Under the hood

When a job is submitted to Whiffle Wind, it enters a queuing system. The length of the queue varies based on the volume of ongoing tasks, but we aim for a maximum waiting time of one day. To maintain system efficiency and fairness, a maximum of two jobs per user can be processed concurrently.

1. Gathering user input

The first step in the processing chain is ingestion of user input, where the system gathers all necessary details from the user, including: the simulation period, locations and types of turbines, locations and heights of met masts, heights for Wind Resource Grid (WRG) and Fields (horizontal slices).

1. Calculating LES domain

Next, the domain size of the large-eddy simulation (LES) domain is defined as a bounding box around the turbine and met mast coordinates, keeping a spatial margin according to Whiffle’s best practices.

Wind farms can be toggled on or off for inclusion in the LES domain, allowing for flexibility in the calculation. When a wind farm is included in the LES calculation, its turbines are parametrised as actuator disks in both the LES and MESO simulation. If a wind farm is excluded, its turbines are only present in the MESO domain.

Some turbines in wind farms excluded from the LES may still fall within the bounding box (i.e. LES domain) applied around LES-included neighbouring wind farms, and will therefore be included as well. This proximity-based inclusion improves the accuracy of the output without incurring additional costs.

By default, the horizontal resolution of the LES domain is 100 metres and the vertical resolution is 25 metres at ground level, with a 2% growth rate starting at 300 metres above the highest elevation level in the domain. The LES has open in-and outflow boundary conditions and is nested with a one-way coupling in Whiffle’s in-house mesoscale model. The mesoscale domain has a fixed horizontal domain size of 256 kilometres by 256 kilometres and a resolution of 2 kilometres.

1. External data sources

The external data sources required as input by Whiffle’s LES model are retrieved for the user location, including: ERA5 Large-scale Numerical Weather Prediction data, CORINE Land Cover, Ensemble DEM and the Global Canopy Map.

1. Job queue

Now that all input data is in place and domain settings are well-defined, the simulation period is split into N jobs (1 job per day), with N the total number of days. The batch of jobs is distributed over a cluster of GPUs and for each individual job The Whiffle LES Model is run with a simulation spin-up time of 6 hours for the mesoscale model, and 1 hour for the LES model.

• Depending on the user input, either or all of the following modules within the LES model are activated: the parametrisation of wind turbines, sampling of meteo-conditions at all met mast locations and heights, histogram creation of wind speed and wind direction for WRG output.

• Depending on the user-specified heights, a linear spatial interpolation between LES model levels is carried out. Below the lowest model level, a wall-model consistent extrapolation towards the ground is implemented.

• Depending on the user-specified locations, a nearest-neighbour selection of LES grid points is performed.

• 10 minute averages are recorded during the simulation.

1. Long term correction

Whiffle’s Long term correction routine is carried out to provide the Annual Energy Production.

1. Output

The WRG file is compiled from the histograms, the met mast CSVs are compiled from output of the the terrain-following sample points and the turbines CSV is compiled from the actuator disk output.