Run on supercomputers, mesoscale models are a class of fluid dynamics models used to predict the time series of natural weather over a limited area. Because they were originally designed to represent weather phenomena at scales between “global” (>2000 km, e.g., a jet stream that spans a continent) and “micro” (<2 km, e.g., a tornado), they are known as “mesoscale models” (Orlanski, 1975). They solve the physical equations that govern the time-evolution of wind and weather within a 3-dimensional fluid.
Mesoscale models were originally developed by scientists at universities and national labs in the 1960s and 70s, and today nearly every nation relies on them for accurate daily weather forecasting. In the 1990s, meteorologists first employed mesoscale modeling for short-term operational wind energy forecasting. It was then recognized that mesoscale models could be used not just to predict the short-term future but to recreate the long-term past. Consultants started using retrospective mesoscale modeling at high spatial resolution (100-400 meter, or microscale, grid spacing) for project-scale wind flow modeling in support of financeable wind energy assessment studies. The reasons for doing so are compelling, as described in the next section.
Key to the ability to perform these retrospective simulations was the advent of global reanalysis datasets such as the NCAR/ NCEP Reanalysis Project, ERA40, ERA-Interim, CFSR, MERRA, MERRA2, and ERA5. These research datasets provided global gridded weather variables at hourly time resolution, going back 40+ years. Although they are too coarse (grid spacing of several 10s of km) to be used directly for project-scale wind flow mapping, they can be used to drive retrospective mesoscale model simulations at high resolution, by providing the necessary initial and boundary conditions the mesoscale models require to simulate the real weather of the past.
Why Does Mesoscale Modeling Work for Wind Energy Assessment?
Linear/CFD models are applied with an assumed uniformity of wind flow on the horizontal edges and vertical top of the model domain, and the model’s job is to adjust the hub-height wind flow across the project due only to the effects of finescale topography and land use across the site. If there are any larger-scale affects occurring that violate that assumption of uniformity, their influence will not be captured. As it turns out, there are many such effects, and they can impose significant gradients or other patterns of wind speed across the site. Fortunately, all of them are represented by mesoscale weather prediction models. What are these effects?
- Orographically altered flows that arise from mountains, valleys, gaps, etc. that lie outside of the immediate project vicinity.
- Realistic representation of the effects of variation in atmospheric stability, from unstable to strongly stable (e.g., +10°C per 100 meters vertical stratification).
- Sea-breeze patterns arising from coastal features outside the project vicinity.
- Storm tracks, monsoon circulations, thunderstorms, weather fronts, or other recurring large-scale weather phenomenon. Note that many of these phenomena are dependent on accurate representation of water phase changes, clouds, precipitation, atmospheric radiation, and planetary boundary layer processes, which only mesoscale models include.
- Vertically propagating orographic gravity waves. This highly underappreciated but very important topographic flow phenomenon leads to deceleration of flow windward of a ridge and acceleration leeward of it, even for very subtle rises in terrain. Only mesoscale models have the flow dynamics capable of supporting this phenomenon.
- Transience. Linear/CFD are typically run multiple times for different combinations of steady, uniform flow conditions. As such, the simulated flow field contains no realistic wind fluctuations, which are the norm rather than the exception at many wind sites, and can affect the spatial wind resource.
Finally, it is important to state that on-site measurements are still considered “ground truth” for wind resource at the project site, and the description above should not be taken to mean that mesoscale models preclude the need for on-site measurements. Various techniques have been developed to spatially bias-correct mesoscale model output with on-site observations, retaining the best of both worlds: the ground truth of the observations and the weather-driven spatial wind patterns produced by the mesoscale model.
Model resolution
Mesoscale models can be run at a wide range of grid resolutions, including microscale (≤ 400 m), and now even delving into the “large eddy simulation” (LES) range (grid spacing ≤ 100 m). A practical approach is to use sufficiently fine resolution to capture phenomena at the scale of typical turbine spacing (a few hundred meters), and then downscale the result to the turbine pad (10s of m grid spacing) with a simplified model. Such downscaling can produce accurate wind maps as long as the “leap” from the mesoscale model grid (e.g. 200 m) to the downscaled resolution (e.g., 30 m) is not too large (preferably < a factor of 10).
