# coding=UTF-8
"""A collection of raster and vector algorithms and utilities."""
import collections
import functools
import logging
import math
import os
import pprint
import queue
import re
import shutil
import sys
import tempfile
import threading
import time
import warnings
import numpy
import numpy.ma
import rtree
import scipy.interpolate
import scipy.ndimage
import scipy.signal
import scipy.signal.signaltools
import scipy.sparse
import shapely.ops
import shapely.prepared
import shapely.wkb
from osgeo import gdal
from osgeo import gdal_array
from osgeo import gdalconst
from osgeo import ogr
from osgeo import osr
from . import geoprocessing_core
from .geoprocessing_core import DEFAULT_CREATION_OPTIONS
from .geoprocessing_core import DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS
from .geoprocessing_core import DEFAULT_OSR_AXIS_MAPPING_STRATEGY
from .geoprocessing_core import INT8_CREATION_OPTIONS
# This is used to efficiently pass data to the raster stats worker if available
if sys.version_info >= (3, 8):
import multiprocessing.shared_memory
GDAL_VERSION = tuple(int(_) for _ in gdal.__version__.split('.'))
[docs]class ReclassificationMissingValuesError(Exception):
"""Raised when a raster value is not a valid key to a dictionary.
Attributes:
msg (str) - error message
missing_values (list) - a list of the missing values from the raster
that are not keys in the dictionary
"""
def __init__(self, missing_values, raster_path, value_map):
"""See Attributes for args docstring."""
self.msg = (
f'The following {missing_values.size} raster values '
f'{missing_values} from "{raster_path}" do not have corresponding '
f'entries in the value map: {value_map}.')
self.missing_values = missing_values
super().__init__(self.msg)
LOGGER = logging.getLogger(__name__)
# Used in joining finished TaskGraph Tasks.
_MAX_TIMEOUT = 60.0
_VALID_GDAL_TYPES = (
set([getattr(gdal, x) for x in dir(gdal.gdalconst) if 'GDT_' in x]))
_LOGGING_PERIOD = 5.0 # min 5.0 seconds per update log message for the module
_LARGEST_ITERBLOCK = 2**16 # largest block for iterblocks to read in cells
[docs]class TimedLoggingAdapter(logging.LoggerAdapter):
"""A logging adapter to restrict logging based on a timer.
The objective is to have a ``logging.LOGGER``-like object that can be
called multiple times in rapid successtion, but with log messages only
propagating every X seconds.
This object is helpful for creating consistency in logging callbacks and is
derived from the python stdlib ``logging.LoggerAdapter``.
"""
def __init__(self, interval_s=_LOGGING_PERIOD):
"""Initialize the timed logging adapter.
Args:
interval_s (float): The logging interval, in seconds. Defaults to
``_LOGGING_PERIOD``.
"""
logging.LoggerAdapter.__init__(self, LOGGER, extra=None)
self.interval = interval_s
self.last_time = time.time()
[docs] def log(self, level, msg, *args, **kwargs):
"""Log a ``LogRecord``.
Args:
level (int): The logging level.
msg (str): The log message.
args (list): The user-defined positional arguments for the log
message.
kwargs (dict): The user-defined keyword arguments for the log
message.
Returns:
``None``.
"""
# The stacklevel arg to logging was introduced to logging in python
# 3.8 and there isn't a clear way to spoof this in python 3.7, so
# ignore.
if sys.version_info >= (3, 8):
# Don't override user-defined stacklevel if present.
if 'stacklevel' not in kwargs:
# Based on logging internals, 3 is the expected stack depth.
kwargs['stacklevel'] = 3
# Python 3.11 modified the stacklevel argument to be more
# consistent with the behavior in the warnings module.
# https://github.com/python/cpython/issues/89334
if sys.version_info >= (3, 11):
kwargs['stacklevel'] -= 1
now = time.time()
if now >= self.last_time + self.interval:
self.last_time = now
self.logger.log(level, msg, *args, **kwargs)
[docs]def raster_calculator(
base_raster_path_band_const_list, local_op, target_raster_path,
datatype_target, nodata_target,
calc_raster_stats=True, use_shared_memory=False,
largest_block=_LARGEST_ITERBLOCK, max_timeout=_MAX_TIMEOUT,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Apply local a raster operation on a stack of rasters.
This function applies a user defined function across a stack of
rasters' pixel stack. The rasters in ``base_raster_path_band_list`` must
be spatially aligned and have the same cell sizes.
Args:
base_raster_path_band_const_list (sequence): a sequence containing:
* ``(str, int)`` tuples, referring to a raster path/band index pair
to use as an input.
* ``numpy.ndarray`` s of up to two dimensions. These inputs must
all be broadcastable to each other AND the size of the raster
inputs.
* ``(object, 'raw')`` tuples, where ``object`` will be passed
directly into the ``local_op``.
All rasters must have the same raster size. If only arrays are
input, numpy arrays must be broadcastable to each other and the
final raster size will be the final broadcast array shape. A value
error is raised if only "raw" inputs are passed.
local_op (function): a function that must take in as many parameters as
there are elements in ``base_raster_path_band_const_list``. The
parameters in ``local_op`` will map 1-to-1 in order with the values
in ``base_raster_path_band_const_list``. ``raster_calculator`` will
call ``local_op`` to generate the pixel values in ``target_raster``
along memory block aligned processing windows. Note any
particular call to ``local_op`` will have the arguments from
``raster_path_band_const_list`` sliced to overlap that window.
If an argument from ``raster_path_band_const_list`` is a
raster/path band tuple, it will be passed to ``local_op`` as a 2D
numpy array of pixel values that align with the processing window
that ``local_op`` is targeting. A 2D or 1D array will be sliced to
match the processing window and in the case of a 1D array tiled in
whatever dimension is flat. If an argument is a scalar it is
passed as as scalar.
The return value must be a 2D array of the same size as any of the
input parameter 2D arrays and contain the desired pixel values
for the target raster.
target_raster_path (string): the path of the output raster. The
projection, size, and cell size will be the same as the rasters
in ``base_raster_path_const_band_list`` or the final broadcast
size of the constant/ndarray values in the list.
datatype_target (gdal datatype; int): the desired GDAL output type of
the target raster.
nodata_target (numerical value): the desired nodata value of the
target raster.
calc_raster_stats (boolean): If True, calculates and sets raster
statistics (min, max, mean, and stdev) for target raster.
use_shared_memory (boolean): If True, uses Python Multiprocessing
shared memory to calculate raster stats for faster performance.
This feature is available for Python >= 3.8 and will otherwise
be ignored for earlier versions of Python.
largest_block (int): Attempts to internally iterate over raster blocks
with this many elements. Useful in cases where the blocksize is
relatively small, memory is available, and the function call
overhead dominates the iteration. Defaults to 2**20. A value of
anything less than the original blocksize of the raster will
result in blocksizes equal to the original size.
max_timeout (float): amount of time in seconds to wait for stats
worker thread to join. Default is _MAX_TIMEOUT.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to
geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
Raises:
ValueError: invalid input provided
"""
if not base_raster_path_band_const_list:
raise ValueError(
"`base_raster_path_band_const_list` is empty and "
"should have at least one value.")
# It's a common error to not pass in path/band tuples, so check for that
# and report error if so
bad_raster_path_list = False
if not isinstance(base_raster_path_band_const_list, (list, tuple)):
bad_raster_path_list = True
else:
for value in base_raster_path_band_const_list:
if (not _is_raster_path_band_formatted(value) and
not isinstance(value, numpy.ndarray) and
not (isinstance(value, tuple) and len(value) == 2 and
value[1] == 'raw')):
bad_raster_path_list = True
break
if bad_raster_path_list:
raise ValueError(
"Expected a sequence of path / integer band tuples, "
"ndarrays, or (value, 'raw') pairs for "
"`base_raster_path_band_const_list`, instead got: "
"%s" % pprint.pformat(base_raster_path_band_const_list))
# check that any rasters exist on disk and have enough bands
not_found_paths = []
gdal.PushErrorHandler('CPLQuietErrorHandler')
base_raster_path_band_list = [
path_band for path_band in base_raster_path_band_const_list
if _is_raster_path_band_formatted(path_band)]
for value in base_raster_path_band_list:
if gdal.OpenEx(value[0], gdal.OF_RASTER) is None:
not_found_paths.append(value[0])
gdal.PopErrorHandler()
if not_found_paths:
raise ValueError(
"The following files were expected but do not exist on the "
"filesystem: " + str(not_found_paths))
# check that band index exists in raster
invalid_band_index_list = []
for value in base_raster_path_band_list:
raster = gdal.OpenEx(value[0], gdal.OF_RASTER)
if not (1 <= value[1] <= raster.RasterCount):
invalid_band_index_list.append(value)
raster = None
if invalid_band_index_list:
raise ValueError(
"The following rasters do not contain requested band "
"indexes: %s" % invalid_band_index_list)
# check that the target raster is not also an input raster
if target_raster_path in [x[0] for x in base_raster_path_band_list]:
raise ValueError(
"%s is used as a target path, but it is also in the base input "
"path list %s" % (
target_raster_path, str(base_raster_path_band_const_list)))
# check that raster inputs are all the same dimensions
raster_info_list = [
get_raster_info(path_band[0])
for path_band in base_raster_path_band_const_list
if _is_raster_path_band_formatted(path_band)]
geospatial_info_set = set()
for raster_info in raster_info_list:
geospatial_info_set.add(raster_info['raster_size'])
if len(geospatial_info_set) > 1:
raise ValueError(
"Input Rasters are not the same dimensions. The "
"following raster are not identical %s" % str(
geospatial_info_set))
numpy_broadcast_list = [
x for x in base_raster_path_band_const_list
if isinstance(x, numpy.ndarray)]
stats_worker_thread = None
try:
# numpy.broadcast can only take up to 32 arguments, this loop works
# around that restriction:
while len(numpy_broadcast_list) > 1:
numpy_broadcast_list = (
[numpy.broadcast(*numpy_broadcast_list[:32])] +
numpy_broadcast_list[32:])
if numpy_broadcast_list:
numpy_broadcast_size = numpy_broadcast_list[0].shape
except ValueError:
# this gets raised if numpy.broadcast fails
raise ValueError(
"Numpy array inputs cannot be broadcast into a single shape %s" %
numpy_broadcast_list)
if numpy_broadcast_list and len(numpy_broadcast_list[0].shape) > 2:
raise ValueError(
"Numpy array inputs must be 2 dimensions or less %s" %
numpy_broadcast_list)
# if there are both rasters and arrays, check the numpy shape will
# be broadcastable with raster shape
if raster_info_list and numpy_broadcast_list:
# geospatial lists x/y order and numpy does y/x so reverse size list
raster_shape = tuple(reversed(raster_info_list[0]['raster_size']))
invalid_broadcast_size = False
if len(numpy_broadcast_size) == 1:
# if there's only one dimension it should match the last
# dimension first, in the raster case this is the columns
# because of the row/column order of numpy. No problem if
# that value is ``1`` because it will be broadcast, otherwise
# it should be the same as the raster.
if (numpy_broadcast_size[0] != raster_shape[1] and
numpy_broadcast_size[0] != 1):
invalid_broadcast_size = True
else:
for dim_index in range(2):
# no problem if 1 because it'll broadcast, otherwise must
# be the same value
if (numpy_broadcast_size[dim_index] !=
raster_shape[dim_index] and
numpy_broadcast_size[dim_index] != 1):
invalid_broadcast_size = True
if invalid_broadcast_size:
raise ValueError(
"Raster size %s cannot be broadcast to numpy shape %s" % (
raster_shape, numpy_broadcast_size))
# create a "canonical" argument list that's bands, 2d numpy arrays, or
# raw values only
base_canonical_arg_list = []
base_raster_list = []
base_band_list = []
for value in base_raster_path_band_const_list:
# the input has been tested and value is either a raster/path band
# tuple, 1d ndarray, 2d ndarray, or (value, 'raw') tuple.
if _is_raster_path_band_formatted(value):
# it's a raster/path band, keep track of open raster and band
# for later so we can `None` them.
base_raster_list.append(gdal.OpenEx(value[0], gdal.OF_RASTER))
base_band_list.append(
base_raster_list[-1].GetRasterBand(value[1]))
base_canonical_arg_list.append(base_band_list[-1])
elif isinstance(value, numpy.ndarray):
if value.ndim == 1:
# easier to process as a 2d array for writing to band
base_canonical_arg_list.append(
value.reshape((1, value.shape[0])))
else: # dimensions are two because we checked earlier.
