Lower’s Nebula: A Three-Filter Journey Through Hydrogen, Oxygen, and Sulfur
Processing a faint winter target with 28 hours of data across SV220, Askar D2, and LPRO filters
Revisiting a Faint Winter Target
Lower’s Nebula in Canis Major doesn’t show up on many “best of winter” lists. It’s faint, it’s diffuse, and it lives in a part of the sky dominated by flashier neighbors like the Rosette and IC 410. But with enough integration time and the right filter combination, it rewards patience — and after January’s bird-poop-on-the-lens incident, I was ready for a target that would pay off rather than punish.
I spent 28 hours and 35 minutes collecting photons from my Bortle 8-9 backyard in Houston, split across three filters: the SVBony SV220 v1 (11 hours of Ha+OIII), the Askar Color Magic D2 (12 hours 5 minutes of SII+OIII), and the Optolong L-Pro (5 hours 30 minutes of broadband). All of it through the ZWO FF65 and Touptek ATR2600C on the AM5 mount — the same gear that took the bird strike back in January, now thoroughly cleaned and back in service.
The plan was to process the data two ways: first as a standalone HOO image from just the SV220 to establish a baseline, then as a full three-filter combination producing both HOO and SHO palettes with natural star color borrowed from the LPRO. Here’s how it went.
First Light: The SV220 Standalone
Before combining anything, I wanted to see what the SV220 could do on its own. This is worth doing anytime you have multiple filters on a target — it tells you what each one contributes individually and gives you a reference point for evaluating the combined result.
The linear pipeline was straightforward: GraXpert for gradient correction, then BlurXTerminator in Correct Only mode, SPCC for color calibration, a second full BXT pass for deconvolution, NoiseXTerminator for cleanup. Nothing exotic — but there was one interesting moment during gradient correction worth unpacking.
Running GraXpert at its default smoothing of 0.0, the background model looked alarming under auto-STF — high-contrast blobs of color that seemed to trace nebula structure. My first instinct was that the AI was latching onto real Hα signal and pulling it into the “gradient.” So I tried again at smoothing 0.3, which should blur the model and force it toward a cleaner large-scale gradient. The result looked dramatically smoother.
Except — it turned out I had the interpretation backwards. Checking the actual pixel values (not the STF preview) revealed that the 0.3 model was so heavily blurred it was nearly uniform, and STF was stretching its tiny remaining variations to full saturation to make it look wild. The original 0.0 model was actually the correct one; I was being fooled by the auto-stretch. Lesson filed away: never trust an STF-stretched background model at face value.
From there, the rest of the linear workflow followed the current best-practice sequence. BlurXTerminator’s “Correct Only” mode tightens stars and fixes residual optical aberrations without sharpening yet — this gives SPCC cleaner star detection for photometric matching. SPCC in narrowband mode used the SV220’s specs: 656.30nm at 7nm for red (Hα), and 500.70nm at 7nm for both green and blue (OIII, since an OSC captures OIII across both Bayer channels). A 340×376 pixel background ROI in a clean corner gave the neutralization enough statistical footing to work properly.
With colors calibrated, a second BXT pass with full deconvolution (0.20 stellar, 0.90 nonstellar) pulled out the fine filamentary structure that makes Lower’s visually interesting. NoiseXTerminator at 0.85 denoise in a single iteration cleaned up the linear data without touching detail.
Then came stretching. StarXTerminator separated stars from nebula, and Bills Stretch on the starless at 0.10 / 1.01 / 2.8 struck the balance I was after — dark background, bright nebula, nothing crushed. The Curve value of 1.01 matters here: with almost no highlight compression, faint structure keeps its brightness instead of getting flattened. Higher curve values (I tried 1.03 first) darkened the nebula noticeably. After gentle curves work for contrast and saturation, I recombined with the original SV220 stars via a PixelMath screen blend.
The result was a clean HOO image with genuine OIII signal showing up in Lower’s brighter knots — more than I expected from a single dual-narrowband filter. Worth the baseline test.
Going Deeper: Three Filters, Two Palettes
With the SV220 baseline in the bank, I moved to the real goal: combining all three filters to produce HOO and SHO palettes with LPRO stars. This is where multi-filter OSC imaging actually pays off.
The D2 went through the same linear pipeline as the SV220, but with different SPCC parameters to match its bandpass. Unlike the SV220’s symmetric 7nm bands, the Askar D2 is asymmetric: 8.5nm for SII at 672nm, 6.5nm for OIII at 500.7nm. Askar did this intentionally — the wider SII bandpass helps balance SNR between the typically weaker SII signal and stronger OIII. Getting these numbers right in SPCC matters for accurate color calibration.
