Mapping Saturn using deep learning


Clouds and aerosols provide unique insight into the chemical and physical processes of gas-giant planets. Mapping and characterizing the spectral features indicative of the cloud structure and composition enables an understanding of a planet’s energy budget, chemistry and atmospheric dynamics1,2,3,4. Current space missions to Solar System planets produce high-quality datasets, yet the sheer amount of data obtained often prohibits detailed ‘by hand’ analyses. Current techniques rely mainly on two approaches: identifying the existence of spectral features by dividing the fluxes of two or more spectral channels, and performing full radiative transfer calculations for individual spectra. The first method is not sufficiently accurate and the second is not easily scalable to the entire planetary surface. Here we have developed a deep learning algorithm, PlanetNet, that is able to quickly and accurately map spatial and spectral features across large, heterogeneous areas of a planet. We use PlanetNet to delineate the major components of the 2008 storm on Saturn5, enhancing the scope of the area previously studied and indicating regions that can be probed more deeply with radiative transfer models. Our spectral-component maps indicate compositional and cloud variations of the vast region affected by the storm, showing regions of vertical upwelling, and diminished clouds at the centre of compact substorms.

Access options

Rent or Buy article

Get time limited or full article access on ReadCube.


All prices are NET prices.

Fig. 1: Flowchart of the PlanetNet algorithm.
Fig. 2: Spatial and spectral characteristics of the PlanetNet identified features.
Fig. 3: Cloud distribution as mapped by PlanetNet across six overlapping datasets.
Fig. 4: SR as mapped by PlanetNet across six overlapping datasets.

Data availability

The data analysed in this work are available through the Planetary Data System ( In addition, the data used for training PlanetNet are permanently archived and can be accessed with the permanent link: or the DOI:

Code availability

PlanetNet is publicly available through the UCL-Exoplanets GitHub page ( In addition, the code is permanently archived and can be accessed with the permanent link: or the DOI:


  1. 1.

    Fouchet, T. et al. Stratospheric aftermath of the 2010 storm on Saturn as observed by the TEXES instrument. I. Temperature structure. Icarus 277, 196–214 (2016).

    ADS  Article  Google Scholar 

  2. 2.

    Fletcher, L. N. et al. Thermal structure and dynamics of Saturn’s northern springtime disturbance. Science 332, 1413 (2011).

    ADS  Article  Google Scholar 

  3. 3.

    Barstow, J. K., Irwin, P. G. J., Fletcher, L. N., Giles, R. S. & Merlet, C. Probing Saturn’s tropospheric cloud with Cassini/VIMS. Icarus 271, 400–417 (2016).

    ADS  Article  Google Scholar 

  4. 4.

    Sánchez-Lavega, A. et al. An enduring rapidly moving storm as a guide to Saturn’s equatorial jet’s complex structure. Nature Commun. 7, 13262 (2016).

    ADS  Article  Google Scholar 

  5. 5.

    Baines, K. H. et al. Storm clouds on Saturn: lightning-induced chemistry and associated materials consistent with Cassini/VIMS spectra. Planet. Space Sci. 57, 1650–1658 (2009).

    ADS  Article  Google Scholar 

  6. 6.

    Brown, R. H. et al. The Cassini visual and infrared mapping spectrometer (VIMS) investigation. Space Sci. Rev. 115, 111–168 (2004).

    ADS  Article  Google Scholar 

  7. 7.

    Yu, S. X. & Shi, J. in Proc. Ninth IEEE Int. Conf. on Computer Vision Vol. 1, 313–319 (IEEE, 2003).

  8. 8.

    Wu, H. & Prasad, S. Convolutional recurrent neural networks for hyperspectral data classification. Remote Sens. 9, 298 (2017).

    ADS  Article  Google Scholar 

  9. 9.

    Yang, J., Zhao, Y., Chan, J. C. W. & Yi, C. in 2016 IEEE Int. Geosci. Remote Sensing Symp. 5079–5082 (IEEE, 2016).

