Planet cartography with neural learned regularization

Finding potential life harboring exo-Earths is one of the aims exoplanetary science. Detecting signatures in exoplanets will likely first be accomplished by determining bulk composition planetary atmosphere via reflected/transmitted spectroscopy. However, a complete understanding habitability conditions surely require mapping presence liquid water, continents and/or clouds. Spin-orbit tomography technique that allows us to obtain maps surface around other stars using light scattered surface. We leverage deep learning and propose for which regularization learned from mock surfaces. The solution inverse problem posed as neural network can trained end-to-end with suitable training data. this work use methods based on procedural generation planets, inspired what we found Earth. also consider recovery surfaces persistent cloud cloudy planets. show reliable carried out our approach, producing very compact continents, even when single passband observations. More importantly, if are partially like Earth is, potentially map distribution clouds always occur same position (associated orography sea temperatures) together non-persistent move across This become test perform an exoplanet detection active climate system. For small rocky planets habitable zone their stars, weather system driven considered strong proxy truly conditions.