Cumulus Support

This was my recording today, which has been almost uninterrupted clear skies and bright sunshine from dawn to dusk.

The depressed values in the early part of the day are the result of shadowing from trees and at some point when I've worked out how I'll raise the solar sensor by w few more feet with the wind sensors.

Likewise the depressed values at the evening end of the plot are shadowing by trees and a neighbour's house. This will become less of an issue until summer solstice and then steadily worsen again towards winter !

I'm interested in how the plot exceeds the Max for quite a portion of the day. I've currently recorded 10.6 sunshine hours but presume the Sun Threshold or TransFactor or both might need tweaking? The plot matches pretty closely with that from my PV array and the "expected insolation" value.

I can't remember why I chose the R-S model, other than perhaps it was the easiest. If the Bird model is easy to do, I could consider offering it as an alternative in MX.

Hold your hat Steve, some different data from today on the way....

The Bird model is by far the most complex.

Needed my hat today, but left it at home. Top of my head got a bit sunburnt (close cropped hair). We're going to up to Shetland to try to escape this heat

Getting breezy here now as well.

Anyway, I plotted out today's data against the Ryan-Stolzenbach, Bird, and Bras models, and the winner was.... Bras! About the simplest model.

Bird requires lots of parameters setting, for most of which I could not find a source, especially not a current "today's data" source.

I have found the R-S curve always under estimates the values I actually see at the beginning and end of the day, it stays zero a while after sunrise and back to zero before sunset when there is patently still a fair amount of light around!

The Bras model goes non-zero at sun rise and set, so is closer.

Anyway first graph, all three models and my data from today - "Me" - the graphs are quite large but need to be to see any detail! Note that the sky clarity is better in the morning, I guess particulates increase in the day with surface winds building up, as does moisture with evaporation. Next is a plot of the percentage difference between each model and the data.
So for my sensor, for today's data, the Bras model seems to provide the best fit. Whether the Bird model could be made to fit better by tweaking parameters - probably. But I tried and that was the best I could do.

mcrossley wrote:
I find I have to use a higher transmission value in winter than summer - possibly due to the additional water content in warmer air?

I do the same, as there is a significant mis-match in the theoretical curve vs reality. I'm not sure that it is due to absolute humidity though, as we can have a high DP in winter and low DP in summer sometimes, and the curve doesn't appear to deviate as the DP changes during the day.

Hi,

mikechristelow wrote:The plot matches pretty closely with that from my PV array and the "expected insolation" value.

I suspect that mike's sensor is simply over-reading somewhat; the Fine Offset specification is only +/-15% (and I do wonder if they even achieve that consistently in practice). The apparent morning/afternoon asymmetry might be due to a tilted sensor or incorrect longitude setting for the site? I'm surprised that you can get any useful correlation with PV panels, which are presumably on a sloping roof? Note that Davis actually fit a spirit level to their solar sensors.

Maybe it's a "London" effect, but I can't get curves anywhere near as impressive as any of those in this thread. In particular the insolation appears to "crush" at high sun elevations, which doesn't make much practical sense. That's on two independent systems, but I must admit that one is a FO and the other "home brewed".

My main "complaint" about the R-S / Cumulus theoretical curve is the values at low sun elevations seem far too low; Zero ( <1 watt/m2) about half an hour after sunrise / before sunset seems rather "unlikely". Maybe that also relates to the "Summer / Winter" issue? Of course it's a very difficult calculation/calibration because the "direct" sunlight is heavily reduced by the cosine response of the sensor, so the diffuse light from the sky predominates. Also, maybe it's a "Latitude" thing, which seems to affect us here in the UK much more than examples from Sacramento or Portugal?

Cheers, Alan.

There been some nice curves in this thread recently concentrating on the match between some models and actual observed intraday variation. However with a few years experience here in Albuquerque, I think the more important model tracking error for Cumulus is seasonal rather than intraday. For those who do not wish to modify their trans factor multiple times per year, I think it is wise to choose a trans factor that gives a good match near the equinox, as this probably will come close to giving the minimum year-round error from this particular source, which is substantial.

It would be unwise to choose a new model as being better matched to ground truth without considering both the intraday and the time of year issues.

Odd you get a squashed response Alan, my home brew sensor just uses a simple photo diode and mimicks the Davis output voltages. You'd think the London crud would affect all the readings - more crud the lower the sun gets.

After observation I found I needed to concentrate on the first few hours in the morning and evening otherwise Cumulus would report sun when there wasn’t any. I also discovered the Met Office wasn’t recording sun for the first 25 to 30 minutes first and last thing in the day; which confirmed something I has seen in read in another forum about ignoring the first and last ½ hour of the day for sunshine. I then found a Met Office weather station nearby producing sunshine readings, with a similar elevation and locality i.e. located on another outcrop of the Chiltern Hills 7 ½ NE of my location. These readings were sometimes way different to the supposed East Anglia region I’m meant to be in.

Armed with this information I have spent the last 5 years fine tuning, and I now run the w/m2 readings through a spreadsheet and then amend the dayfile. I'm presently fine tuning the first 5 months where sunshine data was sparse at times. I have a formula that uses transmission factor 90% and sun threshold 57%, together with a minimum VP2 value and a minimum transmission value. Additionally from observation I have an additional formula that deals with obstructions.

Below are the comparisons of my sun verses Rothamstead and East Anglia for 2017, 2016 and 2015 WIP revision. You will also see how my location affects readings as in Mar and April 2017 and Jan 2016. Though it also affects Hemel Hempstead as well; in winter we can have 50mm of snow on the North side but none on the South side and yesterday they had some heavy showers whilst we had sunshine.