Cloud Catalog from Mars Orbiter Laser Altimeter / Mars Global Surveyor Data Using Machine Learning Algorithms

Abstract

<p><span> </span><span>In the development of Mars climate models, modeling clouds is an important challenge, and especially for CO</span>_{<span>2</span>}<span> clouds</span><span>. This is due to the complexity of th</span><span>e</span><span> atmospheric </span><span>processes involved</span><span> that may imply rethinking microphysical theories, but also to the scarcity of observations. In the la</span><span>te</span><span> 90&#8217;s, Mars Orbiter Laser Altimeter was one of the instrument</span><span>s</span><span> aboard the Mars Global Surveyor spacecraft. Its first goal was to build a precise map of Mars&#8217; topography </span><span>through laser altimetry but its sensitivity </span><span>allowed for cloud observations as well</span><span> . Th</span><span>us,</span><span> previous studies (Neumann & al. 2003 Ivanov & Muhlemann 2001) have </span><span>shown</span><span> that some </span><span>laser </span><span>returns were cloud signatures coming from the atmosphere. However, </span><span>at that time, the </span><span>huge amount of data </span><span>was analysed using simple</span><span> dis</span><span>t</span><span>inction criteria.</span></p><p><span> We use K-means clustering algorithms to compu</span><span>ta</span><span>tionally analyse MOLA data. In order to optimise the method, we first determine the best observed parameters to distinguish the different kinds of returns (surface, noise and clouds). </span><span>The b</span><span>est number of clusters is determined using three independent methods : elbow, silhouette score and gap statistics. The method is test</span><span>ed</span><span> on a restricted sample (10&#160;% of the dataset) and then applied to the </span><span>full raw</span><span> data</span><span>set</span><span>. </span><span>Once </span><span>that </span><span>cloud cluster identified, we can plot spatial and temporal distributions of t</span><span>he cloud return</span><span>s </span><span>and compare them</span><span> with previous results.</span></p><p><span> As mentioned by Neumann & al. (2003), the product of surface reflectivity </span><span>and two-way transmissivity of the atmosphere </span><span>appears as the best parameter </span><span>discriminating</span><span> between surface and cloud returns. A unique number of cluster</span><span>s</span><span> </span><span>(6) </span><span>is identified by all three optimisation methods. Among those clusters, one clearly identifies cloud returns, while others </span><span>represent</span><span> noise and surface returns. </span><span>Our methods</span><span> allows us to </span><span>identify</span><span> more clouds than previous studies. Our cloud distribution remain</span><span>s</span><span> coherent with the ones given in </span><span>previous</span><span> studies, showing the viability of our method. We </span><span> will present</span><span> a catalog of cloud </span><span>returns </span><span>coming from MOLA data. </span><span>We are now working </span><span>to separate different kinds of clouds within these returns (absorptive and reflective clouds, CO</span>_{<span>2</span>}<span> / water clouds, dust &#8230;) using machine learning algorithms and a recent MOLA surface reflectivity map (Heavens & al. 2016).</span></p>