// the find
fatiando/verde
Processing and gridding spatial data, machine-learning style
Verde is a Python library for gridding and interpolating spatial data (bathymetry, GPS, geophysics surveys) using a scikit-learn-style fit/predict interface. It's aimed at earth scientists who need spline-based or nearest-neighbor gridding with cross-validation, not general-purpose GIS work.
The scikit-learn-inspired API means anyone who's used sklearn's Pipeline/cross_val_score can pick this up fast — it reuses those mental models instead of inventing new ones. It handles the annoying parts of real survey data specifically: blocked means/medians for decimating dense point clouds, trend removal, and weighted spline fitting with proper cross-validation via model_selection.py. Test suite includes image-comparison baseline tests (matplotlib figure diffing) plus codecov tracking, and it's backed by a JOSS paper, so the math has had outside review.
It's explicitly a small, single-purpose library from one academic group (Fatiando a Terra) — no evidence of broader industry adoption or a large contributor base outside that project. The README itself admits larger-than-memory datasets aren't supported yet, so it won't scale past what fits in RAM. It only does 2D surface gridding (with vector data support), so if you need 3D volumetric interpolation or irregular meshes you're out of luck. Last major version push aside, the core gridder set is narrow (spline, KNN, scipy-based) compared to dedicated geostatistics libraries like PyKrige or GSTools.