Automatic Tracking and Stereo Reconstruction of Deep-Convective Cloud Tops

Stereo photography provides a promising tool for observing cloud lifecycles and four-dimensional trajectories of cloud surfaces. With an accurately calibrated pair of cameras, the 3D positions and velocities of cloud features can be reconstructed with high spatial and temporal resolutions. Although stereo-photogrammetric cloud observation dates back more than one hundred years, it required a vast amount of manual effort until digital photography opened up the possibility of automating reconstruction and image-analysis processes. A number of recent studies have performed stereo reconstructions of clouds by calibrating the cameras with the positions of known landmarks, such as topographic features or the positions of stars. We extend stereo photogrammetry of clouds to oceanic settings with a calibration method that does not require landmarks. We validate the accuracy of our calibration method by comparing the cloud base heights obtained from the stereo reconstruction to data from a colocated ceilometer, and by comparing the stereo-camera winds to data from nearby radiosondes. In this work, we use image processing techniques as an effort to automate cloud feature extraction and tracking. Automated tracking and matching enables the collection and processing of a huge amount of data from a series of video frames in seconds. In addition, we are developing a method to automatically match corresponding cloud contours in stereo pairs. This method is used to track the 3D positions and velocities of turrets in deep convection reaching heights over 10 km off the coast of Miami, Florida. This work demonstrates the potential of stereo photogrammetry for analyzing deep-convective dynamics.