The LPRO needed a completely different SPCC configuration. It’s a broadband light-pollution filter, not narrowband, so the “Narrowband filters mode” checkbox has to come off, and the QE curve changes to “Sony Color Sensor G-UVIRcut.” The LPRO’s broadband signal gave me dramatically more matched stars in SPCC than either narrowband filter — worth noting for anyone doing color calibration on dim narrowband data.
With all three linear masters calibrated, denoised, and ready, the combination work began. ChannelExtraction on the SV220 gave me an Hα master (red channel) and two OIII channels (green and blue Bayer pixels). The same on the D2 produced an SII master and two more OIII channels. That’s four separate OIII extractions across the two filters — which turned into one of the most useful tricks in this whole workflow.
Averaging all four OIII channels via PixelMath produced a combined OIII master with noticeably better signal-to-noise than any single extraction alone:
(OIII_sv220_g + OIII_sv220_b + OIII_d2_g + OIII_d2_b) / 4For a target where OIII is the weaker signal (which describes most Hα-dominant nebulae), combining OIII data across multiple filters makes a real visible difference. This alone might be worth running a second duo-band filter for.
ChannelCombination built the palettes. HOO took Hα into the red channel and OIII_combined into both green and blue — the classic bicolor mapping. SHO (the Hubble palette) put SII in red, Hα in green, and OIII_combined in blue.
Bills Stretch handled both combined images at 0.10 / 1.03 / 2.8. The HOO came out immediately gorgeous — the extra OIII data from combining two filters revealed structure that had only been hinted at in the single-filter version. Subtle teal threads weaving through the Hα reds, especially in the nebula’s brighter knots.
SHO took more work. It always does. Hα sits in the green channel in SHO, and Hα is typically the strongest signal, so SHO images start out wildly green-dominant. A few SCNR passes at 0.5 amount tamed the green cast, and selective saturation boosts in red and blue brought out the classic gold-and-teal Hubble look. The reward was a completely different emotional register for the same target — deep oranges from SII setting off cool OIII blues where the previous HOO had been all pink and cyan. The final SHO result is available on AstroBin with full technical details.
The Final Touch: Stars That Look Like Suns
Narrowband stars are a known problem. Duo-band filters pass only two narrow emission bands, and stars — which are broadband thermal emitters — come through looking magenta, washed out, or otherwise unnatural. It’s the narrowband tax you pay for nebula contrast.
This is exactly where the LPRO earned its place in the stack. Stretching the LPRO through Bills Stretch and running StarXTerminator gave me a star layer with natural stellar colors — real differentiation between hot blue stars and cool red ones, the kind of stars that make an astrophotograph look like an actual night sky instead of a processed render.
Swapping the LPRO stars into both the HOO and SHO recombinations transformed both finals. A simple PixelMath screen blend — ~((~LPRO_stars)*(~starless)) — was all it took. Stars went from washed-out magenta blobs to varied stellar temperatures scattered naturally across the Hα clouds. This is something I’ll do for every narrowband project going forward where I have any broadband data to draw from.
Lessons Worth Keeping
Trust pixel values, not stretched previews. Auto-STF can make a perfectly good GraXpert background model look like garbage, or vice versa. When evaluating gradient correction (or really any linear-stage work), check actual pixel values. The visual preview is lying to you about contrast.
SPCC bandwidths are filter-specific. The SV220 is 7nm symmetric, the D2 is 8.5nm SII / 6.5nm OIII asymmetric, and the LPRO isn’t narrowband at all. Each one needs its own SPCC configuration. Saving these as process icons in PixInsight is worth the five minutes it takes.
Combine OIII across filters when you can. If you shoot with two duo-bands that both pass OIII, averaging all four Bayer-extracted OIII channels produces visibly cleaner combined data than either filter alone. Free SNR from math you already have the data for.
Run BXT in Correct Only mode before SPCC. Tighter stars mean more photometric matches mean better color calibration. Then run BXT again for full deconvolution after SPCC is done.
Use broadband stars for narrowband finals. The difference between narrowband stars and LPRO stars is genuinely dramatic. Worth the extra hour or two of broadband integration just for the star layer.
What’s Next
One thing I didn’t do on this project, but want to try next time: using the LPRO starless as a luminance layer via LRGBCombination. The idea is to layer the LPRO’s high-SNR broadband detail underneath the narrowband color information, potentially pulling out even more fine structure than either palette alone can show. A future post, probably after the next clear stretch of nights.
For now, Lower’s Nebula stands as proof that the quieter corners of the winter sky reward patience and the right workflow. 28 hours of integration, three filters, two palettes, and one genuinely surprising target.
Clear skies!