Secret Harbour XBeach Surrogate Model Training and Testing Simulations

Full description

This collection contains the results of XBeach numerical model simulations used to develop a hybrid surrogate model. The dataset was created to train and evaluate Machine Learning based surrogate models capable of predicting XBeach outputs given specific input conditions. The training and testing datasets were extracted from a historical hindcast using the Maximum Dissimilarity Algorithm (MDA) to ensure representativeness across different forcing conditions. Each MDA case corresponds to a specific time step from the historical hindcast but is treated as an independent scenario, meaning the dataset does not preserve temporal continuity.

The dataset is divided in "train" and "test" folders containing 1000 representative cases each. The wave boundary conditions are applied at 7 locations along the offshore boundary. At these locations, integrated wave parameters (significant wave height, peak period, peak direction, directional wave spreading) are used to construct a JONSWAP spectrum to provide the wave forcing conditions for the XBeach model. Water level forcing is provided by the linear addition of TPXO9.2 predicted tides and monthly sea surface height from the ECMWF-ORAS5 global ocean reanalysis. Bathymetry and nearshore topography for the model domain are based on 10m gridded digital elevation models (DEMs) derived from Fugro LADS Corporation airborne LiDAR surveys performed in 2009.

Each folder contains:
- jonswap{1..7}.inp --> The jonswap files used as wave boundary forcing for the XBeach simulations. The file loclist.txt contains the location of each of these offshore locations and is used by the model to read the wave boundary condition files appropriately.
- params.txt --> Model configuration file.
- XBlog.txt, XBwarnnig.txt --> Output log files from the XBeach runs.
- xboutput_SecretHarbour_MDA{1..1000}_{train,test}.nc --> XBeach output files. The length of the simulations is 3 hours, with a time step of 1800 seconds.
Lineage: The dataset was generated by running XBeach simulations for 2000 cases selected using the Maximum Dissimilarity Algorithm (MDA). A historical wave hindcast from 1980 to 2019 was used to derive input conditions, with 1000 representative cases extracted from 1980-2015 for training and 1000 cases from 2015-2019 for testing. Each case was run in XBeach surf-beat mode to obtain the corresponding output variables, which were stored for use in training and validating the surrogate model.