base_canonical_arg_list.append(value)
elif isinstance(value, tuple):
base_canonical_arg_list.append(value)
else:
raise ValueError(
"An unexpected ``value`` occurred. This should never happen. "
"Value: %r" % value)
# create target raster
if raster_info_list:
# if rasters are passed, the target is the same size as the raster
n_cols, n_rows = raster_info_list[0]['raster_size']
elif numpy_broadcast_list:
# numpy arrays in args and no raster result is broadcast shape
# expanded to two dimensions if necessary
if len(numpy_broadcast_size) == 1:
n_rows, n_cols = 1, numpy_broadcast_size[0]
else:
n_rows, n_cols = numpy_broadcast_size
else:
raise ValueError(
"Only (object, 'raw') values have been passed. Raster "
"calculator requires at least a raster or numpy array as a "
"parameter. This is the input list: %s" % pprint.pformat(
base_raster_path_band_const_list))
if datatype_target not in _VALID_GDAL_TYPES:
raise ValueError(
'Invalid target type, should be a gdal.GDT_* type, received '
'"%s"' % datatype_target)
# create target raster
raster_driver = gdal.GetDriverByName(raster_driver_creation_tuple[0])
try:
os.makedirs(os.path.dirname(target_raster_path))
except OSError:
pass
target_raster = raster_driver.Create(
target_raster_path, n_cols, n_rows, 1, datatype_target,
options=raster_driver_creation_tuple[1])
target_band = target_raster.GetRasterBand(1)
if nodata_target is not None:
target_band.SetNoDataValue(nodata_target)
if base_raster_list:
# use the first raster in the list for the projection and geotransform
target_raster.SetProjection(base_raster_list[0].GetProjection())
target_raster.SetGeoTransform(base_raster_list[0].GetGeoTransform())
target_band.FlushCache()
target_raster.FlushCache()
try:
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
block_offset_list = list(iterblocks(
(target_raster_path, 1), offset_only=True,
largest_block=largest_block))
if calc_raster_stats:
# if this queue is used to send computed valid blocks of
# the raster to an incremental statistics calculator worker
stats_worker_queue = queue.Queue()
exception_queue = queue.Queue()
if sys.version_info >= (3, 8) and use_shared_memory:
# The stats worker keeps running variables as a float64, so
# all input rasters are dtype float64 -- make the shared memory
# size equivalent.
block_size_bytes = (
numpy.dtype(numpy.float64).itemsize *
block_offset_list[0]['win_xsize'] *
block_offset_list[0]['win_ysize'])
shared_memory = multiprocessing.shared_memory.SharedMemory(
create=True, size=block_size_bytes)
else:
stats_worker_queue = None
if calc_raster_stats:
# To avoid doing two passes on the raster to calculate standard
# deviation, we implement a continuous statistics calculation
# as the raster is computed. This computational effort is high
# and benefits from running in parallel. This queue and worker
# takes a valid block of a raster and incrementally calculates
# the raster's statistics. When ``None`` is pushed to the queue
# the worker will finish and return a (min, max, mean, std)
# tuple.
LOGGER.info('starting stats_worker')
stats_worker_thread = threading.Thread(
target=geoprocessing_core.stats_worker,
args=(stats_worker_queue, len(block_offset_list)))
stats_worker_thread.daemon = True
stats_worker_thread.start()
LOGGER.info('started stats_worker %s', stats_worker_thread)
pixels_processed = 0
n_pixels = n_cols * n_rows
# iterate over each block and calculate local_op
for block_offset in block_offset_list:
# read input blocks
offset_list = (block_offset['yoff'], block_offset['xoff'])
blocksize = (block_offset['win_ysize'], block_offset['win_xsize'])
data_blocks = []
for value in base_canonical_arg_list:
if isinstance(value, gdal.Band):
data_blocks.append(value.ReadAsArray(**block_offset))
# I've encountered the following error when a gdal raster
# is corrupt, often from multiple threads writing to the
# same file. This helps to catch the error early rather
# than lead to confusing values of ``data_blocks`` later.
if not isinstance(data_blocks[-1], numpy.ndarray):
raise ValueError(
f"got a {data_blocks[-1]} when trying to read "
f"{value.GetDataset().GetFileList()} at "
f"{block_offset}, expected numpy.ndarray.")
elif isinstance(value, numpy.ndarray):
# must be numpy array and all have been conditioned to be
# 2d, so start with 0:1 slices and expand if possible
slice_list = [slice(0, 1)] * 2
tile_dims = list(blocksize)
for dim_index in [0, 1]:
if value.shape[dim_index] > 1:
slice_list[dim_index] = slice(
offset_list[dim_index],
offset_list[dim_index] +
blocksize[dim_index],)
tile_dims[dim_index] = 1
data_blocks.append(
numpy.tile(value[tuple(slice_list)], tile_dims))
else:
# must be a raw tuple
data_blocks.append(value[0])
target_block = local_op(*data_blocks)
if (not isinstance(target_block, numpy.ndarray) or
target_block.shape != blocksize):
raise ValueError(
"Expected `local_op` to return a numpy.ndarray of "
"shape %s but got this instead: %s" % (
blocksize, target_block))
target_band.WriteArray(
target_block, yoff=block_offset['yoff'],
xoff=block_offset['xoff'])
# send result to stats calculator
if stats_worker_queue:
# guard against an undefined nodata target
if nodata_target is not None:
target_block = target_block[target_block != nodata_target]
target_block = target_block.astype(numpy.float64).flatten()
if sys.version_info >= (3, 8) and use_shared_memory:
shared_memory_array = numpy.ndarray(
target_block.shape, dtype=target_block.dtype,
buffer=shared_memory.buf)
shared_memory_array[:] = target_block[:]
stats_worker_queue.put((
shared_memory_array.shape, shared_memory_array.dtype,
shared_memory))
else:
stats_worker_queue.put(target_block)
pixels_processed += blocksize[0] * blocksize[1]
timed_logger.info(
'%s %.1f%% complete',
os.path.basename(target_raster_path),
float(pixels_processed) / n_pixels * 100.0)
LOGGER.info('100.0% complete')
if calc_raster_stats:
LOGGER.info("Waiting for raster stats worker result.")
stats_worker_thread.join(max_timeout)
if stats_worker_thread.is_alive():
LOGGER.error("stats_worker_thread.join() timed out")
raise RuntimeError("stats_worker_thread.join() timed out")
payload = stats_worker_queue.get(True, max_timeout)
if payload is not None:
target_min, target_max, target_mean, target_stddev = payload
# In Cython 3.0.0+, taking a square root may return a complex.
# Using only the real component of the complex value mimics the
# C behavior that we expect from our stats worker.
if isinstance(target_stddev, complex):
target_stddev = target_stddev.real
target_band.SetStatistics(
float(target_min), float(target_max), float(target_mean),
float(target_stddev))
target_band.FlushCache()
except Exception:
LOGGER.exception('exception encountered in raster_calculator')
raise
finally:
# This block ensures that rasters are destroyed even if there's an
# exception raised.
base_band_list[:] = []
base_raster_list[:] = []
target_band.FlushCache()
target_band = None
target_raster.FlushCache()
target_raster = None
if calc_raster_stats and stats_worker_thread:
if stats_worker_thread.is_alive():
stats_worker_queue.put(None, True, max_timeout)
LOGGER.info("Waiting for raster stats worker result.")
stats_worker_thread.join(max_timeout)
if stats_worker_thread.is_alive():
LOGGER.error("stats_worker_thread.join() timed out")
raise RuntimeError(
"stats_worker_thread.join() timed out")
if sys.version_info >= (3, 8) and use_shared_memory:
LOGGER.debug(
f'unlink shared memory for process {os.getpid()}')
shared_memory.close()
shared_memory.unlink()
LOGGER.debug(
f'unlinked shared memory for process {os.getpid()}')
# check for an exception in the workers, otherwise get result
# and pass to writer
try:
exception = exception_queue.get_nowait()
LOGGER.error("Exception encountered at termination.")
raise exception
except queue.Empty:
pass
[docs]def array_equals_nodata(array, nodata):
"""Check for the presence of ``nodata`` values in ``array``.
The comparison supports ``numpy.nan`` and unset (``None``) nodata values.
Args:
array (numpy array): the array to mask for nodata values.
nodata (number): the nodata value to check for. Supports ``numpy.nan``.
Returns:
A boolean numpy array with values of 1 where ``array`` is equal to
``nodata`` and 0 otherwise.
"""
# If nodata is undefined, nothing matches nodata.
if nodata is None:
return numpy.zeros(array.shape, dtype=bool)
# comparing an integer array against numpy.nan works correctly and is
# faster than using numpy.isclose().
if numpy.issubdtype(array.dtype, numpy.integer):
return array == nodata
return numpy.isclose(array, nodata, equal_nan=True)
[docs]def choose_dtype(*raster_paths):
"""
Choose an appropriate dtype for an output derived from the given inputs.
Returns the dtype with the greatest size/precision among the inputs, so
that information will not be lost.
Args:
*raster_paths: series of raster path strings
Returns:
numpy dtype
"""
dtypes = [get_raster_info(path)['numpy_type'] for path in raster_paths]
return numpy.result_type(*dtypes)
[docs]def choose_nodata(dtype):
"""
Choose an appropriate nodata value for data of a given dtype.
Args:
dtype (numpy.dtype): data type for which to choose nodata
Returns:
number to use as nodata value
"""
try:
return float(numpy.finfo(dtype).max)
except ValueError:
return int(numpy.iinfo(dtype).max)
[docs]def raster_map(op, rasters, target_path, target_nodata=None, target_dtype=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Apply a pixelwise function to a series of raster inputs.
The output raster will have nodata where any input raster has nodata.
Raster inputs are split into aligned blocks, and the function is
applied individually to each stack of blocks (as numpy arrays).
Args:
op (function): Function to apply to the inputs. It should accept a
number of arguments equal to the length of ``*inputs``. It should
return a numpy array with the same shape as its array input(s).
rasters (list[str]): Paths to rasters to input to ``op``, in the order
that they will be passed to ``op``. All rasters should be aligned
and have the same dimensions.
target_path (str): path to write out the output raster.
target_nodata (number): Nodata value to use for the output raster.
Optional. If not provided, a suitable nodata value will be chosen.
target_dtype (numpy.dtype): dtype to use for the output. Optional. If
not provided, a suitable dtype will be chosen.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to
geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS. If the
``target_dtype`` is int8, the ``PIXELTYPE=SIGNEDBYTE`` option will
be added to the creation options tuple if it is not already there.
Returns:
``None``
"""
nodatas = []
for raster in rasters:
raster_info = get_raster_info(raster)
if raster_info['n_bands'] > 1:
LOGGER.warning(f'{raster} has more than one band. Only the first '
'band will be used.')
nodatas.append(raster_info['nodata'][0])
# choose an appropriate dtype if none was given
if target_dtype is None:
target_dtype = choose_dtype(*rasters)
# choose an appropriate nodata value if none was given
# if the user provides a target nodata,
# check that it can fit in the target dtype
if target_nodata is None:
target_nodata = choose_nodata(target_dtype)
else:
if not numpy.can_cast(target_nodata, target_dtype):
raise ValueError(
f'Target nodata value {target_nodata} is incompatible with '
f'the target dtype {target_dtype}')
driver, options = raster_driver_creation_tuple
gdal_type, type_creation_options = _numpy_to_gdal_type(target_dtype)
options = list(options) + type_creation_options
def apply_op(*arrays):
"""Apply the function ``op`` to the input arrays.
Args:
*arrays: numpy arrays with the same shape.
Returns:
numpy array
"""
result = numpy.full(arrays[0].shape, target_nodata, dtype=target_dtype)
# make a mask that is True where all input arrays are valid,
# and False where any input array is invalid.
valid_mask = numpy.full(arrays[0].shape, True)
for array, nodata in zip(arrays, nodatas):
valid_mask &= ~array_equals_nodata(array, nodata)
# mask all arrays to the area where they all are valid
masked_arrays = [array[valid_mask] for array in arrays]
# apply op to the masked arrays in order
result[valid_mask] = op(*masked_arrays)
return result
raster_calculator(
[(path, 1) for path in rasters], # assume the first band
apply_op,
target_path,
gdal_type,
target_nodata,
raster_driver_creation_tuple=(driver, options))
[docs]def raster_reduce(function, raster_path_band, initializer, mask_nodata=True,
largest_block=_LARGEST_ITERBLOCK):
"""Cumulatively apply a reducing function to each block of a raster.
