My cart
Validation
MOTIVATION: We thought long and hard about what we would like to have, as technical analysts using weather data in different ways for bespoke projects with different meteorology around the world. We then designed an internal system to create the weather data we need to fulfill the requirements for a given project. Once we built it, we were so happy with the result that we thought we should share this capability with the world, to help not only technical analysts, but anyone interested in weather that needs good weather information for any purpose, anywhere. We think you will like it, just like we do.
À votre réussite (to your success) - Bureau Veritas Renewables Technical Advisory Team
1. Introduction: Minute-by-minute weather conditions anywhere, anytime, over the last 60 years.
Have you ever wondered: What the day-to-day weather was like at the house your grandparents grew up in or when man first landed on the moon or when weather ruined your plans on an important day? Or how snowy it has been in a spot you are planning to build a remote cabin? Or the wind speed and wind direction and operating environment in a remote area where you are considering building a wind-energy facility? Or the weather conditions 1000 feet above ground for say, building a skyscraper?
VMTpro (Virtual Meteorological Time series professional) is a customizable, user-friendly, project-point- specific, bespoke mesoscale modeling weather time-series simulation, individually run on the supercomputing cloud, for any latitude-longitude coordinate and height-above-ground at your service. The user literally has the power of state-of-the-art numerical weather prediction back-casting at their fingertips.
With VMTpro, you can re-create detailed minute-by-minute weather conditions anywhere on Earth over the last 60 years. The complexity of atmospheric science, calculus, physics, thermodynamics and supercomputing made simple and delivered in support creative decision making – simply amazing.
Your results are not “pre-computed”. The results are not already recorded in a database. They are “from scratch”, just for your needs, with site specific weather model set-up focused purely on your location.
2. The Advantage of Flexibility for Site-Specific Needs: Aiming for Accuracy
The weather-variable time series are individually calculated on the cloud supercomputer only after you submit your request. Text file(s) are returned to you by email in a convenient text-file time-series, columnated format within, typically, 24 hours. You choose what you want, from time intervals of 1 to 60 minutes, the grid spacing granularity of 1 to 3 km, the variables desired (if not standard variables, including TI, snow, stability, shear, veer, see the order page for all variables included and optional), and the number of heights from 2-1000 meters above ground for any coordinate you request.
The output file you receive provides time series of the local windspeed, temperature, precipitation, stability, and other temporal weather characteristics of a site by harnessing the power of WRF. 1 Output is rapidly generated in parallel on high-performance-computing (HPC) platforms for time series spanning months to years, depending on user-specified run parameters. VMTpro is flexible, allowing users the power to specify even the mesoscale model spatial resolution to use for applications where higher resolution might be advantageous (e.g., in complex terrain or because of other complex meteorological or site-specific factors). VMTpro’s ability to allow users to vary the mesoscale model spatial resolution is an advantage that competing tools do not offer. Once completed on the cloud computing system, VMTpro results are delivered to the user in a common text-file format so that derived time-series products and use cases can be easily addressed.
3. Validation Versus ERA5 Reanalysis: Improved correlation and RMSE
To validate VMTpro time-series output, we conducted a global study using twenty-two (22) test sites across four continents, where measured met-tower data and ERA5 reanalysis output are compared with VMTpro results. [ERA5 is an hourly time series of back-looking weather data supplied publicly by the European Centre for Medium-Range Weather Forecasting, ECMWF.] Validation site locations were chosen for their varied and complementary characteristics, spanning different elevation ranges, terrain- complexity metrics, inland/coastal locations, turbulence intensities, mean-wind speeds, and atmospheric stability conditions. For all validation tests, co-located VMTpro typically year-long timeseries were compared against quality-controlled concurrent measured 10-minute-interval meteorological tower data. Our validation focused on hourly and daily average performance relative to the measurements.
