, L Rusu, M Bernardino and P Pilar, Centre for Marine Technology and Engineering (CENTEC), Instituto Superior Técnico, Technical University of Lisbon, Portugal An operational system based on numerical models for wind and wave forecasting in the Portuguese coastal area is presented. It is based on the wave models WAM (WAve prediction Model) and SWAN (Simulating WAves Nearshore), driven by the wind fields provided by the atmospheric models GFS (Global Forecast System) from NCEP (National Centers for Environmental Prediction) and MM5 (Mesoscale Meteorological Model), respectively.the system is operational since October 2008 and runs automatically on Linux and Windows environments. Forecast products for four days ahead are produced daily. Graphical outputs of the wave parameters (fields and temporal series) are made available in a restricted domain internet site. The validation of the results provided by the forecasting system is carried out by performing comparisons with both wind and wave measurements. LEAD AUTHOR S BIOGRAPHY Carlos Guedes Soares received a MSc from the Massachusetts Institute of Technology (MIT) and a Phd from the Norwegian Institute of Technology in Trondheim. He is currently Professor at Instituto Superior Tecnico (IST), which is the Faculty of Engineering of the Technical University of Lisbon, where he heads the Centre for Marine Technology and Engineering. INTRODUCTION The economic activities in the coastal waters of continental Portugal, especially in the neighbourhood of the important harbours, have had increasing importance in the last years. For planning their activities, the port authorities need to have available reliable information about the sea state several days in advance, as there is the need to anticipate wave conditions that could interfere in ships arriving and leaving the harbours. Furthermore, while wave data are being collected in-situ by waverider buoys at various coastal locations, there is always the need for knowledge of the wave conditions in a wide geographical area through which ship traffic will be occurring. For this purpose the use of the numerical models becomes essential for predicting operationally the wave conditions. Generally speaking, weather and wave forecasts provided by national and international centres are made on a global scale and do not take into account properly the specific characteristics of the local areas. Being an interface between ocean and land, the coastal environment presents characteristics that make numerical modelling more complex and difficult than in the open ocean. Global meteorological models, due to their own structure, philosophy and use of horizontal resolution, are not always the most appropriate tools for such a local forecast. This is valid also for the global wave models when predicting wave conditions in coastal environments with complex bathymetric conditions. An example of the limitations of such models is not resolving islands which can have significant effects. 1,2 Bidlot et al 3 provide a good review of the performance of the operational systems at some of the main meteo-oceanographic centres. This situation justifies the development of an operational system for forecasting wave conditions, based on regional models implemented in high resolution areas, in Volume 4 No Journal of Operational Oceanography 17