This effectively reduces the entire raster to a single value, but it works
by blocks to be memory-efficient.
The ``function`` signature should be ``function(aggregator, block)``, where
``aggregator`` is the aggregated value so far, and ``block`` is a flattened
numpy array containing the data from the block to reduce next.
``function`` is called once on each block. On the first ``function`` call,
``aggregator`` is initialized with ``initializer``. The return value from
each ``function`` call is passed in as the ``aggregator`` argument to the
subsequent ``function`` call. When all blocks have been reduced, the return
value of the final ``function`` call is returned.
Example:
Calculate the sum of all values in a raster::
raster_reduce(lambda total, block: total + numpy.sum(block),
(raster_path, 1), 0)
Calculate a histogram of all values in a raster::
def add_to_histogram(histogram, block):
return histogram + numpy.histogram(block, bins=10)[0]
raster_reduce(add_to_histogram, (raster_path, 1), numpy.zeros(10))
Calculate the sum of all values in a raster, excluding nodata::
nodata = pygeoprocessing.get_raster_info(raster_path)['nodata'][0]
def sum_excluding_nodata(total, block):
return total + numpy.sum(block[block != nodata])
raster_reduce(sum_excluding_nodata, (raster_path, 1), 0)
Args:
function (func): function to apply to each raster block
raster_path_band (tuple): (path, band) tuple of the raster to reduce
initializer (obj): value to initialize the aggregator for the
first function call
mask_nodata (bool): if True, mask out nodata before aggregating. A
flattened array of non-nodata pixels from each block is passed to
the ``function``. if False, each block is passed to the
``function`` without masking.
largest_block (int): largest block parameter to pass to ``iterblocks``
Returns:
aggregate value, the final value returned from ``function``
"""
aggregator = initializer
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
pixels_processed = 0
raster_info = get_raster_info(raster_path_band[0])
x_size, y_size = raster_info['raster_size']
n_pixels = x_size * y_size
for (_, block) in iterblocks(raster_path_band,
largest_block=largest_block):
if mask_nodata:
data = block[~array_equals_nodata(
block, raster_info['nodata'][raster_path_band[1] - 1])]
else:
data = block.flatten()
aggregator = function(aggregator, data)
pixels_processed += block.size
timed_logger.info(
f'{raster_path_band[0]} reduce '
f'{pixels_processed / n_pixels * 100:.1f}%% complete'
)
LOGGER.info('100.0%% complete')
return aggregator
[docs]def align_and_resize_raster_stack(
base_raster_path_list, target_raster_path_list, resample_method_list,
target_pixel_size, bounding_box_mode, base_vector_path_list=None,
raster_align_index=None, base_projection_wkt_list=None,
target_projection_wkt=None, mask_options=None,
vector_mask_options=None, gdal_warp_options=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS,
osr_axis_mapping_strategy=DEFAULT_OSR_AXIS_MAPPING_STRATEGY,
working_dir=None):
"""Generate rasters from a base such that they align geospatially.
This function resizes base rasters that are in the same geospatial
projection such that the result is an aligned stack of rasters that have
the same cell size, dimensions, and bounding box. This is achieved by
clipping or resizing the rasters to intersected, unioned, or equivocated
bounding boxes of all the raster and vector input.
Args:
base_raster_path_list (sequence): a sequence of base raster paths that
will be transformed and will be used to determine the target
bounding box.
target_raster_path_list (sequence): a sequence of raster paths that
will be created to one-to-one map with ``base_raster_path_list``
as aligned versions of those original rasters. If there are
duplicate paths in this list, the function will raise a ValueError.
resample_method_list (sequence): a sequence of resampling methods
which one to one map each path in ``base_raster_path_list`` during
resizing. Each element must be one of
"near|bilinear|cubic|cubicspline|lanczos|mode".
target_pixel_size (list/tuple): the target raster's x and y pixel size
example: (30, -30).
bounding_box_mode (string): one of "union", "intersection", or
a sequence of floats of the form [minx, miny, maxx, maxy] in the
target projection coordinate system. Depending
on the value, output extents are defined as the union,
intersection, or the explicit bounding box.
base_vector_path_list (sequence): a sequence of base vector paths
whose bounding boxes will be used to determine the final bounding
box of the raster stack if mode is 'union' or 'intersection'. If
mode is 'bb=[...]' then these vectors are not used in any
calculation.
raster_align_index (int): indicates the index of a
raster in ``base_raster_path_list`` that the target rasters'
bounding boxes pixels should align with. This feature allows
rasters whose raster dimensions are the same, but bounding boxes
slightly shifted less than a pixel size to align with a desired
grid layout. If ``None`` then the bounding box of the target
rasters is calculated as the precise intersection, union, or
bounding box.
base_projection_wkt_list (sequence): if not ``None``, this is a
sequence of base projections of the rasters in
``base_raster_path_list``. If a value is ``None``, the projection
is read directly from the raster. Use this argument if there are
rasters with no projection defined, but the projections are known.
target_projection_wkt (string): if not ``None``, this is the desired
projection of all target rasters in Well Known Text format, and
target rasters will be warped to this projection. If ``None``,
the base SRS will be passed to the target.
mask_options (dict): optional, if not None, this is a
dictionary of options to use an existing vector's geometry to
mask out pixels in the target raster that do not overlap the
vector's geometry. Keys to this dictionary are:
* ``'mask_vector_path'`` (str): path to the mask vector file.
This vector will be automatically projected to the target
projection if its base coordinate system does not match the
target.
* ``'mask_layer_name'`` (str): the layer name to use for masking.
If this key is not in the dictionary the default is to use
the layer at index 0.
* ``'mask_vector_where_filter'`` (str): an SQL WHERE string.
This will be used to filter the geometry in the mask. Ex: ``'id
> 10'`` would use all features whose field value of 'id' is >
10.
* ``'mask_raster_path'`` (str): Optional. the string path to where
the mask raster should be written on disk. If not provided, a
temporary file will be created within ``working_dir``.
vector_mask_options (dict): optional. Alias for ``mask_options``.
This parameter is deprecated and will be removed in a future
version of ``pygeoprocessing``.
gdal_warp_options (sequence): if present, the contents of this list
are passed to the ``warpOptions`` parameter of ``gdal.Warp``. See
the `GDAL Warp documentation
<https://gdal.org/api/gdalwarp_cpp.html#_CPPv415GDALWarpOptions>`_
for valid options.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
osr_axis_mapping_strategy (int): OSR axis mapping strategy for
``SpatialReference`` objects. Defaults to
``geoprocessing.DEFAULT_OSR_AXIS_MAPPING_STRATEGY``. This parameter
should not be changed unless you know what you are doing.
working_dir=None (str): if present, the path to a directory within
which a temporary directory will be created. If not provided, the
new directory will be created within the system's temporary
directory.
Returns:
None
Raises:
ValueError
If any combination of the raw bounding boxes, raster
bounding boxes, vector bounding boxes, and/or vector_mask
bounding box does not overlap to produce a valid target.
ValueError
If any of the input or target lists are of different
lengths.
ValueError
If there are duplicate paths on the target list which would
risk corrupted output.
ValueError
If some combination of base, target, and embedded source
reference systems results in an ambiguous target coordinate
system.
ValueError
If ``mask_options`` is not None but the
``mask_vector_path`` is undefined or doesn't point to a valid
file.
ValueError
If ``pixel_size`` is not a 2 element sequence of numbers.
"""
if vector_mask_options is not None:
warnings.warn('The vector_mask_options parameter is deprecated and '
'will be removed in a future release of '
'pygeoprocessing. Please use mask_options instead.',
DeprecationWarning)
mask_options = vector_mask_options
# make sure that the input lists are of the same length
list_lengths = [
len(base_raster_path_list), len(target_raster_path_list),
len(resample_method_list)]
if len(set(list_lengths)) != 1:
raise ValueError(
"base_raster_path_list, target_raster_path_list, and "
"resample_method_list must be the same length "
" current lengths are %s" % (str(list_lengths)))
unique_targets = set(target_raster_path_list)
if len(unique_targets) != len(target_raster_path_list):
seen = set()
duplicate_list = []
for path in target_raster_path_list:
if path not in seen:
seen.add(path)
else:
duplicate_list.append(path)
raise ValueError(
"There are duplicated paths on the target list. This is an "
"invalid state of ``target_path_list``. Duplicates: %s" % (
duplicate_list))
# we can accept 'union', 'intersection', or a 4 element list/tuple
if bounding_box_mode not in ["union", "intersection"] and (
not isinstance(bounding_box_mode, (list, tuple)) or
len(bounding_box_mode) != 4):
raise ValueError("Unknown bounding_box_mode %s" % (
str(bounding_box_mode)))
n_rasters = len(base_raster_path_list)
if ((raster_align_index is not None) and
((raster_align_index < 0) or (raster_align_index >= n_rasters))):
raise ValueError(
"Alignment index is out of bounds of the datasets index: %s"
" n_elements %s" % (raster_align_index, n_rasters))
_assert_is_valid_pixel_size(target_pixel_size)
# used to get bounding box, projection, and possible alignment info
raster_info_list = [
get_raster_info(path) for path in base_raster_path_list]
# get the literal or intersecting/unioned bounding box
if isinstance(bounding_box_mode, (list, tuple)):
# if it's a sequence or tuple, it must be a manual bounding box
LOGGER.debug(
"assuming manual bounding box mode of %s", bounding_box_mode)
target_bounding_box = bounding_box_mode
else:
# either intersection or union, get list of bounding boxes, reproject
# if necessary, and reduce to a single box
if base_vector_path_list is not None:
# vectors are only interesting for their bounding boxes, that's
# this construction is inside an else.
vector_info_list = [
get_vector_info(path) for path in base_vector_path_list]
else:
vector_info_list = []
raster_bounding_box_list = []
for raster_index, raster_info in enumerate(raster_info_list):
# this block calculates the base projection of ``raster_info`` if
# ``target_projection_wkt`` is defined, thus implying a
# reprojection will be necessary.
if target_projection_wkt:
if base_projection_wkt_list and \
base_projection_wkt_list[raster_index]:
# a base is defined, use that
base_raster_projection_wkt = \
base_projection_wkt_list[raster_index]
else:
# otherwise use the raster's projection and there must
# be one since we're reprojecting
base_raster_projection_wkt = raster_info['projection_wkt']
if not base_raster_projection_wkt:
raise ValueError(
"no projection for raster %s" %
base_raster_path_list[raster_index])
# since the base spatial reference is potentially different
# than the target, we need to transform the base bounding
# box into target coordinates so later we can calculate
# accurate bounding box overlaps in the target coordinate
# system
raster_bounding_box_list.append(
transform_bounding_box(
raster_info['bounding_box'],
base_raster_projection_wkt, target_projection_wkt))
else:
raster_bounding_box_list.append(raster_info['bounding_box'])
# include the vector bounding box information to make a global list
# of target bounding boxes
bounding_box_list = [
vector_info['bounding_box'] if target_projection_wkt is None else
transform_bounding_box(
vector_info['bounding_box'],
vector_info['projection_wkt'], target_projection_wkt)
for vector_info in vector_info_list] + raster_bounding_box_list
target_bounding_box = merge_bounding_box_list(
bounding_box_list, bounding_box_mode)
if mask_options:
# ensure the mask exists and intersects with the target bounding box
if 'mask_vector_path' not in mask_options:
raise ValueError(
'mask_options passed, but no value for '
'"mask_vector_path": %s', mask_options)
mask_vector_info = get_vector_info(
mask_options['mask_vector_path'])
if 'mask_vector_where_filter' in mask_options:
# the bounding box only exists for the filtered features
mask_vector = gdal.OpenEx(
mask_options['mask_vector_path'], gdal.OF_VECTOR)
mask_layer = mask_vector.GetLayer()
mask_layer.SetAttributeFilter(
mask_options['mask_vector_where_filter'])
mask_bounding_box = merge_bounding_box_list(
[[feature.GetGeometryRef().GetEnvelope()[i]
for i in [0, 2, 1, 3]] for feature in mask_layer],
'union')
mask_layer = None
mask_vector = None
else:
# if no where filter then use the raw vector bounding box
mask_bounding_box = mask_vector_info['bounding_box']
mask_vector_projection_wkt = mask_vector_info['projection_wkt']
if mask_vector_projection_wkt is not None and \
target_projection_wkt is not None:
mask_vector_bb = transform_bounding_box(
mask_bounding_box, mask_vector_info['projection_wkt'],
target_projection_wkt)
else:
mask_vector_bb = mask_vector_info['bounding_box']
