Course on Cloud Native GIS

This course was produced and run by a group called thriveGEO for Yale University - it focused on the fundamentals of cloud-computing and specialized datastores/file formats for spatio-temporal data.

One major challenge for any researcher is data discovery. The STAC specification provides a common and extensible way to describe geospatial information, improving discoverability. A STAC contains multiple levels of metadata that allow you to find a specific data product and query that product so that you receive just the data you need.

This ability to query is critical for performant cloud-native geospatial analysis because data latency is the workflow killer. Moving the appropriate amount of data (potentially in parallel) instead of downloading a whole the dataset saves time and resources.

In order for this all to work effectively, you need file formats like COG and geoparquet that allow users to slice the data into small parts and libraries like DASK that improve parallelization. With appropriately sized chunks and well defined, parallel tasks, a research can to analyses over large swaths of space and time efficiently.

Finally, the course talked about different analysis platforms that range from custom built cloud environments to plug-and-play systems. All of the plug-and-play platforms that were presented immediately raised concerns about vendor lock and reproducibility.

The cloud-native geospatial space is still a developing area and will likely progress significantly in the coming years. The STAC ecosystem could benefit from better search funcitonality by surfacing lower level metadata for each catalog. From a FAIR research perspective, some aspects cloud native are wonderful - e.g. coding queries to access data and running analysis on “not your machine” - while others - lack of versioning and potential for vendors to dispappear - give me pause.