DSCOVR Options

[bkellysky]

Nice image, but something must be off with the colors- the shadow looks brown but should be neutrally colored, as the moon will block all wavelengths equally.

John

There's a lot of discussion upthread of DSCOVR images from the 2017 eclipse. The colors produced by the way they combine the data from different wavelengths makes a difference from what the human eye would see. Even the GOES visible wavelength photos showed a very dark area along the path when the Sun wasn't totally eclipsed at any point in this annular eclipse.

[fredk]

That image is the "enhanced" version. The "natural" version doesn't look quite as bad:

https://epic.gsfc.nasa.gov/archive/natural/...31014165817.png

Looking at the pixel values the blue channel drops to precisely zero in the penumbra, giving the brown hue. I suspect somewhere in the processing (combining the raw channels into the colour version) they've clipped the low end of the blue (and green) channels. Their processing probably wasn't designed for such dark levels since the Earth is normally (almost) fully sunlit from L1!

Actually limb darkening of the sun is wavelength dependent, with shorter wavelengths more attenuated than long near the rim. So around annularity the illumination should actually be shifted to the red. But most of the visible penumbra is still lit by some of the central area of the sun's disk, so I wouldn't expect a noticable effect in these images.

But near the narrow path of annularity the illumination should be shifted somewhat to the red. It would be interesting to quantify that. Sometimes people report a change in the "character" of the light near annularity - maybe the colour shift contributes to that. I didn't notice a colour shift at either the 2023 or the 2012 annular eclipses, just the dimming. Of course the eye would adapt to such a slow colour shift.

[scalbers]

Nice analysis by fredk. The DSCOVR Team reportedly did make some color improvements to the natural color web images a few years ago, though this eclipse image shows some improvements could still help. Perhaps looking at the available calibrated radiance / reflectance data would give some further insight. As mentioned it's unusual for DSCOVR to have to deal with color of low intensity regions except right near the limb at times. Sometimes I like to look for colors of clouds along the barely visible terminator. Hopefully scattered light doesn't vary with wavelength in the camera system.

The various effects including limb darkening of various wavelengths are somewhat accounted for (and something I hope to improve) in my simulated sky and Earth images. Just subjectively I think the redder color shows up visually when experiencing a deep partial eclipse. If it didn't the combination of bluer color and dimmer light would look more unnatural and different from a normal sunset sequence of lighting.

In this GOES animation the land looks a bit redder, more than the clouds though here also it isn't a perfectly true color processing.

https://col.st/4fqw2

[StargazeInWonder]

All very interesting. For what it's worth, this is the second or third eclipse that I experience as relatively deep but not total (77% in this case) under cloudy skies and subjectively, there was no noticeable color shift to the red.

[fredk]

Perhaps looking at the available calibrated radiance / reflectance data would give some further insight. As mentioned it's unusual for DSCOVR to have to deal with color of low intensity regions except right near the limb at times. Sometimes I like to look for colors of clouds along the barely visible terminator. Hopefully scattered light doesn't vary with wavelength in the camera system.

Thanks for the links. Yeah the raw epic data are 32-bit floating-point, and have no problem seeing deep into the penumbra. Here's an example during the eclipse - this is the 443nm channel, clipped on the fully-lit clouds to show detail in the shadow:

But converting three visible channels to a colour image is non-trivial, unless the team has laid out the details for their procedure somewhere.

The various effects including limb darkening of various wavelengths are somewhat accounted for (and something I hope to improve) in my simulated sky and Earth images.

Do you mean by "limb darkening" the extra absorption/scattering near the Earth's limb as viewed from the sun?

Just subjectively I think the redder color shows up visually when experiencing a deep partial eclipse. If it didn't the combination of bluer color and dimmer light would look more unnatural and different from a normal sunset sequence of lighting.

Interesting thought. I'll have to pay more attention to colours if I make it to the path of April's eclipse...

[scalbers]

Nice to see there's a good presentation in the blue channel of the DSCOVR image data. I'm unsure if the details of the DSCOVR processing are widely available. I've done my own version of this processing in the past and some approaches can be a bit tricky. The image below is such a past case. I'd be happy to expound upon this in further detail as for example from this post.



Presently I'm trying out my simulated DSCOVR images for this eclipse case, ironing out some software wrinkles on that front.

The limb darkening color dependence is related to the color of the sun's limb. The limb is redder due to the slant paths when viewing into the incandescent solar atmosphere. One reaches opacity at a higher photospheric altitude near the limb. The effective radiating temperature is thus cooler, producing a redder black body color and lower intensity. Of course the limb color becomes predominant when we are illuminated by a crescent sun.

Looking at the limb of the Earth during a solar eclipse would make a double effect and hopefully will be accounted for

[fredk]

There are some details on the production of colour images from EPIC raw data in this pdf document.

For the level 1b data here:
https://opendap.larc.nasa.gov/opendap/DSCOV...0/contents.html
the channels are transformed to the same viewpoint so you won't get the colour fringing due to the Earth's rotation.

I was wondering about why you were treating limb darkening because I would've guessed you would simply use some standard illuminant for the sun. I guess you meant you were actually trying to incorporate the colour shift due to limb darkening because you are simulating eclipses?

[scalbers]

Interesting to see this document that seems to be associated with their color processing improvements from a few years ago. Good to see they mention the CIE color processing. In the past the H5 file images weren't corrected yet for Earth's rotation as can be seen in the image from two posts up. I wonder if this is newly being done in the H5 file or in another version of the files shown, or simply later on in their processing for the L1b and web images?

The "brown eclipse" issue may stem from a lack of eclipse support in Rayleigh corrections employed with both DSCOVR and GOES image processing. An alternative to the Rayleigh correction the DSCOVR team is using could be a logarithmic interpolation between the observed bands to construct the full spectrum as I've been considering in the context of sky simulations.

Yes I try to account for solar limb darkening when simulating eclipses, from space-based and ground-based or high-altitude views. The DSCOVR perspective view simulation is something I'm ironing out at present - here is a first look based on land surface and 3D atmospheric analysis data fully independent of DSCOVR.

[scalbers]

Animated comparison of simulated Earth (left) and actual DSCOVR images over several hours for the annular eclipse:

https://stevealbers.net/allsky/cases/dscovr...e/animation.png

[fredk]

Nice simulation. No obvious reddening in the penumbra in your simulation, as you'd expect with areas of the central sun's disk contributing significant illumination except very near the centreline.