  10. 10.

    Zhu, X. X. et al. Deep learning in remote sensing: a comprehensive review and list of resources. IEEE Geosci. Remote Sens. Mag. 5, 8–36 (2017).

    Article  Google Scholar 

  11. 11.

    Jolliffe, I. Principal Component Analysis (Springer, 2002).

  12. 12.

    Gurnett, D. A. et al. Radio and plasma wave observations at Saturn from Cassini’s approach and first orbit. Science 307, 1255–1259 (2005).

    ADS  Article  Google Scholar 

  13. 13.

    Fischer, G. et al. Analysis of a giant lightning storm on Saturn. Icarus 190, 528–544 (2007).

    ADS  Article  Google Scholar 

  14. 14.

    Fischer, G. et al. Atmospheric electricity at Saturn. Space Sci. Rev. 137, 271–285 (2008).

    ADS  Article  Google Scholar 

  15. 15.

    Baines, K. H., Carlson, R. W. & Kamp, L. W. Fresh ammonia ice clouds in Jupiter. I. Spectroscopic identification, spatial distribution, and dynamical implications. Icarus 159, 74–94 (2002).

    ADS  Article  Google Scholar 

  16. 16.

    Sromovsky, L. A., Baines, K. H. & Fry, P. M. Saturn’s great storm of 2010–2011: evidence for ammonia and water ices from analysis of VIMS spectra. Icarus 226, 402–418 (2013).

    ADS  Article  Google Scholar 

  17. 17.

    Bishop, C. M. Pattern Recognition and Machine Learning (Information Science and Statistics) (Springer, 2006).

  18. 18.

    Bengio, Y. Learning Deep Architectures for AI Vol. 2 (Now Publishers, 2009).

  19. 19.

    Goodfellow, I., Bengio, Y. & Courville, A. Deep Learning (MIT Press, 2016);

  20. 20.

    Abadi, M. et al. TensorFlow: Large-scale Machine Learning on Heterogeneous Systems (TensorFlow, 2015);

  21. 21.

    Yu, S. X. & Shi, J. in Proc. Ninth IEEE Int. Conf. Computer Vision (ICCV 03) Vol. 2, 313 (IEEE Computer Society, 2003).

  22. 22.

    Ng, A. Y., Jordan, M. I. & Weiss, Y. in Advances in Neural Information Processing Systems 849–856 (MIT Press, 2001).

  23. 23.

    Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Machine Learning Res. 12, 2825–2830 (2011).

    MathSciNet  MATH  Google Scholar 

  24. 24.

    Luxburg, U. A tutorial on spectral clustering. Stat. Comput. 17, 395–416 (2007).

    MathSciNet  Article  Google Scholar 

  25. 25.

    Griffith, C. A. et al. A corridor of exposed ice-rich bedrock across Titan’s tropical region. Nat. Astron. (2019).

Download references


This project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement number 758892, ExoAI) and under the European Union’s Seventh Framework Programme (FP7/2007–2013)/ERC grant agreement numbers 617119 (ExoLights). I.P.W. acknowledges funding by the Science and Technology Funding Council (STFC) (grants ST/K502406/1 and ST/P000282/1) and support from Microsoft Azure for Research cloud computing. C.A.G. is funded by the University of Arizona.

Author information




I.P.W. developed the PlanetNet algorithm and conducted the data analysis. C.A.G. provided the calibrated Cassini/VIMS data and led the spectral interpretation of the data.

Corresponding author

Correspondence to I. P. Waldmann.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Journal peer review information: Nature Astronomy thanks Mario D’Amore and the other anonymous reviewer(s) for their contribution to the peer review of this work.

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Supplementary Information

Supplementary Figures 1–14, Supplementary Table 1

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Waldmann, I.P., Griffith, C.A. Mapping Saturn using deep learning. Nat Astron 3, 620–625 (2019).

Download citation


Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing