{
  "id": 593681,
  "title": "How is transit depth determined for 0.7µm from FGS1's measurements?",
  "url": "/competitions/ariel-data-challenge-2025/discussion/593681",
  "author_name": "Nishan Mann",
  "post_date": "2025-07-30T02:27:09.962000",
  "votes": 1,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Hello everyone, this is what I know so far</p>\n<ul>\n<li>From <a href=\"https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook\" target=\"_blank\">https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook</a>, we see that transit depth for 0.7µm is extracted from FGS1 measuresments and the rest from AIRS's measurements.</li>\n<li>Using calibration notebook <a href=\"https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data\" target=\"_blank\">https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data</a>, the processed data ends up with dimensions (num planets, time steps, wavelength index, space index) For e.g  FGS1: (71, 187, 32, 32) and AIRS-CH0: (71, 187, 282, 32) </li>\n<li>In <code>wavelengths.csv</code> there are 283 wavelengths. So I safely assume that 1 wavelength (<code>wl_1</code>) comes from FGS1 and 282 from AIRS.</li>\n</ul>\n<p>To extract the transit depth for 282 AIRS wavelengths we can:</p>\n<ul>\n<li>use either the image (187, λ, 32) or transit curve (187, λ) obtained by summing over space dimension to infer the transit depth as a function of λ .</li>\n</ul>\n<p>How is this done when it comes to using FGS1's transit data? My guess is:</p>\n<ul>\n<li>use either a) whole cube (187, 32, 32) or b) transit curve (187, ) obtained by summing over space and wavelengths to infer the transit depth for  0.7µm<br>\nIs it a) or b) or something else</li>\n</ul>",
  "messages": [
    {
      "id": 3258221,
      "postDate": "2025-07-30T11:09:31.963Z",
      "content": "<p>For the FGS1 sensor, 32 by 32 refers only to the spatial matrix (it’s not related to wavelengths, since this sensor has only one wavelength — as you mentioned, 0.7).<br>\nI’ve seen several approaches:</p>\n<ul>\n<li>Averaging over the entire matrix (32×32)</li>\n<li>Averaging only over the central part of the matrix (for example, 8×8 or 16×16, etc.). If you look at the first link (perfect-eda-doesn-t-exist) at the end, you’ll see that the signal is mainly captured from the central pixels of this matrix.</li>\n<li>… Some other option?</li>\n</ul>\n<p>Apparently, the best choice can be determined experimentally.</p>",
      "rawMarkdown": "For the FGS1 sensor, 32 by 32 refers only to the spatial matrix (it’s not related to wavelengths, since this sensor has only one wavelength — as you mentioned, 0.7).\nI’ve seen several approaches:\n- Averaging over the entire matrix (32×32)\n- Averaging only over the central part of the matrix (for example, 8×8 or 16×16, etc.). If you look at the first link (perfect-eda-doesn-t-exist) at the end, you’ll see that the signal is mainly captured from the central pixels of this matrix.\n- ... Some other option?\n\nApparently, the best choice can be determined experimentally.",
      "votes": 1,
      "replies": [
        {
          "id": 3258274,
          "postDate": "2025-07-30T13:28:34.177Z",
          "content": "<p>Ah, thanks for the clarification on the dimensions of the FGS1 sensor data Pavel. I had misunderstood!</p>",
          "rawMarkdown": "Ah, thanks for the clarification on the dimensions of the FGS1 sensor data Pavel. I had misunderstood!"
        },
        {
          "id": 3274833,
          "postDate": "2025-08-25T13:49:46.567Z",
          "content": "<p>yes - the PSF of FGS1 sensor is not uniform in shape (if you plot one of the series of images you will know what i mean), how to fully capture that information (or useful signal against the background), could be important.</p>",
          "rawMarkdown": "yes - the PSF of FGS1 sensor is not uniform in shape (if you plot one of the series of images you will know what i mean), how to fully capture that information (or useful signal against the background), could be important.\n",
          "votes": 1
        }
      ]
    },
    {
      "id": 3258056,
      "postDate": "2025-07-30T02:27:09.963Z",
      "content": "<p>Hello everyone, this is what I know so far</p>\n<ul>\n<li>From <a href=\"https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook\" target=\"_blank\">https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook</a>, we see that transit depth for 0.7µm is extracted from FGS1 measuresments and the rest from AIRS's measurements.</li>\n<li>Using calibration notebook <a href=\"https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data\" target=\"_blank\">https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data</a>, the processed data ends up with dimensions (num planets, time steps, wavelength index, space index) For e.g  FGS1: (71, 187, 32, 32) and AIRS-CH0: (71, 187, 282, 32) </li>\n<li>In <code>wavelengths.csv</code> there are 283 wavelengths. So I safely assume that 1 wavelength (<code>wl_1</code>) comes from FGS1 and 282 from AIRS.</li>\n</ul>\n<p>To extract the transit depth for 282 AIRS wavelengths we can:</p>\n<ul>\n<li>use either the image (187, λ, 32) or transit curve (187, λ) obtained by summing over space dimension to infer the transit depth as a function of λ .</li>\n</ul>\n<p>How is this done when it comes to using FGS1's transit data? My guess is:</p>\n<ul>\n<li>use either a) whole cube (187, 32, 32) or b) transit curve (187, ) obtained by summing over space and wavelengths to infer the transit depth for  0.7µm<br>\nIs it a) or b) or something else</li>\n</ul>",
