Live and Forecast Accuracy

Bias and error validation of Solcast data against surface measurements
Live and Forecast

The purpose of this analysis is to enable users to estimate Solcast real-time and forecast accuracy for their site(s) prior to subscription or integration effort. It is based on 12 months of recent Solcast forecast and estimated actuals data, compared to surface measured actuals from high-quality measurement sites.

Also included are error statistics, and benchmarks against common alternatives. Users focused on historical data or TMYs will find a separate, multi-year analysis on the Solcast website. This document analysis irradiance (GHI, DNI, GTI) and power production (fixed-tilt and single-axis tracking).

For brevity, a selection of commonly used statistics are included, summarised by climate type. For a full range of statistics, at site level, and to request the raw data, please contact the Solcast team.

Live and Forecast data products

Solcast operates a global cloud tracking and forecasting system, using near-real-time satellite imagery from 11 weather satellites, and weather data from 7 numerical weather prediction (NWP) models. These inputs are used to make irradiance and PV power data, which is distributed via the Solcast API (Application Programming Interface), which enables automated, synchronous data requests for any point on earth.

This analysis focuses on live and forecast data, from -7 to +14 days from now. The live and forecast data products deliver PV power, irradiance, and weather data globally, at time granularities of 5, 10, 15, 30 and 60 minutes. Coverage is all non-polar continental areas and nearby islands, with spatial resolution of 2km and data updates every 5 to 15 minutes.

Accuracy verification methodology

Accuracy is typically the most important product attribute amongst commercial users of the Solcast API. Customer evaluations or trials can produce definitive results; however they require a large amount of resource, expertise, and elapsed time to gather and analyse the data, and to quality control the surface measurements.

Measurement site selection

Accuracy is typically the most important product attribute amongst commercial users of the Solcast API. Customer evaluations or trials can produce definitive results; however they require a large amount of resource, expertise, and elapsed time to gather and analyse the data, and to quality control the surface measurements.

Sites with very high elevation (over 2000 metres) have been excluded due to limited applicability to solar energy applications. Otherwise, all sites with data available during the analysis period are included. Across these three networks a total of 33 sites are included. For climate type categorisation, two dimensions are used: (1) the latitude zone (Tropical/Subtropical being latitudes equatorward of 35 degrees, and Temperate being latitudes between 35 and 60 degrees north and south (no sites poleward of 60 degrees are included); and (2) site climate type based on annual average precipitation (Humid being sites of greater than 750mm per year, Arid sites are less than 350mm per year, and Semi-Arid sites are between 350mm and 750mm per year) based on the CPC Merged Analysis of Precipitation (CMAP) from NOAA.

Solcast does not perform site-specific adaptation of its satellite-derived data (i.e. measurements from a site are not used in the dataset for that site), and the period of this analysis is not used in any regional algorithmic calibrations.

Map of measurement sites included in the analysis. Of the 33 included sites, a total of 18 are designated Tropical/Subtropical (7 Humid, 1 Semi-Arid, 10 Arid). A total of 15 are designated Temperate (8 Humid, 4 Semi-Arid, and 3 Arid).

Map of measurement sites included in the analysis. Of the 33 included sites, a total of 18 are designated Tropical/Subtropical (7 Humid, 1 Semi-Arid, 10 Arid). A total of 15 are designated Temperate (8 Humid, 4 Semi-Arid, and 3 Arid).
Benchmark data

Two commonly available benchmarks are included, to put accuracy performance in context, and to assess the marginal value of Solcast data. For day-ahead forecasts and real-time estimated actual Live data, the benchmark is the Global Forecast System (GFS) model from NOAA, a commonly-used global weather forecast model. For simplicity, GFS data is taken at +24 hours ahead, even for intraday comparisons, because global weather models such as GFS are not available until ~5 hours after real time. For intra-day forecasts (i.e. +1 and +2 hours), we also include a benchmark called Smart Persistence, which is created by persisting the last hour's clearness index (i.e. the level of cloudiness is persisted, but clear-sky irradiance updates). This means the Smart Persistence benchmark is aware of sun position, and does not, for example, persist irradiance values after sunset).

Verification analysis method

During the 12-month period covering April 2020 to March 2021 inclusive, measurement data is 88% complete on average. All data were converted to hourly means prior to analysis. Forecast horizons available from the Solcast archive at the time were +1 hour, +2 hours and day-ahead only. A range of error statistics were calculated following "Metrics for Evaluating the Accuracy of Solar Power Forecasting" (NREL, 2013), including normalisation by AC capacity in the case of PV power. For irradiance statistics, normalisation uses the mean observed irradiace at that site.

