How Wind and Solar Forecasts Shape ISO Operations

When wind and solar were small fractions of the U.S. generation mix, accurate forecasting of their hourly output was a nice-to-have. Today it is operationally critical. SPP, ERCOT, and CAISO regularly serve more than half of their daytime energy from variable renewables on favorable days, and a forecast bust on either resource translates directly into emergency dispatch decisions and price volatility.

This article walks through how ISOs build wind and solar forecasts, where the main error sources originate, and how operators use the forecasts to plan day-ahead unit commitment and real-time dispatch.

A typical wind farm forecast is built from numerical weather model output (most commonly the High-Resolution Rapid Refresh, the HRRR, plus the Rapid Refresh and the GFS) downscaled to hub-height wind speeds at each turbine location. The wind speeds are then converted to power output using the manufacturer's turbine-specific power curve, which maps a given wind speed at hub height to the megawatt output of the machine.

Aggregating across hundreds or thousands of turbines in a wind farm averages out a substantial amount of turbine-level forecast noise. Aggregating across a regional fleet (for example, all wind farms in West Texas) further smooths the signal. Per public DOE / National Renewable Energy Laboratory studies of operational ISO forecasting (nrel.gov), modern day-ahead wind forecast errors at the regional scale typically fall in the single-digit-percent mean-absolute-error range as a fraction of installed capacity, which is impressive given the underlying meteorological uncertainty.

The hardest forecasts are around frontal passages — when a cold front sweeps across the wind belt, output can ramp up or down by 10–15 GW within a few hours. Even with high-resolution models, forecast timing of frontal passages can be off by an hour or two in either direction, which is enough to materially shift dispatch decisions for the following operating period.

Solar forecasting is built from cloud forecasts (numerical weather model output plus satellite-derived nowcasts), aerosol load forecasts (smoke and dust), and the deterministic solar geometry calculation. The geometry is exact — the sun's elevation and azimuth at every minute of every day for every latitude and longitude is known — so the forecasting challenge reduces to predicting cloud cover and aerosols.

Cloud forecasts at the day-ahead horizon rely on numerical weather models. At the intraday and real-time horizons, geostationary satellite imagery from GOES-East and GOES-West is the dominant input, with cloud advection and growth modeled from successive satellite frames. Per the same public DOE / NREL operational forecasting literature, modern intraday solar forecasts can achieve low-single-digit-percent mean-absolute errors as a fraction of installed capacity for one-hour-ahead predictions in clear-sky regions like the Desert Southwest.

The hardest forecasts are during convective cloud development on summer afternoons in regions like Texas and the Southeast. A forecast of mostly clear skies that misses an unexpected pop-up thunderstorm can swing solar output by hundreds of megawatts within minutes. Smoke from western wildfires has periodically derated solar output across multiple ISO footprints for days at a time, and aerosol forecasts have not historically been a strong suit of standard NWP models.

ISO operators run a centralized renewable forecasting service that produces hourly or sub-hourly forecasts of every utility-scale wind and solar resource within the footprint. CAISO's renewable forecasting is built in-house. ERCOT contracts with multiple vendor forecasts and uses an ensemble. PJM, MISO, and SPP combine vendor and in-house forecasting.

The day-ahead unit commitment process uses the day-ahead renewable forecast as a fixed input — the system commits enough thermal capacity to meet the forecast load minus the forecast renewable output, plus required reserve margins. If the renewable forecast turns out to be too high in real-time (less actual renewable than forecast), the system has to dispatch additional thermal capacity from quick-start units or call on reserves. If the forecast is too low (more actual renewable than forecast), the system has to back down or shut off thermal units, sometimes incurring significant minimum-run constraints.

Forecast errors translate into ancillary service requirements. ISOs procure regulation, spinning reserve, and non-spinning reserve in part to cover forecast error in load and renewables. As renewable penetrations grow, the reserve requirements grow, and the cost of carrying those reserves shows up in every market clearing.

The largest day-ahead renewable forecast errors come from a small number of weather patterns. Frontal passage timing errors are the dominant source for wind. Convective cloud development is the dominant source for solar. Wildfire smoke is an irregular but increasingly important source of error for solar across the Western Interconnection. Tropical systems can shut down coastal wind for days during the summer hurricane season in the Gulf and Atlantic.

Cold-weather icing is an underappreciated source of wind forecast error. When freezing rain or freezing fog accretes on turbine blades, the blades cannot maintain their aerodynamic profile, and turbines either trip offline or run at significantly reduced capacity. Standard wind forecasting models do not always pick up the icing risk in their predicted output, and ISOs have to overlay icing forecasts separately.

Weather Workbench surfaces the input weather variables that drive wind and solar generation at the ISO level: city-scale wind speed forecasts from NWS gridpoint data and city-scale sky cover forecasts as a solar proxy. The Wind & Solar Outlook panels on each relevant ISO detail page categorize forecast wind speeds and sky cover into qualitative tiers — Low / Moderate / High / Very High for wind, and Excellent / Good / Reduced / Poor for solar.

These are rough heuristic indicators rather than calibrated capacity-factor forecasts. They do not incorporate the turbine-by-turbine or panel-by-panel forecasting that ISO operators rely on. For trading and operational decisions, the relevant ISO's official renewable forecasting output is the authoritative source. Weather Workbench is positioned as a quick situational-awareness layer above those primary sources.

Model documentation, satellite imagery, and renewable forecasting practice references in this article are drawn from the public-use federal and ISO sources listed below.