Coarse cells

imageserver_export.aggregate

From a fine raster to coarse cells: the mean and the share of valid pixels.

A 100 m imperviousness raster feeds a model on 1 km cells: each cell wants the mean of its 10x10 pixels, and how many of them were real data. Sentinel values (254 "unclassifiable", 255 "no data") must stay out of the mean, or a coastal cell half in the sea comes out 128 % sealed.

The raster is read one strip of cells at a time, so a whole country at 10 m never has to fit in memory.

Aggregate dataclass

Coarse cells on grid: mean (NaN with no valid pixel) and valid_fraction.

valid_fraction is over the whole cell: a cell that sticks out past the raster's edge counts the missing part as not valid.

aggregate

aggregate(path: Path, factor: int, *, valid_range: tuple[float, float] | None = None) -> Aggregate

Cells of factor x factor pixels of the GeoTIFF at path.

A pixel counts if it lies in valid_range; without it, if it is not the file's nodata. The cell grid starts at the raster's top-left corner, so for cells to fall on an existing grid (the 1 km EEA reference grid, say) export on a grid snapped to it: Grid.snap(..., res=100) with bounds on whole kilometres.