Example VMTpro hourly time-series output, for a 20-day period, is shown in Figure 1 for a 90-m high met-tower at one of the validation locations.2 VMTpro windspeed amplitude and fluctuations track the met-tower observations, with 1-hr and 24-hr correlations of R2(VMT, 1-hr) = 0.883 and R2(VMT, 24-hr) = 0.961, respectively. These correlation results represent substantial improvements over the corresponding ERA5 results for this location, which are R2(ERA5, 1-hr) = 0.826 and R2(ERA5, 24-hr) = 0.935.3 Similarly, the root-mean- squared-error differences (RMSE) between measured and modeled met-tower wind speeds are better for VMTpro than they are for ERA5. 4 VMTpro 1-hour and 24-hour wind-speed errors for this site are RMSE(VMT, 1-hr) = 2.8 m/s and RMSE(VMT, 24-hr) = 1.8 m/s, while the corresponding ERA5 wind-speed errors are significantly higher, with RMSE(ERA5, 1-hr) = 5.6 m/s and RMSE(ERA5, 24-hr) = 5.0 m/s.
These results are typical, and show distinct improvement, as shown in this table:
| Model | R2 - daily | R2 - hourly | RMSE - daily | RMSE - hourly |
| VMTpro | 0.91 | 0.80 | 1.3 m/s | 2.2 m/s |
| ERA5 | 0.88 | 0.76 | 2.3 m/s | 3.0 m/s |
Referring to the Table, the average 1-hr and 24-hr VMTpro correlations across all validation sites are R2(VMT, 1-hr) = 0.80 and R2(VMT, 24-hr) = 0.91, compared to ERA5’s lower average correlations of R2(ERA5, 1-hr) = 0.76 and R2(ERA5, 24-hr) = 0.88. Similarly, the corresponding 1-hr and 24-hr average RMSE differences across all twenty- two sites are RMSE(VMT, 1-hr) = 2.2 m/s and RMSE(VMT, 24-hr) = 1.3 m/s, compared to the higher ERA5 RMSE differences of RMSE(ERA5, 1-hr) = 3.0 m/s and RMSE(ERA5, 24-hr) = 2.3 m/s. VMTpro significantly reduces wind-speed errors and therefore requires less bias correction to actual on-site data, giving materially greater confidence in the results of subsequent analysis. It is common and recommended practice to bias- correct (or use Model Output Statistics “MOS” or machine-learning techniques) with raw mesoscale model wind speed results for the purpose of wind energy applications (this is an available Bureau Veritas service also).
While VMTpro provides the user with the ability to refine WRF’s mesoscale-model spatial resolution (1, 2, or 3 km grid spacing) to better resolve local meteorological behavior, this option isn’t always needed, and the added expense of smaller grid spacing may not be necessary for your application. For example, VMTpro runs that use a 3km spatial resolution produces lower wind-speed errors than ERA5 for 68% of the validation sites we studied. The percentage of sites with lower VMTpro wind-speed errors increases as WRF’s grid is refined, so that more than 90% of test sites exhibit lower wind-speed errors for VMTpro when a 1km grid is specified. For 3 km grid spacing, average daily wind-speed correlation-improvement is +0.02 over ERA5; for 2 km it is +0.04 and nearly the same at 1 km grid spacing. On average, RMSE improves steadily with smaller grid spacing (higher resolution) from 2.4 to 2.3 to 2.2 m/s for 3, 2, and 1 km grid spacing, respectively – and all of these improve on (are lower than) ERA5 RMSE by 0.8 to 1 m/s.
For this reason, we recommend testing VMTpro output for short exploration runs before ordering long time series – simply click the inexpensive “Trial Order” button to order exploratory orders, if necessary. It’s your choice.
1. The Weather Research and Forecast model (WRF) is a state-of-the-art, open-source numerical-weather-prediction system used for both research and operational forecasting. It was developed by university scientists through a partnership with the U.S. National Center for Atmospheric Research (NCAR) and the U.S. National Oceanic and Atmospheric Administration (NOAA).
2. This test-site is located in South Africa. The terrain is complex, with an elevation range of 359 meters. It has a mean windspeed of <ws> = 8m/s, and a measured TI15 = 0.069.
3. R2 correlation reports how well-aligned and in-sync two time series are, with higher values being better, and R2 =1 being the maximum and best value possible.
4. RMSE reports the root-mean-squared difference between two time series, with lower values being better, and RMSE=0 representing the minimum and best value possible.
