Fig. 3
From: 3D volumetric tomography of clouds using machine learning for climate analysis
Necessity of multi-view data for cloud tomography. [Left] Clouds differ by a single voxel (pointed out by a black arrow). Here is a visualization of \({\varvec{\beta }}\) by MIP at \(45^\circ\) off-nadir for a single test cloud. [Right] Blue: the prior probability distribution from which \(\beta\) at the voxel is drawn. Green: the sharply peaked true posterior of \(\beta\) in this voxel of a test cloud by 10 views. Other lines plot inferences of the posterior probability distribution for different numbers of views. As this number increases, \({\hat{P}}(\beta |\mathbf{y})\rightarrow P^\text{true}(\beta |\mathbf{y})\), which is sharply peaked around the true value.