3 order to be able to provide reliable information about the sea states and wind conditions in the neighbourhood of the major harbours. A real interest in implementing such joint forecasting systems exists in various places of the world. Some examples are systems developed for the Mediterranean Sea, 4 for the North and Baltic Seas, 5 around the Korean peninsula, 6 or for the Spanish coasts. 7 In Portugal, the operational system reported here has been developed with capabilities to provide high resolution forecast products for wind and wave conditions in the neighbourhood of major Portuguese ports, initially chosen to be Sines and Leixões. The operational implementation of the system was made in the framework of the project MARPORT (Development of a Wave Prediction System for the Portuguese Ports) and it is based on wind and wave models for global and local scales, whose nesting application for the Portuguese coast was tested and validated in various hindcast studies. 8,9,10 This paper describes how the system operates and provides validation of its initial results against measured data. THE FORECASTING SYSTEM The meteorological model The Fifth-Generation NCAR/Penn State Mesoscale Model MM5, 11,12 is a limited-area, terrain-following sigma-coordinate model designed to predict mesoscale and regional-scale atmospheric circulation. MM5 version 3.7, which is used in the system developed, has the ability to run simultaneously multiple nested grids using a two-way nesting scheme. For two-way interaction the nesting relationship should always be 3:1. The wider grid has a resolution of 38km and represents the interface between the global and regional levels. The areas with higher resolution were set to maintain the relationship of nesting and also to obtain detailed information in the locations of interest, ie, the coastal sectors adjacent to the principal ports of Portugal. Details about the MM5 parameterisation, both as regards grid information and physical options, are presented in Table 1 while the four geographical spaces with various spatial resolutions considered for the MM5 model simulations are presented in Fig 1. The MM5 model forcing is carried out with meteorological fields at different levels provided by the NOAA Atmospheric Model GFS. The model allows for a broad set of meteorological fields at different scales and with the temporal resolution that is desired. Nevertheless, since the wind fields at 10m height are of interest in forcing the wave models, only this parameter is extracted. The temporal resolution chosen to generate wind fields is six hours. The implementation of the MM5 model in the Portuguese coast has been already validated in various hindcast studies, when evaluations of the model results in different areas were made. 13,10,14 Wave models To assess the wave conditions in the Portuguese coastal environment, two state-of-the-art spectral wave models Dynamics Vertical resolution Horizontal resolution (km) Primitive equation, non-hydrostatic 30 sigma levels Grid dimension Global terrain and land use resolution Radiation Surface processes Planetary boundary layer Sea surface temperature Convection Explicit moisture 10 min 5 min 2 min Dudhia scheme for short and long wave radiation. Radiation effects due to clouds are considered. Multi-layer soil diffusion scheme MRF scheme SST is not updated during model integration Grell scheme on all grids Simple ice scheme Table 1: Details of the grids and the physics options used in the NCAR MM5 model were used in order to follow the wave generation and propagation from deep-ocean to the coastal environment. These are the WAM Cycle 4 wave generation model (WAMDI Group 15 ) improved version that allows for twoway nesting 16 and the SWAN version used in the coastal environment, both for wave generation and for nearshore wave transformation. The WAM model is a third-generation wave model that solves the spectral energy transport equation explicitly without any assumption on the shape of the wave spectrum. It represents the physics of the wave evolution for the full set of degrees of freedom of a 2D wave spectrum. The present WAM implementation covers the entire North Atlantic basin. 9 Some details concerning the bathymetric grids in the system focusing towards the Portuguese nearshore are presented in Table 2 and Fig 2. Around the Azores Archipelago and Canary Islands higher resolution in the geographical space is used (0.25º) to account better for the sheltering effect of those islands, which was identified by Ponce de León and Guedes Soares. 1 The input conditions necessary for the WAM model consist of the wind fields at 10m and the ice fields. The number of frequencies used to describe the wave spectrum is 25 (with Hz the lowest frequency) and the number of directions is 24. The generation model time steps used are 300s (5 min) for propagation and 900s (15 min) for the source. As regards the process of coastal transformation of the waves the SWAN (acronym from Simulating WAves Nearshore 17 ) is used. This is a third-generation phase-averaged wave model for the simulation of waves in waters of deep, intermediate and finite depth. The model solves the action balance equation over a finite difference grid system 18 Journal of Operational Oceanography Volume 4 No

4 Fig 1:The geographical spaces considered for the MM5 model simulations Grid Coarse 1 Coarse 2 Medium Fine1 Fine2 Grid spacing (long. lat.) South North West East Wind GFS/NCEP wind fields with 1 1 spatial resolution and 6h temporal resolution Ice GFS/NCEP ice fields with 1 1 spatial resolution and 6h temporal resolution Table 2: Bathymetric grid details and inputs used in WAM model (in Spherical or Cartesian coordinates) and uses typical formulations for wave growth by wind, wave dissipation by whitecapping, and quadruplet nonlinear interactions. The SWAN model simulates also physical processes associated with the wave transformation in intermediate and shallow water depth, such as depth-induced wave refraction, depthlimited breaking, bottom friction, diffraction and nonlinear triad interactions. A large SWAN area that covers the west Iberian coast, as presented in Fig 3, was nested into the WAM model and will be used as a general driver for the coastal simulations. Wave spectra with 0.5º spatial resolution and 1h time resolutions coming from WAM are provided as boundary conditions for SWAN. The bathymetry used in this large SWAN computational domain has a resolution of 0.05º in longitude and 0.1º in latitude, with 100 cells in both directions and corresponds with the computational grid of the model. The SWAN simulations are performed in the nonstationary mode using the second-order upwind numerical scheme with third-order diffusion S&L, (Stelling and Leendertse 18 ). Following the results obtained by Rusu et al, 8 the Janssen 19 parameterisation is used for the wind growth, while for the quadruplet wave-wave interactions the fully explicit computations of the non-linear transfer with Discrete Interaction Approximation (DIA) per sweep is used. After various tests to balance better the computational efficiency and the numerical accuracy, a 20min computational time step was selected. The numerical accuracy was Volume 4 No Journal of Operational Oceanography 19

5 Fig 2: Bathymetry grid definition for five nested grids considered for the WAM model simulations Fig 3:The geographical space of the SWAN simulations 20 Journal of Operational Oceanography Volume 4 No

