Learning how to capture M31 from rainy, Bortle 6 Belfast

I pointed a Seestar S30 Pro at the Andromeda Galaxy from Belfast, where the sky is Bortle 6 and most nights it rains. Below is the best picture I had on each date since first light, then what each change did on its own.

Timeline

Each step is the best picture I had on that date, made with that date's frames and code. Before is the previous step, and frames and minutes are running totals. Steps that came out worse stay in, with a note on what went wrong. Drag the line, or tab to it and use the arrow keys.

2026-09-08 · processing

Step 01: My first pipeline

regenerated from the 8 Sep code (commit 0aaa0ac)

running total 427 frames · 71 min

Before (left): First light: Seestar live stack; total 479 fr · 80 min; 17.0″ · SNR 19.4
After (right): First pipeline, 8 Sep; total 427 fr · 71 min; 14.4″ · SNR 11.1

What changed: I copied the 7/8 Sep subs off the scope and ran my own first pipeline on them: frame QA, Siril drizzle stacking weighted towards the sharpest frames, and GraXpert's AI background removal.
What it did: Stars shrank from 17.0″ to 14.4″, but SNR at my eight test points in the disc fell from 19.41 to 11.07. Culling and sharpness weighting threw signal away, and the AI background model took part of the halo with it.

aligned on 301 stars, median residual 0.21 px

2026-09-09 · processing

Step 02: After the review

regenerated from the 9 Sep code (commit 9f622a1)

running total 467 frames · 78 min

Before (left): First pipeline, 8 Sep; total 427 fr · 71 min; 14.4″ · SNR 11.1
After (right): After the review, 9 Sep; total 467 fr · 78 min; 15.5″ · SNR 16.5

What changed: The next day I had the pipeline reviewed against this night's data and acted on what it measured: no trend-based frame culling, weighting by star count instead of sharpness, a plain plane fit for the background, and star removal with Dark Star, including a fix for the dark ring it left around the core.
What it did: With 467 frames instead of 427, SNR went from 11.07 to 16.48 (+49 %) and stars grew from 14.4″ to 15.5″. The stack is bit for bit what today's pipeline makes from this night.

aligned on 301 stars, median residual 0.21 px

2026-10-16 · processing + new night

Step 03: A second night, under a full moon

saved run history, 16 Oct (commit 63e7708 + uncommitted changes)

running total 995 frames · 2 h 46 min

Before (left): After the review, 9 Sep; total 467 fr · 78 min; 15.5″ · SNR 16.5
After (right): Two nights, 16 Oct; total 995 fr · 2 h 46 min; 13.3″ · SNR 19.0

What changed: I added 529 frames from 25/26 Sep, shot under a 99 % moon while the rest were still on the scope, and stacked both nights frame by frame with noise weighting.
What it did: SNR rose from 16.48 to 18.96 and stars tightened from 15.5″ to 13.3″, from 166 minutes stacked.

aligned on 338 stars, median residual 0.20 px

2026-10-17 · processing + new night

Step 04: Three nights

today's pipeline, 17 Oct (commit 6d0dcc5)

running total 1,841 frames · 5 h 07 min

Before (left): Two nights, 16 Oct; total 995 fr · 2 h 46 min; 13.3″ · SNR 19.0
After (right): Three nights, 17 Oct; total 1,841 fr · 5 h 07 min; 11.9″ · SNR 23.4

What changed: I added 16/17 Oct and the rest of 25/26 Sep, then reprocessed all three nights with today's pipeline, which also changed the colour calibration.
What it did: 1841 frames and 307 minutes took SNR from 18.96 to 23.37 and stars from 13.3″ to 11.9″.

aligned on 338 stars, median residual 0.13 px

Watch the stack grow, step by step →

What each change did

I reprocessed these today so each comparison isolates one change. Before is on the left and after is on the right, aligned on the same stars.

Stage 00: Measure, don't guess

2026-09-07

Before (left): First pipeline, 8 Sep; 427 fr · 71 min; 14.4″ · SNR 11.1
After (right): 7/8 Sep, today's pipeline; 467 fr · 78 min; 15.5″ · SNR 16.5

What changed: Reprocessed today: the 7/8 Sep frames through my first pipeline and through today's, which keeps almost every frame, weights them by star count and fits the background with a flat plane instead of an AI model.
What it did: On the same night SNR went from 11.07 to 16.48 (+49 %), and stars grew from 14.4″ to 15.5″.

