Deck 17: Advanced Tutorial 2 —> Raster Transformations (Technical)

What's the core gotcha when calling transform() on raster vs. vector data?

transform() on raster data does NOT apply the translation component (a residual Translation stays in the metadata); the same call on vector data fully bakes in the transformation (ends up Identity).

Why is this raster/vector asymmetry not actually a bug, per the docs?

It's due to a real difference between the on-disk NGFF model and the in-memory Xarray model → alignment is still correct when interpreted through spatialdata-plot, which applies both correctly.

What function physically pads a translation into the pixel grid using zeros?

rasterize(element, min_coordinate=..., max_coordinate=..., target_unit_to_pixels=...)

What function transforms every element in a SpatialData object to share the same consistent transformation?

transform_to_data_extent(sdata, coordinate_system=..., target_unit_to_pixels=...)

Once transform_to_data_extent() has been applied, what direct slicing operations become possible?

Coordinate-based slicing on both raster (.sel(), xarray) and vector (.cx[], geopandas) data without needing coordinate-system bookkeeping.

What function computes a transformation directly between two named coordinate systems (not just element-to-system)?

get_transformation_between_coordinate_systems(sdata, system_a, system_b)

How does get_transformation_between_coordinate_systems() work internally?

It builds a graph of known transformations and their inverses, then finds a path between the two requested systems.

What method lets you build an axis-aware affine matrix from a transformation object, even with mismatched input/output axes?

transformation.to_affine_matrix(input_axes=..., output_axes=...)