Shading, soiling, tracker malfunction, a slightly misaligned sensor: problems in ground measurements are easy to miss, and they quietly shift every result built on the data. In fact, all the ground-measurement datasets we analyzed had quality issues. When the input is flawed, no amount of downstream analysis can fix it, and the uncertainty carries into site adaptation, performance evaluation, forecasting, and due diligence.

In this webinar, you’ll learn which problems to look for in ground measurements, why Excel and custom scripts often miss them, and how comparing measurements against a solar resource model or a nearby sensor reveals issues a simple time series hides.

Watch the recording to see the workflow live in Solargis Analyst, the tool our own analysts use to work with ground-measured data.

  • Duration: 40 mins
  • Related service: Solargis Analyst
  • Recorded: October 27th, 2026

Questions from attendees#

Full question: In these statistics (100% of all datasets have issues), was Solargis involved in the project on the ground, i.e. did you know the exact topography of the installation, or how the sensor tilt really was in real life? Often the theoretical tilt of PV modules from a CAD file or design is not consistent with what is installed on the ground, so many pyranometer values are flagged as faulty, when in fact they are perfectly aligned with the PV modules. So you can't just throw away these data as false due to theoretical considerations or clear sky models.

Answer: Thank you, we fully agree that the as-built geometry often differs from the design, and that this must be respected in QC.

Two clarifications to the statistic:

1) "100% of datasets have issues" means that every dataset had at least one QC finding of any type – e.g. values below/above physical limits, shading, dew/frost, soiling, static (stuck) values, timestamp shifts, logger outages, misalignment. A single occurrence was enough. It does not mean that the datasets were unusable or that large parts of them were removed.

2) Solargis is not involved in field quality checks, so we have no independent knowledge of the as-built geometry. We can only rely on the documentation provided to us. The customer is therefore responsible for providing a verified configuration of the PV system and of the solar instruments (sensor tilt and azimuth, tracker configuration, horizon, site photos, maintenance logs).

We take these documents as the ground truth and carry out automatic and manual (expert) tests. These tests reveal issues both in the measured data and in the provided meteo station and PV configuration.

Misalignment is flagged when the data show the typical signature (e.g. systematic morning/afternoon asymmetry on clear-sky days, consistent with other sensors and the model) and the deviation from the declared configuration exceeds a threshold.

If a pyranometer is aligned with the as-built modules and this geometry is declared, its data pass. If the declared metadata are wrong (e.g. sensor declared horizontal but installed tilted), the flag is correct with respect to the declaration.

In such a case, the finding points to an error in the provided configuration rather than in the sensor. We then clarify the as-built geometry with the customer and re-evaluate the data. 

Solargis Analyst costs 1,400 EUR per year. It’s desktop software, so one licence covers one user on one computer. If you need more seats, please reach out to us and we can discuss further what would work for your team.

Yes. Analyst can work with any pyranometer model. If an instrument is not in the list of known instruments, you can define it yourself in the settings; this definition is stored only in your application. If you send us feedback about the missing instrument, we will add it to Analyst so it becomes available to all users in the next release.

Full question: If you do not have people on the ground you need to be careful with how you treat data. Indeed the point of reliable reference cannot be treated only theoretically, since we always have unique things on the ground that may change the results. 

How can you deal with this issue, especially things like tilt variation across a plant?

Answer: 

QC of tilted (POA/GTI) sensors depends strongly on the metadata provided, which is why we treat each station separately: for every meteo station we request its own sensor orientation and, for trackers, the full tracker configuration (tracker type, axis tilt and azimuth, rotation limits, backtracking, ground cover ratio). Tilt variation across the plant (e.g. terrain-following trackers) is handled this way – each sensor is checked against its own declared geometry, not a plant-wide design value.

If the declared configuration does not match reality (e.g. axis tilt declared as 0° while the trackers are tilted), deviations above the threshold are correctly flagged as misalignment relative to the declaration. In such cases the solution is to correct the metadata, not to accept the data blindly – the as-built values (ideally from inclinometers, tracker controller logs or site survey) are then used and the QC is repeated.

There is no fixed limit on the number of sensors – in our tests we processed a dataset with 70 pyranometers, only computer resources (CPU and RAM). 

Full question: Is there a standard way to clean the recorded data? All the plant operators are relying on different methodologies to clean the recordings. 

Answer: There is no single binding standard yet, which is why approaches differ between operators.

The commonly used references are:

  • BSRN quality-control recommendations (physically possible and extremely rare limits, comparison between components),
  • the IEA PVPS Task 16 / NREL "Best Practices Handbook for the Collection and Use of Solar Resource Data for Solar Energy Applications" (4rd edition, 2024), currently the most complete reference,
  • IEC 61724-1, which defines data-quality checks for PV plant monitoring (range checks, missing and stuck values, etc.).

In practice, good QC combines automatic tests (physical limits, consistency between GHI, DHI and DNI, comparison with clear-sky and satellite data, comparison between redundant sensors) with expert visual review, and documents every flag.

The Z-score can be a useful supplementary test when you have many comparable sensors (e.g. several POA sensors of the same orientation across a large plant), but we would not use it as the main method:

  • With few sensors it barely works: with 2 or 3 sensors the Z-score can mathematically never exceed about 1–1.4, so the usual thresholds of 2–3 never trigger.
  • With two sensors it cannot tell which one is wrong.
  • The mean and standard deviation are themselves distorted by the faulty sensor; robust statistics (median and MAD, "modified Z-score") are better.
  • Errors common to all sensors (shared soiling, the same logger or timestamp problem, same calibration error) are invisible to inter-sensor comparison, so an independent reference such as clear-sky and satellite data is still needed.
  • On large plants, cloud shadows cause real differences between sensors, so the test should be limited to stable (clear-sky) conditions or longer aggregation.