Below are shown raw model (non-downscaled) mean 100-m wind speed maps over an area approximately 20 km x 20 km (Fig. 1), with model resolutions of 600 m (left) and 200 m (right) grid spacing. Below those maps are the same output downscaled to 30 m using a 1-arcsecond digital elevation model (Fig. 2). The raw model output clearly shows the effect of the higher resolution from the mesoscale model (Fig. 1 right compared to left). The extra detail in the downscaled model wind field is also clearly evident (Fig. 2 compared to Fig. 1). The two downscaled maps look similar to each other, yet differences can still be seen. The map downscaled from 200 m is likely more accurate, because a smaller downscaling “leap” was taken (factor of 6.7, compared to 20).
Mesoscale Model error analysis
ArcVera compiled a mesoscale modeling and met tower dataset, to explore the relationships of mesoscale modeling cross-prediction errors to other project and model attributes. The dataset consisted of about 105 project runs with two to ten met towers per project.
Table 1 depicts the relationship between mesoscale model grid spacing and cross-prediction error (uncertainty). Not surprisingly, the error decreases with decreasing grid spacing.
Table 1. Mesoscale Model Wind-Speed Accuracy
(1) Mean absolute deviation
(2) Standard deviation
(3) Number of samples
The relationship between cross prediction error of the mesoscale model and the degree of site complexity is explored in Table 2. For a given wind farm project area, we define terrain complexity as follows:
Table 2. Terrain Complexity and Wind-Speed Accuracy, All Resolutions. Abbreviations as in Table 1.
The average cross-prediction errors increase steadily with increasing terrain complexity. For comparison, cross prediction errors for linear and CFD models reported in the literature show standard deviations of 4% in simple terrain and 10% in complex terrain.
Finally, we looked at two very different countries for which ArcVera had large and roughly equally sized samples, USA and Brazil. USA is midlatitude, with winds driven by a variety of weather regimes; whereas Brazil is a tropical nation with winds driven primarily by the Atlantic easterly trades impinging upon terrain features in the northeastern region. Only complex-terrain sites (TEV > 75 m) were included here for both countries. The cross prediction error is very similar (Table 4), indicating that mesoscale modeling, when applied consistently and driven by a global reanalysis data set, provides consistent accuracy across very different wind resource environments.
Table 3. Model Wind-Speed Accuracy in the USA vs. Brazil, Complex Terrain. Abbreviations as in Table 1.
Conclusion
Mesoscale models are eminently capable of producing accurate, detailed wind-resource maps of project areas using microscale (100-300 m) grid spacing. There are compelling reasons why this high-resolution mesoscale modeling approach to wind flow mapping should yield more accurate spatial wind maps than linear flow or computational fluid dynamics methods. Those approaches do not account for the broad range of spatial scales and rich spectrum of meteorological phenomena and thermodynamics that greatly influence a site’s wind climate. Other studies confirm this improved performance to be the case.
The uncertainty of mesoscale modeling for wind resource assessment was examined via cross prediction error statistics for a large global dataset of wind projects and meteorological towers. Uncertainty was found to vary directly with mesoscale model grid spacing and terrain complexity. However, uncertainty was very nearly the same for two geographically different countries (USA and Brazil), indicating that mesoscale modeling provides consistent performance for a wide variety of wind resource types.
References
Orlanski, I., 1975: A rational subdivision of scales for atmospheric processes. Bull. Amer. Meteor. Soc., 56, 527–530.
- Gregory S. Poulos, PhD - CEO, Principal Atmospheric Scientist, ArcVera Renewables
- Mark Stoelinga, PhD - Lead, Atmospheric Science Innovation and Application, ArcVera Renewables
Please share!
If you want to cooperate with us and would like to reuse some of our content, please contact: contact@energetica-india.net.
Next events
Renewable Energy India Expo 2025
October 30, 2025
The Renewable Energy India Expo (REI Expo) showcases India’s achievements in renewable energy and industry progress. In its remarkable 17th edition, the event brought together top brands and over 800 suppliers, featuring a diverse range of prod...
