Understanding the science of low-light photography
Four things to understand about night sky photography.
This is a post that will explain some of the technical aspects of night photography. You will learn how these 4 factors: light sensitivity, digital noise, movement and external conditions, affect the quality of your images. This post will not cover the artistic aspects of composition, foreground illumination, and focus stacking.
Summary
Can you photograph the stars with any camera?
Structure of a (Digital) Single Lens Reflex camera
Four keys to make better pictures: 1 - Light sensitivity 2 - Digital noise 3 - Movement 4 - Light pollution
Conclusion & General recommendations to make use of the points above
Can ANY Camera Photograph The Stars?
Absolutely! We should keep in mind that until quite recently, even high-end research astronomy equipment was no better than some current consumer products. In fact a professional photo camera may not be the best tool to make accurate observations of stellar objects, since a large part of the signals that stars emit can not be detected by a CMOS sensor. The ‘visible’ part of the electro-magnetic spectrum, that humans can see before their range degrades with age, spans from near UV to far red (~420 – 630 nm wavelength), that's 0.0035% of the entire spectrum. Within this range the sensitivity to light of a camera sensors is not the same at every wavelength of the light. The sensitivity to green light is highest for tow reasons:
- The quantum efficiency (QE) of the sensor differs by wavelength, with pure green light (555nm wavelength) the max QE varies between 65 and 82%, but most of the time, it will be much lower.
- An RGB sensor is composed of a grid of photosites (commonly referred to as "pixels") with different color filters. To produce an image that resembles what a human perceives, the photosites are generally composed of 50% of green-filtered photosites and the remaining is shared by blue- and red-filtered photosites and by special contrast-detecting photosites used by the auto-focus module. Additionally, consumer CMOS sensors are covered with an extra filter that reduces the light transmittance in the red and infra-red spectrum to reduce the blur and color shift that would result from the presence of infra-red radiation when a hot object or a heat source (like the sun) is near the camera.
So to answer the question: all things considered, yes, any camera can be used to produce an image of the visible light that reaches the earth. And we will see in the next section that consumer cameras have a range of properties that will affect their light sensitivity. However, beyond technical differences, I will argue that the external conditions play a more important role than the sensors.
Single Lens Reflex Cameras
The photo camera is a complex optic system with two main parts: the feedback system and the image aquisition system. The first one, composed of a mirror, a prism and various filters is made to provide a feedback to the user on what the focussed image looks like. This part is absent from a mirror-less camera where the feedback is provided by a low resolution recording of the image that reaches the sensor and is projected onto a miniature screen either in the view finder or on the back display of the camera. The second part of the system is the sensor itself and the various translucid elements that deform the light path in order to focus it (to converge it) onto the spatial plane where the sensor is located. The nature and number of the optic elements is of importance for the night-photographer, since it directly affects the light transmittance of the optic system, and thus it’s sensitivity to light. Because of the deformation of the light path that the optic elements are responsible for, they might accidentally introduce geometric inhomogeneities that make up the image aberrations of each particular lens model.
Light Sensitivity
Lens Apperture
There is no perfect lens, but among night photography adepts, it is generally accepted that bright lenses (‘fast’ lenses in photo jargon) are better. A lens is made of stacked glass or plastic elements, it blocks out part of the light that travels through it, before it can reach the sensor. Constructors generally provide a standard estimate of the amount of light that is blocked by a lens. This amount is derived from the ‘f’ number: a value of f1.2 means that 1/1.2 = 83% of the light travels through the lens and reaches the sensor. At f4, only 25% of the light goes through. This means that the type of lens and in particular its aperture has the most significant impact on the light sensitivity of a camera system. Arguably more than the sensor.
Every f-stop corresponds to approximately half the amount of light of the previous value (f6.2 = 16.1%) Read the section about movement bellow, to understand how the maximal aperture of a lens affects the amount of movement in a long exposure photograph. In our experience, the amount of time it takes to correctly expose an image of the stars with a typical full frame sensor (like any pre 2009 non back-side illuminated Nikon 24x35mm model) makes lenses with a maximal aperture >2.8 less than optimal as they impose exposure times that are too long to adequately freeze the relative motion of the stars across the field of view.
Sensors
There are many commercially available imaging systems to choose from, but as seen in this figure, they come with very different sensor sizes (and this is only a portion of the range of sizes! There are smaller and larger ones). There have been many manufacturers of camera sensors, many of which have created their own sensors. Still, every sensor relies on comparable technology and they all have a number of properties in common.