Full question: It's clear that there is a lot of value in having two Class-A pyranometers on a weather stations (for comparative purposes). Are there any other measurements that has value to duplicate? Like ambient temperature? Maybe something else?

Answer: Yes, redundancy is valuable beyond GHI, but quality comes before quantity. 

A redundant sensor helps only if it does not share the same failure: in practice, many problems come from the data logger (time shifts, gaps, wrong scaling), and sensors connected to one logger usually share them.

  • Ambient temperature and relative humidity: one instrument installed correctly according to WMO guidelines (proper height, radiation shield with ventilation, no influence of nearby objects such as inverters, buildings or hot surfaces) is better than two or more poorly installed ones. If a second sensor is added, it is most useful when placed at a slightly different location (a few tens of metres apart), ideally on a separate logger or channel. This helps to identify local effects and sensor faults (shield soiling, ventilation failure, drift).

 

  • Wind speed: ideally at two heights. The 10 m measurement is the standard for comparison with meteorological models and for site adaptation. A measurement close to module/tracker height is less relevant for resource assessment, but it carries important information for tracker operation (wind gusts, stow positions).

 

  • Instead of a pure duplicate, measuring a different component often adds more value: GHI together with DHI (and DNI) allows the closure test GHI = DHI + DNI·cos(zenith), one of the strongest QC checks available. A pyranometer next to a reference cell adds a sensor with different failure modes.

 

  • For GHI specifically, we see a QC benefit in adding a third sensor of a different type to the two Class A pyranometers: the two Class A instruments provide the reference, and the third one shows whether the behavior is similar. 

    A silicon photodiode pyranometer works well for this (e.g. the sensor used in rotating shadowband irradiometers, or a stand-alone unit). The photodiode sensor is less sensitive to soiling and dew and responds faster, but it is less accurate (spectral and cosine response, temperature dependence) and is usually not heated or ventilated. 

    Because the two sensor types react differently to the same event, comparing them helps us distinguish whether a deviation in the Class A data (including one affecting both Class A sensors) is caused by a sensor issue (e.g. soiling or dew on the dome) or by a real atmospheric phenomenon. It serves as a QC companion, not as a replacement for the Class A pyranometers.

Yes. Clear-sky irradiance is calculated from the position of the station (latitude, longitude and elevation) and the date and time.

Correct time zone and timestamp convention (start/end/centre of the interval) are therefore essential – a time shift is one of the most common issues we find. 

  • Plane-of-array (POA/GTI): yes – the clear-sky irradiance is transposed to the sensor plane using the declared tracker configuration.
  • Temperature: no, there is no physical "clear-sky" equivalent for air temperature. 

Full question: How accurate is the Solargis Analyst analysis theoretical/satellite data compared to class 1 pyranometers actual data? Was the analysis done for latest years of 2023-2025?  

Answer: The clear-sky values in Analyst are not an estimate of the actual irradiance – they represent theoretical cloudless conditions and are used as an envelope for plausibility checks (e.g. detecting time shifts, misalignment or values that are physically implausible). 

Comparing them with measurements as an accuracy metric would therefore not be meaningful, and they were not validated in that sense.

The Solargis satellite-based time series is validated against high-quality ground measurements; results for GHI and DNI are published in our knowledge base: https://kb.solargis.com/docs/validation-of-solar-irradiance-data

Solargis was also included in the independent IEA PVPS Task 16 "Worldwide Benchmark of Modelled Solar Irradiance Data" (2023).

The validation applies to the current model version. We do not have a public study limited to the years 2023–2025. 

No – the Analyst license covers the software. Solargis model time series for your site are a separate data product. Please fill out this form to get in touch with our sales team or write an email to contact@solargis.com and we will prepare an offer for you.

Automated cleaning has the potential to improve measurement reliability, especially at remote and dusty sites where manual cleaning cannot be frequent.

However, we do not have enough experience with such installations yet to say under which conditions it is an improvement.

Points we would look at:

  • the cleaning itself can create artefacts (spikes or drops during wiping/spraying, water droplets or streaks, residues from hard water, scratches on the dome),
  • every cleaning event should be logged with a timestamp, so that affected records can be flagged and the cleaning effect verified,
  • it does not replace regular inspection, levelling and recalibration.

For a measurement to be accepted as a bankable reference, the documented maintenance procedure and traceable calibration are what matters most.

We would be glad to evaluate data from such an installation if you have it.

Keep reading

Impact of Extreme Weather Conditions on PV Projects
Webinars

Impact of Extreme Weather Conditions on PV Projects

Our expert team gives actionable tips on how to avoid the negative surprises of extreme weather and shares some of our customers' best practices.

Solargis Analyst New Features and Improvements (version 1.4.2)
Webinars

Solargis Analyst New Features and Improvements (version 1.4.2)

The new version of Solargis Analyst (version 1.4.2) includes an updated database structure that is crucial for all upcoming new features and parameters.

How satellite model data help with storage applications
Webinars

How satellite model data help with storage applications

In this webinar, we will present how 1-minute granularity data can also help with optimal sizing of batteries and replacement strategy— even during the initial design phase.