No-data & gap filling¶

Manage the cells that don't hold real data:

• change_no_data_value — change the no-data sentinel (e.g. -9999 → NaN).
• fill — set every in-domain cell to a constant value.
• fill_gaps — interpolate holes in one raster using another raster's valid domain (nearest-neighbour).

Setup¶

In [1]:

%matplotlib inline

import shutil
import tempfile
from pathlib import Path

import numpy as np

DATA = Path('../../../examples/data')
WORK = Path(tempfile.mkdtemp(prefix='pyramids-t2-'))
DATA.is_dir(), WORK.is_dir()

%matplotlib inline import shutil import tempfile from pathlib import Path import numpy as np DATA = Path('../../../examples/data') WORK = Path(tempfile.mkdtemp(prefix='pyramids-t2-')) DATA.is_dir(), WORK.is_dir()

Out[1]:

(True, True)

In [2]:

from pyramids.dataset import Dataset

ds = Dataset.read_file(DATA / 'acc4000.tif')
ds.shape, ds.no_data_value

from pyramids.dataset import Dataset ds = Dataset.read_file(DATA / 'acc4000.tif') ds.shape, ds.no_data_value

2026-07-11 14:41:53 | INFO | pyramids.base.config | Logging is configured.

Out[2]:

((1, 13, 14), (-3.4028230607370965e+38,))

The source raster — no-data cells render as blanks against the valid domain.

In [3]:

ds.plot(band=0, title="Source raster", cmap="viridis")

ds.plot(band=0, title="Source raster", cmap="viridis")

Out[3]:

<cleopatra.array_glyph.ArrayGlyph at 0x7f52bf21cad0>

/home/runner/work/pyramids/pyramids/.pixi/envs/docs/lib/python3.14/site-packages/matplotlib/colors.py:816: RuntimeWarning: overflow encountered in multiply
  xa *= self.N

Change the no-data sentinel — change_no_data_value¶

Re-label the sentinel that marks empty cells.

In [4]:

relabelled = ds.change_no_data_value(-1.0)
ds.no_data_value[0], '->', relabelled.no_data_value[0]

relabelled = ds.change_no_data_value(-1.0) ds.no_data_value[0], '->', relabelled.no_data_value[0]

Out[4]:

(-3.4028230607370965e+38, '->', np.float64(-1.0))

Fill the domain — fill¶

Replace every valid cell with a constant (no-data cells stay no-data).

In [5]:

filled = ds.fill(5.0)
valid = filled.read_array()
float(np.nanmax(np.where(valid == filled.no_data_value[0], np.nan, valid)))

filled = ds.fill(5.0) valid = filled.read_array() float(np.nanmax(np.where(valid == filled.no_data_value[0], np.nan, valid)))

Out[5]:

5.0

Interpolate holes — fill_gaps¶

fill_gaps(mask, src_array) fills cells that are no-data in the source but valid in the mask dataset, using nearest-neighbour interpolation. Below, a 5×5 grid with two punched holes is repaired against a hole-free mask.

In [6]:

ND = -9999.0
base = np.arange(1, 26, dtype='float32').reshape(5, 5)
holed = base.copy()
holed[2, 2] = ND
holed[1, 3] = ND
src = Dataset.create_from_array(
    holed, top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND
)
mask = Dataset.create_from_array(
    base, top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND
)
repaired = np.asarray(src.fill_gaps(mask, src.read_array())).squeeze()
int((holed == ND).sum()), '->', int((repaired == ND).sum()), '| repaired[2,2] =', float(
    repaired[2, 2]
)

ND = -9999.0 base = np.arange(1, 26, dtype='float32').reshape(5, 5) holed = base.copy() holed[2, 2] = ND holed[1, 3] = ND src = Dataset.create_from_array( holed, top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND ) mask = Dataset.create_from_array( base, top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND ) repaired = np.asarray(src.fill_gaps(mask, src.read_array())).squeeze() int((holed == ND).sum()), '->', int((repaired == ND).sum()), '| repaired[2,2] =', float( repaired[2, 2] )

Out[6]:

(2, '->', 0, '| repaired[2,2] =', 14.0)

Before & after: src has two punched holes (the blank cells); the repaired grid is hole-free. The repaired array is wrapped back into a Dataset so it can be plotted the same way.

In [7]:

src.plot(band=0, title="With gaps (src)", cmap="viridis")

src.plot(band=0, title="With gaps (src)", cmap="viridis")

Out[7]:

<cleopatra.array_glyph.ArrayGlyph at 0x7f52bf1d9f90>

In [8]:

repaired_ds = Dataset.create_from_array(
    repaired.astype('float32'), top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND
)
repaired_ds.plot(band=0, title="Gaps filled (fill_gaps)", cmap="viridis")

repaired_ds = Dataset.create_from_array( repaired.astype('float32'), top_left_corner=(0, 0), cell_size=1.0, epsg=4326, no_data_value=ND ) repaired_ds.plot(band=0, title="Gaps filled (fill_gaps)", cmap="viridis")

Out[8]:

<cleopatra.array_glyph.ArrayGlyph at 0x7f52b6d5b890>

Notes¶

• change_no_data_value and fill take inplace=True to mutate the dataset in place.
• fill_gaps is the building block behind aligned crops that repair edge gaps.
• See also: Crop & mask.