Problems in HelioClim-3
|Last update: Oct. 2016|
The glitter is the reflection of the sun on the Earth surface in the direction of the satellite, as depicted in Fig. 1:
This phenomenon is observed over oceans, clouds, but also over icy and sandy surfaces.
As Heliosat-2 (the model used to derived the HelioClim-3 radiation values from the Meteosat images) associates the quantity of white observed in the image with the quantity of clouds in the atmosphere, the model fails in correctly estimating the HelioClim-3 values by returning erroneously very low radiation values for these locations.
Fig. 1: MSG composite image, 2006-03-29 04:00 UT, © Eumetsat
Impact on the HelioClim-3 values: since 2004, a glitter correction has been set that correctly treats watered areas. Unfortunatly, all the other glitter cases have not been correctly.
For instance, Fig. 2 illustrates the impact of glitter over clouds on the HelioClim-3 data compared with in-situ measurements (pyranometer). This 2-D histogram, which represents the hourly HelioClim-3 data versus the corresponding hourly in-situ measurements for a site located in Brazil after Quality Check for Apr. 2013 to Dec. 2014. The glitter effect affects the data for low sun elevation angles in the evenings from December to February, which results in a cloud of wrong points close to the x-axis (black oval).
Bug between 2009 and 2011
| In 2009: a bug has been introduced in the glitter correction over watered areas. The following regions have been erroneously set to 0 depending on the instant in the year (Fig. 3): || |
Fig. 3: Between 2009 and 2011
Fig. 4: Since 2011
Consequences on the data: users reported erroneously low values along sea shores, and in particular in Agadir (Fig. 5) and in Namibia (Fig. 6). Indeed, the black rectangle was also affecting the Meteosat pixels along sea shores.
If you retrieve a time series in such areas, you will receive low values between 28 Jan. 2009 until April 2011!
Fig. 5: HelioClim-3 values for Agadir,
Fig. 6: HelioClim-3 values for Namibia,
In Heliosat-2, each HC3 radiation pixel value is computed using the "proportion of cloud" or "proportion of white" observed in this pixel in a combination of two channels located in the visible part of the spectra. This "proportion" is computed as an attenuation coefficient of the clear-sky value. When the snow covers the ground under a sunny weather, Heliosat-2 is confusing the snow on the ground with a large cloud coverage. The HelioClim-3 radiation is thus drastically underestimated in snowy areas (the consequence is that the electrical production of a solar power plant covered with the snow and the corresponding HelioClim-3 values are coherent. But the counterpart is that you will not be able so far to rely on the HelioClim-3 to know if the panels need to be cleaned of from snow).
The first step to solve this problem is to flag the HelioClim-3 data when there is snow on the ground to warn the user that he might receives underestimated values. To produce this flag, an alternative source should be exploited, such as the product available over the northern hemisphere from the National Snow and Ice Data Center (NSIDC). This institute provides 1-km (24576*24576 px, Dec. 2014 onwards), 4-km (6144*6144 px, 23 Feb. 2014 onwards) and 24-km products that could be exploited to complete HelioClim-3 delivery.
To definitively solve this issue, other Meteosat channels (infra-red and water vapor) must be exploited.
This snow issue has been regularly reported by our users. Please read the two following conversations:
I contemplate purchasing an annual subscription to one of our services. However, I compared monthly irradiation values over an inclined (48°) plane from PVGIS (Classic PVGIS or Climate-SAF PVGIS) and from SoDa (HelioClim-3) for the year 2005 for a site in southern Germany (48°, 12°), the irradiation values are strongly lower in your data (half of what is retrieved in PVGIS whatever the database). What is the origin of this discrepancy? Alexey Mineev at REN Solar AS, Norway
Our answer => The PVGIS irradiation values are monthly irradiation values averaged over the years available: over Europe, from 1981 to 1990 for Classic-PVGIS and from 1985 to 2005 and from June 2006 to May 2010 for Climate-CMSAF PVGIS. The data he used is from a specific year 2005 in the database HC3, thus the comparison between averaged values and not-averaged ones should differ but be of the same magnitude. However the values were (rounded):
- monthly HC3 (2005, in kWh/m2): 21, 21, 73, 135, 156, 164, 147, 131, 124, 100, 20, 11
- Climate-CMSAF PVGIS (average): 42, 65, 114, 152, 153, 161, 156, 149, 120, 88, 49, 37
The values are close for almost half a year, but the values are strongly underestimated in HelioClim-3 during the winter period. After a verification, in 2005 the snow cover has been reported by Meteorological Ground stations which confirmed our idea. The answer of this Customer was: The difference during the winter time looks discouraging to us and this is the area of our further interest to look into.
I am a student of renewable energies and currently I write my bachelor thesis. In the bachelor thesis I evaluate yield prognosis for pv systems and compare predicted yield values with those measured during the operation of the pv systems. For this purpose I need irradiation data for the last years. Would your data suit my needs? J. W., Berlin
Our answer => The document in pdf format accessible here describes exactly the project of this student. More precisely, the influence of snow shading of pv systems is often underestimated. His purpose was to investigate the yield losses through snow shading of pv systems. After the explanations relative to the limits of the HelioSat-2 model, here was my response: As a consequence, for you to be able to carry out your research in good conditions and even if it means that we lose a potential customer, I do not recommend you to use the HelioClim data in order to determine the yield losses through snow shading.
On 20 March 2015 (http://astro.ukho.gov.uk/eclipse/0112015/), an eclipse has generated a shadow on the Meteosat Second Generation images. The consequence is that the HelioClim-3 irradiation values will be strongly overestimated during this lapse of time. Please, pay attention to DISCARD the corresponding slots in the retrieved time series.
Heliosat-2 requises a mask for the distinction offshore/inshore. This mask is derived from a Digital Elevation Model. The problem is that the spatial resolution of the DEM has evolved with the time, from 10 km with Terrain Base to 90 m with the current SRTM base. Twice in the past, the mask has been modified to exploit the most updated information in terms of DEM. This leads to inhomogeneous radiation information along the coasts. Indeed, it might happen that in the past, a given pixel has been stated as located off-shore, then inland, and then again off-shore. This problem is still pending in our base, resulting in a few additional percents of errors along the sea shores.
When the sun is low above the horizon, Heliosat faces a few difficulties in correctly assessing the radiation at ground level. Indeed, the path of the sun rays is longer which engenders errors. A threshold has been consequently applied in Heliosat-2 below which the HelioClim-3 radiation values are not computed. This threshold has been set to 12°.