Quartz Solar combines satellite imagery, numerical weather forecasts (NWPs), topographic data and real-time PV readings in cutting edge machine learning algorithms to produce highly accurate forecasts of solar energy generation from minutes to hours ahead.

The machine learning prediction uses Open Climate Fix’s proprietary PVNet model and takes seconds to run, so ensuring forecasts are always current. The forecasts are updated by re-running the algorithm as new data arrives - most frequently every five minutes as new satellite images are taken.

The estimation of actual PV production comes from Sheffield Solar’s PV_Live service. PV_Live is used by National Grid Electricity System Operator (ESO) as the official estimate of solar PV electricity generation in Great Britain (GB). PV_Live produces two estimates of PV generation: intra-day an initial estimate is made with an approximately 30-minute lag. This estimate uses a smaller sample of PV system readings. Overnight a larger set of half-hourly PV system generation figures is used to create an updated (and day behind) PV generation estimate. Both of these figures are available in the Quartz Solar API and via the UI. The Quartz Solar algorithm is trained to forecast the day-behind PV_Live figures as they are the most accurate estimates.

The algorithm and historical data sources are open source, ensuring we capture input from as many people as possible and continuously improve the algorithm. Code can be found on https://github.com/openclimatefix.

Other key forecast properties:

• Forecasts are produced in half-hourly time steps, from the next half hour out to eight (8) hours ahead.
• The geographic extent is all of Great Britain (GB).
• Forecasts are produced at the GB National and regional level. The regions used are the Grid Supply Points (GSPs), as defined by National Grid ESO.