# Calling `merge_bounding_box_list` will raise an ValueError if the
# bounding box of the mask and the target do not intersect. The
# result is otherwise not used.
_ = merge_bounding_box_list(
[target_bounding_box, mask_vector_bb], 'intersection')
if raster_align_index is not None and raster_align_index >= 0:
# bounding box needs alignment
align_bounding_box = (
raster_info_list[raster_align_index]['bounding_box'])
align_pixel_size = (
raster_info_list[raster_align_index]['pixel_size'])
# adjust bounding box so lower left corner aligns with a pixel in
# raster[raster_align_index]
for index in [0, 1]:
n_pixels = int(
(target_bounding_box[index] - align_bounding_box[index]) /
float(align_pixel_size[index]))
target_bounding_box[index] = (
n_pixels * align_pixel_size[index] +
align_bounding_box[index])
if mask_options:
# Create a warped VRT.
# This allows us to cheaply figure out the dimensions, projection, etc.
# of the target raster without actually warping the entire raster to a
# GTiff.
temp_working_dir = tempfile.mkdtemp(dir=working_dir, prefix='mask-')
warped_vrt = os.path.join(temp_working_dir, 'warped.vrt')
warp_raster(
base_raster_path=base_raster_path_list[0],
target_pixel_size=target_pixel_size,
target_raster_path=warped_vrt,
resample_method='near',
target_bb=target_bounding_box,
raster_driver_creation_tuple=('VRT', []),
target_projection_wkt=target_projection_wkt,
base_projection_wkt=(
None if not base_projection_wkt_list else
base_projection_wkt_list[0]),
gdal_warp_options=gdal_warp_options)
# Convert the warped VRT to a GTiff for rasterization.
if 'mask_raster_path' in mask_options:
mask_raster_path = mask_options['mask_raster_path']
else:
# Add the mask raster path ot the vector mask options to force
# warp_raster to use the existing raster mask.
mask_raster_path = os.path.join(temp_working_dir, 'mask.tif')
mask_options['mask_raster_path'] = mask_raster_path
new_raster_from_base(
warped_vrt, mask_raster_path, gdal.GDT_Byte, [0], [0])
# Rasterize the vector onto the new GTiff.
rasterize(mask_options['mask_vector_path'],
mask_raster_path, burn_values=[1],
where_clause=(
mask_options['mask_vector_where_filter']
if 'mask_vector_where_filter' in mask_options
else None))
for index, (base_path, target_path, resample_method) in enumerate(zip(
base_raster_path_list, target_raster_path_list,
resample_method_list)):
warp_raster(
base_path, target_pixel_size, target_path, resample_method,
target_bb=target_bounding_box,
raster_driver_creation_tuple=(raster_driver_creation_tuple),
target_projection_wkt=target_projection_wkt,
base_projection_wkt=(
None if not base_projection_wkt_list else
base_projection_wkt_list[index]),
mask_options=mask_options,
gdal_warp_options=gdal_warp_options)
LOGGER.info(
'%d of %d aligned: %s', index+1, n_rasters,
os.path.basename(target_path))
LOGGER.info("aligned all %d rasters.", n_rasters)
if mask_options:
shutil.rmtree(temp_working_dir, ignore_errors=True)
[docs]def new_raster_from_base(
base_path, target_path, datatype, band_nodata_list,
fill_value_list=None, n_rows=None, n_cols=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Create new raster by coping spatial reference/geotransform of base.
A convenience function to simplify the creation of a new raster from the
basis of an existing one. Depending on the input mode, one can create
a new raster of the same dimensions, geotransform, and georeference as
the base. Other options are provided to change the raster dimensions,
number of bands, nodata values, data type, and core raster creation
options.
Args:
base_path (string): path to existing raster.
target_path (string): path to desired target raster.
datatype: the pixel datatype of the output raster, for example
gdal.GDT_Float32. See the following header file for supported
pixel types:
http://www.gdal.org/gdal_8h.html#22e22ce0a55036a96f652765793fb7a4
band_nodata_list (sequence): list of nodata values, one for each band,
to set on target raster. If value is 'None' the nodata value is
not set for that band. The number of target bands is inferred
from the length of this list.
fill_value_list (sequence): list of values to fill each band with. If
None, no filling is done.
n_rows (int): if not None, defines the number of target raster rows.
n_cols (int): if not None, defines the number of target raster
columns.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
"""
base_raster = gdal.OpenEx(base_path, gdal.OF_RASTER)
if n_rows is None:
n_rows = base_raster.RasterYSize
if n_cols is None:
n_cols = base_raster.RasterXSize
driver = gdal.GetDriverByName(raster_driver_creation_tuple[0])
local_raster_creation_options = list(raster_driver_creation_tuple[1])
numpy_dtype = _gdal_to_numpy_type(datatype, local_raster_creation_options)
base_band = base_raster.GetRasterBand(1)
block_size = base_band.GetBlockSize()
# It's not clear how or IF we can determine if the output should be
# striped or tiled. Here we leave it up to the default inputs or if its
# obviously not striped we tile.
if not any(
['TILED' in option for option in local_raster_creation_options]):
# TILED not set, so lets try to set it to a reasonable value
if block_size[0] != n_cols:
# if x block is not the width of the raster it *must* be tiled
# otherwise okay if it's striped or tiled, I can't construct a
# test case to cover this, but there is nothing in the spec that
# restricts this so I have it just in case.
local_raster_creation_options.append('TILED=YES')
if not any(
['BLOCK' in option for option in local_raster_creation_options]):
# not defined, so lets copy what we know from the current raster
local_raster_creation_options.extend([
'BLOCKXSIZE=%d' % block_size[0],
'BLOCKYSIZE=%d' % block_size[1]])
# make target directory if it doesn't exist
try:
os.makedirs(os.path.dirname(target_path))
except OSError:
pass
base_band = None
n_bands = len(band_nodata_list)
target_raster = driver.Create(
target_path, n_cols, n_rows, n_bands, datatype,
options=local_raster_creation_options)
target_raster.SetProjection(base_raster.GetProjection())
target_raster.SetGeoTransform(base_raster.GetGeoTransform())
base_raster = None
for index, nodata_value in enumerate(band_nodata_list):
if nodata_value is None:
continue
target_band = target_raster.GetRasterBand(index + 1)
try:
target_band.SetNoDataValue(nodata_value.item())
except AttributeError:
target_band.SetNoDataValue(nodata_value)
target_raster.FlushCache()
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
pixels_processed = 0
n_pixels = n_cols * n_rows
if fill_value_list is not None:
for index, fill_value in enumerate(fill_value_list):
if fill_value is None:
continue
target_band = target_raster.GetRasterBand(index + 1)
# some rasters are very large and a fill can appear to cause
# computation to hang. This block, though possibly slightly less
# efficient than ``band.Fill`` will give real-time feedback about
# how the fill is progressing.
for offsets in iterblocks((target_path, 1), offset_only=True):
shape = (offsets['win_ysize'], offsets['win_xsize'])
fill_array = numpy.full(shape, fill_value, dtype=numpy_dtype)
pixels_processed += (
offsets['win_ysize'] * offsets['win_xsize'])
target_band.WriteArray(
fill_array, offsets['xoff'], offsets['yoff'])
timed_logger.info(
f'filling new raster {target_path} with {fill_value} '
f'-- {float(pixels_processed)/n_pixels*100.0:.2f}% '
f'complete')
target_band = None
target_band = None
target_raster = None
[docs]def create_raster_from_vector_extents(
base_vector_path, target_raster_path, target_pixel_size,
target_pixel_type, target_nodata, fill_value=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Create a blank raster based on a vector file extent.
Args:
base_vector_path (string): path to vector shapefile to base the
bounding box for the target raster.
target_raster_path (string): path to location of generated geotiff;
the upper left hand corner of this raster will be aligned with the
bounding box of the source vector and the extent will be exactly
equal or contained the source vector's bounding box depending on
whether the pixel size divides evenly into the source bounding
box; if not coordinates will be rounded up to contain the original
extent.
target_pixel_size (list/tuple): the x/y pixel size as a sequence
Example::
[30.0, -30.0]
target_pixel_type (int): gdal GDT pixel type of target raster
target_nodata (numeric): target nodata value. Can be None if no nodata
value is needed.
fill_value (int/float): value to fill in the target raster; no fill if
value is None
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
"""
if target_pixel_type not in _VALID_GDAL_TYPES:
raise ValueError(
f'Invalid target type, should be a gdal.GDT_* type, received '
f'"{target_pixel_type}"')
# Determine the width and height of the tiff in pixels based on the
# maximum size of the combined envelope of all the features
vector = gdal.OpenEx(base_vector_path, gdal.OF_VECTOR)
shp_extent = None
for layer_index in range(vector.GetLayerCount()):
layer = vector.GetLayer(layer_index)
for feature in layer:
try:
# envelope is [xmin, xmax, ymin, ymax]
feature_extent = feature.GetGeometryRef().GetEnvelope()
if shp_extent is None:
shp_extent = list(feature_extent)
else:
# expand bounds of current bounding box to include that
# of the newest feature
shp_extent = [
f(shp_extent[index], feature_extent[index])
for index, f in enumerate([min, max, min, max])]
except AttributeError as error:
# For some valid OGR objects the geometry can be undefined
# since it's valid to have a NULL entry in the attribute table
# this is expressed as a None value in the geometry reference
# this feature won't contribute
LOGGER.warning(error)
layer = None
target_srs_wkt = vector.GetLayer(0).GetSpatialRef().ExportToWkt()
vector = None
if shp_extent is None:
raise ValueError(
f'the vector at {base_vector_path} has no geometry, cannot '
f'create a raster from these extents')
create_raster_from_bounding_box(
target_bounding_box=[
shp_extent[0], shp_extent[2], shp_extent[1], shp_extent[3]
],
target_raster_path=target_raster_path,
target_pixel_size=target_pixel_size,
target_pixel_type=target_pixel_type,
target_srs_wkt=target_srs_wkt,
target_nodata=target_nodata,
fill_value=fill_value,
raster_driver_creation_tuple=raster_driver_creation_tuple
)
[docs]def create_raster_from_bounding_box(
target_bounding_box, target_raster_path, target_pixel_size,
target_pixel_type, target_srs_wkt, target_nodata, fill_value=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Create a raster from a given bounding box.
