In this document:
Validation study that evaluates Solargis accuracy of Global Horizontal Irradiation (GHI) and PV power output (PVOUT) forecasts across different climate zones, forecast horizons and data resolutions is presented.
Overview
As photovoltaic (PV) power becomes an increasing share of the global energy mix, its inherent variability introduces operational challenges. PV generation depends on weather conditions, time of day, and seasonal patterns, making accurate forecasting of Global Horizontal Irradiation (GHI) and PV power output (PVOUT) essential for reliable system operation.
Accurate forecasts are critical for multiple stakeholders:
Grid operators use them to balance supply and demand and maintain system stability.
PV asset operators rely on forecasts to meet contractual commitments and avoid financial penalties.
Energy traders use them to anticipate market dynamics and optimize decisions
Forecasting also supports battery storage management and maintenance planning during low-production periods.
Validation study details
This validation study assesses the accuracy of GHI and PVOUT forecasts by comparing them against satellite-based reference data. It evaluates forecast performance across different time horizons, locations, climate conditions, and data resolutions.
The analysis includes both short-term (“hour-ahead”) and longer-term (“day-ahead”) forecasts, based on data from 2025 to reflect current operating conditions. The study has a global scope, covering multiple sites across all continents and focusing on fixed photovoltaic systems.
Forecast accuracy is evaluated using standard metrics to ensure transparency and comparability, with results further grouped by major climate zones (temperate, cold, arid, tropical, and polar) to highlight performance under different environmental conditions.
Scope and methodology
Locations: 153 global sites across all continents.
Horizons: Evaluates "hour-ahead" (nowcasting) and "day-ahead" (planning) horizons.
Climatic analysis: Results are categorized into five major zones: Temperate, Cold, Arid, Tropical, and Polar.
Metrics: Standard indicators include Bias, Mean Absolute Deviation (MAD), and Root Mean Square Deviation (RMSD).
Validation study parameters | |
|---|---|
Geographical scope | Global |
Data parameters | GHI, PVOUT |
Data resolution | 15min, hourly |
Calculated indicators | Bias, MAD, RMSD |
Number of validation sites | 153 |
Table 1: Validation study parameters
GHI validation: "hour-ahead" horizon
GHI (Global Horizontal Irradiation) represents total solar irradiance on a horizontal surface and is the primary driver of PVOUT. In Solargis, “hour-ahead” forecasts are generated using the Cloud Motion Vector (CMV) model, which tracks cloud movement. This short-term horizon falls into the nowcasting category.
The tables below summarize GHI accuracy for hourly and 15-minute resolutions (see the Annex for metric calculation formulas).
GHI | hourly | 15min |
|---|---|---|
Bias | -0.18% | -0.18% |
MAD | 4.87% | 5.39% |
RMSD | 8.58% | 9.58% |
Table 2: GHI accuracy statistics comparing hourly and 15-minute values (hour-ahead)
GHI, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | -0.18% | 0.11% | -0.48% | -0.21% | -0.02% | 0.49% |
MAD | 4.87% | 6.88% | 4.16% | 4.42% | 4.19% | 6.33% |
RMSD | 8.58% | 11.05% | 7.80% | 7.84% | 7.09% | 10.81% |
Table 3: GHI accuracy by climate zone for hourly resolution (hour-ahead)
GHI, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | -0.18% | 0.11% | -0.48% | -0.21% | -0.02% | 0.49% |
MAD | 5.39% | 7.53% | 4.61% | 4.93% | 4.66% | 7.01% |
RMSD | 9.58% | 12.14% | 8.80% | 8.83% | 7.97% | 12.02% |
Table 4: GHI accuracy by climate zone for 15-minute resolution (hour-ahead)
Key findings
Forecast accuracy is higher for hourly data due to temporal aggregation, which smooths short-term variability. Accuracy also varies by climate zone, reflecting differences in atmospheric stability and cloud dynamics.
