---

name: verde description: | Spatial data gridding and interpolation with a machine-learning style API. Process geographic and Cartesian point data onto regular grids. Use when Claude needs to: (1) Grid scattered spatial data onto regular grids, (2) Interpolate point data using splines, linear, or cubic methods, (3) Process geographic coordinates with projections, (4) Reduce large datasets using block averaging, (5) Remove polynomial trends from spatial data, (6) Cross-validate gridding parameters, (7) Create processing pipelines with Chain, (8) Grid vector data like GPS velocities.

Verde - Spatial Data Gridding

Quick Reference

import verde as vd

# Basic gridding
spline = vd.Spline()
spline.fit(coordinates, values)  # coordinates = (lon, lat) tuple
grid = spline.grid(spacing=0.1)  # Returns xarray Dataset

# Access result
elevation = grid.elevation.values

# Save output
grid.to_netcdf('output.nc')

Key Classes

Class	Purpose
Spline	Bi-harmonic spline interpolation (smooth, good extrapolation)
Linear	Delaunay triangulation (fast, no extrapolation)
Cubic	Cubic interpolation (medium smoothness)
Chain	Pipeline of processing steps
BlockReduce	Decimate data to block means/medians
Trend	Polynomial trend fitting and removal
Vector	Grid 2-component vector data

Essential Operations

Grid Scattered Data

coordinates = (longitude, latitude)  # Tuple of 1D arrays
values = elevation  # 1D array

spline = vd.Spline()
spline.fit(coordinates, values)
grid = spline.grid(spacing=0.1, data_names=['elevation'])

Project to Cartesian

import pyproj

projection = pyproj.Proj(proj='merc', lat_ts=data_lat.mean())
proj_coords = projection(longitude, latitude)

spline = vd.Spline()
spline.fit(proj_coords, values)
grid = spline.grid(spacing=1000)  # 1000m spacing

Block Reduce Large Datasets

import numpy as np

reducer = vd.BlockReduce(reduction=np.median, spacing=0.1)
coords_reduced, values_reduced = reducer.filter(coordinates, values)

Remove Trend Before Gridding

trend = vd.Trend(degree=2)  # Quadratic
trend.fit(coordinates, values)
residuals = values - trend.predict(coordinates)

# Grid residuals, then add trend back

Processing Pipeline

chain = vd.Chain([
    ('trend', vd.Trend(degree=1)),
    ('reduce', vd.BlockReduce(np.median, spacing=0.05)),
    ('spline', vd.Spline())
])
chain.fit(coordinates, values)
grid = chain.grid(spacing=0.01)

Cross-Validation

spline = vd.Spline()
scores = vd.cross_val_score(spline, coordinates, values, cv=5)
print(f"Mean R2: {scores.mean():.3f}")

Mask Far from Data

grid = spline.grid(spacing=0.1)
mask = vd.distance_mask(coordinates, maxdist=0.2, grid=grid)
grid_masked = grid.where(mask)

Grid Parameters

Parameter	Description
spacing	Grid cell size (same units as coordinates)
region	(west, east, south, north) bounds
shape	(n_north, n_east) grid dimensions
adjust	'spacing' or 'region' - which to adjust for exact fit

Gridder Comparison

Gridder	Speed	Smoothness	Extrapolation
Spline	Medium	High	Good
Linear	Fast	Low	None
Cubic	Fast	Medium	None

Common Issues

Issue	Solution
Poor extrapolation	Use distance_mask() to mask far from data
Slow with large data	Use BlockReduce first
Regional trends	Remove with Trend before gridding
Wrong spacing	Check coordinate units (degrees vs meters)

References

• Gridders - Available gridders and parameters
• Cross-Validation - Parameter tuning methods

Scripts

• scripts/grid_data.py - Grid scattered data to NetCDF