Args:
target_bounding_box (tuple): a 4-element iterable of (minx, miny,
maxx, maxy) in projected units matching the SRS of
``target_srs_wkt``.
target_raster_path (string): The path to where the new raster should be
created on disk.
target_pixel_size (tuple): A 2-element tuple of the (x, y) pixel size
of the target raster. Elements are in units of the target SRS.
target_pixel_type (int): The GDAL GDT_* type of the target raster.
target_srs_wkt (string): The SRS of the target raster, in Well-Known
Text format.
target_nodata (float): The nodata value of the target raster, or
``None`` if no nodata value is to be set.
fill_value=None (number): If provided, the value that the target raster
should be filled with.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Returns:
``None``
"""
if target_pixel_type not in _VALID_GDAL_TYPES:
raise ValueError(
f'Invalid target type, should be a gdal.GDT_* type, received '
f'"{target_pixel_type}"')
bbox_minx, bbox_miny, bbox_maxx, bbox_maxy = target_bounding_box
driver = gdal.GetDriverByName(raster_driver_creation_tuple[0])
n_bands = 1
# determine the raster size that bounds the input bounding box and then
# adjust the bounding box to be that size
target_x_size = int(abs(
float(bbox_maxx - bbox_minx) / target_pixel_size[0]))
target_y_size = int(abs(
float(bbox_maxy - bbox_miny) / target_pixel_size[1]))
x_residual = (
abs(target_x_size * target_pixel_size[0]) -
(bbox_maxx - bbox_minx))
if not numpy.isclose(x_residual, 0.0):
target_x_size += 1
y_residual = (
abs(target_y_size * target_pixel_size[1]) -
(bbox_maxy - bbox_miny))
if not numpy.isclose(y_residual, 0.0):
target_y_size += 1
if target_x_size == 0:
LOGGER.warning(
"bounding_box is so small that x dimension rounds to 0; "
"clamping to 1.")
target_x_size = 1
if target_y_size == 0:
LOGGER.warning(
"bounding_box is so small that y dimension rounds to 0; "
"clamping to 1.")
target_y_size = 1
raster = driver.Create(
target_raster_path, target_x_size, target_y_size, n_bands,
target_pixel_type, options=raster_driver_creation_tuple[1])
raster.SetProjection(target_srs_wkt)
# Set the transform based on the upper left corner and given pixel
# dimensions.
x_source = bbox_maxx if target_pixel_size[0] < 0 else bbox_minx
y_source = bbox_maxy if target_pixel_size[1] < 0 else bbox_miny
raster_transform = [
x_source, target_pixel_size[0], 0,
y_source, 0, target_pixel_size[1]]
raster.SetGeoTransform(raster_transform)
# Fill the band if requested.
band = raster.GetRasterBand(1)
if fill_value is not None:
band.Fill(fill_value)
# Set the nodata value.
if target_nodata is not None:
band.SetNoDataValue(float(target_nodata))
band = None
raster = None
[docs]def interpolate_points(
base_vector_path, vector_attribute_field, target_raster_path_band,
interpolation_mode):
"""Interpolate point values onto an existing raster.
Args:
base_vector_path (string): path to a shapefile that contains point
vector layers.
vector_attribute_field (field): a string in the vector referenced at
``base_vector_path`` that refers to a numeric value in the
vector's attribute table. This is the value that will be
interpolated across the raster.
target_raster_path_band (tuple): a path/band number tuple to an
existing raster which likely intersects or is nearby the source
vector. The band in this raster will take on the interpolated
numerical values provided at each point.
interpolation_mode (string): the interpolation method to use for
scipy.interpolate.griddata, one of 'linear', near', or 'cubic'.
Return:
None
"""
source_vector = gdal.OpenEx(base_vector_path, gdal.OF_VECTOR)
point_list = []
value_list = []
for layer_index in range(source_vector.GetLayerCount()):
layer = source_vector.GetLayer(layer_index)
for point_feature in layer:
value = point_feature.GetField(vector_attribute_field)
# Add in the numpy notation which is row, col
# Here the point geometry is in the form x, y (col, row)
geometry = point_feature.GetGeometryRef()
point = geometry.GetPoint()
point_list.append([point[1], point[0]])
value_list.append(value)
point_array = numpy.array(point_list)
value_array = numpy.array(value_list)
# getting the offsets first before the raster is opened in update mode
offset_list = list(
iterblocks(target_raster_path_band, offset_only=True))
target_raster = gdal.OpenEx(
target_raster_path_band[0], gdal.OF_RASTER | gdal.GA_Update)
band = target_raster.GetRasterBand(target_raster_path_band[1])
nodata = band.GetNoDataValue()
geotransform = target_raster.GetGeoTransform()
for offset in offset_list:
grid_y, grid_x = numpy.mgrid[
offset['yoff']:offset['yoff']+offset['win_ysize'],
offset['xoff']:offset['xoff']+offset['win_xsize']]
grid_y = grid_y * geotransform[5] + geotransform[3]
grid_x = grid_x * geotransform[1] + geotransform[0]
# this is to be consistent with GDAL 2.0's change of 'nearest' to
# 'near' for an interpolation scheme that SciPy did not change.
if interpolation_mode == 'near':
interpolation_mode = 'nearest'
raster_out_array = scipy.interpolate.griddata(
point_array, value_array, (grid_y, grid_x), interpolation_mode,
nodata)
band.WriteArray(raster_out_array, offset['xoff'], offset['yoff'])
[docs]def align_bbox(geotransform, bbox):
"""Pad a bounding box so that it aligns with the grid of a given geotransform.
Ignores row and column rotation.
Args:
geotransform (list): GDAL raster geotransform to align to
bbox (list): bounding box in the form [minx, miny, maxx, maxy]
Returns:
padded bounding box in the form [minx, miny, maxx, maxy]
Raises:
ValueError if invalid geotransform or bounding box are provided
"""
# NOTE: x_origin and y_origin do not necessarily equal minx and miny.
# If pixel_width is positive, x_origin = minx
# If pixel_width is negative, x_origin = maxx
# If pixel height is positive, y_origin = miny
# If pixel height is negative, y_origin = maxy
x_origin, pixel_width, r_x, y_origin, r_y, pixel_height = geotransform
if r_x != 0 or r_y != 0:
LOGGER.warning('Row and/or column rotation supplied to align_bbox '
'will be ignored')
if pixel_width == 0 or pixel_height == 0:
raise ValueError('Pixel width and height must not be 0')
if bbox[2] < bbox[0] or bbox[3] < bbox[1]:
raise ValueError('bbox must be in order [minX, minY, maxX, maxY]')
if ((pixel_width > 0 and bbox[0] < x_origin) or
(pixel_height > 0 and bbox[1] < y_origin) or
(pixel_width < 0 and bbox[0] > x_origin) or
(pixel_height < 0 and bbox[1] > y_origin)):
raise ValueError('bbox must fall within geotransform grid')
return [
x_origin + abs(pixel_width) * math.floor(
(bbox[0] - x_origin) / pixel_width * numpy.sign(pixel_width)),
y_origin + abs(pixel_height) * math.floor(
(bbox[1] - y_origin) / pixel_height * numpy.sign(pixel_height)),
x_origin + abs(pixel_width) * math.ceil(
(bbox[2] - x_origin) / pixel_width * numpy.sign(pixel_width)),
y_origin + abs(pixel_height) * math.ceil(
(bbox[3] - y_origin) / pixel_height * numpy.sign(pixel_height))]
[docs]def zonal_statistics(
base_raster_path_band, aggregate_vector_path,
aggregate_layer_name=None, ignore_nodata=True,
polygons_might_overlap=True, include_value_counts=False,
working_dir=None):
"""Collect stats on pixel values which lie within polygons.
This function summarizes raster statistics including min, max,
mean, and pixel count over the regions on the raster that are
overlapped by the polygons in the vector layer. Statistics are calculated
in two passes, where first polygons aggregate over pixels in the raster
whose centers intersect with the polygon. In the second pass, any polygons
that are not aggregated use their bounding box to intersect with the
raster for overlap statistics.
Statistics are calculated on the set of pixels that fall within each
feature polygon. If ``ignore_nodata`` is false, nodata pixels are considered
valid when calculating the statistics:
- 'min': minimum valid pixel value
- 'max': maximum valid pixel value
- 'sum': sum of valid pixel values
- 'count': number of valid pixels
- 'nodata_count': number of nodata pixels
- 'value_counts': number of pixels having each unique value
Note:
There may be some degenerate cases where the bounding box vs. actual
geometry intersection would be incorrect, but these are so unlikely as
to be manually constructed. If you encounter one of these please create
an issue at https://github.com/natcap/pygeoprocessing/issues with the
datasets used.
Args:
base_raster_path_band (tuple or list[tuple]): base raster
(path, band) tuple(s) to analyze. If a list of multiple raster
bands is provided, they must be aligned (all having the same
bounding box, geotransform, and projection).
aggregate_vector_path (string): a path to a polygon vector whose
geometric features indicate the areas in
``base_raster_path_band`` to calculate zonal statistics.
aggregate_layer_name (string): name of vector layer that will be
used to aggregate results over. If set to None, the first layer
in the DataSource will be used as retrieved by ``.GetLayer()``.
Note: it is normal and expected to set this field at None if the
aggregating vector dataset has a single layer as many do,
including the common 'ESRI Shapefile'.
ignore_nodata: if true, then nodata pixels are not accounted for when
calculating min, max, count, or mean. However, the value of
``nodata_count`` will always be the number of nodata pixels
aggregated under the polygon.
polygons_might_overlap (boolean): if True the function calculates
aggregation coverage close to optimally by rasterizing sets of
polygons that don't overlap. However, this step can be
computationally expensive for cases where there are many polygons.
Setting this flag to ``False`` directs the function rasterize in
one step.
include_value_counts (boolean): If True, the function tallies the
number of pixels of each value under the polygon. This is useful
for classified rasters but could exhaust available memory when run
on a continuous (floating-point) raster. Defaults to False.
working_dir (string): If not None, indicates where temporary files
should be created during this run.
Return:
If `base_raster_path_band` is a tuple, the return value is a nested
dictionary of stats for that raster band. Top-level keys are the
aggregate feature FIDs. Each nested FID dictionary then contains
statistics about that feature: 'min', 'max', 'sum', 'count',
'nodata_count', and optionally 'value_counts'. Example::
{
0: {
'min': 0,
'max': 14,
'sum': 42,
'count': 8,
'nodata_count': 1,
'value_counts': {
2: 5,
4: 1,
14: 2
}
}
}
If `base_raster_path_band` is a list of tuples, the return value is
a list of the nested dictionaries described above. Each dictionary in
the list contains the stats calculated for the corresponding raster
band in the `base_raster_path_band` list.
Raises:
ValueError
if ``base_raster_path_band`` is incorrectly formatted, or if
not all of the input raster bands are aligned with each other
RuntimeError
if the aggregate vector or layer cannot be opened
"""
# Check that the raster path/band input is formatted correctly
multi_raster_mode = isinstance(base_raster_path_band, list)
if not multi_raster_mode:
base_raster_path_band = [base_raster_path_band]
for path_band_tuple in base_raster_path_band:
if not _is_raster_path_band_formatted(path_band_tuple):
raise ValueError(
'`base_raster_path_band` not formatted as expected. Expected '
f'(path, band_index), received {repr(path_band_tuple)}')
# Check that all input rasters are aligned. This should hold true if they
# have the same projection, geotransform, and bounding box.
for attr in ['geotransform', 'bounding_box', 'projection_wkt']:
vals = set()
for path, _ in base_raster_path_band:
raster_info = get_raster_info(path)
vals.add(str(raster_info[attr]))
if len(vals) > 1:
raise ValueError(
'All input rasters must be aligned. Multiple values of '
f'"{attr}" were found among the input rasters: {vals}')
# Check that the aggregate vector and layer can be opened
aggregate_vector = gdal.OpenEx(aggregate_vector_path, gdal.OF_VECTOR)
if aggregate_vector is None:
raise RuntimeError(
f"Could not open aggregate vector at {aggregate_vector_path}")
if aggregate_layer_name is None:
aggregate_layer = aggregate_vector.GetLayer()
else:
aggregate_layer = aggregate_vector.GetLayerByName(aggregate_layer_name)
if aggregate_layer is None:
raise RuntimeError(
f"Could not open layer {aggregate_layer_name} of {aggregate_vector_path}")
# Define the default/empty statistics values
# These values will be returned for features that have no geometry or
# don't overlap any valid pixels.
def default_aggregate_dict():
default_aggregate_dict = {
'min': None, 'max': None, 'count': 0, 'nodata_count': 0, 'sum': 0}
if include_value_counts:
default_aggregate_dict['value_counts'] = collections.Counter()