      "rawMarkdown": "Hello everyone, this is what I know so far\n- From https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook, we see that transit depth for 0.7µm is extracted from FGS1 measuresments and the rest from AIRS's measurements.\n- Using calibration notebook https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data, the processed data ends up with dimensions (num planets, time steps, wavelength index, space index) For e.g  FGS1: (71, 187, 32, 32) and AIRS-CH0: (71, 187, 282, 32) \n- In `wavelengths.csv` there are 283 wavelengths. So I safely assume that 1 wavelength (`wl_1`) comes from FGS1 and 282 from AIRS.\n\nTo extract the transit depth for 282 AIRS wavelengths we can:\n- use either the image (187, λ, 32) or transit curve (187, λ) obtained by summing over space dimension to infer the transit depth as a function of λ .\n\nHow is this done when it comes to using FGS1's transit data? My guess is:\n- use either a) whole cube (187, 32, 32) or b) transit curve (187, ) obtained by summing over space and wavelengths to infer the transit depth for  0.7µm\nIs it a) or b) or something else",
      "votes": 1
    }
  ],
  "comments": [
    {
      "id": 3258221,
      "author_name": "Pavel Orlov",
      "author_url": "",
      "post_date": "2025-07-30T11:09:31.963000",
      "content": "<p>For the FGS1 sensor, 32 by 32 refers only to the spatial matrix (it’s not related to wavelengths, since this sensor has only one wavelength — as you mentioned, 0.7).<br>\nI’ve seen several approaches:</p>\n<ul>\n<li>Averaging over the entire matrix (32×32)</li>\n<li>Averaging only over the central part of the matrix (for example, 8×8 or 16×16, etc.). If you look at the first link (perfect-eda-doesn-t-exist) at the end, you’ll see that the signal is mainly captured from the central pixels of this matrix.</li>\n<li>… Some other option?</li>\n</ul>\n<p>Apparently, the best choice can be determined experimentally.</p>",
      "votes": 1,
      "replies": [
        {
          "id": 3258274,
          "author_name": "Nishan Mann",
          "author_url": "",
          "post_date": "2025-07-30T13:28:34.177000",
          "content": "<p>Ah, thanks for the clarification on the dimensions of the FGS1 sensor data Pavel. I had misunderstood!</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 3274833,
          "author_name": "Gordon Yip",
          "author_url": "",
          "post_date": "2025-08-25T13:49:46.567000",
          "content": "<p>yes - the PSF of FGS1 sensor is not uniform in shape (if you plot one of the series of images you will know what i mean), how to fully capture that information (or useful signal against the background), could be important.</p>",
          "votes": 1,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3258221": "For the FGS1 sensor, 32 by 32 refers only to the spatial matrix (it’s not related to wavelengths, since this sensor has only one wavelength — as you mentioned, 0.7).\nI’ve seen several approaches:\n- Averaging over the entire matrix (32×32)\n- Averaging only over the central part of the matrix (for example, 8×8 or 16×16, etc.). If you look at the first link (perfect-eda-doesn-t-exist) at the end, you’ll see that the signal is mainly captured from the central pixels of this matrix.\n- ... Some other option?\n\nApparently, the best choice can be determined experimentally.",
    "3258056": "Hello everyone, this is what I know so far\n- From https://www.kaggle.com/code/lordpatil/perfect-eda-doesn-t-exist/notebook, we see that transit depth for 0.7µm is extracted from FGS1 measuresments and the rest from AIRS's measurements.\n- Using calibration notebook https://www.kaggle.com/code/gordonyip/calibrating-and-binning-ariel-data, the processed data ends up with dimensions (num planets, time steps, wavelength index, space index) For e.g  FGS1: (71, 187, 32, 32) and AIRS-CH0: (71, 187, 282, 32) \n- In `wavelengths.csv` there are 283 wavelengths. So I safely assume that 1 wavelength (`wl_1`) comes from FGS1 and 282 from AIRS.\n\nTo extract the transit depth for 282 AIRS wavelengths we can:\n- use either the image (187, λ, 32) or transit curve (187, λ) obtained by summing over space dimension to infer the transit depth as a function of λ .\n\nHow is this done when it comes to using FGS1's transit data? My guess is:\n- use either a) whole cube (187, 32, 32) or b) transit curve (187, ) obtained by summing over space and wavelengths to infer the transit depth for  0.7µm\nIs it a) or b) or something else"
  }
}