Loading...
Loading...
Answers to frequently asked questions about wind energy, resource assessment, and AI-powered predictions.
A good onshore wind farm capacity factor is typically 25-45%, while offshore wind farms achieve 40-55%. The global average for onshore wind is approximately 34%, but this varies significantly by region. Sites in the US Great Plains, Patagonia, and the North Sea consistently achieve capacity factors above 40%. Capacity factor depends on wind resource quality, turbine technology selection (particularly specific power), hub height, and operational losses including wake effects and curtailment. A site with a capacity factor below 25% is generally considered marginal for commercial development at current equipment costs.
Onshore wind farms are built on land and represent the most mature, cost-effective form of wind energy with over 900 GW installed globally. Offshore wind farms are built in bodies of water, typically on the continental shelf, and benefit from stronger, more consistent winds with less turbulence. Offshore capacity factors are 10-15 percentage points higher than onshore on average, but capital costs are 2-3 times higher due to marine foundations, subsea cabling, and specialized installation vessels. Offshore wind is growing rapidly, with global installed capacity exceeding 75 GW and projects expanding into floating foundation technology for deeper water sites.
Onshore wind farm costs typically range from $1.0-1.6 million per MW of installed capacity, meaning a 100 MW project costs $100-160 million. Offshore wind farms cost $2.5-5.0 million per MW, putting a 500 MW offshore project at $1.25-2.5 billion. These figures include turbines (which represent 60-70% of onshore costs), foundations, electrical infrastructure, grid connection, development costs, and construction. Operating costs add $10-25/MWh over the project lifetime. The levelized cost of energy (LCOE) for new onshore wind is now $30-50/MWh in most markets, making it competitive with or cheaper than new fossil fuel generation.
LCOE (Levelized Cost of Energy) is the total lifetime cost of building and operating a power plant divided by its total energy output, expressed in dollars per megawatt-hour ($/MWh). LCOE allows direct comparison between different generation technologies on a common basis. For wind energy, LCOE depends primarily on the capacity factor, capital cost, financing terms, and operating costs over a typical 25-30 year project life. Global average onshore wind LCOE has fallen from over $100/MWh in 2010 to approximately $30-50/MWh in 2025, driven by larger turbines, taller towers, and economies of scale in manufacturing and installation.
Wind farms are typically designed for a 25-30 year operational life, though many projects are now extending to 35 years with component refurbishment. Turbine foundations and electrical infrastructure often have longer physical lifespans than the turbines themselves. At the end of the initial design life, operators can choose to decommission the site, extend operations with maintenance investments, or repower with newer, larger turbines on existing foundations. Repowering is increasingly common at legacy sites, as modern turbines can produce 2-3 times more energy than the units they replace using the same land area and grid connection.
Wind resource assessment (WRA) is the process of characterizing the wind conditions at a potential wind farm site to estimate how much energy it will produce. A comprehensive WRA includes measuring or modeling wind speed, direction, shear, turbulence, and their variations across seasons and years. The assessment progresses through stages: desktop screening using reanalysis data and modeling tools, followed by on-site measurement campaigns using meteorological masts or LiDAR devices for bankable energy estimates. The entire process from initial screening to bankable assessment typically takes 2-4 years using traditional methods.
Wind resource assessments draw on multiple data sources. Reanalysis datasets like ERA5 (ECMWF) and MERRA-2 (NASA) provide global hourly wind data spanning decades, forming the backbone of desktop studies. On-site measurements from meteorological masts (60-120m tall towers with calibrated anemometers) and remote sensing devices (LiDAR and SoDAR) provide ground-truth data at the specific site. Satellite-derived wind data supplements offshore assessments. Terrain data from digital elevation models (like Copernicus DEM) characterizes topographic effects on wind flow. WindAI combines ERA5, MERRA-2, and Copernicus DEM data with its machine learning model trained on real production data from 289 wind farms.
ERA5 is the fifth-generation atmospheric reanalysis dataset produced by the European Centre for Medium-Range Weather Forecasts (ECMWF). It provides hourly estimates of atmospheric variables including wind speed, direction, temperature, and pressure on a 31 km global grid from 1940 to near-present. ERA5 combines historical weather observations with numerical weather prediction models to create a physically consistent, gap-free record of past atmospheric conditions. It is the most widely used reanalysis dataset in the wind industry for long-term resource characterization, measure-correlate-predict (MCP) analysis, and model training. WindAI extracts 6 ERA5 variables across 16 grid points for each assessment location.
P50 and P90 are exceedance probability metrics used to express uncertainty in wind energy production estimates. P50 means there is a 50% probability that actual production will exceed the stated value — it represents the median or best estimate. P90 means there is a 90% probability of exceeding the value — it is a conservative estimate used by lenders for debt sizing. The gap between P50 and P90 typically ranges from 10-20% of the P50 value, depending on the sources of uncertainty. Banks typically finance projects based on P90 values to ensure debt service coverage even in below-average wind years. WindAI reports both P50 and P90 estimates for every assessment.
The accuracy of wind resource assessments varies significantly by method and stage. Desktop studies using mesoscale modeling typically achieve annual energy prediction accuracy of plus or minus 10-15%. On-site measurement campaigns with 12+ months of data reduce uncertainty to plus or minus 5-8% at the P50 level. Machine learning approaches like WindAI, trained on real production data, achieve annual prediction errors of 2.1-7.8% (RMSE 0.1477, R-squared 0.777) against operational wind farm output. No method eliminates uncertainty entirely — even after years of measurement, inter-annual wind variability of 6-10% persists. The key is matching assessment accuracy to the decision being made at each project stage.
