Answer
Solar forecasting software predicts how much power a plant will make, from 30 minutes (Reuniwatt InstaCast) to 14 days (Solargis Forecast, Solcast) ahead. Long-term P50 and P90 yield figures come from simulation, not forecasting. The commercial tools do not publish list prices, so you ask each vendor for a quote.
A forecast that looks 30 minutes ahead and one that looks 14 days ahead are different products, even when both are sold as solar forecasting software. This guide compares seven vendor offerings by horizon, method and price model, then explains where P50 and P90 yield estimates fit.
Last updated: November 2026
Three jobs get mixed up under this name. Energy forecasting predicts the output of a plant that is already running. Simulation estimates the yearly yield (energy output) of a design before it is built. Monitoring measures what the plant produced. This guide compares forecasting tools first. Then it explains where simulation and P50/P90 fit.
SurgePV publishes this guide and sells solar design and proposal software with energy-yield modeling. Every tool in the table below is an operational forecast service from another company. The table is not a ranking. Horizons, methods and plan details come from each vendor’s own pages, read on 28 October 2026. Accuracy claims on those pages are the vendors’ own.
Need the design side of the yield estimate?
Forecast feeds run a plant that exists. For a design that does not exist yet, see how SurgePV models yield on your own project.
Book a DemoChecking cost first? See the SurgePV pricing page.
How do the forecasting tools compare?
The rows run from the shortest horizon a vendor lists to the longest.
| Tool | Forecast horizon (vendor’s page) | Method (vendor’s page) | Positioned for | Price model |
|---|---|---|---|---|
| Reuniwatt InstaCast | 30-minute horizon at 1-minute steps, updated every minute | Images from an on-site sky imager, with machine-learning post-processing | Plant control systems and off-grid microgrids | Not published; contact form |
| Reuniwatt HourCast | 10 to 15 minutes up to 6 hours ahead, resolution down to 10 minutes | Satellite cloud motion vectors, weather models and AI | Single sites, portfolios or territories | Not published; contact form |
| Quartz Solar Forecast (Open Climate Fix) | 0 to 48 hours for a single PV site | GFS, ICON or ECMWF weather models | Open-source model; the project says a commercial forecast is also available | Open source; commercial forecast on request |
| Solargis Forecast | Nowcasting up to 3 hours; forecasts to 14 days (Professional, Enterprise) or 7 days (Basic) | Satellite cloud motion vectors for nowcasting; a blend of IFS, GFS, ICON and HRRR weather models for later horizons | Basic: day-ahead trading and smaller portfolios. Professional: intraday trading, grid balancing, utility-scale, trackers. Enterprise: aggregators, large portfolios, grid operators | Basic, Professional and Enterprise plans are all “Let’s talk” |
| Solcast forecast data and PV power models | Forecast data from now to +14 days; 5 to 60 minute steps and 5-minute, 15-minute or 6-hour updates, depending on plan | Rooftop, Advanced and Premium PV power models. Premium is trained on your own generation data (6 months minimum, 1 year recommended) | Rooftop model for smaller assets (not suitable above 1 MW, per Solcast’s energy-data pricing page); Premium for utility-scale assets | Request a quote; free requests at your own locations for trials |
| Reuniwatt DayCast | Up to 10 days ahead (the same page also says “up to a whole week ahead”) | Weather models tuned for cloudiness, plus AI and statistical models | Single sites, portfolios or territories | Not published; contact form |
| Meteologica | Not fixed: the page says granularity, range and update frequency adapt to the customer | Latest weather models, calibrated with generation and availability data | Wind and solar asset owners, energy traders, and grid operators and utilities (for demand forecasts) | Not published; contact sales |
- Reuniwatt InstaCast0.5 h 30 minutes
- Solargis nowcasting3 h
- Reuniwatt HourCast6 h
- Quartz Solar Forecast48 h 2 days
- Solargis Forecast Basic168 h 7 days
- Reuniwatt DayCast240 h 10 days
- Solargis Pro and Enterprise336 h 14 days
- Solcast336 h 14 days
Source: each vendor's own product pages listed in the table (Meteologica does not publish a fixed horizon), checked 28 October 2026.
Key Takeaway
Pick the horizon from the decision you need to make. If it happens within an hour, look at sky-imager and satellite products. If it happens a day or more ahead, weather-model blends matter more.
On free options: Solcast says you can make 10 free Rooftop PV Power Forecast Model data requests at your own locations. It also says you can make unlimited Advanced PV Power Model requests at its “unmetered” locations. Open Climate Fix publishes the Quartz Solar Forecast model as open source. It has a hosted API documented at open.quartz.solar. An API is a way for other software to connect.