6 increased by setting the iteration number from 1 (which is the default value for no stationary simulations) to 4. In the spectral space for the SWAN simulations 36 directions and 30 frequencies logarithmically spaced from Hz to 0.6Hz were considered. System implementation The forecasting system that incorporates the models MM5 and WAM is implemented in a cluster of computers, using Linux64 (debian) operating system and runs daily (each morning). In the case of the MM5 model the option of parallel processing through software MPICH is used. Programs to achieve an automatic acquisition of the meteorological fields from the NCEP site and to perform their subsequent format conversion were developed. The output data from WAM and MM5 models transformed in ASCII format are also automatically processed using the MATLAB environment and transferred to SWAN as input. The wind fields computed by the MM5 model in the coarse domain are used to force the SWAN. In order to obtain wind field data with a regular spatial resolution of 0.5º, these fields are interpolated using a triangle-based linear interpolation method. The SWAN model runs in Windows environment, the data transfer from Linux to Windows environment being performed fully automatically. A MATLAB script is used for the post-processing tasks and to obtain graphical outputs. Another set of scripts manage different files and images and move them to the archive site and to the intranet web server. The wave models are initialised daily with conditions kept when the run of the previous day is finished. As regards the MM5 model this runs daily but because no hot file it is used the model starts 72h before the specific date. This system has been operational continuously since October 2008 and runs automatically on Linux and Windows environments producing a four-day forecast daily. Nevertheless in order to provide more reliable results, from its initial implementation the system incorporated several improvements, especially as regards the physical processes activated, as well as in relationship with the initiation of each four-day cycle of predictions. In the form described herewith the system has been working since December Fig 4: Significant wave height fields and wave vectors in the SWAN domain, 14/1/2010-h00, forecast 1 day Products The implemented forecasting system provides information about the spatial distribution of the main wave parameters (significant wave height and mean direction) in the Portuguese continental coastal environment, with a temporal resolution of six hours. Moreover, it also provides time series forecast for four days ahead with 3h time steps of some wave parameters significant wave height, mean and peak period, Wave measurements OOOOOOOO OOOOOOOO OOOOOOOO OOOOOOOO OOOOOOOO OOOOOOOO Wind measurements VVVV VVVV VVVV VVVV VVVV VVVV Forecast day xxxxxxxx oooooooo # # # # # # # # ΔΔΔΔΔΔΔΔ 2 xxxxxxxx oooooooo # # # # # # # # ΔΔΔΔΔΔΔΔ 3 xxxxxxxx oooooooo # # # # # # # # ΔΔΔΔΔΔΔΔ Table 3:The illustrative scheme of the results of the predictions ( x forecast 1 day, o forecast 2 days, # forecast 3 days, Δ forecast 4 days) in relationship with the measurements (O wave measurements, V wind measurements) Volume 4 No Journal of Operational Oceanography 21

7 (a) (b) Fig 5: Direct output of the forecasting system, time series for the significant wave height, mean and peak period for a four-day forecast simulated at the buoys locations, (time interval 14/1/2010-h00 18/1/2010-h00) a) Sines; b) Leixões 22 Journal of Operational Oceanography Volume 4 No

8 mean and peak direction obtained for locations near the ports of Sines and Leixões. Figs 4 and 5 illustrate some examples of the information made available to ports through the restricted access website. EVALUATION OF THE SYSTEM RESULTS A two-month period (7 Jan 28 Feb 2010) is considered in the present work to illustrate the results of the forecasting system. During this time interval, wind measurements obtained at the meteorological station of the Sines port ( 8.940ºW/37.975ºN) were available together with wave data obtained from two wave rider type directional buoys located offshore the Sines port ( ºW/ ºN) and Leixões port ( ºW/ ºN) operating at approximately 97m water depth and 83m, respectively. The locations of the two buoys maintained by the Hydrographical Institute of the Portuguese Navy are indicated in Fig 3. With the wind and wave forecast performed for four days, an evaluation of the results is made for each day of predictions. An illustrative scheme of the relationship between measurements and predictions is presented in Table 3. Hence, the wind and wave parameters will be organised in four different series. The first series contains predictions with a temporal extension of up to 24h (marked with the symbol x in Table 3) and is denoted as forecast day 1. The second series includes predictions for periods between 24h and 48h (marked o in the Table) and is denoted as forecast day 2. In the same way are defined the series of forecast day 3 and 4. They contain predictions between 48h and 72h and between 72h and 96h, and are denoted in Table 3 by # and Δ, respectively. This representation can give an image of the quality of wave and wind predictions as a function of the distance in time from the moment of the prediction. To evaluate the MM5 results, time series of wind speed were extracted from the coarse domain at the position of the meteorological station and compared with the wind observations. The direct comparisons between the four series of wind velocities simulated by MM5 model and wind measurements are presented in Fig 6, while the results of the statistical analyses are presented in Table 4. Only the data coming from the coarse wind domain were analysed, due to the fact that these data were used in forcing the SWAN model and hence only these data will influence the quality of the results provided by the wave model. It can be observed from the statistical analysis that in the first day of predictions the wind speed is slightly underestimated by the MM5 model (positive bias of 0.01m/s) while for the other days this parameter is overestimated (negative bias). The root mean square error (RMSE) presents similar values in the first three days of forecast (around 2), and only in the last day is this greater, at In relation with the scatter index (SI) and the correlation coefficient (r), the values from the first day are the best (smaller for SI and greater for r). As expected the quality of the predictions decreases once the time from the zero moment of the prediction increases and this fact is also indicated by the evolutions of the scatter index and of the correlation coefficient. n=198 Bias RMSE SI r forecast Vw (m/s) day day day day Table 4: Statistical results of the wind velocities for the period 7/1/2010-h00 28/2/2010-h18 Fig 6: Direct comparison for the wind velocity simulated by the MM5 model in the coarse area against the measurements at the meteorological station, time period 7/1/2010-h00 28/2/2010-h18 (198 data points) Volume 4 No Journal of Operational Oceanography 23