WindAI uses a deep neural network trained on over 10 million hourly observations from 289 operational wind farms across 8 countries. When you provide a location's coordinates and turbine specifications, WindAI extracts 391 features from ERA5 reanalysis data, MERRA-2 data, and Copernicus DEM terrain data — including wind speeds at multiple heights, temperature, pressure, boundary layer characteristics, terrain elevation, slope, and roughness. These features are fed through a 6-layer neural network that outputs hourly capacity factor predictions for an entire year (8,760+ hours), from which mean capacity factor, AEP, P50/P90, monthly profiles, and diurnal patterns are derived.
WindAI uses a 6-layer fully connected deep neural network with an architecture of 391 input features mapped through layers of 768, 256, 256, 128, and 64 neurons to a single output (predicted capacity factor). The network uses GELU activation functions, batch normalization, and dropout regularization. It was trained on 10M+ hourly observations from 289 wind farms using mean squared error loss with the Adam optimizer. The model achieves an RMSE of 0.1477 and R-squared of 0.777 on held-out test plants that were never seen during training.
WindAI was trained on data from 289 operational wind farms spanning 8 countries: Australia, Brazil, United Kingdom, Belgium, Denmark, Canada, New Zealand, and the United States. These farms represent a diverse range of terrain types (flat plains, rolling hills, coastal sites, mountain ridgelines), climate zones (tropical, temperate, subarctic), and turbine technologies (40+ turbine models from all major manufacturers). The training dataset contains over 10 million hourly observations of actual wind farm production paired with concurrent atmospheric and terrain features.
WindAI achieves an RMSE of 0.1477 and R-squared of 0.777 when validated against held-out test wind farms that were not included in the training data. In terms of annual energy prediction, WindAI's errors range from 2.1% to 7.8% compared to actual operational production. This accuracy is comparable to or better than traditional consultant desktop studies (which typically have 10-15% uncertainty) and approaches the accuracy of on-site measurement campaigns, at a fraction of the cost and time. Performance is strongest for sites in regions well-represented in the training data.
A single WindAI assessment costs $49.99 and delivers a complete report including mean capacity factor, annual energy production (AEP), P50 and P90 estimates, monthly capacity factor profiles, diurnal patterns, and 8,760+ hourly predictions. Your first 5 assessments are free. Bulk pricing is available for developers screening large portfolios. By comparison, a consultant desktop study costs $5,000-$15,000 per site and takes 2-6 weeks, while a meteorological mast campaign costs $50,000-$150,000 and requires 12-18 months. WindAI delivers results in 2-5 minutes.
Wind shear is the change in wind speed with height above the ground. Near the surface, friction slows the wind, so speeds increase with altitude — a phenomenon described by the power law: V2 = V1 times (H2/H1) raised to the power of alpha, where alpha is the shear exponent. Typical shear exponents range from 0.10 over open water to 0.25 over forested land. Wind shear matters because modern turbine hub heights of 100-170 meters are well above typical measurement heights (10-80m), so accurate shear characterization is critical for extrapolating wind speeds and estimating energy production. Small errors in the shear exponent can change AEP estimates by 5-15%.
Increasing hub height exposes the turbine rotor to stronger winds, and because wind power is proportional to the cube of wind speed, even modest speed increases yield significant energy gains. Moving from 80m to 120m hub height typically increases energy production by 10-20%, depending on the wind shear profile. In areas with high surface roughness (forests, suburban terrain), the benefit is greater because shear is stronger. However, taller towers cost more and face logistical constraints — road weight limits, crane availability, and aviation permitting. The optimal hub height balances incremental energy revenue against incremental tower cost for each specific site.
Specific power is a turbine's rated power divided by its rotor swept area, measured in watts per square meter (W/m2). It is the single most important metric for matching a turbine to a site's wind resource. Low specific power turbines (200-300 W/m2) have large rotors relative to their generator size, producing higher capacity factors at low-to-medium wind sites. High specific power turbines (350-450 W/m2) are designed for high-wind sites where the smaller rotor still reaches rated output frequently. Choosing the wrong specific power can reduce AEP by 10-20% compared to the optimal selection. Most modern onshore turbines have specific power in the 250-350 W/m2 range.
The power law is the most commonly used equation for estimating wind speed at one height based on a measurement at another height: V2 = V1 times (H2/H1) raised to the power alpha, where V1 is the known wind speed at height H1, V2 is the estimated speed at height H2, and alpha is the shear exponent. For example, with a shear exponent of 0.2, a wind speed of 7 m/s at 50m extrapolates to approximately 8.05 m/s at 100m. The shear exponent varies with surface roughness, atmospheric stability, and time of day. The logarithmic law is an alternative that explicitly accounts for roughness length and is considered more physically rigorous in the surface layer.
When a wind turbine extracts kinetic energy from the wind, the air downstream moves slower and is more turbulent — this is the wake. Turbines located behind others in the prevailing wind direction receive this degraded airflow, producing less energy and experiencing higher fatigue loads. Wake losses typically reduce total wind farm output by 5-15% compared to the sum of what each turbine would produce in isolation. The magnitude depends on turbine spacing (measured in rotor diameters), wind farm layout, wind direction distribution, and atmospheric stability. Larger modern rotors have made wake management more important, and computational wake models (Jensen, Fuga, Gaussian) are used during layout optimization.