aligned on 292 stars, median residual 0.21 px

Stage 01: The full moon night

2026-09-07, 2026-09-25 · nights 2026-09-07 + 2026-09-25

Before (left): 7/8 Sep, today's pipeline; 467 fr · 78 min; 15.5″ · SNR 16.5
After (right): 25/26 Sep alone; 998 fr · 2 h 46 min; 10.6″ · SNR 13.4

What changed: Reprocessed today with the same pipeline: the dark 7/8 Sep night against all of 25/26 Sep, under a 99 % moon.
What it did: Moonlight raised the frame background to 1882 ADU from 1326, so 166 minutes gave SNR 13.41 where 78 dark minutes gave 16.48. Stars came out at 10.6″ against 15.5″.

aligned on 292 stars, median residual 0.17 px

Stage 02: A ring around the core

2026-09-25

Before (left): 25/26 Sep, old core protection; 998 fr · 2 h 46 min; 10.6″ · SNR 13.4
After (right): 25/26 Sep alone; 998 fr · 2 h 46 min; 10.6″ · SNR 13.4

What changed: Reprocessed today on all of 25/26 Sep: star removal with the old core protection, a fixed brightness threshold, against the new one, set relative to the nucleus.
What it did: The old threshold left a dark ring 2.2′ from the nucleus, where the brightness dipped 16.0 %. With the new one no dip is deeper than 1.6 %.

aligned on 328 stars, median residual 0.11 px

Stage 03: Stacking nights

2026-09-07, 2026-09-25, 2026-10-16 · nights 2026-09-07 + 2026-09-25 + 2026-10-16

Before (left): 7/8 Sep, today's pipeline; 467 fr · 78 min; 15.5″ · SNR 16.5
After (right): Three nights, 17 Oct; 1,841 fr · 5 h 07 min; 11.9″ · SNR 23.4

What changed: Reprocessed today: my best single night by SNR, 7/8 Sep, against all three nights stacked frame by frame with noise weighting.
What it did: 1841 frames over 307 minutes lifted SNR from 16.48 to 23.37 (+42 %). Minutes alone would predict +99 %, but the moonlit frames are noisier.

aligned on 292 stars, median residual 0.17 px

Stage 04: Fixing the lavender

2026-09-07, 2026-09-25 · nights 2026-09-07 + 2026-09-25

Before (left): Two nights, NOMAD colour; 995 fr · 2 h 46 min; 13.3″ · SNR 19.0
After (right): Two nights, current colour; 995 fr · 2 h 46 min; 13.3″ · SNR 18.9

What changed: Reprocessed today on the two-night stack: Siril's photometric calibration against the NOMAD catalogue, as I first ran it, against local Gaia with a correction to the spectrophotometric balance.
What it did: The disc's blue-to-green ratio went from 1.368 to 0.988 and red-to-green from 1.079 to 1.202, which takes out the lavender cast.

aligned on 337 stars, median residual 0.12 px

The nights

Most nights here are cloudy. These are the clear ones, with Belfast's weather for the hours I was imaging and what the frames recorded.

2026-09-07 23:21–01:14 UTC 7/8 Sep 2026

frames

467 × 10 s

FWHM

16.2″

moon

12 %

gusts

28 mph

cloud

4 %

dropped

22 %

QA out

12

Evan and Dad watched a movie but kept coming out to check the stacked image off the scope on the Seestar app.

23:21–01:14 UTC · 479 × 10 s · moon 12 % below horizon · cloud 4 % · gusts 28 mph · 22 % dropped
2026-09-25 21:35–03:43 UTC 25/26 Sep 2026

frames

999 × 10 s

FWHM

11.7″

moon

99 %

gusts

29 mph

cloud

14 %

dropped

34 %

QA out

250

Session with Quinny and Erin.

21:35–03:43 UTC · 1249 × 10 s · moon 99 % up to 34° · cloud 14 % · gusts 29 mph · 34 % dropped
2026-10-16 19:45–21:22 UTC 16/17 Oct 2026

frames

376 × 10 s

FWHM

10.3″

moon

33 %

gusts

15 mph

cloud

3 %

dropped

27 %

QA out

8

Evan and Dad watched half of Fall. 6/10.

19:45–21:22 UTC · 384 × 10 s · moon 33 % below horizon · cloud 3 % · gusts 15 mph · 27 % dropped

Clear nights

Belfast's cloud for every night since I started. The dark squares are the clear ones.