What Each Horizon Is Used For
Solargis splits its service into two methods. Nowcasting means very short-term forecasting. Solargis nowcasting runs up to 3 hours ahead. It uses cloud motion vector models on real-time satellite imagery, with updates every 5 to 15 minutes. These models track how clouds move. Forecasts from there out to 14 days blend several weather models into one consensus. Reuniwatt describes the same split with three products. They use sky-imager images for the next 30 minutes, satellite images for the next few hours, and weather models beyond that.
Solargis ties nowcasting to grid balancing and intraday trading. Grid balancing keeps power supply and demand level. Intraday trading is buying and selling power within the same day. Solargis lists day-ahead trading and maintenance scheduling as uses for the longer forecasts. If your decision happens within an hour, look at the sky-imager and satellite products. If it happens a day or more ahead, the weather-model blends matter more.
Latest Update
On 3 February 2026 Solargis added hail forecasting to Solargis Forecast for the continental United States. It is available on the Professional and Enterprise plans. It updates hourly for subscribed locations and gives a lead time of up to 48 hours (Solargis announcement).
How do forecasting, simulation and monitoring differ?
A tool called “forecasting software” may do one of three jobs. Ask which one before you compare features.
- Forecasting answers “what will this plant produce in the next 15 minutes, day or week?” It needs live weather inputs. For the best results it also needs your plant’s own generation history. Solcast’s Premium model, for example, asks for 6 months of generation data at minimum.
- Simulation answers “what should this design produce in a typical or conservative year?” It runs before construction on historical weather data. The result feeds financial models, permits and lender reports. Installers meet it inside design software. For more, see solar simulation software. The simulation accuracy comparison is also useful.
- Monitoring answers “what did this plant actually produce, and is that normal?” It compares measured output with an expected value.
Most installers need simulation and monitoring, not a forecast feed. A forecast feed matters when someone sells power into a market, balances a grid or dispatches storage. To dispatch storage is to decide when a battery charges and discharges. Solcast’s pricing page says its Rooftop model is not suitable for assets above 1 MW. So even forecast vendors split their products by plant size.
P50 and P90 Explained
P50 and P90 are exceedance levels of annual energy yield, not forecasts. An exceedance level is a value that the real result is expected to beat a set share of the time. Solargis defines P50 as the annual yield that has a 50% probability of being exceeded. It defines P90 as the yield expected to be exceeded 90% of the time, which makes it the more conservative figure. Solargis also warns that a Pxx value is not a statistical percentile. The 90th percentile is the value below which 90% of data falls. P90 is a level expected to be exceeded with 90% probability.
Assume annual yield follows a normal distribution (a bell-shaped spread around the average). Solargis calls this common practice. Then every Pxx value follows from P50 and the standard deviation (a measure of how widely results spread):
| Estimate | Formula (Solargis) | Probability of exceedance |
|---|---|---|
| P75 | P50 - 0.675 x standard deviation | 75% |
| P90 | P50 - 1.282 x standard deviation | 90% |
| P95 | P50 - 1.645 x standard deviation | 95% |
| P99 | P50 - 2.326 x standard deviation | 99% |
Total uncertainty combines three sources by root-sum-square. This assumes they are independent. The three sources are the irradiance model (sunlight data), the PV simulation and year-to-year weather variability. Root-sum-square means you square each one, add them and take the square root. For the variability term, Solargis recommends at least 10 years of good historical data (Solargis, 24 September 2025).
Worked example (illustrative inputs, not a real site). Assume a P50 of 10,000 MWh per year. Assume uncertainty terms of 2% (irradiance model), 3% (simulation) and 6% (interannual variability). Interannual variability is the change from year to year.
- Total uncertainty: the square root of (2² + 3² + 6²) = the square root of (4 + 9 + 36) = the square root of 49 = 7%.
- Standard deviation: 7% of 10,000 MWh = 700 MWh.
- P75: 10,000 - 0.675 x 700 = 10,000 - 472.5 = 9,527.5 MWh.
- P90: 10,000 - 1.282 x 700 = 10,000 - 897.4 = 9,102.6 MWh, which is 8.97% below P50.
- P95: 10,000 - 1.645 x 700 = 10,000 - 1,151.5 = 8,848.5 MWh.
- P99: 10,000 - 2.326 x 700 = 10,000 - 1,628.2 = 8,371.8 MWh.
