Nearest-neighbour resampling

Nearest-neighbour resampling assigns each output pixel the value of the closest source grid cell, with no interpolation. The result preserves the exact data values and shows crisp, block-like cell boundaries — ideal for visualising the native structure of the source grid.

Use NearestNeighbour when:

• you want to inspect the raw grid cell structure of your data, or

• your field contains discrete categories (e.g. land-use, cloud type) where averaging between cells would be meaningless.

For smooth, interpolated rendering, see the Bilinear notebook.

Example: 2 m temperature over Europe

We will use a sample GRIB file containing 2 m temperature on a regular lat/lon grid.

import earthkit.data as ekd

import earthkit.plots as ekp

data = ekd.from_source("sample", "test.grib").to_fieldlist()
data.ls()

	parameter.variable	time.valid_datetime	time.base_datetime	time.step	vertical.level	vertical.level_type	ensemble.member	geography.grid_type
0	2t	2020-05-13 12:00:00	2020-05-13 12:00:00	0 days	0	surface	0	regular_ll
1	msl	2020-05-13 12:00:00	2020-05-13 12:00:00	0 days	0	surface	0	regular_ll

Default pixel count

Calling NearestNeighbour() with no arguments uses 1000 × 1000 pixels. Each pixel is coloured with the value of whichever source grid cell its centre falls closest to, so you can clearly see the boundaries between cells.

from earthkit.plots.resample import NearestNeighbour

chart = ekp.Map()

chart.contourf(data, resample=NearestNeighbour())

chart.coastlines()
chart.gridlines()
chart.legend()

chart.show()

Comparing Bilinear and NearestNeighbour

The difference between the two methods is most visible at low resolution. Here we use a coarse pixel count to make the individual cells clearly visible.

from earthkit.plots.resample import Bilinear

figure = ekp.Figure(rows=1, columns=2, domain="Europe")

ax = figure.add_map()
ax.contourf(data, resample=Bilinear(80))
ax.title("Bilinear(80) — smooth")

ax = figure.add_map()
ax.contourf(data, resample=NearestNeighbour(80))
ax.title("NearestNeighbour(80) — sharp cells")

figure.coastlines()
figure.legend(location="right")

figure.show()

Controlling pixel count and resolution

Like Bilinear, you can set a fixed pixel count or use NearestNeighbour.at_resolution(dx) to specify the pixel spacing in degrees. A finer resolution retains the cell-boundary sharpness while showing more cells.

figure = ekp.Figure(rows=1, columns=2, domain="Europe")

# Fixed pixel count
ax = figure.add_map()
ax.contourf(data, resample=NearestNeighbour(nx=200, ny=100))
ax.title("nx=200, ny=100")

# Resolution-based: one pixel per 0.5 degrees
ax = figure.add_map()
ax.contourf(data, resample=NearestNeighbour.at_resolution(0.5))
ax.title("at_resolution(0.5°)")

figure.coastlines()
figure.legend(location="right")

figure.show()

NOTE: NearestNeighbour is only valid for regular rectilinear source grids. For curvilinear grids it automatically falls back to Bilinear. For HEALPix or reduced Gaussian grids, first regrid with Regrid and then apply NearestNeighbour via a Chain.