loading page

Gaining a global perspective on the surface composition of Venus from orbit through near infrared observations -- with a little help from machine learning approaches
  • +3
  • Jörn Helbert,
  • Melinda Dyar,
  • Alessandro Maturilli,
  • Mario D'Amore,
  • Sabrina Ferrari,
  • Indhu Varatharajan
Jörn Helbert
German Aerospace Center DLR Berlin

Corresponding Author:[email protected]

Author Profile
Melinda Dyar
Mount Holyoke College
Author Profile
Alessandro Maturilli
German Aerospace Center DLR Berlin
Author Profile
Mario D'Amore
German Aerospace Center DLR Berlin
Author Profile
Sabrina Ferrari
University of Pavia
Author Profile
Indhu Varatharajan
German Aerospace Center DLR Berlin
Author Profile

Abstract

Venus is the most Earth-like of the terrestrial planets, though very little is known about its surface composition. Thanks to recent advances in laboratory spectroscopy and spectral analysis techniques, this is about to change. Although the atmosphere prohibits observations of the surface with traditional imaging techniques over much of the EM spectral range, five transparent windows between ~0.86 µm and ~1.18 µm occur in the atmosphere’s CO2 spectrum. New high temperature laboratory spectra from the Planetary Spectroscopy Laboratory at DLR show that spectra in these windows are highly diagnostic for surface mineralogy [1]. The Venus Emissivity Mapper (VEM) [2] builds on these recent advances VEM is the first flight instrument specially designed to focus solely on mapping Venus’ surface using the windows around 1 µm. Operating in situ from Venus orbit, VEM will provide a global map of composition as well as redox state of the surface, enabling a comprehensive picture of surface-atmosphere interaction on Venus. VEM will return a complex data set containing surface, atmospheric, cloud, and scattering information. Total planned data volume for a typical mission scenario exceeds 1TB. Classical analysis techniques have been successfully used for VIRTIS on Venus Express [3-5] and could be employed with the VEM data. However, application of machine learning approaches to this rich dataset is vastly more efficient, as has already been confirmed with laboratory data. Binary classifiers [6] demonstrate that at current best estimate errors, basalt spectra are confidently discriminated from basaltic andesites, andesites, and rhyolite/granite. Applying the approach of self-organizing maps to the increasingly large set of laboratory measurements allows searching for additional mineralogical indicators, especially including their temperature dependence. [1] Dyar M. D. et al. 2017 LPS XLVIII, #1512. [2] Helbert, J. et al. 2016. San Diego, CA, SPIE. [3] Smrekar, S.E., et al. Science, 2010 328(5978), 605-8. [4] Helbert, J., et al., GRL, 2008 35(11). [5] Mueller, N., et al., JGR, 2008 113. [6] Dyar M. D. et al. 2017 LPS XLVIII, #3014.