9 The evaluation of the SWAN model results was accomplished by comparing the main wave parameters coming from the model (Hs and Tm) against the same parameters resulted from the measurements at the buoys Sines and Leixões. The time step was 3h and the measurements considered for the validation of the system are those available on the internet site of the Hydrographical Institute of the Portuguese Navy. Nevertheless, there are also some periods when although predictions were performed operationally the results were not compared because the buoys were not operational (as for example the case of the Leixões buoy in January 2010). From the analyses of Figs 7 and 8 that illustrate comparisons for the significant wave height and mean periods, model simulations against measurements at the two buoys, a reasonable agreement between the predictions of the system and the measurements can be observed. Fig 7: Direct comparisons SWAN Sines buoy, significant wave height (Hs) and mean period (Tm), time interval 7/1/2010-h09 28/2/2010-h21 (285 data points) Fig 8: Direct comparisons SWAN Leixões buoy, significant wave height (Hs) and mean period (Tm), time interval 2/2/2010-h11 28/2/2010-h21 (134 data points) 24 Journal of Operational Oceanography Volume 4 No

10 The quality of the results provided by the forecasting system was also analysed statistically. For the case of the Sines buoy, the simulated Hs are greater than the measurements for all the four temporal series considered, as indicated by the negative biases presented in Table 5. For the first two days of forecast the Hs bias is close to zero and also there are small values for the statistical parameters RMSE and SI (less than 0.5 and 0.2, respectively) while the correlation coefficient is greater than 0.9. These indicate a good quality of the forecast products provided by the system. The same observations concerning the Hs statistics are valid in the case of the Leixões buoy (Table 6), with the observation that there is a slight Hs underestimation in the first two days of forecast (positive biases less than 0.04m). n=285 Bias RMSE SI r forecast Hs (m) Tm (m) day day day day day day day day Table 5: Statistical results of the wave parameters at Sines buoy, time interval 7/1/ h 28/2/ h In relation with the mean period for both buoys the statistical parameters are rather similar (slightly better for the Sines buoy). The simulated values are greater than the measurements (negative bias) with values less than 0.46s for Sines and less than 1.02s at Leixões. For both locations SI is less than 0.2 for all the situations considered which can be considered in general a good approach. The correlation coefficient has values between 0.85 and 0.9 for Sines but is less than 0.85 for Leixões. From a qualitative point of view the results from the last two days of forecast are less accurate. This concerns both locations and is indicated by the statistical and the graphical results. The tendency of Hs underestimation is indicated also by the scatter plots related with this parameter presented in Fig 9 for Sines and in Fig 10 for Leixões. The linear regressions adjusted to each dataset show also that, in general, the extreme values are underestimated. n=134 Bias RMSE SI r forecast Hs (m) Tm (m) day day day day day day day day Table 6: Statistical results of the wave parameters at Leixões buoy, time interval 2/2/ h 28/2/ h Fig 9: Hs scatter plots at Sines buoy for the time interval 7/1/2010-h09 28/2/2010-h21 Volume 4 No Journal of Operational Oceanography 25