Actuals power production data was derived by converting measured irradiance to power using Solcast's PV power model, which is based on the open source pvlib python tool with proprietary extensions, and adding noise in the irradiance to power conversion. The approach of simulating PV production rather than using actual plant data was chosen for the following reasons: (1) allows irradiance and power statistics to be available for the same sites; (2) ensures transparency and repeatability of results by users; (3) ensures a broad geographical spread of sites; (4) avoids conflation from plant availability and curtailment on accuracy statistics (in any case, users can specify real-time plant conditions in API requests). The plant specifications used are the Solcast defaults for 10MW Utility Scale sites, including an inverter loading ratio of 1.30. A noise model was used to model the error in the PV power model itself, in order to make accuracy results realistic. The noise model was trained on PV plant measurements and corresponding Solcast PV model estimates from ten sites globally where full plant specifications have been given to the PV model. PV Power and GTI, which use a Hay transposition model are not calculated unless a site has both GHI and DNI measurement.

This analysis excludes probabilistic statistics, because most users only require deterministic statistics. Users interested in probabilistic measures may note that forecast data from the Solcast API does include probabilistic 10th and 90th percentile values, which are dynamically generated using a combination of satellite cloud tracking and a range of NWP models.

Nocturnal zero values are included in this verification for hourly measures; users interested in daytime results only for hourly values can approximate results via doubling the error values listed here. For daily total energy values, there is no difference in statistics owing to inclusion of nocturnal values, so the statistics presented here remain directly applicable.

Accuracy verification results

PV Power production forecast accuracy

The following tables show statistics for the Mean Absolute Percentage Error (MAPE), as defined in the above-mentioned NREL 2013 paper, of power production for single 10MW PV plants with horizontal single axis tracking. The statistics are grouped by climate type and latitude zone. For a complete range of statistics and PV system types, and results for specific measurement sites, please contact Solcast.

MAPE (%) errors of Day-ahead PV Power Forecasts

Measure: Daily total energy production (single-axis-tracking)
Site type Data source Day-ahead (+24hr) error (%)
Mean & range
Global average (all 33 sites) Solcast 5.9%
(2.1% to 9.4%)
Smart Persistence 13.2%
(2.7% to 21.2%)
GFS 7.7%
(2.4% to 13.2%)
Tropical/Subtropical, Arid & Semi-Arid (11 sites) Solcast 4.4%
(2.1% to 5.9%)
Smart Persistence 7.6%
(2.7% to 9.4%)
GFS 5.8%
(2.4% to 10.0%)
Tropical/Subtropical, Humid (7 sites) Solcast 7.6%
(6.1% to 9.4%)
Smart Persistence 15.7%
(7.9% to 20.6%)
GFS 10.4%
(5.3% to 13.2%)
Temperate, Arid & Semi-Arid (7 sites) Solcast 6.0%
(4.4% to 7.1%)
Smart Persistence 11.1%
(4.0% to 17.5%)
GFS 8.1%
(6.1% to 10.5%)
Temperate, Humid (8 sites) Solcast 6.5%
(4.9% to 9.2%)
Smart Persistence 16.6%
(12.1% to 21.2%)
GFS 8.0%
(6.1% to 10.5%)

