DJI Air 2S Dynamic Range and ISO Performance Testing
To be honest, I was a bit disappointed with the image quality from the Air 2S after my first outing, but then I realized how far I was having to push the images in Lightroom (+2.5 for exposure and +75 for shadows) to handle the extreme dynamic range of the landscapes. That realization made me feel much better about the capabilities of the camera but also led me to want to develop a better understanding of how the camera performs. Unfortunately, none of the camera review sites have bothered to do any sort of real, in-depth image quality analysis on this camera (mostly they just repeat DJI’s marketing line “Air 2S Melds Incredible Image Quality With Unmatched Flight Performance”). So I decided to do some testing of my own.
I’ve taken several series of images under different conditions to see what I can learn. And I’m now very impressed with the capabilities of this little flying camera (especially after capturing these images). This started out as an ISO test but accidentally ended up also being a dynamic range test (that’s what happens when you take a camera outside). What I learned is that when pushing a camera sensor to its limits at one extreme (noise in the shadows), we also end up testing the limits at the other extreme (clipping in the highlights) because dynamic range is directly affected by ISO.
Caveat: these tests are not scientific nor controlled but I think they do provide some useful insights for how to get the best images from this camera in real shooting situations.
Basic ISO Comparison
I captured a series of photos with the aircraft hovering in my front yard. The sun came up over the trees just as I getting started so portions of the house were in direct sunlight while most of the scene was in shadow. This scene was therefore very challenging for the sensor and thus quite instructive for how to best utilize the camera for capturing scenes with strong lighting.
Single Image DNG
These photos were captured in single shot DNG mode, autofocus, 0 EV exposure compensation, and fixed ISO. The camera was allowed to select the shutter speed for each image based on the metered exposure. The comparison images below are ~200% crops minimally processed in Lightroom with Sharpening set to zero. To my eye, the three images in the top row are all good, and I think the ISO 800 image would be fine with some noise reduction applied depending on the intended use of the photo. In addition to the visible noise, I also noticed some unusual camera behavior and strange color shifts at ISO 1600 and 3200.
I repeated this same series of exposures twice with exposure compensation set to -1 and -2 EV so the exposures were deprived of light. At ISO 100, I saw no discernable noise in the shadows in the 0 and -1 EV exposures; some noise is visible in the -2 EV exposure, comparable to the shadow noise at ISO 400 at 0 EV shown above.
For single exposures in low light situations with shutter speed limitations, ISO 400 produces good/acceptable images and ISO 800 produces acceptable images (with proper sharpening and noise reduction). I would still try to avoid ISO 1600 or 3200.
In August 2021, TechRadar published an article stating that they were surprised by the low noise in the DNG raw files produced by the Air 2S, so they reached out to DJI who confirmed that “temporal denoising technology is used on the raw files”. The story is quite vague on exactly what their reviewers noticed and what they discussed with DJI, but I have to say that I’m not seeing any evidence of heavy-handed noise reduction being applied in the higher ISO images.
5-exposure HDR AEB
I also shot a series of 5-exposure AEB images at each ISO with 0 EV exposure compensation, then merged the images to HDR in Lightroom. At ISO 100, there is no noise-this is the recommended option for landscape photos with the Air 2S. ISO 200 and 400 have minimal noise, and ISO 800 looks good.
Although I’m not entirely sure of the use case but, with AEB, ISO 1600 can produce good images, and ISO 3200 is acceptable although it appears there is additional processing applied to the DNG files at ISO 1600 and 3200. I’ll have to repeat this test for some nighttime photos to see what works the best. I would expect highlight clipping to become a problem at these higher ISOs for most night shots with artificial lights in the scene.
Dynamic Range Testing
As I mentioned in the intro, the sun came out as I started this test, and I didn’t realize just how intense the light was on the front of the house. The title image for this post is the single shot photo captured at ISO 100 with -2 EV exposure compensation and is the only photo in which the highlights on the curtains in the front windows are not completely blown out. Even in the 5-exposure AEB at ISO 100, the highlights are blown.
The three photos above were all processed minimally and made to look equally exposed. In the -2 EV image, the highlights are very bright but not clipped, and there is still good detail in the shadows (although the shadow detail is much better in the 0 EV exposure). In the -1 EV exposure, the white areas in the windows are blown out, while in the 0 EV exposure the image is also blown out where the sunlight is striking the brick.
In high dynamic range scenes, this camera will overexpose highlights, even with exposure compensation set to -1 EV. Always use 5-shot AEB with some negative exposure compensation dialed in for good measure.
For more on why ISO invariance matters, take a look at this post.
The Sony IMX283/IMX383 sensors are ISO invariant in the sense that there are no additional details lurking in the deep shadows to be revealed by increasing the ISO; this is clearly demonstrated in the data at photonstophotos.net. However, at least as implemented in the DJI Air 2S, it is not ISO invariant like some APS-C and full frame cameras where the exposure can be pushed 4 or 5 stops with no noise penalty.
In my testing, it looks like an exposure at ISO 100 can be pushed about 3 stops and produce an equivalent image to an exposure captured at ISO 800 (see comparisons below). At +4 EV (ISO 1600 equivalent), color noise and additional luminance artifacts became prevalent in the shadows in the ISO 100 exposure, but highlight areas that were completely blown in the ISO 1600 exposure looked great. At ISO 1600 and above, noise suppression is better handled by increasing ISO in the camera rather than increasing exposure in Lightroom; the tradeoff is the loss of ~5 stops of dynamic range.
For low light aerial landscape photography, it seems there will likely always be some highlights in the image either twilight on the horizon or city lights depending on the scene. For these types of images, I think it is crucial to preserve detail in those bright areas, particularly for cityscapes, so for most situations it seems that shooting at ISO 100 is the best option (for still photos, obviously not for videos) so that maximum dynamic range is retained. If there are no highlights, perhaps for a moonlit rural landscape, then I would probably try ISO 1600 )but not higher).
SmartPhoto is unreliable, unpredictable, and produces crap. Do not use for daylight scenes. I shot the image below using SmartPhoto-this image probably has the most blown highlights of any of the photos I captured for this test. There is definitively nothing “Smart” about this photo!