return default_aggregate_dict
# Create a copy of the aggregate vector with the FID copied into a
# persistent attribute. Do this before reprojecting because converting to
# GPKG can change the FIDs.
# If the aggregate vector already contains a field called 'original_fid',
# this will break and GDAL will emit a warning.
temp_working_dir = tempfile.mkdtemp(dir=working_dir)
copy_path = os.path.join(temp_working_dir, 'vector_copy.gpkg')
target_vector = ogr.GetDriverByName('GPKG').CreateDataSource(copy_path)
target_layer_id = 'disjoint_vector'
target_layer = target_vector.CreateLayer(
name=target_layer_id,
srs=aggregate_layer.GetSpatialRef(),
geom_type=ogr.wkbPolygon)
fid_field_name = 'original_fid'
target_layer.CreateField(ogr.FieldDefn(fid_field_name, ogr.OFTInteger))
valid_fid_set = set()
aggregate_stats_list = [{} for _ in base_raster_path_band]
original_to_new_fid_map = {}
for feature in aggregate_layer:
fid = feature.GetFID()
# Initialize the output data structure:
# a list of zonal_stats dicts, one for each raster input
for stats_dict in aggregate_stats_list:
stats_dict[fid] = default_aggregate_dict()
geom_ref = feature.GetGeometryRef()
if geom_ref is None:
LOGGER.warning(
f'Skipping feature with FID {fid} because it has no geometry')
continue
valid_fid_set.add(fid)
feature_copy = ogr.Feature(target_layer.GetLayerDefn())
feature_copy.SetGeometry(geom_ref.Clone())
feature_copy.SetField(fid_field_name, fid)
target_layer.CreateFeature(feature_copy)
original_to_new_fid_map[fid] = feature_copy.GetFID()
target_layer, target_vector, feature, feature_copy = None, None, None, None
geom_ref, aggregate_layer, aggregate_vector = None, None, None
# Reproject the vector to match the raster projection
target_vector_path = os.path.join(temp_working_dir, 'reprojected.gpkg')
reproject_vector(
base_vector_path=copy_path,
target_projection_wkt=raster_info['projection_wkt'],
target_path=target_vector_path,
layer_id=target_layer_id,
driver_name='GPKG',
copy_fields=True)
target_vector = gdal.OpenEx(target_vector_path, gdal.OF_VECTOR)
try:
bbox_intersection = merge_bounding_box_list(
[raster_info['bounding_box'],
get_vector_info(target_vector_path)['bounding_box']
], 'intersection')
except ValueError as err:
if 'Bounding boxes do not intersect' in repr(err):
LOGGER.error(
f'Aggregate vector {aggregate_vector_path} does not intersect '
'the input raster(s)')
if multi_raster_mode:
return aggregate_stats_list
else:
return aggregate_stats_list[0]
else:
# this would be very unexpected to get here, but if it happened
# and we didn't raise an exception, execution could get weird.
raise
# Expand the intersection bounding box to align with the nearest pixels
# in the original raster
aligned_bbox = align_bbox(raster_info['geotransform'], bbox_intersection)
# Clip base rasters to their intersection with the aggregate vector
LOGGER.info('Clipping rasters to their intersection with the vector')
target_raster_path_band_list = []
for i, (base_path, band) in enumerate(base_raster_path_band):
raster_info = get_raster_info(path)
if (raster_info['datatype'] in {gdal.GDT_Float32, gdal.GDT_Float64}
and include_value_counts):
LOGGER.warning(
f'Value counts requested on a floating-point raster: {path}. '
'This can cause excessive memory usage if the raster has '
'continuous values.')
target_path = os.path.join(temp_working_dir, f'{i}.tif')
target_raster_path_band_list.append((target_path, band))
gdal.Warp(
destNameOrDestDS=target_path,
srcDSOrSrcDSTab=base_path,
format='GTIFF',
# specify the original pixel size because warp doesn't necessarily
# preserve it by default. resolution should always be positive
xRes=abs(raster_info['pixel_size'][0]),
yRes=abs(raster_info['pixel_size'][1]),
outputBounds=aligned_bbox,
callback=_make_logger_callback("Warp %.1f%% complete %s"))
# Calculate disjoint polygon sets
if polygons_might_overlap:
LOGGER.info('calculating disjoint polygon sets')
# Only consider polygons that overlap the rasters
# Use the original vector to be sure that the correct FIDs are returned
disjoint_fid_sets = calculate_disjoint_polygon_set(
aggregate_vector_path, bounding_box=bbox_intersection)
else:
disjoint_fid_sets = [valid_fid_set]
# Rasterize each disjoint polygon set onto its own raster layer
fid_nodata = numpy.iinfo(numpy.uint16).max
fid_raster_paths = []
for i, disjoint_fid_set in enumerate(disjoint_fid_sets):
fid_raster_path = os.path.join(temp_working_dir, f'fid_set_{i}.tif')
fid_set_str = ", ".join(str(fid) for fid in disjoint_fid_set)
gdal.Rasterize(
destNameOrDestDS=fid_raster_path,
srcDS=target_vector,
allTouched=False,
attribute=fid_field_name,
noData=fid_nodata,
outputBounds=aligned_bbox,
xRes=abs(raster_info['pixel_size'][0]), # resolution must be > 0
yRes=abs(raster_info['pixel_size'][1]),
format='GTIFF',
outputType=gdal.GDT_UInt16,
creationOptions=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS[1],
callback=_make_logger_callback(
f'rasterizing disjoint polygon set {i + 1} of '
f'{len(disjoint_fid_sets)} set %.1f%% complete (%s)'),
where=(f'{fid_field_name} IN ({fid_set_str})'))
fid_raster_paths.append(fid_raster_path)
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
# Calculate statistics for each raster and each feature
# working block-wise through the rasters
for i, (raster_path, band) in enumerate(target_raster_path_band_list):
LOGGER.info('calculating stats on raster '
f'{i} of {len(target_raster_path_band_list)}')
# fetch the block offsets before the raster is opened for writing
offset_list = list(iterblocks((raster_path, band), offset_only=True))
nodata = get_raster_info(raster_path)['nodata'][band - 1]
data_source = gdal.OpenEx(raster_path, gdal.OF_RASTER)
data_band = data_source.GetRasterBand(band)
found_fids = set() # track FIDs found on at least one pixel
for set_index, fid_raster_path in enumerate(fid_raster_paths):
LOGGER.info(
f'disjoint polygon set {set_index + 1} of '
f'{len(fid_raster_paths)}')
fid_raster = gdal.OpenEx(fid_raster_path, gdal.OF_RASTER)
fid_band = fid_raster.GetRasterBand(1)
for offset_index, offset in enumerate(offset_list):
timed_logger.info(
"%.1f%% done calculating stats for polygon set %s on raster %s",
offset_index / len(offset_list) * 100, set_index, i)
fid_block = fid_band.ReadAsArray(**offset)
data_block = data_band.ReadAsArray(**offset)
# Update stats for each FID found in this block of data
for fid in numpy.unique(fid_block):
if fid == fid_nodata:
continue
found_fids.add(fid)
# get the pixels that fall within the feature
feature_data = data_block[fid_block == fid]
nodata_mask = array_equals_nodata(feature_data, nodata)
aggregate_stats_list[i][fid]['nodata_count'] += nodata_mask.sum()
if ignore_nodata:
feature_data = feature_data[~nodata_mask]
if feature_data.size == 0:
continue
# compute stats
if aggregate_stats_list[i][fid]['min'] is None:
# initialize min/max to an arbitrary data value
aggregate_stats_list[i][fid]['min'] = feature_data[0]
aggregate_stats_list[i][fid]['max'] = feature_data[0]
aggregate_stats_list[i][fid]['min'] = min(
feature_data.min(), aggregate_stats_list[i][fid]['min'])
aggregate_stats_list[i][fid]['max'] = max(
feature_data.max(), aggregate_stats_list[i][fid]['max'])
aggregate_stats_list[i][fid]['sum'] += feature_data.sum()
aggregate_stats_list[i][fid]['count'] += feature_data.size
if include_value_counts:
# .update() here is operating on a Counter, so values are
# ADDED, not replaced.
aggregate_stats_list[i][fid]['value_counts'].update(
dict(zip(*numpy.unique(
feature_data, return_counts=True))))
fid_band, fid_raster = None, None
# Handle edge cases: features that have a geometry but do not
# overlap the center point of any pixel will not be captured by the
# method above.
unset_fids = valid_fid_set.difference(found_fids)
x_origin, pixel_width, _, y_origin, _, pixel_height = data_source.GetGeoTransform()
# subtract 1 because bands are 1-indexed
raster_nodata = get_raster_info(raster_path)['nodata'][band - 1]
target_layer = target_vector.GetLayerByName(target_layer_id)
for unset_fid in unset_fids:
# Look up by the new FID
# FIDs in target_layer may not be the same as in the input layer
unset_feat = target_layer.GetFeature(original_to_new_fid_map[unset_fid])
unset_geom_ref = unset_feat.GetGeometryRef()
geom_x_min, geom_x_max, geom_y_min, geom_y_max = unset_geom_ref.GetEnvelope()
unset_geom_ref, unset_feat = None, None
if pixel_width < 0:
geom_x_min, geom_x_max = geom_x_max, geom_x_min
if pixel_height < 0:
geom_y_min, geom_y_max = geom_y_max, geom_y_min
xoff = int((geom_x_min - x_origin) / pixel_width)
yoff = int((geom_y_min - y_origin) / pixel_height)
win_xsize = int(numpy.ceil(
(geom_x_max - x_origin) / pixel_width)) - xoff
win_ysize = int(numpy.ceil(
(geom_y_max - y_origin) / pixel_height)) - yoff
# clamp offset to the side of the raster if it's negative
if xoff < 0:
win_xsize += xoff
xoff = 0
if yoff < 0:
win_ysize += yoff
yoff = 0
# clamp the window to the side of the raster if too big
if xoff + win_xsize > data_band.XSize:
win_xsize = data_band.XSize - xoff
if yoff + win_ysize > data_band.YSize:
win_ysize = data_band.YSize - yoff
if win_xsize <= 0 or win_ysize <= 0:
continue
# here we consider the pixels that intersect with the geometry's
# bounding box as being the proxy for the intersection with the
# polygon itself. This is not a bad approximation since the case
# that caused the polygon to be skipped in the first phase is that it
# is as small as a pixel. There could be some degenerate cases that
# make this estimation very wrong, but we do not know of any that
# would come from natural data. If you do encounter such a dataset
# please email the description and datset to jdouglass@stanford.edu.
unset_fid_block = data_band.ReadAsArray(
xoff=xoff, yoff=yoff, win_xsize=win_xsize, win_ysize=win_ysize)
unset_fid_nodata_mask = array_equals_nodata(
unset_fid_block, raster_nodata)
if ignore_nodata:
unset_fid_block = unset_fid_block[~unset_fid_nodata_mask]
if unset_fid_block.size == 0:
aggregate_stats_list[i][unset_fid]['min'] = 0
aggregate_stats_list[i][unset_fid]['max'] = 0
aggregate_stats_list[i][unset_fid]['sum'] = 0
else:
aggregate_stats_list[i][unset_fid]['min'] = numpy.min(
unset_fid_block)
aggregate_stats_list[i][unset_fid]['max'] = numpy.max(
unset_fid_block)
aggregate_stats_list[i][unset_fid]['sum'] = numpy.sum(
unset_fid_block)
aggregate_stats_list[i][unset_fid]['count'] = unset_fid_block.size
aggregate_stats_list[i][unset_fid]['nodata_count'] = numpy.sum(
unset_fid_nodata_mask)
if include_value_counts:
# .update() here is operating on a Counter, so values are ADDED,
# not replaced.
aggregate_stats_list[i][unset_fid]['value_counts'].update(dict(
zip(*numpy.unique(unset_fid_block, return_counts=True))))
# Convert counter object to dictionary
if include_value_counts:
for key, value in aggregate_stats_list[i].items():
aggregate_stats_list[i][key]['value_counts'] = dict(
aggregate_stats_list[i][key]['value_counts'])
LOGGER.info(
f'all done processing polygon sets for {os.path.basename(aggregate_vector_path)}')
# dereference gdal objects
data_band, data_source = None, None
disjoint_layer, target_layer, target_vector = None, None, None
shutil.rmtree(temp_working_dir)
if multi_raster_mode:
return aggregate_stats_list
else:
return aggregate_stats_list[0]
[docs]def get_vector_info(vector_path, layer_id=0):
"""Get information about an GDAL vector.
Args:
vector_path (str): a path to a GDAL vector.
layer_id (str/int): name or index of underlying layer to analyze.
Defaults to 0.
Raises:
ValueError if ``vector_path`` does not exist on disk or cannot be
opened as a gdal.OF_VECTOR.
Return:
raster_properties (dictionary):
a dictionary with the following key-value pairs:
* ``'projection_wkt'`` (string): projection of the vector in Well
Known Text.
* ``'bounding_box'`` (sequence): sequence of floats representing
the bounding box in projected coordinates in the order
[minx, miny, maxx, maxy].
* ``'file_list'`` (sequence): sequence of string paths to the files
that make up this vector.
