MSPC: Overview:

MSPC is a free-to-use geocomputing environment created and provided by Microsoft. It offers:

  1. a Data Catalog with a lot of commonly used datasets ready to go (and worked examples to use as a quick-start);
  2. a Hub for doing all of your computing work;
  3. really thorough Documentation;
  4. a friendly and responsive Discussions where you can seek help from the growing user community.

The Hub is built using JupyterHub, and provides environments to working in Python (CPU and GPU) and R notebooks, as well as QGIS. The Python environment is based on the PanGeo stack (including the powerful combination of core packages: xarray, geopandas, and dask).

The Hub is set up with dask, allowing you to create and use cloud clusters to easily (well, relatively easily! see details below…) parallelize large or demanding jobs. (Note: Parallel computing adds extra levels of debugging: not just your main analysis code, but also the parallel perfomance of that code and the configuration and resource-use of your cluster, using new tools such as the dask dashboard. Thus, save yourself the work and don’t parallelize until you know you need to!)

The Data Catalog, and the workflows for which MSPC was designed, use the SpatioTemporal Asset Catalog (STAC) specification. This specification creates JSON files that index arbitrarily complex, spatiotemporal datasets (think single-band image files, nested within single-date, multiband image acquisitions, nested within long-term imagery archives, nested within multi-archive catalogs of data…). The actual data files are stored as BLOBs (binary large objects) in cloud-optimized formats (namely, Cloud Optimized Geotiffs (COGs) in Azure’s westeurope region (where the MSPC computing happens). This whole setup allows analyses to, with minimal code, slice and dice the data along space-time axes (e.g., “give me this dataset for these 2 years and within this lat-lon bounding box”), then stream just the necessary spatiotemporal portions of the data into the analysis at (and not before) the time a final computed result is requested.

All in all, this cloud-native workflow is staking out the future of geocomputation in our fields. It resolves a number of limitations of the old, bare-metal way of using servers, but brings with it a new set of tools and a new learning curve. Read on to start learning the ins and outs and start getting things done!

Quick-start:

  1. Request access.
  2. Peruse the Data Catalog and choose a dataset of interest.
  3. Open that dataset’s Example Notebook (e.g.).
  4. Launch the Hub.
  5. Start copying, pasting, and playing with code from the Example Notebook!
  6. Check out this excellent, 1-hour tutorial by Tom Augspurger, MSPC developer.

Details: