"Imaging" the sky

Shoot a high-powered pulsed laser at it, and do a time resolved measurement of how much comes back, and when. The pulse won't get far into dense cloud.

Sending up a small fleet of dirigible balloons could give you more complete information, but would be even more expensive.

I guess you want it somewhat more detailed than this >

Sky temperature difference measured with Peltier can give you a rough all-sky estimate at night. In urban areas, there's enough street light to detect clouds from reflection.

Open access satellite imagery is also available, but low latency stuff has low resolution.

I've got an open souce Indi-Allsky camera, which counts the number of visible stars. Moon, Sun, automatic exposure and auroras will mess up that number in addition to clouds, though.

-- mikko

In places with enough street lamps you can measure the 2x mains frequency backscattered light as a proxy clear sky detector.

That or differential temperature sensors one facing skywards and one facing the ground. Real observatories these days tend to measure total water vapour along line of sight with a Lidar or similar kit.

Here is one example paper the links from it may provide inspiration:

As I mentioned to Mikko, that would only work for those areas that are thusly illuminated. E.g., I would get no data from the areas above the deserts north, south east and west of town. And, limited data from the "settled" areas as well (city ordinance discourages up-facing light emissions as they interfere with the local observatories)

But, can you isolate this *directionally*? I.e., are there clouds to my west? East? Is that storm coming up from the Gulf of California? Or, will it veer eastward?

What sort of cost for that instrumentation?

As I mentioned to Mikko (elsewhere this post), can I, instead, look in specific directions and check for expected objects/features? E.g., if I went looking for Betelgeuse at 12:47PM on Month, Day, Year and failed to see "it" (or, a light source that resembles what it looked like yesterday, the day before and the day before that), can I deduce that there is cloud cover along that sight line?

And, if I assume a particular ceiling height, infer where that cloud cover exists relative to my current location?

Then, check for it again (in a slightly different part of the sky owing to the earth's rotation) an hour later, etc.

And, do this for some number of such objects?

At night the clouds would appear warmer than space, so an infra-red scanning system would give you a picture of the clouds. I don't know how you would calibrate it, because the apparent cloud temperature would depend on air temperature and IR radiation from the earth.

But the cloud cover wouldn't instantaneously change from observations made during the preceding daylight/twilight periods. I.e., "THIS is what cloud cover looks like and THAT is what clear skies look like, in the Ir" (assuming both are present)

Think about how "you" can tell the condition of the skies without even looking upward: "It's not *bright* enough for 1PM" or "The light is the wrong 'color'"...

I wonder it it would also show the water vapour tracks left by jets?

John :-#)#

What do you actually want to do? I find meteosat images the most reliable for knowing about cloud cover and they image regularly.

It is strongly biassed towards the zenith. Clear sky looks like a wall at about -80C whilst cloud base temperatures are typically around freezing point (+/-10 degrees at least in habitable latitudes).

That and dewpoint is usually all we care about.

Dunno. We had one for total water vapour that monitored along the line of sight for some OH bond emission or other. It went crazy when we were observing a natural bright OH maser source with the telescope. W3(OH)

Easier to image with enough sensitivity to catch all stars down to magnitude 5 which with a modern ccd is about 10s exposure (less if you have a really fast lens). Your problem is going to be with artificial satellites and aircraft streaks crossing the field.

It is probably easier to use a whole sky camera and look for the dark patches where the clouds are. Sky is actually bright by comparison if you are in an astronomcally dark region. This is from a reputable UK maker ~$900- there are much cheaper clones around and with bigger CCDs (but slower optics but that probably doesn't matter for your use).

You will also have trouble with the moon/sun lighting up both clouds and sky and potentially blinding your imager.

What do you actually want to do? Knowing that I might be able to suggest a free non-hardware alternative (like the meteosat images).

In the UK there can be hourly changes in the cloud cover from clear to

I think you might have to measure daylight and darkness observations differently. The apparent temperature of the clouds will increase when they are in sunlight - but so will the apparent temperature of the space in between them, which is filled with illuminated dust particles.

Recently I haven't even needed to look out the window to guess at the weather. We had a period of several days of continuous thick black cloud cover when the ambient light level was so low I needed to switch on lights to see around the house. If you are only interested in the oveall cloud cover, rather than mapping the clouds, an ORP12 and a 1.5v battery would work in daytime.

I don't want to rely on any "external sources". That means you have to count on them being "available" for decades and update your "scrape" whenever they make a change to the data they present along with its form/format.

Ask yourself how *you* use your observations of the sky. Will I need an umbrella? Will I have to water the yard today? Will I need to water the rose bushes TWICE today? Will it be an overcast (i.e., warmer) overnight? Will I get peak performance from my solar array?