Arid regions achieve the best performance due to stable, clear-sky conditions.
Temperate and cold climates show moderate errors driven by variable cloud cover and frontal systems.
Tropical and polar regions exhibit the highest errors, caused by convective activity, rapid cloud formation, low sun angles, and seasonal extremes.
In general, forecast accuracy decreases as atmospheric variability increases.
The map below shows validation sites where Solargis compared forecasts against satellite-based GHI. It is interactive—clicking on a site displays its metadata and corresponding hourly accuracy statistics.
Map 1: Interactive map showing validation sites and site-specific GHI accuracy statistics.
GHI validation: "day-ahead" horizon
“Day-ahead” GHI forecasts are generated using Numerical Weather Prediction (NWP) models and represent the most widely used forecast horizon due to electricity market scheduling requirements.
The tables below summarize overall accuracy for hourly and 15-minute data (temporal aggregation effect remains visible, with better performance for hourly resolution).
GHI | Hourly | 15min |
|---|---|---|
Bias | 0.56% | 0.57% |
MAD | 6.15% | 6.61% |
RMSD | 10.11% | 10.89% |
Table 5: GHI accuracy statistics comparing hourly and 15-minute values (day-ahead)
GHI, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | 0.56% | 1.00% | -0.15% | 0.74% | 0.80% | 1.86% |
MAD | 6.15% | 8.36% | 5.24% | 5.74% | 5.53% | 7.27% |
RMSD | 10.11% | 12.80% | 9.00% | 9.43% | 8.89% | 11.85% |
Table 6: GHI accuracy by climate zone for hourly resolution (day-ahead)
GHI, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | 0.57% | 1.01% | -0.14% | 0.74% | 0.81% | 1.86% |
MAD | 6.61% | 8.98% | 5.66% | 6.16% | 5.91% | 7.90% |
RMSD | 10.89% | 13.73% | 9.82% | 10.13% | 9.50% | 12.88% |
Table 7: GHI accuracy by climate zone for 15-minute resolution (day-ahead)
Key findings
Accuracy is lower compared to “hour-ahead” forecasts due to the longer prediction horizon. However, climate-based patterns remain consistent:
Arid regions achieve the best performance due to stable conditions
Tropical regions perform worst due to convective cloud formation.
Polar regions also show lower accuracy, partly influenced by limited sample size,
Temperate and cold climates show moderate errors.
The map below illustrates spatial accuracy patterns for “day-ahead” GHI (hourly data). Clicking on a site displays its detailed statistics.
Map 2: Global accuracy patterns for Day-Ahead GHI predictions based on hourly data.
PVOUT validation: "hour-ahead" horizon
PVOUT represents the electrical power produced by a PV system from sunlight. While GHI is the primary input determining available solar energy, PVOUT also depends on PV system configuration, including module tilt and orientation, installed capacity, shading and system losses.
Validation statistics are based on standardized PV configurations:
Geometry: Fixed (single angle)
Azimuth: 0° or 180°, depending on hemisphere
Tilt: Optimal for location
Installed Capacity: 10,000 kW
Module Technology: CSI
Detailed PV parameters for all sites can be found in the Annex.
