Bogus UVC explained

Time to set the record straight?

Sorry if it this all sounds a bit technical. But it’s not too hard really, and I think you’ll it very interesting if you can wade your way through it. With apologies to my friends at Gigahertz Optics. It was only by using data from one of their instruments that I was able illustrate the cause of the bogus UVC problem ...

You may recall me writing last year about some spuriously high ‘measurements’ of UVC. I challenged the authors, and have finally heard back from them.

Recall that I’d suggested they repeat their measurements in sunlight, but with a piece of glass covering the entrance port to their instrument. Glass transmits nothing at wavelengths below 300 nm, so any ‘signal’ seen at those shorter wavelengths must be due to stray light. Sadly, they’ve advised me that they’re unable to carry out that definitive experiment because their instrument expert is no longer with us. I just hope the realisation of his error after my post wasn’t a contributing factor …. I’m sad too that there’s nobody else there who can do the work.

I thought that perhaps I could try doing it for them using a diode array spectrograph manufactured by Gigahertz Optics in Germany that my colleagues at NIWA Lauder had purchased in 2020. It’s optics - details can be found here - are similar to those used by these groups who’ve claimed to measure high levels of UVC in sunlight.

When I checked, I found our instrument is currently out of action, so despaired that we might be out of luck. But I discovered - after a bit of badgering - that a couple of my colleagues (Alex Geddes and John Robinson) had serendipitously made the crucial measurements shortly after they’d purchased it. They’d used it to measure the transmission of a couple of glass cover-plates that were being used in on the facia of ovens. The clients were concerned about damage from UV transmitted by the glass.

The samples weren’t ideal. Neither was transparent. Both were darkly pigmented. But one was cloudier than the other, so my colleagues labelled them as ‘clear’ and ‘smoky’. Even at visible wavelengths, they transmitted only ~25 percent of the radiation (compared with common window glass that transmits more than 90 percent).

A sequence of measurements was made at Lauder with the Gigahertz spectrograph on the afternoon of 13 November 2020. The conditions weren’t optimal either. Ideally, the measurements should be made near noon in the middle of summer when the sun is as close as possible to being directly overhead. But these were made an hour-and-a-half after solar noon in spring when the sun elevation angle was only 50 degrees (i.e., solar zenith angle, SZA = 40°). Ozone levels were also relatively high during that spring period, so the UVI was only 6, less than half the peak values seen at the site.

But it turned out the samples and the observation conditions were good enough for the job in hand, as you’ll see …

Spectra of sunlight through two samples of glass were bracketed by spectra without any glass in the path. Having the measurements carried out at Lauder was crucial because we could then also compare the results with those measured at the same time using our state-of-the-art ‘UV4’ double monochromator: an instrument that’s especially designed to minimise stray light problems caused by the steep drop-off in irradiance at short wavelengths. It complies to the demanding standards set by the international Network for the Detection of Atmospheric Composition Change (NDACC), so data from it therefore represents the ‘truth’.

The Gigahertz spectra (black lines) are compared with this UV4 spectrum (blue line) below. I’ve also included the spectrum of sunlight measured using a satellite-borne instrument outside Earth’s atmosphere (i.e., the ‘extra-terrestrial’ sunlight, as shown by the red curve). The green curve and (mainly hidden underlying) yellow curve show the spectra of sunlight measured through the two glass samples. These last two include upward corrections (of about a factor of 4) to bring the peak values into line with the black and blue curves at the longest wavelengths available.

With those corrections, the spectra are indistinguishable from each other at wavelengths greater than 400 nm. The two black curves are virtually indistinguishable at all wavelengths, showing that any changes over the ten-minute period were small. Similarly, it’s hard to distinguish any differences between these and the blue curve (though to get such good agreement I must admit that I had to apply a downward correction factor of 0.88 to the Gigahertz data to bring into line with the high quality UV4 data). The shortest wavelength measured by the UV4 instrument is at 285 nm. At Earth’s surface there’s no need to measure at shorter wavelengths because there’s never anything measurable below about 290 nm (despite those claims to the contrary). Data are, however, available from the general-purpose Gigahertz instrument right down to 200 nm. Values are close to zero for all wavelengths less than 300 nm, but, if you look very carefully, you can see some slightly higher values near the shortest wavelengths. Tellingly, that the flare-up near 200 nm is similar with-and-without the glass samples in place, even though glass doesn’t transmit at wavelengths below 300 nm.

