Hello. I have noticed some issues with the CAMS radiation time series data (CAMS solar radiation time-series). When querying all-sky data for some European locations, the GHI variable has a strange/incorrect profile for many days in the last months, in which the morning hours have a smooth curve and the afternoon hours seem to have too much radiation squeezed in. The issue seems to stem from incorrect/incomplete BHI and DHI components (maybe also BNI), and has implications for solar forecasting and grid integration which is the primary purpose of this dataset.
Have any users noticed this before? And is this simply due to incomplete/noisy satellite data (i.e., annoying but to be expected occasionally)? or something more fundamental? More details below.
Example request (ADS manual request ID a8c98b8d-13ae-4583-9ee8-54fae9f550c1):
dataset = "cams-solar-radiation-timeseries"
request = {
"sky_type": "observed_cloud",
"location": {"longitude": 12.2, "latitude": 53.3},
"altitude": ["-999."],
"date": ["2024-10-27/2024-11-06"],
"time_step": "1hour",
"time_reference": "universal_time",
"format": "csv"
}
client = cdsapi.Client()
client.retrieve(dataset, request).download()
Example of day with expected profile:
Example of day with incorrect profile:
1 Like
Hi Tom. Thanks for this. I’ll see if I can get an expert to take a look and respond to you.
Luke.
Hi Tom. This is the reply from our expert:
I had a quick look at this situation. On 29.10. at 10 UTC we have a thick low cloud over the area of interest in North-Western Germany.
So, yes, direct radiation is expectedly zero in such a cloud condition. So, columns BNI and BHI are zero.
Furthermore, we have a large optical thickness and low sun conditions in winter. This reduced the available global radiation to very small values. The GHI values are very low, indeed, but this is ok in winter time under thick clouds.
Below are our pyranometer ground based observations taken by our own institute’s roof top in Oldenburg on 29 October 2024. We have roughly the same cloud conditions on that day here in Oldenburg (as you can see in the satellite images, Oldenburg is on a similar latitude as the point of interest, only in North-West instead of North-East in Germany). Therefore we can at least compare the values roughly with the location of interest. Our observation was also well below 50 W/m2 before 9 UTC and between 50 and 100 W/ms until 13 UTC.
So, for me this looks all fine. Basically it is a dark November day. And yes, street lights turn on if GHI is under 50 W/m2, it is really dark at the moment in Northern Germany if it is cloudy.
On the 2.11. again, we have a cleared up situation in North-East Germany, and then the radiation is back, as the user reports.
Hi Luke. Thanks for the detailed reply… the comparison with ground-based observations is useful, and the weather interpretation makes sense with the observational data. However we are still unsure about some aspects of the profile of the CAMS data. I include some plots here that illustrate the concern.
In the images that follow we use CAMS radiation time-series with 1 minute resolution. Note that units are different from the ground-based irradiance plot shown above, but we can ignore the scale and focus on the profile
In the CAMS data set, the GHI (all-sky) profile on cloudy days is often a very smooth curve in the morning hours (green arrow):
Zoom into single days:
The profile generally shows good agreement with ground-based obs at Oldenburg, but the ground-based GHI does not have the smooth profile in the morning hours (which is to be expected-- real-life noise-- it is the CAMS data that we are not sure about).
This behaviour is becoming more pronounced in November: the morning hours often have a very smooth curve and the afternoon hours seem to have too much radiation “squeezed in” to too few hours. The plot below shows 4 days from the last week:
In summary, the cloudy conditions → zero direct radiation → low GHI situation makes perfect sense in both ground-based and CAMS data, but it is the often-observed smoothness of the morning profile in the CAMS data that is puzzling. Of course CAMS is a different product to the ground based obs (we are comparing apples and pears, similar but different), so we do not expect perfect agreement, but the CAMS profile continues to concern us.
Could there be something in the generation of the CAMS datasets that leads to this smooth profile? It may be that is an artefact of some modelling/processing step in the generation of the variables from the satellite data, but it seems way too regular and smooth for comfort.
We will look into some other ground-based observations for a deeper analysis and share soon, but if you have anything more information on the CAMS dataset itself then I would be glad to discuss further.
Thanks again for a thorough response 
Tom
Hi Tom. I’ve asked our expert to take a look at the points you’ve raised.
1 Like
Hi Tom,
we had already a short discussion with some of the CAMS Radiation Service developers. We have no quick answer ready, we need to investigate a bit further and look into all used input data. I have some ideas what it could be, but this needs assessment and some coding/checks.
Thank you for the moment to make your point clearer to us.
Best regards
Marion Schroedter-Homscheidt, CAMS Radiation Service Manager
