Operational Wave Model Validation SWAN Objective Skill assesment

Operational Wave Model Validation SWAN

Objective • Skill assesment of Operational Wave Model (CSM - North Sea) Method • • • A hindcast, for year 2009 Comparison of wave parameters betweeen Operational Model result and observed (wave bouy) wave parameters In addition to bulk-wave parameter, wind sea and swell wave properties will also be examined (spectral partitioning algorithm is used) Model : SWAN Observed data : Wave spectrum data from RWS (thanks to Ivo)

SWAN model set up Observed data The data consist of : Komen et al -Energy spectrum -Wave direction and spreading Forcing : -Hirlam Wind (per 3 hour) -Output of WW 3 Global as swell boundary Three buoys are used for validation -Eierlandse Gat -Europlatform -K 13 platform

Building ‘ 2 D wave spectrum’ from RWS field data • • (Quasi) 2 D wave spectra can be reconstructed from field spectral data from RWS, using “spreading function”. Parameter given by the field data : – Energy as function of frequency E(f) – Mean Wave Direction as function of frequency θ 0 – Wave spreading as function of frequency (for NDF standard deviation) • (Quasi) 2 D wave spectrum can be constructed using following expression : Since : Therefore : • Normal distribution spreading function : A routine is built to construct ‘quasi’ 2 D spectra out of RWS field data (based on Alfons Smalle’s routine)

Apply Spectral Partitioning Literatures • • • Tracy, B. A. , E. -M. Devaliere, et al. (2007) Hanson, J. L. , B. A. Tracy, et al. (2009) Portilla et al 2008 Hanson and Phillips 2001 etc An algorithm is built to differentiate between wind sea and swell – wave separation 1. 2. 3. Directional wave spectrum is partitioned using so called “watershed algorithm” (Matlab has ‘watershad. m’ function in Image Processing Toolbox) Partitions then identified (as wind sea) using Wave Age criteria (Komen et. al 1984) The rest of partitions belongs to Swell Wind Seas are calssified based on Wave Age criteria : Angle between Wind and Wind Sea direction Wind speed

An example of wave separation result Station : K 13 APFM Period : 20 June – 30 June

Hindcast for year 2009 1. SWAN model is run for 1 year period (yr 2009) 2. SWAN output (2 D wave spectrum) are stored 3. Spectral partitioning is applied for both model output and observed spectra 4. Wave parameters (wave height, peak period, and mean direction) are produced for both wind sea and swell 5. Comparison of these parameters for every month 6. Model skill is represented by statistical measures (Bias, RMS error)

2009 hindcast Result Monthly Wind Sea and Swell heigth error measures RMS error Normalized RMS error

2009 hindcast Result Monthly Wind Sea and Swell heigth error measures Bias 1. For wave heigth and wave direction, model performs reasonably good 2. For Peak period, buoy Eierlandse Gat shows distinct different tendency, especially for month of May to August. 3. Slight tendency of underestimation for June, while overestimation for April

2009 hindcast Result Problem with field data RMS error -Energy spectrum data fro buoy Eierlandse Gat, gives unreasonbly high number (not 999 nor other easily recognized numbers)

K 13 APFM April

K 13 APFM April, 02 -13 • Wind waves are slightly overestimated for waves greater than 2 m (arround 9 th of April), which corresponds to slight overestimation in the wind speed during this time • Swells are predicted with good skill

K 13 APFM June

K 13 APFM June, 01 -15 • Computed wind waves are in good agreement with observation, with slight tendency of underestimation • Swells are predicted well with tendency of underestimation • Swells and wind waves existance are almost equally dominant

K 13 APFM November

K 13 APFM November, 02 -14 • Computed wind waves are in good agreement with observation • Swells heigth are predicted well with tendency of slight underestimation • Wind sea energy is dominant

Summary • Wave separation algorithm has been created (developing. . ) • An extensive wave model validation has been carried out for Operational CSM model – – Model performs very good reproducing Wind Sea Model performs very good reproducing swell The forcing, Hi. RLAM wind fields are in good agreement with observations Consistent tendency of slight underestimation for swell Next. . . • To add more materials/analysis to the curent result, so it can be decent enough for a publication • More details on the wave separation algorithm • More details on the analysis of the validation