What all camera sensors have in common
1- first, they rely on mostly the same technology, which means that they can be compared based on their attributes (like surface area of the sensor and of the photosites)
2- Sensors are made of photosites (inaccurately called ‘pixels’ because they produce images that are composed of little squares called ‘picture elements’ or in short ‘pixels’). When sensors are larger, they have either more photosites, or BIGGER photosites.
3- The proportion of the number vs. the size of the photosites defines the compromise between light sensitivity and spatial resolution.
These properties of camera sensors allow us to make the following predictions: when we have a camera system with a small sensor, it is best to have a small number of ‘pixels’, because they will be larger, and as we will see bellow, this allows for better light sensitivity.
Higher resolution = worse low-light capability
The latest cameras have been released with ever increasing resolutions up to 100 millions of pixels but the sensor size is still 24 by 32 mm and for this reason the pixels are always smaller with every resolution increase. This resolution race is AT THE COST of sensitivity because of the inverse relationship between spatial resolution and light sensitivity. Here is one piece of evidence:
This figure shows how the light sensitivity of a CMOS sensor depends on the pixel size. It plots the MPE30 in lux per second.(= The number of lux per sec to produce signal to noise ratio (SNR) equal to or greater than 30) This means that the intensity of the signal is 30db higher than that of the noise. This introduces a new notion: the noise. (Read the next section find out more)
This is how you interpret this figure:
If pixel size is larger, it takes less light, or less time to achieve a good signal-to-noise ratio
1 lux = 1 lumen / m2. Estimating how many photons are hitting the sensor when the illuminance is 1 lux, is not a trivial calculation because the lux is a photometric measure of energy, rather than a count, and thus the the number or photons in a lux vary with the spectral distribution of the light.
(If you are interested in reading an example of such calculation, here is one: shorturl.at/gmFNT )
On the graph: with 5.2 μm pixels (so a surface area of 27μm2 ), requires much less light than a sensor with 2.6μm to produce a usable image. For a phone sensor (<2 μm photosites) you need even more light, and single images are dominated by the background noise. On a side note: small pixels are not only a BAD thing, in fact spatial resolution is much higher when pixels are smaller! Conclusion: sensors with big pixels have better sensitivity to light, but lower spatial resolution. Pixel sizes are not everything though: the overall size of the sensor matters too, but for other reasons: the larger sensors are less impacted by movement. With the same focal length, a typical hand shaking of a few millimeters in amplitude would produce a much more blurry image on a phone than on a medium format, if there wasn’t an image stabilization module in the phone. Also, in a theoretical example where the sensor was perfect and could capture every photon that hits it, without creating noise, it is easy to conceive that, at any exposure duration a larger sensor would collect more light than one with less surface area.
Noise
Every imaging system produces ‘noise’ which can be simply defined as non-specific signal, or signal that the camera is recording but which does not come from the phenomenon one is observing (in the case of photography: the light reflected from the scene that reaches the sensor). Noise can be generated by heat, stray light, or imperfections in the sensor itself that produce false signal during the read-out. In astrophotography, but also in microscopy and virtually any situation where one has to compromise with low luminance levels, a common method to reduce the noise is multi-image stacking. If the phenomenon that one is observing has a duration long enough to take several images, this is a very efficient way to control the non-specific signal. There are many sources and types of noise that add up to degrade the quality of our images. But for simplicity I will group them in 2 broad categories:
Random noise: comes from the physical conditions in which the picture is taken, (ambient temperature etc), it manifests itself through hot pixels randomly distributed, and generally, of short duration. Because those hot pixels are rarely seen in the same place on two consecutive images, they can be suppressed by averaging as little as 3 to 5 images (depending on the application, one will use the average of 5 images, or like in confocal microscopy, the lowest value that a pixel had across 3-10 images). The rationale is that real objects stay visible the whole time, in a give location (if they don’t move). In photography, movement is not negligible, objects' positions change slightly between images, so averaging is a ‘safer’ approach.