9 of 46 nights were usable (mean cloud <= 30 % between 21:00-03:00 UTC), and I imaged 3 of them. 0 % 50 % 100 % cloud usable imaged

Cloud cover by night
Night	Cloud %	Usable	Imaged
1 Sept	66	no	no
2 Sept	79	no	no
3 Sept	100	no	no
4 Sept	36	no	no
5 Sept	99	no	no
6 Sept	88	no	no
7 Sept	22	yes	yes
8 Sept	11	yes	no
9 Sept	100	no	no
10 Sept	47	no	no
11 Sept	97	no	no
12 Sept	100	no	no
13 Sept	81	no	no
14 Sept	50	no	no
15 Sept	41	no	no
16 Sept	61	no	no
17 Sept	68	no	no
18 Sept	98	no	no
19 Sept	13	yes	no
20 Sept	97	no	no
21 Sept	57	no	no
22 Sept	100	no	no
23 Sept	94	no	no
24 Sept	100	no	no
25 Sept	14	yes	yes
26 Sept	51	no	no
27 Sept	55	no	no
28 Sept	85	no	no
29 Sept	100	no	no
30 Sept	1	yes	no
1 Oct	93	no	no
2 Oct	39	no	no
3 Oct	65	no	no
4 Oct	100	no	no
5 Oct	100	no	no
6 Oct	37	no	no
7 Oct	18	yes	no
8 Oct	100	no	no
9 Oct	56	no	no
10 Oct	6	yes	no
11 Oct	100	no	no
12 Oct	95	no	no
13 Oct	58	no	no
14 Oct	100	no	no
15 Oct	6	yes	no
16 Oct	13	yes	yes

Where the frames went

The scope skips some exposures and my QA throws out a few more. On the moonlit night a third of the slots never became frames.

7/8 Sep 2026

467

613

25/26 Sep 2026

998

1,929

16/17 Oct 2026

376

527

One slot is one 10 s exposure the scope should have taken. Bar ends show the total slots. stacked rejected (QA, plus 1 the stacker left out) dropped by the scope

Exposure slots by night
Night	Slots	Saved	Dropped by the scope	Rejected	Stacked
7/8 Sep 2026	613	479	134	12	467
25/26 Sep 2026	1,929	1,249	680	251	998
16/17 Oct 2026	527	384	143	8	376

Share dropped against city-centre gusts, one point per night-hour (r = 0.15 over 13 hours). No relationship: I don't know yet why the scope drops frames.

Gusts against drop rate by hour
Night, hour (UTC)	Gusts, mph	Dropped	Slots
7/8 Sep 2026, 23:00	26	20 %	206
7/8 Sep 2026, 00:00	26	23 %	330
7/8 Sep 2026, 01:00	26	20 %	77
25/26 Sep 2026, 21:00	25	23 %	131
25/26 Sep 2026, 22:00	28	25 %	329
25/26 Sep 2026, 23:00	29	24 %	322
25/26 Sep 2026, 00:00	29	28 %	261
25/26 Sep 2026, 01:00	28	35 %	327
25/26 Sep 2026, 02:00	27	51 %	325
25/26 Sep 2026, 03:00	26	59 %	234
16/17 Oct 2026, 19:00	13	22 %	78
16/17 Oct 2026, 20:00	13	28 %	322
16/17 Oct 2026, 21:00	10	29 %	127

Focus through the night

On 7/8 Sep the stars grew all night while the focuser never moved. Autofocus mid-session is the fix I haven't tried yet.

7/8 Sep 2026 13.0″ → 17.0″

25/26 Sep 2026 10.0″ → 12.3″

16/17 Oct 2026 10.2″ → 10.4″

Median star size (FWHM) through each night, from the frames QA kept, on one shared scale. Hours count from the first frame. Lower is sharper. 7/8 Sep drifted 4.0″ with the focuser never moving; 16/17 Oct held within 0.2″.

Star size at the start and end of each night
Night	First 15 min	Last 15 min	Night median
7/8 Sep 2026	13.0″	17.0″	16.2″
25/26 Sep 2026	10.0″	12.3″	11.6″
16/17 Oct 2026	10.2″	10.4″	10.3″

Signal against time

With nothing else changed, signal-to-noise grows with the square root of time, which is the faint curve. Points under it lost something along the way. Points over it gained more from processing than the extra minutes alone would give.

Filled points are at or above the √t line through step 01; hollow ones fall more than 5 % short of it. The solid line is first light, the scope's own stack.

SNR by step
Step	Minutes	SNR	√t expectation
01 First pipeline, 8 Sep	71	11.1	11.1
02 After the review, 9 Sep	78	16.5	11.6
03 Two nights, 16 Oct	166	19.0	16.9
04 Three nights, 17 Oct	307	23.4	23.0