MAPE (%) errors of Intraday & Day-Ahead PV Power Forecasts

Measure: Hourly energy production (single-axis-tracking), nocturnal zeros included
Site type Data source +1 hours ahead error (%)
Mean & range
+2 hours ahead error (%)
Mean & range
Day-ahead (+24hr) error (%)
Mean & range
Global average (all 33 sites) Solcast 2.9%
(1.4% to 5.0%)
4.1%
(2.5% to 6.2%)
5.1%
(2.9% to 8.2%)
Smart Persistence 4.0%
(2.0% to 6.3%)
5.2%
(2.6% to 7.4%)
9.0%
(3.8% to 14.0%)
GFS 6.0%
(3.3% to 9.2%)
Tropical/Subtropical, Arid & Semi-Arid (11 sites) Solcast 2.3%
(1.4% to 2.9%)
3.1%
(1.9% to 3.8%)
3.9%
(2.9% to 5.0%)
Smart Persistence 3.0%
(2.0% to 3.7%)
4.1%
(2.6% to 5.1%)
5.8%
(3.8% to 8.6%)
GFS 4.6%
(3.7% to 5.9%)
Tropical/Subtropical, Humid (7 sites) Solcast 3.0%
(2.4% to 3.8%)
4.5%
(3.2% to 5.6%)
6.1%
(4.5% to 7.0%)
Smart Persistence 4.4%
(3.0% to 5.3%)
5.8%
(3.7% to 6.9%)
10.1%
(3.8% to 8.6%)
GFS 7.3%
(4.6% to 8.5%)
Temperate, Arid & Semi-Arid (7 sites) Solcast 3.8%
(2.8% to 5.0%)
4.8%
(3.5% to 6.1%)
5.3%
(3.5% to 6.1%)
Smart Persistence 4.9%
(3.4% to 6.3%)
6.0%
(3.9% to 7.3%)
10.0%
(5.8% to 11.4%)
GFS 6.1%
(3.3% to 7.8%)
Temperate, Humid (8 sites) Solcast 2.9%
(2.4% to 3.8%)
4.2%
(3.4% to 6.2%)
5.5%
(4.1% to 8.2%)
Smart Persistence 4.3%
(3.3% to 5.5%)
5.5%
(4.3% to 7.4%)
10.8%
(7.8% to 14.0%)
GFS 6.4%
(4.8% to 9.2%)
Irradiance real-time and forecast accuracy

The following tables show statistics for the normalised Root Mean Square Error (nRMSE), as defined in the above-mentioned NREL 2013 paper, of GHI irradiance. For a full range of statistics (including DNI and GTI), at site level, and to request the raw data, please contact the Solcast team.

nRMSE (%) errors of Intraday & Day-Ahead Irradiance Forecasts & Estimated Actuals

Measure: GHI hourly average, daytime only, scaled by mean observation
Site type Data source Real-time
Mean & range
+1 hour
Mean & range
+2 hour
Mean & range
+24 hour
Mean & range
Global average (all 33 sites) Solcast 15.6%
(5.5% to 32.4%)
16.6%
(6.0% to 36.6%)
22.6%
(9.2% to 45.2%)
27.4%
(8.4% to 58.5%)
Smart Persistence N/A 22.9%
(8.1% to 41.0%)
29.2%
(11.5% to 46.9%)
47.9%
(10.7% to 72.7%)
GFS 33.9%
(8.5% to 76.3%)
Tropical/Subtropical, Arid & Semi-Arid (11 sites) Solcast 10.4%
(5.5% to 14.4%)
10.8%
(6.0% to 15.3%)
14.2%
(9.2% to 19.4%)
16.7%
(8.4% to 24.1%)
Smart Persistence N/A 14.3%
(8.1% to 19.6%)
18.6%
(11.5% to 26.6%)
26.6%
(10.7% to 44.0%)
GFS 19.4%
(9.5% to 33.8%)
Tropical/Subtropical, Humid (7 sites) Solcast 17.2%
(9.6% to 32.4%)
19.3%
(10.3% to 36.6%)
27.3%
(15.0% to 45.2%)
34.8%
(19.1% to 58.5%)
Smart Persistence N/A 26.5%
(13.9% to 41.0%)
33.1%
(17.65 to 46.9%)
55.4%
(25.6% to 72.6%)
GFS 43.9%
(22.4% to 76.3%)
Temperate, Arid & Semi-Arid (7 sites) Solcast 20.0%
(9.9% to 26.1%)
20.5%
(10.4% to 26.7%)
25.2%
(13.2% to 31.8%)
28.0%
(14.5% to 35.9%)
Smart Persistence N/A 26.7%
(13.0% to 33.0%)
33.1%
(16.1% to 41.6%)
52.0%
(24.1% to 67.4%)
GFS 35.8%
(16.0% to 50.9%)
Temperate, Humid (8 sites) Solcast 17.1%
(12.2% to 19.8%)
18.3%
(14.5% to 21.7%)
26.4%
(22.7% to 33.8%)
33.1%
(28.0% to 43.0%)
Smart Persistence N/A 27.1%
(23.6% to 31.6%)
35.2%
(31.9% to 41.7%)
63.0%
(51.6% to 72.7%)
GFS 41.0%
(35.5% to 51.0%)

Live and Forecast Data Products

Live and Forecast API
The complete suite of irradiance and weather data required for effective monitoring, operation, and forecasting for rooftops and large-scale solar farms.
Grid Aggregations
Estimating the aggregate power for hundreds of thousands of PV sites in a single value to improve load forecasting, manage your VPP, or beat the market.