Geometric aspects of stereoscopic spaceborne imaging of dynamic clouds in the CLOUD experiment

There are gaps of knowledge regarding cloud dynamics. This is partly due to lack of global-scale high resolution data on three-dimensional flow fields, specifically updrafts that drive cloud convection. Such data can be inferred from tracking cloud-envelope evolution. Towards this, a dedicated CLOUD experiment is planned in the C3IEL mission (by the Israeli and French Space Agencies). CLOUD involves three or two small satellites at low earth orbit, hundreds of kilometers apart. As they overpass clouds, they will stereoscopically image clouds simultaneously, and track cloud evolution during four minutes. The planned nadir field of view is 80km, at 20m nadir resolution. The very large baseline, high off-nadir view angles involved during overpass, and Earth curvature deviate significantly from small-angle stereoscopy common in computer vision. Moreover, in these large angles, the appearance of cloud features vary significantly across viewpoints and time. These aspects challenge geometric self-calibration, stereo triangulation and ultimately estimation of cloud envelope flow. We present analysis of these challenges and simulations. Large eddy simulations create dynamic cloud fields. Through these clouds, radiative transfer is operated. The radiance is projected to perspective images, at perturbed platform poses in orbit, to challenge computer vision. Flow is then estimated based on non-rigid matching of estimated cloud envelopes.