In building multiple large geospatial pipelines, I got increasingly frustrated with the amount of bookwork that pollutes the scientific intent of the code. This is both due to spatial bookkeeping, aligning pixels, ensuring map projections match, etc., and for managing hardware resources: chunking data for memory puproses, CPU or GPU paths for extracting parallelism, etc.

To this end I've spent the last few years building a library, called Yirgacheffe, to back the pipelines I build that takes care of both of those concerns, and provides a numpy-like, pandas-like interface for working with raster and vector geospatial datasets when doing spactial analysis.

Hiding all the bookwork away means the pipeline code intent is more obviously aligned with the method, and robustness is increased by relying on well tested code behind the scenes.

This talk will give an overview of Yirgacheffe's declarative nature, where it can help you with your pipelines, and where it still isn't the best choice and needs more work/help.