| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
by Roger N. Clark
Choosing ISO for astrophotography and very low light imaging requires multiple things to be considered, including finest detail digitization, minimizing fixed pattern and pseudo-fixed pattern noise, and maximizing dynamic range. Together there is a relative optimum. Ignoring any one may produce inferior results.
The Night Photography Series:
Contents
Introduction
ISO is Post Sensor Gain
Dual Gain Sensors
Finding Sensor Gain
Exposure Time
Minimize Fixed Pattern Noise and Pseudo Fixed Pattern Noise
Dynamic range
Summary
Very low light astrophotography requires multiple things to consider to provide the best results. One can find online web sites and forum posts that tell the optimum ISO. Unfortunately, most do not consider all the parameters affecting results. Add to that, the additional variable of what one can achieve with different exposure times.
Common "knowledge" on the internet is maximize exposure time and low ISO to achieve the greatest dynamic range after considering the drop in sensor read noise with increasing ISO. Dynamic range is maximum signal / noise floor, and it is commonly cited that the noise floor is sensor read noise. But that is not correct in astrophotography because the noise floor is read noise plus noise from skyglow (light pollution plus airglow), plus noise from dark current (all added in quadrature). Noise from skyglow and dark current increases with exposure time. And usually not considered is effects of pattern noise.
Question: What ISO and sub-exposure time to use? For example, low ISO for longer exposures, like 5 minutes or high ISO and shorter exposures, like 1 minute?
In astrophotography, one usually makes multiple exposures and averages them together in a process called stacking. The term stacking originates from film and darkroom days where people would stack multiple negatives together and put the stack in an enlarger to make a print, thus reducing noise. Today, stacking is done digitally when processing digital camera images.
There are multiple factors to consider when choosing ISO. First, I'll assume the same total exposure time regardless of what the ISO and individual exposure times are set to. Note that changing ISO does not change the amount of light collected; only exposure time and aperture area change the amount of light collected. For more information on ISO, see: What is ISO on a digital camera?
The factors to consider are:
1) digitize the finest details. Gain in electrons / DN should be less than 1. Less than 0.5 is better. DN = Data Number, the values in the raw file.
2) High enough ISO to minimize problems from fixed pattern noise (e.g. banding) and pseudo-fixed pattern noise (patterns like banding that may change after each exposure or a few exposures).
3) Dynamic range. Dynamic range is maximum signal / noise floor.
4) For dual gain sensors, the ISO with the higher gain state is usually a better choice because pattern noise will be less.
The following gives more details.
Many web sites and even camera manufacturers call ISO sensitivity. But ISO is a post sensor gain. ISO does not change how much light is collected by the sensor; only exposure time and aperture does that. ISO selects the range of signal intensities that get sent to the digitizer. (Silicon sensors, as in CMOS and CCD sensors are analog devices.) For more details, see What is ISO on a digital camera? ISO Myths and Digital Cameras.
ISO is gain, so what gain should one use? It is not simply an ISO, because different camera at the same ISO may have different gains., and gain at a given ISO approximately correlates with pixel size. For example, gains are given in Table 1 for several cameras.
Table 1 Camera Gains
gain in electrons / DN (DN = Data Number in the raw file)
ISO Canon 6D Canon 7D Mark II Canon 90D
(6.58 micron (4.09 micron (3.29 micron
pixels) pixels) pixels)
----------------------------------------------------------------
100 5.95 2.74 1.84
200 2.97 1.34 0.92
400 1.49 0.67 0.458
800 0.74 0.34 0.232
1600 0.37 0.168 0.118
3200 0.18 0.084 0.060
6400 0.093 0.042 0.030
----------------------------------------------------------------
Note: gains were measured by the author from
Clarkvision Reviews.
One may read online about "unity gain" where one DN = 1 electron. But the problem is that the sensor is analog, and analog-to-digital conversion electronics also has noise, especially in the lowest bit. One may also read that as long a gain is adequately digitizing the noise, there is no need to go higher in gain. For example, if read noise = 2 electrons, digitize with a gain of 2 electrons / DN. So what gain to use? The key is digitizing very faint signals much smaller than one DN. In astrophotography, the signals on faint objects are, typically smaller than read noise and smaller than 1 photon in one exposure!
