Migrating to Rasterio 1.0

affine.Affine() vs. GDAL-style geotransforms

One of the biggest API changes on the road to Rasterio 1.0 is the full deprecation of GDAL-style geotransforms in favor of the affine library. For reference, an affine.Affine() looks like:

affine.Affine(a, b, c,
              d, e, f)

and a GDAL geotransform looks like:

(c, a, b, f, d, e)

Fundamentally these two constructs provide the same information, but the Affine() object is more useful.

Here’s a history of this feature:

  1. Originally, functions with a transform argument expected a GDAL geotransform.
  2. The introduction of the affine library involved creating a temporary affine argument for rasterio.open() and a src.affine property. Users could pass an Affine() to affine or transform, but a GDAL geotransform passed to transform would issue a deprecation warning.
  3. src.transform remained a GDAL geotransform, but issued a warning. Users were pointed to src.affine during the transition phase.
  4. Since the above changes, several functions have been added to Rasterio that accept a transform argument. Rather than add an affine argument to each, the transform argument could be either an Affine() object or a GDAL geotransform, the latter issuing the same deprecation warning.

The original plan was to remove the affine argument + property, and assume that the object passed to transform is an Affine(). However, after further discussion it was determined that since Affine() and GDAL geotransforms are both 6 element tuples users may experience unexplained errors and outputs, so an exception is raised instead to better highlight the error.

Before 1.0b1:

  • rasterio.open() will still accept affine and transform, but the former now issues a deprecation warning and the latter raises an exception if it does not receive an Affine().
  • If rasterio.open() receives both affine and transform a warning is issued and transform is used.
  • src.affine remains but issues a deprecation warning.
  • src.transform returns an Affine().
  • All other Rasterio functions with a transform argument now raise an exception if they receive a GDAL geotransform.

Tickets

  • #86 - Announcing the plan to switch from GDAL geotransforms to Affine().

  • #763 - Implementation of the migration and some further discussion.

    Beginning in 1.0b1:

  • In rasterio.open “affine” will no longer be an alias for the transform keyword argument.

  • Dataset objects will no longer have an affine property.

  • The transform keyword argument and property is always an instance of the Affine class.

I/O Operations

Methods related to reading band data and dataset masks have changed in 1.0.

Beginning with version 1.0b1, there is no longer a read_mask method, only read_masks. Datasets may be opened in read-write “w+” mode when their formats allow and a warning will be raised when band data or masks are read from datasets opened in “w” mode.

Beginning with 1.0.0, the “w” mode will become write-only and reading data or masks from datasets opened in “w” will be prohibited.

Deprecated: rasterio.drivers()

Previously users could register GDAL’s drivers and open a datasource with:

import rasterio

with rasterio.drivers():

    with rasterio.open('tests/data/RGB.byte.tif') as src:
        pass

but Rasterio 1.0 contains more interactions with GDAL’s environment, so rasterio.drivers() has been replaced with:

import rasterio
import rasterio.env

with rasterio.Env():

    with rasterio.open('tests/data/RGB.byte.tif') as src:
        pass

Tickets

  • #665 - Deprecation of rasterio.drivers() and introduction of rasterio.Env().

Removed: src.read_band()

The read_band() method has been replaced by read(), which allows for faster I/O and reading multiple bands into a single numpy.ndarray().

For example:

import numpy as np
import rasterio

with rasterio.open('tests/data/RGB.byte.tif') as src:
    data = np.array(map(src.read_band, (1, 2, 3)))
    band1 = src.read_band(1)

is now:

import rasterio

with rasterio.open('tests/data/RGB.byte.tif') as src:
    data = src.read((1, 2, 3))
    band1 = src.read(1)

Tickets

  • # 83 - Introduction of src.read().
  • #96, #284 - Deprecation of src.read_band().

Removed: src.read_mask()

The src.read_mask() method produced a single mask for the entire datasource, but could not handle producing a single mask per band, so it was deprecated in favor of src.read_masks(), although it has no direct replacement.

Tickets

  • #284 - Deprecation of src.read_mask().

Moved: Functions for working with dataset windows

Several functions in the top level rasterio namespace for working with dataset windows have been moved to rasterio.windows.*:

  • rasterio.get_data_window()
  • rasterio.window_union()
  • rasterio.window_intersection()
  • rasterio.windows_intersect()

Tickets

  • #609 - Introduction of rasterio.windows.

Moved: rasterio.tool

This module has been removed completely and its contents have been moved to several different locations:

Tickets

  • #609 - Deprecation of rasterio.tool.

Moved: rasterio.tools

This module has been removed completely and its contents have been moved to several different locations:

Tickets

  • #609 - Deprecation of rasterio.tools.

Removed: rasterio.warp.RESAMPLING

This enum has been replaced by rasterio.warp.Resampling.

Removed: dataset’s ul() method

This method has been replaced by the xy() method.

Signature Changes

For both rasterio.features.sieve() and rasterio.features.rasterize() the output argument has been replaced with out. Previously the use of output issued a deprecation warning.

Deprecation of dataset property set_* and get_* methods

Methods get_crs, set_crs, set_nodatavals, set_descriptions, set_units, and set_gcps are deprecated and will be removed in version 1.0. They have been replaced by fully settable dataset properties crs, nodatavals, descriptions, units, and gcps.

In the cases of units and descriptions, set_band_unit and set_band_description methods remain to support the rio-edit-info command.

Creation Options

Rasterio no longer saves dataset creation options to the metadata of created datasets and will ignore such metadata starting in version 1.0. Users may opt in to this by setting RIO_IGNORE_CREATION_KWDS=TRUE in their environments.