Geoweaver for Automating ML-based High Resolution Snow Mapping Workflow

The high spatiotemporal resolution of snow maps requires the close collaboration of computing devices, big data storage, and a parallel computing framework. To achieve operation-level automation and sustainability, we used Geoweaver to build and manage our machine-learning-driven high-resolution snow mapping workflow. The workflow will take remote sensing images as input, train and test machine learning models on the fly, deploy the trained model as a web service, and record all the provenance information in a database. The use of Geoweaver will significantly reduce the difficulty in training and deploying ML models in snow research, and make the latest techniques like artificial intelligence and high-performance computing friendly to domain scientists, especially those who are not keen on coding. In this work, we will demonstrate how Geoweaver can preserve the historical workflow runs and reproduce the original experiments even the original computing devices are changed. The workflow in Geoweaver is exportable and can be registered and shared on a public repository such as GitHub to increase communication and reduce duplicated efforts in the cryosphere community.