Figure 1 shows a model simulation on extractable detail at different gains. Compare the results to the gains for different ISOs. If we chose a gain (per Table 1 above) of about 0.3 electron / DN gain, we would choose ISO 1600 on the Canon 6D. ISO 800 on the Canon 7D Mark II, and between 400 and 800 on the Canon 90D.
If a sensor has two gains, generally use the higher gain ISO. For example see the Canon R5 here: https://www.photonstophotos.net/Charts/RN_e.htm#Canon%20EOS%20R5_14 which shows a gain change at ISO 400. Using ISO 400 and higher will have substantially lower read noise. At ISO 400, the gain is 1.248 electrons / 1.602 DN which equals 0.78 electron / DN. At ISO 800 the gain would then be 0.39 e / DN. I have used my R5 at ISO 800 and 1600 for astrophotography.
A google search for camera model and sensor gain might find a web page or forum post where someone has derived the gain.
If a camera has had sensor characteristics measured at photonstophotos.net, the gains at each ISO can be derived from the read noise.
First find the read noise in electrons at https://www.photonstophotos.net/Charts/RN_e.htm by selecting the desired camera model and an ISO.
Second find the read noise in DN at https://www.photonstophotos.net/Charts/RN_ADU.htm by selecting the desired camera model and the same ISO as above.
Example: choose a Canon EOS 6D Mark II from each of the above photonstophotos web pages. We see at ISO 1600 the read noise is 3.811 electrons from the first web page and 12.21 DN from the second. The gain is then 3.811 / 12.21 = 0.312 electron /DN at ISO 1600, which is an excellent gain (ISO) for low light astrophotography.
But gain is only the first one of the things to consider when choosing ISO.
I have seen in internet forums the idea to use the longest exposure time to boost signals above the noise. There is an idea that signals below the noise are lost, and especially so if the signal is below 1 DN. These ideas are not correct. In fact, most faint nebulae are fainter than the noise from skyglow (light pollution + airglow), and many signals are below 1 DN in a single exposure.
The surface brightnesses in and around the Great Nebula in Orion, M42, are shown in Figure 2. The number of photons collected per exposure are shown in Table 2.
Table 2 M42 Collected Photons
Canon 90D, 107 mm aperture lens, f/5.6, 1 minute exposures.
Read noise = 1.46 electrons
Sky photons per exposure RGB = 4.44 5.9 2.34 (Bortle 3)
Dark Current ~ 0.1 electrons per 1-minute exposure (T = 16 C).
Noise = N = sqrt ( 1.46*1.46 + 5.9 + 0.1*0.1) = 2.84 electrons (green channel)
Gain = 0.118 electron / raw DN, offset = 2048 DN (241.7 electrons) DN = raw data number
Position Brightness Photons collected
magnitudes per per pixel per
square arc-second 1-minute exposure S/N (Green)
per 1-minute
R G B R G B exposure
A 15.0 14.6 14.7 1145 5060 3385 1780 Trapezium, oxygen emission (teal)
B 18.4 18.2 18.2 50 165 138 58 Oxygen + hydrogen emission (blue)
C 18.7 19.5 19.0 37.1 36.6 42.5 13 Hydrogen emission, pink arm
D 18.6 18.7 18.5 39.3 79 80 28 Reflection nebula, pastel blue arm
E 21.7 21.8 21.2 2.42 4.5 5.8 1.9 Reflection nebula, outer blue ring
F 20.6 20.8 20.4 6.55 11.5 11.9 4.0 Reflection nebula, outer ring 2
i1 22.3 22.9 23.0 1.36 1.67 1.10 0.59 Interstellar dust 1, reddish-brown
i2 23.4 24.5 24.7 0.51 0.40 0.26 0.14 Interstellar dust 2, reddish-brown
The image and Table 2 analysis shows that significant detail is fainter than the sky signal in the Bortle 3 sky. More details would be fainter than the sky signal in higher Bortle skies. In general when stacking tens of images, detail 10 and 20 times lower than sky glow can be brought out, and even signals on the order of 1/10 photon per exposure can show adequately. If hundreds of images were stacked, even lower signals can be brought out with good sensors that have low pattern noise. Note, the above M42 image had no dark, bias or flat frames measured but is a highly calibrated image. See Astrophotography Made Simple for how to calibrate without measuring calibration frames. Calibration frames add random noise.