11 CONCLUSIONS A forecasting system is described for the Portuguese coastal environmental that has been operational since October 2008 and provides daily wave and wind predictions for a time window of four days for the Portuguese ports of Sines and Leixões. The validation of the wind predictions provided by the MM5 meteorological model was accomplished by making comparisons with the wind measurements at the Sines meteorological station, while for the wave parameters delivered by SWAN, wave measurements from the buoys at Sines and Leixões were used. Both direct comparisons of the temporal series and the results of the statistical analysis show in general a good agreement between the wave and wind data provided by the forecasting system and the observed data. Moreover, the level of accuracy of the forecast products are also in agreement with the results provided by similar systems. 7,20 It was also noticed that once increasing the time window of the predictions the quality of the predictions decreases. ACKNOWLEDGEMENTS The work has been performed in the scope of the project MARPORT, Wave Forecast System for Portuguese Harbours, which has been partially funded by the Portuguese Agency for Innovation under the program PRIME-IDEA, contract number 70/ The last three authors have been been funded by the Portuguese Foundation for Science and Technology, the second and third under postdoctoral grants (SFRH/BPD/ 65553/2009 and SFRH/BPD/41063/2007) and the fourth under a doctoral (SFRH/BD/48313/2008) grant. REFERENCES 1. Ponce de Leon S and Guedes Soares C The sheltering effect of the Balearic Islands in the hindcast wave field. Ocean Engineering. 37, Ponce de León S and Guedes Soares C On the sheltering effect of islands in ocean wave models. J. Geophys., 110, C09020, doi: /2004jc002682, 17pp. 3. Bidlot J-R, Holmes DH, Wittmann PA, Lalbeharry R and Chen HS Intercomparison of the performance of operation ocean wave forecasting systems with buoy data. Weather Forecast., 17, , Bertotti L and Cavaleri L Large and small scale wave forecast in the Mediterranean Sea. Nat. Hazards Earth Syst. Sci., 9, Behrens A and Günther H Operational wave prediction of the extreme storms in Northern Europe. Nat. Hazards, 49, Park S, Lee DU and Seo JW Operational wind wave prediction system at KMA. Marine Geodesy, 32 (2), Gómez Lahoz M and Carretero Albiach JC Wave forecasting at the Spanish coasts. Journal of Atmospheric & Ocean Science, 10 (4), Rusu L, Pilar P and Guedes Soares C. 2008a. Hindcast of the wave conditions along the west Iberian coast. Coastal Engineering, 55 (11), Fig 10: Hs scatter plots at Leixões buoy for the time interval 2/2/2010-h11 28/2/2010-h21 26 Journal of Operational Oceanography Volume 4 No

12 9. Pilar P, Guedes Soares C and Carretero JC year wave hindcast for the North East Atlantic European Coast. Coastal Engineering, 55(11), Bernardino M, Rusu L and Guedes Soares C Validation of a wave forecast system for the Portuguese ports. Proceedings of the 5th EuroGOOS Conference, Exeter, UK, Dudhia J, Gill D, Kuo YR, Bourgeois A, Wang W, Bruyere C, Wilson J and Kelly S PSU/NCAR mesoscale modelling system. MM5 modelling system Version 3. NCAR Tech. Notes. 12. Grell GA, Dudhia J and Stauffer DR A description of the fifth-generation Penn State/NCAR mesoscale modelling system (MM5). Tech. Note NCAR/TN 398+STR, NCAR. 13. Rusu L, Bernardino M and Guedes Soares C Influence of wind resolution on the prediction of waves generated in an estuary. Journal of Coastal Research, SI 56, Rusu L, Bernardino M and Guedes Soares C. 2008b. Influence of the wind fields on the accuracy of numerical wave modelling in offshore locations. Proceedings of the 27 th International Conference on Offshore Mechanics and Arctic Engineering, Lisbon, Portugal, ASME, WAMDI Group, The WAM model A third generation ocean wave prediction model. J. Phys. Oceanogr., 18, Gómez Lahoz M and Carretero Albiach JC A two-way nesting procedure for the WAM model: Application to the Spanish Coast. J. Offshore Mechanics and Arctic Engineering, 119, Booij N, Ris RC and Holthuijsen LH A thirdgeneration wave model for coastal regions, 1, Model description and validation. J. Geophys. Res., 104, Stelling GS and Leendertse JJ Approximation of convective processes by cyclic AOI methods. Proceeding 2 nd international conference on estuarine and coastal modelling, ASCE Tampa, Florida, Janssen PAEM Quasi-linear theory of windwave generation applied to wave forecasting. J. Phys. Oceanogr., 21, Bidlot J-R Inter-comparison of operational wave forecasting systems against in-situ observations. Wave Forecasts Verification Project, Report SPA_ETWS_verification Volume 4 No Journal of Operational Oceanography 27