"""
vector = gdal.OpenEx(vector_path, gdal.OF_VECTOR)
if not vector:
raise ValueError(
"Could not open %s as a gdal.OF_VECTOR" % vector_path)
vector_properties = {}
vector_properties['file_list'] = vector.GetFileList()
layer = vector.GetLayer(iLayer=layer_id)
# projection is same for all layers, so just use the first one
spatial_ref = layer.GetSpatialRef()
if spatial_ref:
vector_projection_wkt = spatial_ref.ExportToWkt()
else:
vector_projection_wkt = None
vector_properties['projection_wkt'] = vector_projection_wkt
layer_bb = layer.GetExtent()
layer = None
vector = None
# convert form [minx,maxx,miny,maxy] to [minx,miny,maxx,maxy]
vector_properties['bounding_box'] = [layer_bb[i] for i in [0, 2, 1, 3]]
return vector_properties
[docs]def get_raster_info(raster_path):
"""Get information about a GDAL raster (dataset).
Args:
raster_path (String): a path to a GDAL raster.
Raises:
ValueError
if ``raster_path`` is not a file or cannot be opened as a
``gdal.OF_RASTER``.
Return:
raster_properties (dictionary):
a dictionary with the properties stored under relevant keys.
* ``'pixel_size'`` (tuple): (pixel x-size, pixel y-size)
from geotransform.
* ``'raster_size'`` (tuple): number of raster pixels in (x, y)
direction.
* ``'nodata'`` (sequence): a sequence of the nodata values in the bands
of the raster in the same order as increasing band index.
* ``'n_bands'`` (int): number of bands in the raster.
* ``'geotransform'`` (tuple): a 6-tuple representing the geotransform
of (x orign, x-increase, xy-increase, y origin, yx-increase,
y-increase).
* ``'datatype'`` (int): An instance of an enumerated gdal.GDT_* int
that represents the datatype of the raster.
* ``'projection_wkt'`` (string): projection of the raster in Well Known
Text.
* ``'bounding_box'`` (sequence): sequence of floats representing the
bounding box in projected coordinates in the order
[minx, miny, maxx, maxy]
* ``'block_size'`` (tuple): underlying x/y raster block size for
efficient reading.
* ``'numpy_type'`` (numpy type): this is the equivalent numpy datatype
for the raster bands including signed bytes.
* ``'overviews'`` (sequence): A list of (x, y) tuples for the
number of pixels in the width and height of each overview level of
the raster.
* ``'file_list'`` (sequence): A list of files that make up this raster.
"""
raster = gdal.OpenEx(raster_path, gdal.OF_RASTER)
if not raster:
raise ValueError(
"Could not open %s as a gdal.OF_RASTER" % raster_path)
raster_properties = {}
raster_properties['file_list'] = raster.GetFileList()
projection_wkt = raster.GetProjection()
if not projection_wkt:
projection_wkt = None
raster_properties['projection_wkt'] = projection_wkt
geo_transform = raster.GetGeoTransform()
raster_properties['geotransform'] = geo_transform
raster_properties['pixel_size'] = (geo_transform[1], geo_transform[5])
raster_properties['raster_size'] = (
raster.GetRasterBand(1).XSize,
raster.GetRasterBand(1).YSize)
raster_properties['n_bands'] = raster.RasterCount
raster_properties['nodata'] = [
raster.GetRasterBand(index).GetNoDataValue() for index in range(
1, raster_properties['n_bands']+1)]
# GDAL creates overviews for the whole raster but has overviews accessed
# per band. We assume that all bands have the same overviews.
raster_properties['overviews'] = []
for overview_index in range(raster.GetRasterBand(1).GetOverviewCount()):
overview_band = raster.GetRasterBand(1).GetOverview(overview_index)
raster_properties['overviews'].append((
overview_band.XSize, overview_band.YSize))
# blocksize is the same for all bands, so we can just get the first
raster_properties['block_size'] = raster.GetRasterBand(1).GetBlockSize()
# we dont' really know how the geotransform is laid out, all we can do is
# calculate the x and y bounds, then take the appropriate min/max
x_bounds = [
geo_transform[0], geo_transform[0] +
raster_properties['raster_size'][0] * geo_transform[1] +
raster_properties['raster_size'][1] * geo_transform[2]]
y_bounds = [
geo_transform[3], geo_transform[3] +
raster_properties['raster_size'][0] * geo_transform[4] +
raster_properties['raster_size'][1] * geo_transform[5]]
raster_properties['bounding_box'] = [
numpy.min(x_bounds), numpy.min(y_bounds),
numpy.max(x_bounds), numpy.max(y_bounds)]
# datatype is the same for the whole raster, but is associated with band
band = raster.GetRasterBand(1)
raster_properties['datatype'] = band.DataType
raster_properties['numpy_type'] = _gdal_to_numpy_type(
band.DataType, band.GetMetadata('IMAGE_STRUCTURE'))
band = None
raster = None
return raster_properties
[docs]def reproject_vector(
base_vector_path, target_projection_wkt, target_path, layer_id=0,
driver_name='ESRI Shapefile', copy_fields=True,
target_layer_name=None,
osr_axis_mapping_strategy=DEFAULT_OSR_AXIS_MAPPING_STRATEGY):
"""Reproject OGR DataSource (vector).
Transforms the features of the base vector to the desired output
projection in a new vector.
Note:
If the ESRI Shapefile driver is used, the ``target_layer_name``
optional parameter is ignored. ESRI Shapefiles by definition use the
filename to define the layer name.
Args:
base_vector_path (string): Path to the base shapefile to transform.
target_projection_wkt (string): the desired output projection in Well
Known Text (by layer.GetSpatialRef().ExportToWkt())
target_path (string): the filepath to the transformed shapefile
layer_id (str/int): name or index of layer in ``base_vector_path`` to
reproject. Defaults to 0.
driver_name (string): String to pass to ogr.GetDriverByName, defaults
to 'ESRI Shapefile'.
copy_fields (bool or iterable): If True, all the fields in
``base_vector_path`` will be copied to ``target_path`` during the
reprojection step. If it is an iterable, it will contain the
field names to exclusively copy. An unmatched fieldname will be
ignored. If ``False`` no fields are copied into the new vector.
target_layer_name=None (str): The name to use for the target layer in
the new vector. If ``None`` (the default), the layer name from the
source layer will be used.
osr_axis_mapping_strategy (int): OSR axis mapping strategy for
``SpatialReference`` objects. Defaults to
``geoprocessing.DEFAULT_OSR_AXIS_MAPPING_STRATEGY``. This parameter
should not be changed unless you know what you are doing.
Returns:
None
"""
base_vector = gdal.OpenEx(base_vector_path, gdal.OF_VECTOR)
# if this file already exists, then remove it
if os.path.isfile(target_path):
LOGGER.warning(
"%s already exists, removing and overwriting", target_path)
os.remove(target_path)
target_sr = osr.SpatialReference(target_projection_wkt)
# create a new shapefile from the orginal_datasource
target_driver = ogr.GetDriverByName(driver_name)
target_vector = target_driver.CreateDataSource(target_path)
layer = base_vector.GetLayer(layer_id)
layer_dfn = layer.GetLayerDefn()
# Create new layer for target_vector using same name and
# geometry type from base vector but new projection
layer_name = layer_dfn.GetName()
if target_layer_name is not None:
layer_name = target_layer_name
if driver_name == 'ESRI Shapefile':
target_file_basename = os.path.splitext(
os.path.basename(target_path)[0])
if layer_name != target_file_basename:
LOGGER.warning(
f'Ignoring user-defined layer name {layer_name}. '
f'Defining a layer name is incompatible with the ESRI '
'Shapefile vector format. Use the filename instead or '
'use a different vector format.')
target_layer = target_vector.CreateLayer(
layer_name, target_sr, layer_dfn.GetGeomType())
# this will map the target field index to the base index it came from
# in case we don't need to copy all the fields
target_to_base_field_id_map = {}
if copy_fields:
# Get the number of fields in original_layer
original_field_count = layer_dfn.GetFieldCount()
# For every field that's copying, create a duplicate field in the
# new layer
for fld_index in range(original_field_count):
original_field = layer_dfn.GetFieldDefn(fld_index)
field_name = original_field.GetName()
if copy_fields is True or field_name in copy_fields:
target_field = ogr.FieldDefn(
field_name, original_field.GetType())
target_layer.CreateField(target_field)
target_to_base_field_id_map[fld_index] = len(
target_to_base_field_id_map)
# Get the SR of the original_layer to use in transforming
base_sr = layer.GetSpatialRef()
base_sr.SetAxisMappingStrategy(osr_axis_mapping_strategy)
target_sr.SetAxisMappingStrategy(osr_axis_mapping_strategy)
# Create a coordinate transformation
coord_trans = osr.CreateCoordinateTransformation(base_sr, target_sr)
# Copy all of the features in layer to the new shapefile
target_layer.StartTransaction()
error_count = 0
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
LOGGER.info("starting reprojection")
for feature_index, base_feature in enumerate(layer):
timed_logger.info(
"reprojection approximately %.1f%% complete on %s",
100.0 * float(feature_index+1) / (layer.GetFeatureCount()),
os.path.basename(target_path))
geom = base_feature.GetGeometryRef()
if geom is None:
# we encountered this error occasionally when transforming clipped
# global polygons. Not clear what is happening but perhaps a
# feature was retained that otherwise wouldn't have been included
# in the clip
error_count += 1
continue
# Transform geometry into format desired for the new projection
error_code = geom.Transform(coord_trans)
if error_code != 0: # error
# this could be caused by an out of range transformation
# whatever the case, don't put the transformed poly into the
# output set
error_count += 1
continue
# Copy original_datasource's feature and set as new shapes feature
target_feature = ogr.Feature(target_layer.GetLayerDefn())
target_feature.SetGeometry(geom)
# For all the fields in the feature set the field values from the
# source field
for target_index, base_index in (
target_to_base_field_id_map.items()):
target_feature.SetField(
target_index, base_feature.GetField(base_index))
target_layer.CreateFeature(target_feature)
target_feature = None
base_feature = None
target_layer.CommitTransaction()
LOGGER.info(
"reprojection 100.0%% complete on %s", os.path.basename(target_path))
if error_count > 0:
LOGGER.warning(
'%d features out of %d were unable to be transformed and are'
' not in the output vector at %s', error_count,
layer.GetFeatureCount(), target_path)
layer = None
base_vector = None
[docs]def reclassify_raster(
base_raster_path_band, value_map, target_raster_path, target_datatype,
target_nodata, values_required=True,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Reclassify pixel values in a raster.
A function to reclassify values in raster to any output type. By default
the values except for nodata must be in ``value_map``. Include the
base raster nodata value in ``value_map`` to reclassify nodata pixels
to a valid value.
Args:
base_raster_path_band (tuple): a tuple including file path to a raster
and the band index to operate over. ex: (path, band_index)
value_map (dictionary): a dictionary of values of
{source_value: dest_value, ...} where source_value's type is the
same as the values in ``base_raster_path`` at band ``band_index``.
Cannot be empty and must contain a mapping for all raster values
if ``values_required=True``. If nodata is mapped, nodata will be
reclassified, otherwise nodata will be set to ``target_nodata``.
target_raster_path (string): target raster output path; overwritten if
it exists
target_datatype (gdal type): the numerical type for the target raster
target_nodata (numerical type): the nodata value for the target raster.
Must be the same type as target_datatype. All nodata pixels in
the base raster will be reclassified to this value, unless the
base raster notata values are present in ``value_map``.
values_required (bool): If True, raise a ValueError if there is a
value in the raster that is not found in ``value_map``.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
Raises:
ReclassificationMissingValuesError
if ``values_required`` is ``True``
and a pixel value from ``base_raster_path_band`` is not a key in
``value_map``.