Non-random noise: This types of noise come from the optic system itself, and from the conditions under which the image is captured. Temperature is hart to classify as the occurrence temperature-induced noise is not random, but the physical location on the sensor where ambient temperature occasions hot pixels is. Heat can also be generated by the camera, and in such case it happens in predictable locations. The system-specific noise comes from the architecture of the camera sensor and other components that are always present in a given camera model, the sensor read out can be tuned to control unspecific signal to a minimum level, this is set by the manufacturers. However, the camera-generated noise is not always entirely the same between two units from the same camera model, and additionally varies as a function of camera settings and variable elements such as the lens. For demanding conditions, it is recommended to complete each photography session with dark images, shot with the lens cap closed and at the same settings as the rest of the series (ISO in particular). The average image of several dark images is then subtracted from each photograph of the actual photo series to arrive at a noise adjusted image. This corrects for sensor inhomogeneities. Similarly to sensor aberrations, the lenses are far from perfect and produce inhomogeneous light intensities across the image. This is corrected by making ‘light images’ with lens cap off, but through a special frosted glass filter (or just a white piece of cloth, as used for telescopes). This image allows for a correction of vinietting and sensitivity differences in the sensor (and dust particles).
Movement
In long -exposure photography, movement becomes an important aspect of the image, whether it is part of the esthetic choice or happens by accident. The most obvious and easily avoidable source of movement is the user. It is important to take the pictures in ast steady conditions possible by using a tripod, and a self-timer function in camera. Additionally to user-caused disturbances, environmental conditions also produce unwanted vibrations. As a general rule, you can reduce those by choosing a position low to the ground, protected from the wind. Remove any unnecessary mobile elements from the camera (such as the straps), and do not move near the camera during the exposure as certain soil types can conduct the vibration from the steps to surprisingly important distances.
Earth rotation
Being perfectly immobile relative to the stars is not possible on earth, though: the earth is always rotating, causing the stars to permanently change position relatively to the camera. The speed at which this occurs depends on where the camera is pointed at, the higher the azimuth the faster the drift. In the example below, I took a series of exposures over 15 min and combined them in one image. You can clearly see that the portions of the image that are further away from the north star have much longer star trails than at the center.
The reason this is relevant when we think about the low-light performance of various imaging systems, is because their sensitivity to light will determine how long the user has to expose the sensor to produce an image. In other terms, if the light sensitivity is low (e.g. because using a small sensor, or a lens with a limited maximum aperture), then the exposure time has to be higher. When the subject is not moving, you can make image stacks, and long exposures, but in the case of stelar objects there is a continuous movement that limits how long we can expose our images before the trailing becomes visible. This time depends on the resolution of the final image and on the size of the stars that are projected onto the sensor. If the stars’ projections are infra-resolutive (the projected size is smaller than one pixel) then the apparent size will be between 1 and 5 pixel (there is always some blurring that makes objects look bigger than they are), and the trailing will not be visible, but if the trail is longer than a few pixels, then the elongated shape will start to become recognizable. And this is more likely to happen with high resolution systems (that have smaller pixels) as the projection of an object has to travel less distance on the sensor to be detected by neighboring pixels.
The Imaging System Impacts The (Apparent) Movement Of The Stars
Why choosing a wider lens reduces star trailing
We saw that sensor resolution affects how likely you will see light trails from the moving stars, but that’s not all!
The size of the sensor and the focal length of the lens are just as important. The combination of a large sensor with a wide angle lens produces a great combo for night sky photography, but why? A large sensor is more robust to mild shaking because a little jitter of e.g. half a millimeter covers just a fraction of the image when, on a phone sensor, that same shake would amount to almost a quarter of the sensor's cross section. The same logic applies for mild movement from the subject’s projection on a steady sensor. And if one is using a wide angle lens, then the field of view being much larger than on a longer focal length, the objects are also comparatively smaller. The consequence is that larger fields of view cause objects AND THEIR MOVEMENTS to be proportionally scaled down, causing the star trailing to be less apparent.
Calculating the exposure time limit
A commonly used approximation called the 'rule of 500' is often referred to by photographers.
This rule is meant to give a general approximation of the maximal exposure time that, given the focal length and the sensor size, will be short enough for the star trails to remain (almost) unnoticeable. The rule states that the maximum exposure time (MET) is given by the following equation: MET (in seconds) = 500 / (focal length * (the ratio of the surface area of a full frame over the camera’s sensor surface)) or it can be simplified as follows or a full frame sensor: MET = 500/focal length*1.
For example: with a full frame sensor and a 20mm lens: MET= 25sec, and with the giant Kodak KAF 39000 sensor : MET=500/20mm*0.47 = 53 seconds! This rule has a major flaw, as it omits an important parameter: the pixel size ! It works well with pixel sizes of ~5-6 μm but fails to limit star trailing when the pixels are smaller, and when the sensor size is very different from a full frame. Since the camera models that were rolled out in last few years have had consistently smaller pixel sizes, the rule of 500 is increasingly off.