PVOUT | hourly | 15min |
|---|---|---|
Bias | -0.15% | -0.15% |
MAD | 4.89% | 5.40% |
RMSD | 8.66% | 9.68% |
Table 8: PVOUT accuracy statistics comparing hourly and 15-minute values (hour-ahead)
PVOUT, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | -0.15% | 0.15% | -0.40% | -0.21% | 0.00% | 0.31% |
MAD | 4.89% | 5.94% | 4.08% | 4.82% | 4.97% | 6.71% |
RMSD | 8.66% | 9.62% | 7.72% | 8.63% | 8.73% | 11.54% |
Table 9: PVOUT average bias, MAD, and RMSD variability by climate zone for hourly resolution (hour-ahead)
PVOUT, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | -0.15% | 0.15% | -0.40% | -0.21% | 0.00% | 0.31% |
MAD | 5.40% | 6.49% | 4.51% | 5.36% | 5.50% | 7.38% |
RMSD | 9.68% | 10.57% | 8.69% | 9.72% | 9.74% | 12.79% |
Table 10: PVOUT average bias, MAD, and RMSD variability by climate zone for 15-minute resolution (hour-ahead)
PVOUT, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
MAD | 4.89% | 5.94% | 4.08% | 4.82% | 4.97% | 6.71% |
Min MAD | 1.60% | 4.34% | 1.60% | 3.05% | 4.08% | 5.99% |
Max MAD | 8.61% | 8.24% | 6.99% | 8.61% | 6.90% | 7.45% |
Table11: PVOUT MAD variability (average, min, max) by climate zone for hourly resolution (hour-ahead)
PVOUT, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
MAD | 5.40% | 6.49% | 4.51% | 5.36% | 5.50% | 7.38% |
Min MAD | 1.81% | 4.80% | 1.81% | 3.24% | 4.45% | 6.59% |
Max MAD | 9.15% | 8.80% | 7.58% | 9.15% | 7.54% | 8.19% |
Table12: PVOUT MAD variability (average, min, max) by climate zone for 15-minute resolution (hour-ahead)
Key findings
Accuracy patterns mirror those of GHI. Bias remains near zero across climate zones, but this does not guarantee high accuracy, as positive and negative errors can cancel each other out. MAD is the most relevant metric, reflecting the typical deviation between predicted and actual PVOUT, while RMSD highlights extreme errors but can exaggerate rare outliers.
“Hour-ahead” forecast accuracy is highest in arid zones and lowest in tropical and polar regions. Polar site statistics are less representative due to only two validation locations. To provide a comprehensive view, tables 11 and 12 include also minimum, and maximum MAD values, which help manage forecast expectations by showing both typical performance and variability across sites and climates.
The map below illustrates spatial accuracy patterns for “hour-ahead” PVOUT forecasts (hourly data). Clicking on a site displays its metadata and accuracy statistics.
Map 3: Global accuracy patterns for Hour-Ahead PVOUT predictions.
PVOUT validation: "day-ahead" horizon
The “day-ahead” forecast horizon (24–48 hours ahead) is critical for PVOUT because electricity markets and grid operations are organized around day-ahead scheduling. Generators submit expected production for the following day, and accurate forecasts help avoid imbalance penalties.
Grid operators use these forecasts to plan conventional generation, reserve capacity, and anticipate potential constraints, ensuring system balance and stability. PV plant operators also use day-ahead forecasts for production planning, market participation, and battery storage optimization. Shorter horizons (“hour-ahead” or intra-day) primarily refine these day-ahead schedules for real-time balancing.
Tables below summarize overall and per climate “day-ahead” forecast accuracy for hourly and 15-minute data resolution.