Note we don't expect 100% reliability in these assessments -- just a bit of useful bias to tweek our plans. Combined with a history of THIS site (which bears little resemblance to a site a mile away)

I can look out my windows and see storms on the other side(s) of town. I want to adjust my plans based on the likelihood that they will move towards me (vs. away). I know which way they move at different times of the year so know whether I can expect precipitation from them, if they will later obstruct sunlight falling on an array, etc. I know if I will have to take measures to protect the citrus trees (and other cold-sensitive plants and wildlife) from a "hard freeze" later in the overnight hours. If the PV array will be occluded "later", then I will be more miserly in my use of power *now* to avoid running out when it is less plentiful. etc.

I want a machine to be able to benefit from those same observations without

But I know where the stars are (will be). So, can ignore anything that is "moving too fast" or "isn't where it belongs".

That's why I think naively just looking upwards isn't a winning scenario.

With everything, I want to drive costs down by trading performance. "Can I see some particular object that SHOULD lie above the clouds in that particular part of the sky?"

I'm more worried about how daylight will affect the sensor.

See above. I'm going to play with repurposing a PTZ camera for visible and Ir observations as a starting point. Gimmick the optics so that it is narrowly focused and see how little light/dark it can perceive. And, see if I can modify the electronics to have a longer exposure.

Of course! But, I'm not just checking the skies once a day, etc. I will be watching continuously.

During daylight hours, it is possible to determine which parts of town are seeing "daylight", "cloudy", "precipitation" and virga. My concern isn't with "those areas" but, rather, whether or not the "sky" will shift to bring those conditions to *me*.

But, during daylight, I can distinguish between blue skies and white/grey clouds. I don't have to use the same wavelengths for all my observations.

I can look at the "ground" to determine the *type* of cloud cover and whether it is transitional or not. But, that won't tell me what's on either side of my bit of sky. It won't help me *plan*.

Last Sunday I had parked the van on a local hill and a friend and I were doing exactly what you seem to want to automate. The first problem was the wind direction, which appeared to be different at different heights

- so different cloud layers were moving in different directions.

Then we noticed that there appeared to be a large patch of blue sky appearing but there was no corresponding patch of sunlight on the ground beneath it. We worked out that this was because we could see blue sky through a diagonal gap between two layers - but in the direction of the sunlight, the layers overlapped. This was confirmed by the tops of some of the clouds (presumably in the lower layer) being illuminated by sunshine.

We came to the conclusion that weather was extremely difficult to predict from observations like that, even on a very short-term basis.

But weather forecasters have to (try) to be accurate. All I need to do is improve the odds of my making a "better decision" than I would in the absence of any such data.

If you *lived* on that hill, do you think your opinion might have changed?

It is largely overcast, today. Yet, not over my location. That suggests a higher probability that I would have less output from a solar array than on a "clear skies" day -- even though the sky above me is currently unobstructed.

I.e., using all available solar output (instead of banking some of it) has a greater chance of leading to a problem later in the day. OTOH, I suspect it will be a fair bit warmer, overnight, than it would have, otherwise.

I suspect folks who live "off grid" (relatively common in parts of AZ) make such calculations all the time. Just as folks living off well water likely make note of long term weather patterns to determine the likelihood that they may run out of water or need to have their well redrilled.

Not really, the winds and clouds would be different again the next day. A few things would be obvious, like an approaching squall if the wind was steady from one direction, but there are days when the wind direction at ground level varies throughout the day and approaching weather patterns appear to melt away before they arrive..

The UK is on a turbulent boundary between a marine envirnment with mild moist air to the North and a continental environment with dry air and settled temperature extremes to the South. Anticyclones progress from the South-west to the North-east, bringing swings in wind direction which differ according to whether they pass to the North or the South of the observer.

If the boundary shifts Northwards we get day after day of settled weather and heatwaves in Summer because the anticyclones bring air from the South. If it shifts Southwards, in Winter we get prolonged freeze-ups as they bring air from the North or Northeast.

Living on a hill would allow cloud patterns to be observed at a greater distance but would increase the error in predicting the direction they would travel and the actual weather they would bring.

Heat might be more of a problem. I'd expect the sun to fry a line/band into your sensor of at best noisy and at worst damaged pixels.

If you can find an archive of QUIAG still online they have figured out which cheap webcams can be subverted to do decent long exposures. The Sony CCD chips are generally the best of the bunch. Best I can find now:

If you choose your target stars carefully as in brighter than 3rd magnitude there are about 240 of them in the sky - enough to give you a reasonable coverage of the sky. Shouldn't be hard to get that deep in 4 or 5 seconds even on a mediocre camera (no chance at video rates).

Although you can probably get away with it orange-deep red.

A combination of a hot mirror filter (against hot IR) and a low pass wratten 29 or 25 filter will make blue skies pretty much black and still leave clouds looking bright red. Contrast helps AI interpretation.

Monochrome cameras tend to be cheaper than full colour. Beware that not all colour cameras react well to aggressive optical filters in front of them - Bayer demosaicing can go crazy faced with such weird data.

Weather radar. of course the pre-supposes a functional government or a large capital investment, so it may not work for everyone.

Yes, that should work. But just by having constant or controlled exposure, you can pretty easily determine if there are 'some dots' visible in a given sky segment. This would get rid of alignment issues and should give good enough result for most purposes.

-- mikko