Hi @martin and everyone! I’m a new AI/ML AWG member, an aerospace engineer (MS, Virginia Tech) with an ML/data analysis background from experimental fluid dynamics research. The SAA/GCR separation project caught my eye, and I’d really like to learn more about it. I’m new to radiation dosimetry, but I’m comfortable with time-series data and classification methods and eager to learn the domain from the group. Happy to start small. Would love any guidance on how to jump in.
We’re on the final stretch before the 4 Aug deadline. Let’s finish strong and make this project a success! Every completed day and every uploaded CSV counts.
Hi everyone, I noticed that some SAA passages cross the UTC day boundary—for example, a passage may begin around 23:57 on one day and continue until about 00:03 the next day. Since the tool labels each day separately, this would create two saved intervals even though physically it is one continuous SAA passage.
Should we label each visible segment within its respective day and let them be merged during later processing, or is there another convention we should follow for passages that cross midnight? I wanted to clarify this as we are nearing completion of the manual-labeling task. Thank you!
A huge thank you to everyone who has contributed to labeling SAA passes over the past weeks. Your effort is directly helping us build a solid ground truth dataset for SAA/GCR separation.
Special recognition goes to IdilDogaTurkmen, who labeled data on 44 days, covering 56.4% of all labeled days and logging 56h 18m of labeled time. This is by far the largest contribution to the effort so far. Amazing work!
Thanks as well to everyone else who pitched in:
Volunteer
Days
Rows
% of days
Labeled time
IdilDogaTurkmen
44
18,989
56.4%
56h 18m
ShayanGhosh
10
5,736
12.8%
16h 58m
HitaeshiSehgal
9
10,234
11.5%
30h 14m
ShivamKumarSharma
6
1,615
7.7%
4h 47m
Kakona
5
1,923
6.4%
5h 42m
RaviRanjan
4
4,603
5.1%
13h 37m
Raj taware
1
250
1.3%
0h 44m
Every single labeled pass matters, so thank you all for the time and care you have put into this.
Next, we would like to show how the volunteer labeled SAA passes compare to NASA SRAG’s own SAA crossing data. You can find the comparison here:
The comparison shows good overall agreement, but it also shows that some of our labeled passes will need to be revisited and corrected. We will follow up with details on which entries need a second look.
Thanks again for all your hard work. This dataset would not exist without you.
During data review, we found that many SAA passes are missing.
In addition, we’ve encountered stability issues with the labeling application.
We’re working on resolving both problems. If anyone would like to help with labeling the remaining passes, we’d appreciate it.
At the next main monthly @AIMLawg AWG mtg would you be able to present this project, its background, methods, status, and the ask/what/how many AIML AWG member you are looking for to contribute (and their task/skills/etc)?
Hi Team, next week I will briefly present our work as part of my talk (Asymmetry in SPACEDOS Measurements During SAA Transits) at WRMISS 2026 in Prague.
I will inform the radiation measurement community about our ongoing project.
Hi, thanks for the great catch! Let’s go with your suggestion and label each visible segment within its respective day separately. We can easily merge them later during post-processing.Could you please drop the list of the cases you found here in the chat?
Thank you for the response. I checked through my uploaded CSVs and found four cases where what appears to be one SAA passage was split across the UTC day boundary:
Dec 10 → Dec 11: 23:47:57–23:58:40 / 00:00:05–00:07:10
Dec 12 → Dec 13: 23:51:21–23:58:33 / 00:00:08–00:05:28
Dec 14 → Dec 15: 23:52:32–23:59:41 / 00:07:07–00:12:30
Dec 16 → Dec 17: 23:54:15–23:59:34 / 00:00:06–00:07:17
These are all UTC. I kept the two segments separately in their respective daily CSVs as discussed, so they can be merged during post-processing.
We are starting a new round of manual labeling for the SAA/GCR Separation Using ML project, open until the end of September 2026.
The process is simple: select any day, complete the labeling, export the CSV, and upload it to the shared folder. There is no need to reserve a day, and multiple volunteers can label the same day.
You are welcome to label as many days as your availability allows.
We also welcome anyone interested in contributing to data validation, analysis, ML/model development, software, visualization, or future research activities.
Thank you to everyone contributing to the project!
Just wanted to update that I’ve completed all 78 days for this round and uploaded the CSVs following the naming format.
Let me know if I can help with anything else, data validation, analysis, model development, or anything related to the project, I’d be most happy to help wherever it’s useful.
Hi @odylec and @martin, I’d like to help with the data-validation side of this round. Would a small audit of the exported label intervals be useful—checking start/end ordering, duplicate or overlapping intervals within each annotator’s file, and candidate passage pairs across UTC midnight, including the four cases mentioned above? Different volunteers labeling the same day would be kept separate.
I can return a flagged-interval CSV, a short reproducible Python checker, and a brief note showing exactly which source files and rows were checked. Original annotations would stay intact, with any suggested cross-day pairing left for your review.
I have a local pilot for timestamp and spacing checks on public DosTel1/DosTel2 RadLab data. It exposed unresolved duplicate/cadence questions, which I’ve sent to OSDR support; I’m not interpreting the numerical integration as physical dose.
If this would fill a gap in existing validation, please point me to a small starting batch and preferred place to submit the checker. Happy to adapt to the team’s current approach.
Hi @martin and @odylec, I am a new AWG member and I’m interested in contributing to this project’s data analysis/Ml development/software aspects. I’d appreciate a little guidance on how I can help.