I real life, our recommendation is to aim for half of the MET given by this rule (so in the same example with a 20mm lens on a 24x32 sensor: MET=25s, real world MET=0.5*25= 12.5sec). But remember, the speed of the star drift depends on where you point the camera. If you look to the North you might get very usable images with much longer exposure times than recommended by this rule.
Limiting factor – the light pollution (and space pollution)
Light pollution is the unwanted ambient light that is present in a scene. The most part comes from artificial light sources like street lights or stadiums, homes ect. The moon too is very bright and makes stars appear much fainter. The continuous growth in light output from the cities added to technological progression to make light even brighter causes the rare dark places left to be even more difficult to find and always further from the towns. (I used to have very dark places 30min from home. Now I have to drive 4 hours to find similar darkness).
Unfortunately, as I said in the intro the ambient conditions, in particular the amount of light pollution is more important than any of the factors we mentioned in this article. The best sensor and lens are useless if you shoot on a full moon or near Los Angeles. To explain: the luminance of stellar objects has to be many times more bright than the ambient light for the camera system (and ourselves) to see it. If the camera has very low light sensitivity like a phone, you can compensate by increasing exposure time and stacking images to remove random noise. But if the noise is consistent like the ambient light there is not much you can do. A better camera would only allow you to reduce exposure time but it can't filter stray light from the nearby towns. It will be just as useless as a phone if the light pollution is intense.
Choosing a location
You can consult online tools to look for dark areas in your region such as the dark site finder (https://darksitefinder.com) or the maps from the international dark-sky association (https://www.darksky.org) which lists some of the best locations in each geographical region. The location is an important aspect of night sky photography, as the sky is not only too bright near cities, but also too busy, with sometimes very intense air traffic that can really ruin a sequence. In California where we shoot the most sequences, the towns are so bright and busy that even dark locations like desert areas between Los Angeles and the Nevada border get too much air traffic until late at night. We recommend high elevation national parks, the plane routes are generally far away, and there is no street lights. Additionally, the light particles traveling through the atmosphere to reach you have a shorter distance to cover and suffer less dispersion on the way, increasing the intensity of the signal you will be able to capture. (Beware of wildlife and be prepared for cold nights even during milky way season, though!)
Unfortunately, even in the best location, there is a real plague that can’t be avoided: SATELITES!
There is nothing we can do, the number of satellites currently used by the USA alone (in August 2021) is 2505 with over 2000 of which being commercial satellites. The large and rapidly increasing number of objects orbiting the earth is unfortunately not only a visual nuisance but a significant risk for the satellites themselves, for the future of space exploration and for human property on earth when the space debris re-enter the atmosphere. On Jan 1st, 2021 the Union of concerned scientists (UCS) satellite database (https://www.ucsusa.org/resources/satellite-database) recorded that 3,170 of the 6542 orbiting the planet at the time where not currently operational. The space junk is becoming such a concern that the Inter-Agency Space Debris Coordination Committee (IADC) has commissioned the first space debris removal mission, due to launch in 2025.
(https://www.unoosa.org/oosa/en/ourwork/topics/space-debris/compendium.html)
Impact of light pollution
Coming back to our main topic, light pollution impacts night photographs in two ways:
1 - The added light makes the celestial objects look much fainter in comparison to the background sky, and much more difficult to see.
2 - The overall luminance of the scene can sometimes be so high that it is impossible to expose long enough to see the stars (which are faint and need a long exposure time) before the photos are over exposed.
When the light pollution is mild it is possible to filter some of the artificial light reflected by the atmosphere by using a light pollution filter (that’s generally a band-pass filter that blocks out a portion of the visible light spectrum from orange to light red ~580-610). Note that any celestial object that emits in these wavelengths will be attenuated by the filter.
Conclusion and recommendations
Conclusion and recommendations
This very simplified article about general technical aspects of low-light photography covered a variety of topics. We learned that there are really a lot of different equipment types available for consumer imaging systems but that besides the lenses, the sensor size and pixel size on the the sensors have an important impact on the sensitivity to light. Larger pixels collect more light, and larger sensors suffer less from small amounts of movement. Any camera system can make pictures of the night sky, but there is a limit to how long the exposure can be, because of the earth rotation causes the stars to permanently move relative to us and because of light pollution. Finally, to get the best chances to meet good conditions, we recommend traveling to a location far from cities and air traffic routes and at high elevation. Stay warm and enjoy the view!
And if you are a home owner, please consider a type of light fixture that directs the light downward, where it is needed and help spread the word. Reduce light pollution. https://www.darksky.org/light-pollution/