Brendan started off with a high-level overview of Xarray-Spatial, an open source raster-based spatial analytics library that uses Numba to speed up algorithms and Dask for scalability across different cores and machines.
“The aim of Xarray-Spatial is to provide analysis functions on top of rasters without dependencies on GDAL or GEOS.”
Brendan explained two fundamental types of data you’ll find in the geospatial world:
Raster: grids of pixels that represent continuous phenomena.
Vector: points, lines, and polygons that represent discrete phenomena.
Moving on to what makes Xarray-Spatial so valuable, Brendan highlighted the power of Datashader, a general-purpose rasterization pipeline. Datashader allows users to integrate vector data into raster-based analysis. It also makes operations like rasterization of polygons at the same resolutions as satellite imagery possible, which enables analysis between sources.
Brendan and the makepath team recently looked into what parts of Xarray-Spatial would benefit the most from better GPU support and performance, to leverage the trailblazing work going on at NVIDIA.
Four tools stood out:
Hillshade with real shadows
Focusing on CuPy and CUDA on top of Xarray-Spatial has led to great results.