Longer exposure & lower ISO or shorter exposure & higher ISO - what gives better results when photographing stars?

Question

I've been dabbling with night landscapes and star photography with varying levels of success. I know I don't have the ideal lens (Canon 17-40mm f4 on a Canon 6D body) but I've seen some great shots done with the same gear. I also have a Canon 50mm f1.4 which is great but it's just not wide enough.

I usually shoot at f4, ISO 800-1600, Long Exposure Noise reduction on, 30 - 40 seconds. I find the results very noisy and the stars not bright enough. at 30 seconds stars are not bright enough and at 40 seconds they are already trailing. Here is one of my attempts.

I recently saw some photos (examples here and here) that are captured at shorter exposure times but much higher ISO (5000-6400)

I am thinking perhaps when I am close to 40 seconds the sensor is heating up more which is causing more noise? Specifically for photography stars, is shorter exposure time and higher ISO a better formula?

Answer 1

Noise is a fact of life when it comes to astrophotography, with the exception being stacked deep sky photos taken on a tracking mount (more in a moment).

Your photo is actually very low noise, in the grand scheme of wide field, single-frame astrophotography shots that I have seen...but it also lacks saturation. I think it really comes down to a matter of taste, but ultimately, one way or the other, you will get roughly the same amount of noise in your photographs regardless of the ISO setting. If you wish to achieve the same amount of saturation, you have to do one of two things. You are either going to need to use a higher ISO setting (ISO 3200, maybe even as high as 6400), or you are going to have to boost exposure in post. The vast majority of noise in astrophotography is from photon shot noise, so using a higher ISO is the same as a post-process exposure boost from a noise standpoint.

In your example photo, you have a wide-field, single-frame shot. Your limited to a single frame because of the foreground, unless you resort to more complex trickery where you take multiple frames, cut out the sky, and stack those frames to improve the saturation of the sky. Certainly possible...also a lot of work. Like you, I like astrophotography shots that include some of the landscape in the foreground, so it's worth trying some manual partial stacking to improve your SNR.

Heat is certainly a contributor to noise during long exposures. I am not sure that 40 seconds is long enough to produce so much heat that thermal noise becomes a more significant factor than photon shot noise. Older DSLRs used to have thermal bubbles due to overheating of near off-die components...when taking dark frames, you could clearly see regions at the corners or along edges of the frame that had more noise. I've never seen such an occurrence with my 7D, and there are times when I have taken 40-50 second long exposures at 16mm.

There are ways to reduce the various non-photon sources of noise. Dark frames and Bias frames are two. The use of dark and bias frames is usually only really necessary when doing multiple exposure stacking with a tool like Deep Sky Stacker. Generally speaking, "Long Exposure Noise Reduction" in-camera is really just taking a dark frame that is natively subtracted from the light frame before it is saved to the memory card. A single dark frame will help mitigate some read noise, but not as much as a properly stacked multi-exposure dark frame as explained on DSS's site here.

---

It should be noted that the most important thing in astrophotography is SNR, or signal-to-noise ratio. The higher your per-frame SNR, the better the results...stacked or otherwise. You could take 120 5-second frames, or 5 120-second frames...the five 120 second frames are always going to produce better results. You could even take 500 5-second frames, and the 5 120-second frames are still going to produce a richer result, since the per-frame SNR is much higher. Each frame contains richer, more complete information that you are unlikely to ever fully replicate by stacking much shorter exposures.

The next best way to improve SNR is to move to a camera with bigger pixels. Per-pixel SNR is higher with larger pixels, so on a per-pixel basis, your results should be better, and at higher ISO settings, than with a camera that has smaller pixels. If we were to compare the 1D X and the 7D (both 18mp sensors), the 1D X's larger pixels will each gather 2.6x more light. You are already using the 6D, which is a very good camera for astrophotography thanks to it's large pixels and great high ISO performance. From a pure SNR standpoint (based on sensorgen.info data), the 1D X at ISO 3200 supports ~3x the saturation per pixel, the 6D at ISO 3200 supports ~2x the saturation per pixel, as any one of Canon's 18mp APS-C sensors. The 6D also has the highest quantum efficiency (photon to electron conversion rate) of all Canon DSLRs (as of 2013 - later models have bested it), which means more SNR "bang for the buck".

Since you are already using the best camera you can probably get from Canon for astrophotography purposes, the only other thing you can really do is crank the ISO up. At lower ISO settings, more read noise is present. Particularly with Canon, the more you crank up the ISO, the lower the read noise contribution, to the point where at the highest ISO settings, read noise may be as little as 1.3e- per pixel (well below the flat minimum of ~3e- for the Sony Exmor found in the D800.)

---

Therefor, since boosting exposure post-process is the same as boosting ISO when read noise is so low, to improve the saturation of the sky and brightness of the stars, use a higher ISO setting. You said you use ISO 800-1600. Try ISO 3200, 6400...maybe even 8000. The general idea is to reduce your white point such that the camera uses it's electronics to boost the signal as much as possible before read, to minimize the impact of read noise. It should be noted that boosting exposure of an ISO 800 shot in post such that it resembles an exposure of ISO 6400 would likely result in MORE noise, as read noise at ISO 800 is more than twice as much at the lower ISO setting (5.1e- vs. 2.0e- according to sensorgen.info.)

