raptor-stats

Compute zonal statistics using efficient Raptor (Raster+Vector) methods.

Installation

pip install raptor-stats

Usage

from raptorstats import zonal_stats

stats = zonal_stats("path/to/vector.shp", "path/to/raster.tif", method="scanline")

See the zonal_stats API docs for more details on input types and additional parameters.

Methods

scanline: Scans the raster file once, line by line, and computes all the intersections with the vector layer in a single pass. Suitable for fast one-time run results.
agqt: Builds a QuadTree with precomputed statistics, then combines it with the scanline method to answer queries more efficiently. Suitable for systems that repeatedly query changing vector layers over the same raster.

Performance

Raptor methods performance advantage increases with the size of the input raster and number of features. For example, with an ~1.9 billion pixel raster and 50 features (US states):

See images at https://github.com/simonpedrogonzalez/raptor-stats/blob/main/README.md

For the same raster, on but around 3000 features (US counties):

NOTES:

These tests were made on a i7-8750H (2019) 16GB RAM Linux machine.
The performance of the agqt method depends on the depth of the tree selected and the size of the features.

Credits

Author: Simon Pedro Gonzalez.
This package is based on the following project, where you can read more about the zonal stats problem, methods and performance comparison.
This package API is inspired in the rasterstats package by Matthew Perry.