Lenders use these figures differently from developers. In standard project finance practice, Solargis notes, base-case cash flows are usually built on P50 energy estimates. The gap to P90 widens as uncertainty grows. Lenders cover that gap with several tools: loan-to-value limits, minimum debt service coverage ratios, reserve accounts, dividend limits and sponsor guarantees (Solargis, 8 April 2026). A loan-to-value limit caps the loan against the project’s value. A debt service coverage ratio checks that income covers the loan payments by a set margin. For the glossary definition, see P50/P90.
Estimating yield for a design, not a running plant?
SurgePV models energy yield and finances from your roof design. Bring one project to a guided demo.
Book a Free DemoChecking cost first? See the SurgePV pricing page.
Weather Data Behind Long-Term Yield Estimates
Long-term P50 and P90 figures start from historical weather data, not from a forecast feed. Solargis says its historical time series starts in 1994, 1999 or 2007, depending on the location. A TMY (typical meteorological year) P50 file is a synthetic year built from the typical months of that record. It is small and widely supported, but it leaves out extreme years. Solargis also offers a TMY P90 file. It is built from recombined months so that annual irradiation is close to the P90 value. Solcast lists TMY files built from 2007 to 2025 data at 90 m resolution. It offers P50 only on its TMY Pro plan, and P50, P75, P90 and P95 on TMY Max (Solcast pricing). For how these files are used in design, read solar TMY data.
How to Evaluate Forecasting Software
- Match the horizon to the decision. A 30-minute horizon and a 14-day horizon are different products, as the table shows.
- Check what it needs from you. Some models need your generation history. Others, like Solcast’s Rooftop model, work with or without system specifications.
- Check plant size limits and plan limits. Solargis Basic stops at 7 days and hourly steps. Solcast’s Rooftop model is not for assets above 1 MW.
- Ask for a trial against your own meters. Solcast offers historical forecasts for your locations so you can compare them with your measurements. Both Solcast and Solargis publish validation studies (Solcast, Solargis). Those are the vendors’ own results.
- Ask how accuracy was measured. Request the metric, the horizon, the reference measurements and the climate zones. A single accuracy percentage tells you little.
- Confirm delivery and price in writing. Delivery options listed by these vendors include API, SFTP (secure file transfer) and CSV (a plain spreadsheet file). Prices are quoted on request.
Pro Tip
Ask each vendor for historical forecasts at your own site and compare them with your meter data before you sign. Agree the error metric, horizon and period first, so the comparison cannot be tuned afterwards.
Forecast feeds do not replace a design-stage yield model. Read more about SurgePV’s energy-yield modeling and the generation and financial tool, or book a demo to see yield modeling on your own project.
Frequently Asked Questions
How far ahead can solar forecasting software predict?
It depends on the product. Reuniwatt lists 30 minutes for InstaCast, up to 6 hours for HourCast and up to 10 days for DayCast. Solargis Forecast lists nowcasting (very short-term forecasts) up to 3 hours. It lists forecasts up to 14 days on its Professional and Enterprise plans (7 days on Basic). Solcast lists forecast data out to 14 days. The open-source Quartz Solar Forecast covers 0 to 48 hours for one site.
What is the difference between solar forecasting and solar simulation?
Forecasting predicts what a running plant will produce in the coming minutes, hours or days, using live weather inputs. Simulation estimates the annual yield of a design from historical weather, before the plant is built. Monitoring measures what the plant actually produced.
What is the difference between P50 and P90?
P50 is the annual energy yield with a 50% probability of being exceeded. P90 is the yield with a 90% probability of being exceeded, so it is lower. Solargis describes P90 as the more conservative estimate. It also notes that a Pxx value is an exceedance level, not a statistical percentile.
How is P90 calculated from P50?
Assume annual yield follows a normal distribution. Then P90 equals P50 minus 1.282 times the standard deviation (Solargis). Take a P50 of 10,000 MWh and a 7% standard deviation (700 MWh). P90 is then 10,000 - 1.282 x 700 = 9,102.6 MWh. The inputs are illustrative, not a real site.
Does solar forecasting software cost money?
Mostly yes, and none of the commercial vendors compared here publish list prices. Solcast, Solargis, Reuniwatt and Meteologica sell on request. Solcast offers free data requests at your own locations for trials. Open Climate Fix publishes an open-source model.
Does SurgePV forecast solar production?
SurgePV is solar design and proposal software with energy-yield modeling. The tools in the comparison table are operational forecast services from other companies. That is a different product category. To see SurgePV’s yield modeling on your own project, book a guided demo.
Where this fits
This article is part of SurgePV's Solar Design hub, which works through the topic from first principles to the decisions a project team actually has to make.
