As solar capacity grows, PV operators increasingly need to manage curtailment, negative electricity prices and rapidly changing grid conditions. Short-term forecasting can help, but Masaaki Hasegawa, Director at EKO Instruments, argues that forecast performance ultimately depends on the quality of the irradiance data behind it.
Capturing rapid changes in irradiance
Under changing cloud conditions, solar irradiance can fluctuate within seconds or even fractions of a second. These changes influence inverter loading, battery energy storage system (BESS) dispatch and compliance with grid ramp-rate requirements.
Traditional thermopile pyranometers can have response times of several seconds. According to Hasegawa, this may be sufficient for long-term resource assessment but can smooth out short-lived irradiance peaks and drops that matter for real-time plant operation.
Fast-response instruments can capture these changes with greater temporal resolution. EKO says its MS-80SH pyranometer provides a sub-second response, allowing control systems to detect changing irradiance sooner.
This difference can result in peaks being missed or smoothed by a slower-response pyranometer.
Potential operational benefits include:
Earlier detection of irradiance spikes and drops
More responsive BESS dispatch
Improved ramp-rate management
Better management of inverter clipping
These are particularly relevant as operators move from reactive towards more proactive plant control.
When apparently good data is wrong
Fast measurement alone does not guarantee reliable data. Sensors operate outdoors for long periods and can gradually develop measurement errors.
Hasegawa identifies several potential causes:
Sensor soiling
Calibration drift
Tilt misalignment
Partial shading
Incorrect time settings
Inappropriate averaging intervals
Unit conversion errors
The difficulty is that these problems may not cause an obvious failure. Instead, they can introduce small biases that remain hidden within apparently consistent datasets.
This matters because forecasting models learn from the data supplied to them. If training data contains systematic errors, machine learning models can reproduce those errors in future forecasts.
The potential financial effect is clear. The example using clean training data produces a forecast bias of +1.2 per cent and an RMSE of 18 W/m², while biased training data produces a forecast bias of -11.8 per cent and an RMSE of 97 W/m².
Data quality affects more than forecasting
Irradiance measurements are also used to assess PV performance through measures such as performance ratio and specific yield.
Poor reference data can make it difficult to determine whether a plant is genuinely underperforming or whether the apparent problem comes from the measurement system itself.
Hasegawa argues that this can affect:
Fault detection
Maintenance decisions
Performance assessment
Contractual discussions
Warranty claims
Data quality can therefore become an operational and financial issue rather than simply a measurement concern.
Moving towards continuous validation
ISO 9060 defines requirements for pyranometer performance, but there is no equivalent universal standard for validating irradiance data produced under real operating conditions.
Field data is influenced by local weather, installation quality, maintenance and data acquisition settings, making standardisation difficult.
One response is continuous automated validation. This can combine physical-limit checks, statistical analysis, comparisons between nearby sensors and benchmarking against independent satellite-derived irradiance data.
EKO’s approach combines irradiance measurements, model data, satellite data, industry standards and specialist knowledge within its EKO Q platform to produce a data quality report.
This is a company-specific solution, but it reflects a wider shift towards treating measurement data as something that must be verified rather than simply collected.
As solar plants become more dependent on forecasting, storage and automated control, reliable irradiance data becomes increasingly important. Better algorithms can improve predictions, but their value will remain limited if the measurements supporting them are inaccurate.
For Hasegawa, data integrity is therefore becoming part of the foundation of modern PV operation, influencing everything from real-time control to long-term financial decisions.
For the complete analysis, read the full insight from EKO Instruments on irradiance measurement, data validation and solar forecasting performance in PES Solar: https://pes.eu.com/exclusive-articles/why-data-quality-is-the-missing-link-in-solar-forecasting-performance