To see more closely what’s going on at shorter UVB and UVC wavelengths, I re-plotted exactly the same data using a logarithmic y-axis that ranges over 6 orders of magnitude (below). Each tick mark on the axis represents a factor-of-ten change.

Now you can see that while the Gigahertz spectrograph works fine in sunlight at wavelengths more than 300 nm, it rapidly diverges from the truth (defined by the blue line) at shorter wavelengths. While data from the UV4 spectrometer continues to plummet down by several orders of magnitude below that wavelength, the Gigahertz data begins to increase again and continues to increase all the way back to 200 nm. In fact, by 200 nm all its spectra are higher than the red curve, implying values that are impossibly more than the sunlight arriving outside Earth’s atmosphere. The true irradiance at Earth’s surface is essentially zero at all wavelengths less than 290 nm. By comparing with the UV4 data it’s clear that over the wavelength range 285 to 290 nm, the Gigahertz signal is at least 1000 times too large. And it gets progressively worse at shorter wavelengths where atmospheric absorption becomes larger but the Gigahertz signal increases.

While this instrument isn’t identical to that used by Herndon et al, the behaviour at wavelengths shorter than 300 nm is remarkably similar, as you can check out here. The numbers in our experiment may be slightly more believable than theirs because they exceed the extraterrestrial value only for wavelengths less than 210 nm, whereas theirs exceed it for all wavelengths shorter than 220 nm (or all wavelengths less than 250 nm for the earlier d’Antoni data shown in their same plot). But that may be due to the lower sun elevation angles in our case. In any case, all are about 40 orders of magnitude too high at that wavelength.

Rather than disappearing as it should, the spurious UVC signal is even larger for the spectra with the glass samples in the path! That (unexpected!!) increase may be because of the changed geometry of the light path caused by additional diffusion as it passes through the glass. The wider range of angles entering the instrument means that the stray light contribution will be worse. This argument could also explain why the performance of the instrument shown at Gigahertz web site seems so much better. The fraction of stray light inside would be minimised for overhead sun, especially if the field of view is limited to the direct sun only, as appears to be the case there from their lower-than-expected values at longer wavelengths. The significant signal below 300 nm tells me that the sample spectrum they show was taken at a high sun elevation. But the irradiance never exceeds 0.7 Wm^-2 whereas in the spectra shown above for Lauder, corresponding values were twice as large. My guess is that the field of view in the web page plot was limited to less than the full solar disc.

The ratios are shown below. The black line is just to show that over the ten-minute period of observations, any changes in the radiation were small, especially at wavelengths longer than 320 nm. You can see that the transmission of both glass samples starts to decrease at wavelengths below about 400 nm, reducing to 50 percent transmission by 360 nm and down to less than 10 percent at wavelengths shorter than 340 nm. In reality the transmission reduces to zero by around 330 nm, but it never gets that low here because there’s a bit of stray light that leaks around the edges. There’s also too much stray light bouncing around inside the instrument, as you can see at wavelengths less than 300 nm where things go horribly wrong. There, the transmission appears to increase to become impossibly more than unity. Materials that truly transmit at UV wavelengths are rare and I can tell you they’re much more expensive than the glass samples tested here!

At wavelengths shorter than 300 nm this type of instrument is simply not good enough for measuring UV in sunlight. The bottom line is that while simple diode array spectrographs may work OK down for wavelengths around 300 nm, you shouldn’t believe their results at shorter wavelengths.

Finally, our spurious UVC result is consistent with those from other diode array spectrographs as reported by Herndon et al. But unlike them, we won’t be claiming it as a valid measurement of sunlight transmitted through the atmosphere! 😊

Hopefully there’s still somebody there in that group who can write the retraction.

For my colleagues at Gigahertz Optics, I’d just like to add all this shouldn’t reflect too badly on their impressive instruments. The data on their web site shows that they work well for the wavelength range intended, and a low-wavelength limit around 300 nm is sufficient for many purposes (including evaluating skin damage).

Note added 12 Dec 2024. I should also add (after discussions with Ralf Zuber at Gigahertz) that with care, one do a lot better with their instrument. Firstly, it includes additional filters which can be sequentially switched in to block stray light. Secondly, they have developed sophisticated mathematical procedures to minimise the effects of any remaining stray light. But even with these improvements the irradiance from sunlight at wavelengths below 280 nm is always far below their detection threshold. I’ll write more about this in a later post.