Historical Data
Forecast
Knowledge Center
About
Contact

the times series

A Journey Through Time-Domain Wind Resource Assessment.

2. TIMES METHODOLOGY

By Marta Gil-Bardají

This is the second post in the TIMES Series, where we break down everything about Vortex TIMES. In this post, we’ll discuss how we create the TIMES dataset and give you a glance into the enhancement process. Check out other posts in the series from the table of contents here.

TIMES Generation Overview

We begin with a WRF baseline simulation, which, to ensure cost-effectiveness, avoids explicitly solving the microscale for the entire 10, 20, or 30-year period. The simulation employs a resolution of 300 m, striking a balance between cost and accuracy. Despite being saved every 10 minutes, the output lacks meteorological variability due to the hourly to half-hourly representativeness of WRF at that resolution.

The surface-layer scheme utilized is the Mellor–Yamada–Nakanishi–Niino (MYNN) scheme, calculating turbulent kinetic energy (TKE) for a rough estimation of wind standard deviation (SD).

In parallel, a Vortex-LES simulation explicitly tackles the microscale, directly calculating SD. To manage computational costs, Vortex-LES runs for a subset of the baseline simulation period. Patterns and relationships discovered during the time-matched period between LES and baseline simulations inform improvements to specific aspects of the overall baseline simulation, resulting in the generation of TIMES output.

BASELINE	ENHANCEMENT
300 m mesoscale WRF simulation	100 m microscale WRF-LES simulation
10, 20 or 30 years 10-minute output	6 months 10-minute aggregate output
at one location using ERA5 reanalysis.	at one location using ERA5 reanalysis
the results are stored every 10 minutes to produce a high-temporal resolution baseline series	the 10-minute mean wind and standard deviation are computed from the high frequency output of the model
the standard deviation is derived from the turbulence kinetic energy parametrized by the MYNN scheme	the standard deviation is computed from the high frequency output of the model

Training Process

Vortex-LES and the baseline simulation share a 6-month consecutive concurrent period. Both are modeled using similar technology, using the same reanalysis (ERA5), and utilizing identical topography and land use datasets. The key distinction lies in the Vortex-LES simulation’s explicit solving of the microscale in its last two domains (333 m and 111 m), as opposed to parametrizing it. Consequently, these simulations are highly comparable and coherent, facilitating the identification of systematic differences and patterns even over short periods.

Enhancing Wind Distribution

The 100 m Vortex-LES outperforms the 300 m mesoscale baseline simulation in describing wind distribution. The explicit solving of the microscale and increased resolution enable the detection of more localized patterns. During the concurrent period, wind speed means at each height from both baseline and Vortex-LES are compared. Correction factors are learned for each height, adjusting the baseline values towards the LES results. Additionally, a dampened quantile-mapping strategy is trained to make wind histograms more similar. This robust strategy consistently improves the wind distribution metrics across the entire corrected TIMES dataset.

Improving Sub-hourly Variability

The baseline mesoscale simulation, stored every 10 minutes, tends to be ‘too soft,’ implying that changes in variables between timestamps (ramps) are close to zero. However, observed ramps can be more significant due to increased sub-hourly variability beyond the mesoscale simulation’s capture. A Random Forest machine learning algorithm is employed in training, learning from Vortex-LES the required amount of sub-hourly variability to be added to the baseline simulation. This learning considers predictor variables like time, temperature, wind data values, existing variability, and more. The training ensures that the sub-hourly variability in the TIMES dataset retains physical meaning from the atmospheric model, avoiding being merely statistical noise.

Stay tuned for more insights. Explore the entire series here. Your feedback drives our commitment to advancing atmospheric modeling with Vortex TIMES.

For pricing options, feel free to contact us.

We are open to suggestions about other topics that can be added to the analysis, leave your comments below:

Modeled wind resource data for the wind industry.
At any site around the world. Onshore and offshore.

info@vortexfdc.com