PVOUT | hourly | 15min |
|---|---|---|
Bias | 1.08% | 1.08% |
MAD | 6.27% | 6.70% |
RMSD | 10.49% | 11.24% |
Table 13: PVOUT accuracy statistics comparing hourly and 15-minute values (day-ahead)
PVOUT, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | 1.08% | 1.17% | 0.24% | 1.51% | 1.54% | 1.95% |
MAD | 6.27% | 7.22% | 5.13% | 6.40% | 6.75% | 7.91% |
RMSD | 10.49% | 11.21% | 9.02% | 10.77% | 11.23% | 12.88% |
Table 14: PVOUT average bias, MAD, and RMSD variability by climate zone for hourly resolution (day-ahead)
PVOUT, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
Bias | 1.08% | 1.17% | 0.24% | 1.51% | 1.54% | 1.95% |
MAD | 6.70% | 7.73% | 5.51% | 6.82% | 7.15% | 8.51% |
RMSD | 11.24% | 12.02% | 9.78% | 11.50% | 11.92% | 13.88% |
Table 15: PVOUT average bias, MAD, and RMSD variability by climate zone for 15-minute resolution (day-ahead)
PVOUT, hourly | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
MAD | 6.27% | 7.22% | 5.13% | 6.40% | 6.75% | 7.91% |
Min MAD | 1.82% | 5.15% | 1.82% | 4.15% | 5.32% | 7.39% |
Max MAD | 9.69% | 9.69% | 8.19% | 8.53% | 8.37% | 8.43% |
Table 16: PVOUT MAD variability (average, min, max) by climate zone for hourly resolution (day-ahead)
PVOUT, 15min | Climate | |||||
|---|---|---|---|---|---|---|
All | Tropical | Arid | Temperate | Cold | Polar | |
Number of validation sites | 153 | 33 | 44 | 55 | 19 | 2 |
MAD | 6.70% | 7.73% | 5.51% | 6.82% | 7.15% | 8.51% |
Min MAD | 1.97% | 5.55% | 1.97% | 4.39% | 5.57% | 7.84% |
Max MAD | 10.29% | 10.29% | 8.76% | 8.93% | 8.92% | 9.19% |
Table 17: PVOUT MAD variability (average, min, max) by climate zone for 15-minute resolution (day-ahead)
Key findings
As expected, accuracy is slightly lower than for “hour-ahead” forecasts, but spatial and climate patterns remain consistent: arid regions show the best performance, while tropical and polar regions have the lowest. Tables with average, minimum, and maximum MAD help manage expectations by capturing both typical performance and variability across sites and climates.
The map below displays accuracy patterns for “day-ahead” PVOUT forecasts (hourly data). Clicking on a site provides detailed metadata and forecast accuracy statistics.
Map 4: Global accuracy patterns for Day-Ahead PVOUT predictions.
Conclusions
Climate & predictability: Local atmospheric conditions significantly influence predictability. Arid climates show the lowest errors due to stable radiation conditions, while tropical and polar regions suffer from complex cloud dynamics.
Resolution & horizon: Forecast accuracy deteriorates with increasing horizons and higher temporal resolutions (15-minute data), as hourly data benefits from an aggregation effect where short-term fluctuations are averaged out.
Comprehensive evaluation: Average metrics provide only a partial view. Because accuracy differs significantly based on local conditions, stakeholders should consider both average values and the range of errors (minimum to maximum MAD).
Additional useful information
Forecast accuracy enhancement
In operational environment, PVOUT forecasts can be improved by refining PV configuration parameters. Solargis uses an internally developed PV configuration estimator that analyzes customer-provided measured data to identify the most representative PV parameters (tilt, azimuth, installed capacity, etc.). This data-driven adjustment aligns forecasts with the actual performance of the PV plant, which generally leads to a noticeable improvement in accuracy.
Forecast Accuracy Evaluation report
For any location worldwide including validation sites, Solargis can generate a detailed Forecast Accuracy Evaluation report. These reports are designed to help customers understand the reliability of forecast data for PV plant operation and market participation. Key insights include:
Statistics for multiple horizons (short-term to several days ahead).
Monthly and seasonal analysis of performance variations.
Distribution statistics and percentile-based analysis of forecast deviations.
Exceedance statistics indicating the frequency of large forecast errors.
The example of the report can be found in the Annex.
Annex
Full hourly and 15-min datasets used for accuracy evaluation are uploaded here.
Detailed PV configurations of all validation sites are uploaded here.
Example of the Forecast Accuracy Evaluation report is uploaded here.
Accuracy metrics formulas below:
GHI accuracy formulas | PVOUT accuracy formulas |
|---|---|
Where:
: The -th forecast value. : The i-th historical satellite-derived value. : The number of data pairs (forecast and satellite-derived values) : Mathematical notation for the absolute value. : Mathematical notation for the sum or aggregation of values. Normalization Factors (The key difference):
For GHI: Values are divided by 1000.
For PVOUT: Values are divided by the installed capacity of the PV system.