arXiv subjects

Linda Forster

Publications and source records attributed to Linda Forster.

3 recordsLinked to original sources

arXiv2021

Toward Cloud Tomography from Space using MISR and MODIS: The Physics of Image Formation for Opaque Convective Clouds

3D convective cloud images form via two intertwined radiative diffusion processes. Sunlight starts in the anti-solar direction and ends in toward-sensor ones, but repeated forward-peaked scattering smears the well-collimated beams across_direction_ space. This loss of directional memory in the cloud's "outer shell" (OS) is modeled as a random walk (RW) on the sphere. We show that, for typical cloud phase functions, 5 or 6 scatterings suffice for severe degrading of directionality. Simultaneously, a RW unfolds in_standard_ 3D space where steps are angularly-correlated, hence a drift in the original direction and an associated lateral dispersion. Any distinctive cloud image "feature" originates in the OS, and the shallower the better. That is also why we previously found that the optical depth of the "veiled core" (VC) is ~5. Significant amounts of sunlight thus arrive at the VC as a diffuse irradiance, and leave it even more isotropic. The diffusion limit of 3D radiative transfer (RT) is therefore valid inside the VC. Consequently, the underlying RW in the VC unfolds in 3D space, now with isotropic steps since extinction is scaled back to account for forward scattering. We show that the VC optical thickness controls cloud-scale brightness contrast between the illuminated and self-shaded sides of the cloud. Full cloud image formation thus involves diffusions (i.e., RWs) in both Euclidian and spherical/direction spaces. 3D cloud tomography based on MISR and MODIS multi-angle/-spectral data is an emerging technique in passive VNIR-SWIR sensing that will make judicious use this spatial separation of RT regimes to accelerate forward modeling without significant loss in accuracy.

physics.ao-ph↗

arXiv2020

Multi-Agent Motion Planning using Deep Learning for Space Applications

State-of-the-art motion planners cannot scale to a large number of systems. Motion planning for multiple agents is an NP (non-deterministic polynomial-time) hard problem, so the computation time increases exponentially with each addition of agents. This computational demand is a major stumbling block to the motion planner's application to future NASA missions involving the swarm of space vehicles. We applied a deep neural network to transform computationally demanding mathematical motion planning problems into deep learning-based numerical problems. We showed optimal motion trajectories can be accurately replicated using deep learning-based numerical models in several 2D and 3D systems with multiple agents. The deep learning-based numerical model demonstrates superior computational efficiency with plans generated 1000 times faster than the mathematical model counterpart.

cs.RO↗

arXiv2019

Cloud Tomography from Space using MISR and MODIS: Locating the "Veiled Core" in Opaque Convective Clouds

For passive satellite imagers, current retrievals of cloud optical thickness and effective particle size fail for convective clouds with 3D morphology. Indeed, being based on 1D radiative transfer (RT) theory, they work well only for horizontally homogeneous clouds. A promising approach for treating clouds as fully 3D objects is cloud tomography, and this has been demonstrated for airborne observations. For cloud tomography from space, however, more efficient forward 3D RT solvers are required. Here, we present a path forward, acknowledging that optically thick clouds have "veiled cores." Photons scattered into and out of this deep region do not contribute significant information to the observed imagery about the inner structure of the cloud. We investigate the location of the veiled core for the MISR and MODIS imagers. While MISR provides multi-angle imagery in the visible and near-IR, MODIS includes channels in the short-wave IR, albeit at a single view angle. This combination will enable future 3D retrievals to disentangle the cloud's effective particle size and optical thickness. We find that, in practice, the veiled core is located at an optical distance of $\approx$5 starting from the cloud boundary along the line-of-sight. For MODIS' absorbing wavelengths the veiled core covers a larger volume, starting at smaller optical distances. This result makes it possible to reduce the number of unknowns for the cloud tomographic reconstruction, and opens up new ways to increase the efficiency of the 3D RT solver at the heart of the reconstruction algorithm.

physics.geo-ph↗