![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F11013868%2Faa9c4b9cb5a3eb0e85b951de5549ffc7%2Fbug.png?generation=1757365106586635&alt=media\")
There seems to be a problem with the data, where the order of the spectrum for AIRS and the ground truth are reversed compared to each other. Is this a bug, or is it intentional, and if it is intentional, where is it explained?
\nI made a minimal example showing the issue in this small notebook. The notebook makes a very simplistic calibration for planet 43278385 and then estimates the spectrum with a ratio of values inside and outside the transit window. I also plot the ground truth and the reversed ground truth, and it is clear that the reversed ground truth is a better match. I centered the graphs to make the match clearer.
\nHere is the figure from the notebook:
\n
This is the wavelength vs wavenumber issue. Infrared spectra typically use wavenumber because it's proportional to energy (E=hcν̃), making molecular transitions evenly spaced - more intuitive for analysis. They're inverse: wavenumber=10000/wavelength.
\nWhen wavelength increases, wavenumber decreases, so the spectrum looks flipped. Just reverse one and they'll match.
Yes this is intentional. You can find in the file axis_info.paquet the wavelengths corresponding to axis 2 of the channel AIRS-CH0 and you will see they are reversed compared to the ground truth wavelengths in wavelengths.csv.
", "rawMarkdown": "Yes this is intentional. You can find in the file axis_info.paquet the wavelengths corresponding to axis 2 of the channel AIRS-CH0 and you will see they are reversed compared to the ground truth wavelengths in wavelengths.csv.", "votes": 2, "replies": [ { "id": 3285902, "postDate": "2025-09-08T21:38:07.620Z", "content": "I see. Thanks!
", "rawMarkdown": "I see. Thanks!" } ] }, { "id": 3285891, "postDate": "2025-09-08T21:00:01.620Z", "content": "There seems to be a problem with the data, where the order of the spectrum for AIRS and the ground truth are reversed compared to each other. Is this a bug, or is it intentional, and if it is intentional, where is it explained?
\nI made a minimal example showing the issue in this small notebook. The notebook makes a very simplistic calibration for planet 43278385 and then estimates the spectrum with a ratio of values inside and outside the transit window. I also plot the ground truth and the reversed ground truth, and it is clear that the reversed ground truth is a better match. I centered the graphs to make the match clearer.
\nHere is the figure from the notebook:
\n
This is the wavelength vs wavenumber issue. Infrared spectra typically use wavenumber because it's proportional to energy (E=hcν̃), making molecular transitions evenly spaced - more intuitive for analysis. They're inverse: wavenumber=10000/wavelength.
\nWhen wavelength increases, wavenumber decreases, so the spectrum looks flipped. Just reverse one and they'll match.
Yes this is intentional. You can find in the file axis_info.paquet the wavelengths corresponding to axis 2 of the channel AIRS-CH0 and you will see they are reversed compared to the ground truth wavelengths in wavelengths.csv.
", "votes": 2, "replies": [ { "id": 3285902, "author_name": "Thomas Dueholm Hansen", "author_url": "", "post_date": "2025-09-08T21:38:07.620000", "content": "I see. Thanks!
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "3286357": "This is the wavelength vs wavenumber issue. Infrared spectra typically use wavenumber because it's proportional to energy (E=hcν̃), making molecular transitions evenly spaced - more intuitive for analysis. They're inverse: wavenumber=10000/wavelength.\nWhen wavelength increases, wavenumber decreases, so the spectrum looks flipped. Just reverse one and they'll match.", "3285899": "Yes this is intentional. You can find in the file axis_info.paquet the wavelengths corresponding to axis 2 of the channel AIRS-CH0 and you will see they are reversed compared to the ground truth wavelengths in wavelengths.csv.", "3285891": "There seems to be a problem with the data, where the order of the spectrum for AIRS and the ground truth are reversed compared to each other. Is this a bug, or is it intentional, and if it is intentional, where is it explained?\n\nI made a minimal example showing the issue in [this small notebook](https://www.kaggle.com/code/thomasdueholmhansen/airs-spectrum-indices-should-be-flipped). The notebook makes a very simplistic calibration for planet 43278385 and then estimates the spectrum with a ratio of values inside and outside the transit window. I also plot the ground truth and the reversed ground truth, and it is clear that the reversed ground truth is a better match. I centered the graphs to make the match clearer.\n\nHere is the figure from the notebook:\n\n" } }