Another factor that can influence exposure times and final results is tracking accuracy. If by exposing longer, more frames need to be rejected due to tracking errors, then it is better to shorten exposure times to have fewer rejections. The results above show that faint signals can be imaged even with relatively short total exposure times
Yet another factor is effective rejection of satellites, aircraft, hot or dead pixels when stacking with sigma-clipped average, is the number of frames in the stack. Rejection gets better with more images in the stack. Try for at least 9 images in the stack, but 15, 20 or more is even better. If you only stack 3 or 4, there is less information to reject outliers. In that regard, three ten minute exposures will be less effective to reject transients than 15 two-minute exposures.
If shortening exposure times results in the weakest signals getting clipped to zero on the histogram (left side), raise the ISO (gain).
Pattern noise, both fixed and pseudo-fixed pattern noise decreases with increasing gain (ISO). Pattern noise, for example, banding occurs after the sensor and if after the analog gain stage, its effects will decrease relative to the amplified signal as ISO increases.
Banding pattern noise is illustrated in Figures 3a and 3b.
Pseudo-fixed-pattern noise is noise, such as banding, that may change from frame to frame or after a few frames. Banding patterns are shown changing over a 3-hour period for a Canon 7D Mark II in Figure 4. This is a well behaved camera, especially for the era. The pattern changes only about 1 electron, but that will affect the faintest objects in the scene, as illustrated in Table 2. One way to reduce pseudo-fixed-pattern noise is to stack many frames, but best to start with a low pattern noise camera. When in the market for a new camera, select models that have low pattern and pseudo-fixed-pattern noise.
Another way to see pseudo-fixed-pattern noise is in video mode: select a very high ISO and in a dark room view the camera LCD. One will often see banding that changes from frame to frame.
Dynamic range in an image is maximum signal divided by the noise floor. The noise floor in an astrophoto includes sensor apparent read noise, noise from dark current, and noise from unwanted signal in skyglow (light pollution plus airglow).
The noise floor in a deep sky image is square root (read_noise^2 + skyglow_noise^2 + dark_current_noise^2). Noise in the sky signal equals the square root of the sky signal. Noise in dark current is the square root of the dark current. Thus, the noise floor can also be expressed as:
noise floor = square root (read_noise^2 + sky_signal + dark_current_signal)
Skyglow noise and dark current noise increases with the square root of the exposure time, so longer exposures have less dynamic range. To minimize the noise floor, expose long enough that noise from skyglow and dark current are greater than read noise. At lower ISO's read noise is greater, so exposure times should be longer. There is no magic number here, but most try for skyglow and dark current signal level to be 10x greater than read noise, but even 3x greater only increases random noise from the read noise component by 15%. At 5x greater, it is 10% and 10x greater it is 5%.
As light pollution increases (Bortle level increases), dynamic range decreases for the same exposure time.
Per factors 1 (ISO gain to digitize fine details) and 2 (minimize pattern noise), once ISO has been selected, then exposure time determines dynamic range. If you aim for skyglow noise 10x the read noise, that occurs approximately when the histogram peak (the peak is typically the sky glow) is about 1/3 from left to right on the camera LCD histogram plot. With low read noise under 2 electron, even 1/4 from left to right is fine. The images of M42 in Figure 2 had the skyglow less than 1/8 from left to right, and that works because the sensor has little pattern noise and read noise is very low, 1.46 electrons. Just be sure the toe on the left side of the histogram peak is away from the left side of the plot so that the faintest areas do not get clipped. This is the minimum exposure time. You can go longer, but remember, that reduces dynamic range.
Dynamic range reduces with the square root of the exposure time. The dynamic range of the final image increases with the square root of the number of exposures stacked. For example, 10 one-minute exposures will have a higher dynamic range than two 5-minute exposures, and with good sensors, record the faintest targets similarly well. The short exposure times will also show more star colors due to fewer stars being saturated. Stacking pushes the noise floor down, while keeping the same maximum signal.
Selecting the optimum ISO for a digital camera requires multiple things to be considered, including gain (optimum 0.3 to 0.5 electron / DN to digitize the faintest objects and separate them from other noise sources), high enough ISO to minimize pattern and pseudo-fixed pattern noise, and long enough exposure time so that skyglow noise swamps sensor read noise, all while keeping sufficient dynamic range with individual exposure times.
The Night Photography Series:
| Home | Galleries | Articles | Reviews | Best Gear | Science | New | About | Contact |
First Published May 22, 2025
Last updated May 22, 2025.