HybridCT Dataset -- ground truth, training, predicted data, and model parameters

This dataset consists of three parts: Ground truth data: Microphysical cloud properties from Large Eddy Simulation for LWC (liquid water content) and reff (effective droplet radius) in 80x80x4 km domain with 40 m resolution. Training data: collection of input data and labels for training the HybridCT `emulator` Input: Synthetic multi-angle radiance fields which are generated from the ground truth cloud fields via radiative transfer simulations. The synthetic multi-angle images represent the viewing angle geometry of the Multi-angle Imaging Spectroradiometer (MISR) onboard the Terra spacecraft at its red channel (670 nm). Labels: Multi-angle optical thickness fields directly computed from the ground truth cloud fields. Predicted data: Multi-angle optical thickness fields predicted by the HybridCT `emulator` which serve as input for the `reconstruction`. Model artifacts: parameters for the trained U-Net machine learning model to predict optical thickness maps from multi-angle radiance input.