Identifying and Tracking Cloud Clusters from Satellite Imagery Using Python

Cloud clusters (CCs) are regions of cold cloud-top temperatures that occur on the mesoscale and are associated with intense convection. CCs are often examined in the study of tropical cyclone genesis and mesoscale convective systems (MCSs). One common approach for identifying these CCs is an algorithm that applies a temperature mask on infrared satellite images to find the regions with cold cloud-top temperatures. The CCs are tracked through time by quantifying the amount of overlap between these regions as compared across sequential images. This presentation provides an overview of the Python implementation of this CC identification and tracking algorithm that was part of a study of MCSs over the central United States.

In this implementation, the data is loaded into Xarray Datasets, and the temperature mask is applied. The Scipy Ndimage library then identifies the contiguous regions of cold cloud-top temperatures. These regions are stored as nodes in a NetworkX graph and based on the amount of overlap between regions in sequential images, edges are added to represent CC movement through time. Each connected component within the NetworkX graph is a potential CC. These can then be filtered depending on the criteria being used in the study. Because a 22-year period of GOES-East infrared satellite data were downloaded for the related MCS study, this implementation utilizes the Multiprocessing library within a Snakemake workflow to allow for large collections of satellite images to be processed quickly.

This implementation benefits from the algorithm’s inherent ability to account for merges and splits. It can also be easily adjusted for different defining criteria, making it versatile for use in other future studies.