---

To make things a little clearer, I've diagrammed a hypothetical astrophotography scenario. This scenario assumes a 30 second exposure at f/4, performed once for each ISO setting from 100 through 12800, using a Canon 5D III. The assumption is that a 30s f/4 exposure at ISO 12800 results in the brightest pixels (stars) reaching the "saturation point" (in other words, the brightest stars come out pure white, as any red, green, and blue pixels for those stars reach the maximum charge level). The exact same exposure at all other ISO settings will result in an exposure below the saturation point. Additionally, the difference between read noise and photon shot noise is demonstrated.

In the diagram below, the linear X axis represents each ISO setting, and the logarithmic Y axis represents charge level in electrons (e-). Red and green lines are drawn for each ISO setting, with red representing read noise, and green representing saturation point. The dynamic range is effectively the ratio between saturation point and read noise (green over red). For ISO 100, the saturation point is also the literal maximum photodiode charge level (FWC, or full well capacity). The blue bars represent the signal, and the darker part of the blue bar represents the intrinsic noise in that signal (photon shot noise, which is the square root of the signal.)

Assuming a 30s f/4 exposure that reaches maximum saturation at ISO 12800, the charge of that signal is 520e- (according to sensorgen.info). Therefor, assuming the exact same exposure is used for all other ISO settings...the signal, as well as the photon noise, will be IDENTICAL. (Charge in the photodiode is a product of light over time...which is affected by ONLY aperture and shutter speed.) What changes as we reduce ISO is that read noise begins to rise. Since the scale is logarithmic, ISO settings 800 through 12800 have little difference in read noise (particularly 1600 through 12800). Once we reach ISO 400, read noise starts to rise to the point where it is a greater ratio of the overall signal than photon noise.

The key difference between shooting at ISO 12800, and shooting at ISO 400, is the saturation point (green bars). At ISO 12800, read noise is low, and the signal saturates, so you'll have a bright, colorful image strait out of camera. At ISO 400, the signal is a small fraction (520e-) of the saturation point (18273e-), and will this require a significant boost to exposure in post to look the same as the ISO 12800 shot. If one does shoot at ISO 400 and correct exposure in post, then overall noise constitutes a significant factor of the signal. The read noise floor, below which useful information effectively does not exist, is almost as high as photon shot noise. Such a post-process exposure boost would result in a high degree of banding and color noise, likely right up through the midtones.

For an extreme example, if one were to shoot at ISO 100, read noise becomes the primary contributor of noise (in this particular example...keep in mind, at ISO 100, the image is severely underexposed relative to the saturation point.) Boosting an ISO 100 exposure in this case (which, in order to simulate what the ISO 12800 shot produced, would have to be a SIX STOP BOOST) would result in significant banding and color noise. The following diagram demonstrates how noise, both read and photon shot, are amplified by correcting exposure in post for ISO 100 - 6400, in order to match the ISO 12800 exposure:

Remember that the scale here is logarithmic, so the amount of noise for each successively lower ISO setting is exponentially higher after exposure correction in post.

Answer 2

I'm not even going to try to supplant jrista's very informative and well written answer. He covers the bases of the physics in the camera's imaging pipeline very well. I would like to add an observation that may shed some light on the relationship between stars and noise.

If all the stars in the universe were equally bright as viewed from the surface of the Earth, the night sky would be solid white. Pause a moment and let that sink in. There are very few spots in the sky, even when using the narrowest available field of view, that you can point a highly sensitive telescope (such as the Hubble) to that will not reveal a light source. The most notable "dark" areas of sky are nebulae that block most of the light of stars and galaxies behind them.

It is true that you can do things to increase the SNR that allow you to develop your images in such a way that the stars are brighter compared to the darker sky around them. When you do that, though, you also boost the brightness of dimmer stars that were no brighter than the noise before you made such adjustments and you also raise the level of even dimmer stars that were not even visible at all to the point that they are now producing the same amount of signal as the noise in the image. No matter how good the SNR number is, there will always be some stars that are the same brightness as the noise. The brightest stars, in terms of their brightness as seen from Earth, are the rarest and the dimmest stars are by far the most numerous in the night sky. So in some ways increasing the SNR during image capture and then boosting exposure in post can make the image look noisier! Not because there is more noise in the image. There isn't. But because those very dim stars that you pulled out of the dark background appear to be noise.

I think the secret for single exposure images is in post processing. To be sure, maximize your SNR following jrista's answer when you shoot your images. But also try this in post-processing: Once you've got the brightest stars the way you want them, pull down everything below a certain luminance value all the way to black. Reducing the color saturation will also help to deal with chrominance noise, which is the main culprit I see in your very good example image.

Answer 3

I'd guess this is going to vary from model to model, camera to camera and even based on shooting conditions. I'd hazard that on a cool night where the image sensor is cooled more, you'll have better luck with long exposure while if it is a hot night, the sensor will heat faster and higher ISO may give better results. Stacking like Matt Grum mentioned is an option in some cases too.

Personally, I tend to try to set somewhere in the middle and base the ISO on whatever produces acceptable levels of noise and then use as long an exposure as necessary. On my 5D Mark iii that ends up being somewhere in the 5000-6400 range.

Answer 4

I don't have any data on where the crossover point exists between the advantage of gaining more light and the disadvantage of thermal noise, however you can get the best of both worlds by shooting multiple short exposures and stacking them in software.

There are programs designed to do this for astrophotography that will also align the images in a stack which has the added benefit of avoiding star trails. Take a look at Deep Sky Stacker.