"""
if len(value_map) == 0:
raise ValueError("value_map must contain at least one value")
if not _is_raster_path_band_formatted(base_raster_path_band):
raise ValueError(
"Expected a (path, band_id) tuple, instead got '%s'" %
base_raster_path_band)
raster_info = get_raster_info(base_raster_path_band[0])
nodata = raster_info['nodata'][base_raster_path_band[1]-1]
# If nodata was included in the value_map pop it from our lists
# and handle it separately. Doing this on the off chance nodata is
# a max or min float which can cause floating decimal discrepencies.
value_map_copy = value_map.copy()
nodata_dest_value = target_nodata
if nodata is not None:
for key, val in value_map.items():
if numpy.isclose(key, nodata):
nodata_dest_value = val
del value_map_copy[key]
break
if target_nodata is None and nodata_dest_value is None:
raise ValueError(
"target_nodata was set to None and the base raster nodata"
" value was not represented in the value_map. reclassify_raster"
" does not assume the base raster nodata value should be used"
" as the target_nodata value. Set target_nodata to a valid number"
" and/or add a base raster nodata value mapping to value_map.")
keys = sorted(numpy.array(list(value_map_copy.keys())))
values = numpy.array([value_map_copy[x] for x in keys])
numpy_dtype = _gdal_to_numpy_type(
target_datatype, raster_driver_creation_tuple[1])
def _map_dataset_to_value_op(original_values):
"""Convert a block of original values to the lookup values."""
out_array = numpy.full(
original_values.shape, target_nodata,
dtype=numpy_dtype)
if nodata is None:
valid_mask = numpy.full(original_values.shape, True)
else:
valid_mask = ~numpy.isclose(original_values, nodata)
out_array[~valid_mask] = nodata_dest_value
if values_required:
unique = numpy.unique(original_values[valid_mask])
has_map = numpy.isin(unique, keys)
if not all(has_map):
missing_values = unique[~has_map]
raise ReclassificationMissingValuesError(
missing_values, base_raster_path_band[0], value_map
)
index = numpy.digitize(original_values[valid_mask], keys, right=True)
out_array[valid_mask] = values[index]
return out_array
raster_calculator(
[base_raster_path_band], _map_dataset_to_value_op,
target_raster_path, target_datatype, target_nodata,
raster_driver_creation_tuple=raster_driver_creation_tuple)
[docs]def warp_raster(
base_raster_path, target_pixel_size, target_raster_path,
resample_method, target_bb=None, base_projection_wkt=None,
target_projection_wkt=None, n_threads=None, mask_options=None,
vector_mask_options=None, gdal_warp_options=None, working_dir=None,
use_overview_level=-1,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS,
osr_axis_mapping_strategy=DEFAULT_OSR_AXIS_MAPPING_STRATEGY):
"""Resize/resample raster to desired pixel size, bbox and projection.
Args:
base_raster_path (string): path to base raster.
target_pixel_size (list/tuple): a two element sequence indicating
the x and y pixel size in projected units.
target_raster_path (string): the location of the resized and
resampled raster.
resample_method (string): the resampling algorithm. Must be a valid
resampling algorithm for `gdal.WarpRaster`, one of:
'rms | mode | sum | q1 | near | q3 | average | cubicspline |
bilinear | max | med | min | cubic | lanczos'
target_bb (sequence): if None, target bounding box is the same as the
source bounding box. Otherwise it's a sequence of float
describing target bounding box in target coordinate system as
[minx, miny, maxx, maxy].
base_projection_wkt (string): if not None, interpret the projection of
``base_raster_path`` as this.
target_projection_wkt (string): if not None, desired target projection
in Well Known Text format.
n_threads (int): optional, if not None this sets the ``N_THREADS``
option for ``gdal.Warp``.
mask_options (dict or None): optional. If None, no masking will
be done. If a dict, it is a dictionary of options relating to the
dataset mask. Keys to this dictionary are:
* ``'mask_vector_path'``: (str) path to the mask vector file. This
vector will be automatically projected to the target
projection if its base coordinate system does not match
the target. Where there are geometries in this vector, pixels in
``base_raster_path`` will propagate to ``target_raster_path``.
* ``'mask_layer_id'``: (int/str) the layer index or name to use
for masking, if this key is not in the dictionary the default
is to use the layer at index 0.
* ``'mask_vector_where_filter'``: (str) an SQL ``WHERE`` string
that can be used to filter the geometry in the mask.
Ex: 'id > 10' would use all features whose field value of 'id' is
> 10.
* ``'mask_raster_path'``: (str). If present in the dict, all other
keys in ``mask_options`` are ignored. This string must be
a path to a raster representing a validity mask, where pixel
values of 1 indicate validity. This raster must be in the same
projection and have the same dimensions as the target warped
raster. The general (and easiest) use case for ``warp_raster``
is to use ``'mask_vector_path'`` instead.
vector_mask_options=None (dict): Alias for ``mask_options``.
This option is deprecated and will be removed in a future release
of ``pygeoprocessing``.
gdal_warp_options (sequence): if present, the contents of this list
are passed to the ``warpOptions`` parameter of ``gdal.Warp``. See
the GDAL Warp documentation for valid options.
working_dir (string): if defined uses this directory to make
temporary working files for calculation. Otherwise uses system's
temp directory.
use_overview_level=-1 (int/str): The overview level to use for warping.
A value of ``-1`` (the default) indicates that the base raster
should be used for the source pixels. A value of ``'AUTO'``
will make GDAL select the overview with the resolution that is
closest to the target pixel size and warp using that overview's
pixel values. Any other integer indicates that that overview index
should be used. For example, suppose the raster has overviews at
levels 2, 4 and 8. To use level 2, set ``use_overview_level=0``.
To use level 8, set ``use_overview_level=2``.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
osr_axis_mapping_strategy (int): OSR axis mapping strategy for
``SpatialReference`` objects. Defaults to
``geoprocessing.DEFAULT_OSR_AXIS_MAPPING_STRATEGY``. This parameter
should not be changed unless you know what you are doing.
Return:
None
Raises:
ValueError
if ``pixel_size`` is not a 2 element sequence of numbers.
ValueError
if ``mask_options`` is not None but the
``mask_vector_path`` is undefined or doesn't point to a valid
file.
"""
_assert_is_valid_pixel_size(target_pixel_size)
base_raster_info = get_raster_info(base_raster_path)
if target_projection_wkt is None:
target_projection_wkt = base_raster_info['projection_wkt']
if vector_mask_options is not None:
warnings.warn('The vector_mask_options parameter is deprecated and '
'will be removed in a future release of '
'pygeoprocessing. Please use mask_options instead.',
DeprecationWarning)
mask_options = vector_mask_options
if target_bb is None:
# ensure it's a sequence so we can modify it
working_bb = list(get_raster_info(base_raster_path)['bounding_box'])
# transform the working_bb if target_projection_wkt is not None
if target_projection_wkt is not None:
LOGGER.debug(
"transforming bounding box from %s ", working_bb)
working_bb = transform_bounding_box(
base_raster_info['bounding_box'],
base_raster_info['projection_wkt'], target_projection_wkt)
LOGGER.debug(
"transforming bounding to %s ", working_bb)
else:
# ensure it's a sequence so we can modify it
working_bb = list(target_bb)
# determine the raster size that bounds the input bounding box and then
# adjust the bounding box to be that size
target_x_size = int(abs(
float(working_bb[2] - working_bb[0]) / target_pixel_size[0]))
target_y_size = int(abs(
float(working_bb[3] - working_bb[1]) / target_pixel_size[1]))
# sometimes bounding boxes are numerically perfect, this checks for that
x_residual = (
abs(target_x_size * target_pixel_size[0]) -
(working_bb[2] - working_bb[0]))
if not numpy.isclose(x_residual, 0.0):
target_x_size += 1
y_residual = (
abs(target_y_size * target_pixel_size[1]) -
(working_bb[3] - working_bb[1]))
if not numpy.isclose(y_residual, 0.0):
target_y_size += 1
if target_x_size == 0:
LOGGER.warning(
"bounding_box is so small that x dimension rounds to 0; "
"clamping to 1.")
target_x_size = 1
if target_y_size == 0:
LOGGER.warning(
"bounding_box is so small that y dimension rounds to 0; "
"clamping to 1.")
target_y_size = 1
# this ensures the bounding boxes perfectly fit a multiple of the target
# pixel size
working_bb[2] = working_bb[0] + abs(target_pixel_size[0] * target_x_size)
working_bb[3] = working_bb[1] + abs(target_pixel_size[1] * target_y_size)
reproject_callback = _make_logger_callback(
"Warp %.1f%% complete %s")
warp_options = []
if n_threads:
warp_options.append('NUM_THREADS=%d' % n_threads)
if gdal_warp_options:
warp_options.extend(gdal_warp_options)
mask_vector_path = None
mask_layer_id = 0
mask_vector_where_filter = None
if mask_options:
if 'mask_raster_path' not in mask_options:
# translate pygeoprocessing terminology into GDAL warp options.
if 'mask_vector_path' not in mask_options:
raise ValueError(
'mask_options passed, but no value for '
'"mask_vector_path": %s', mask_options)
mask_vector_path = mask_options['mask_vector_path']
if not os.path.exists(mask_vector_path):
raise ValueError(
'The mask vector at %s was not found.', mask_vector_path)
if 'mask_layer_id' in mask_options:
mask_layer_id = mask_options['mask_layer_id']
if 'mask_vector_where_filter' in mask_options:
mask_vector_where_filter = (
mask_options['mask_vector_where_filter'])
if mask_options:
temp_working_dir = tempfile.mkdtemp(dir=working_dir)
warped_raster_path = os.path.join(
temp_working_dir, os.path.basename(target_raster_path).replace(
'.tif', '_nonmasked.tif'))
else:
# if there is no vector path the result is the warp
warped_raster_path = target_raster_path
base_raster = gdal.OpenEx(base_raster_path, gdal.OF_RASTER)
raster_creation_options = list(raster_driver_creation_tuple[1])
_, type_creation_options = _numpy_to_gdal_type(
base_raster_info['numpy_type'])
raster_creation_options += type_creation_options
gdal.Warp(
warped_raster_path, base_raster,
format=raster_driver_creation_tuple[0],
outputBounds=working_bb,
xRes=abs(target_pixel_size[0]),
yRes=abs(target_pixel_size[1]),
resampleAlg=resample_method,
outputBoundsSRS=target_projection_wkt,
srcSRS=base_projection_wkt,
dstSRS=target_projection_wkt,
multithread=True if warp_options else False,
warpOptions=warp_options,
overviewLevel=use_overview_level,
creationOptions=raster_creation_options,
callback=reproject_callback,
callback_data=[target_raster_path])
if mask_options:
if 'mask_raster_path' in mask_options:
# If the user provided a mask raster, use that directly; assume
# it's been rasterized correctly.
source_raster_info = get_raster_info(warped_raster_path)
source_nodata = source_raster_info['nodata'][0]
def _mask_values(array, mask):
output = numpy.full(array.shape, source_nodata)
valid_mask = (
mask == 1 & ~array_equals_nodata(array, source_nodata))
output[valid_mask] = array[valid_mask]
return output
raster_calculator(
[(warped_raster_path, 1),
(mask_options['mask_raster_path'], 1)],
_mask_values, target_raster_path,
source_raster_info['datatype'], source_nodata)
else:
# If the user did not provide a mask in raster form, we can just
# call down to ``mask_raster``, which will rasterize the vector and
# then mask out pixels in ``warped_raster_path`` for us.
updated_raster_driver_creation_tuple = (
raster_driver_creation_tuple[0],
tuple(raster_creation_options))
mask_raster(
(warped_raster_path, 1),
mask_options['mask_vector_path'],
target_raster_path,
mask_layer_id=mask_layer_id,
where_clause=mask_vector_where_filter,
target_mask_value=None, working_dir=temp_working_dir,
all_touched=False,
raster_driver_creation_tuple=(
updated_raster_driver_creation_tuple))
shutil.rmtree(temp_working_dir)
[docs]def rasterize(
vector_path, target_raster_path, burn_values=None, option_list=None,
layer_id=0, where_clause=None):
"""Project a vector onto an existing raster.
Burn the layer at ``layer_id`` in ``vector_path`` to an existing
raster at ``target_raster_path_band``.
Args:
vector_path (string): filepath to vector to rasterize.
target_raster_path (string): path to an existing raster to burn vector
into. Can have multiple bands.
burn_values (list/tuple): optional sequence of values to burn into
each band of the raster. If used, should have the same length as
number of bands at the ``target_raster_path`` raster. If ``None``
then ``option_list`` must have a valid value.
option_list (list/tuple): optional a sequence of burn options, if None
then a valid value for ``burn_values`` must exist. Otherwise, each
element is a string of the form:
* ``"ATTRIBUTE=?"``: Identifies an attribute field on the features
to be used for a burn in value. The value will be burned into all
output bands. If specified, ``burn_values`` will not be used and
can be None.
* ``"CHUNKYSIZE=?"``: The height in lines of the chunk to operate
on. The larger the chunk size the less times we need to make a
pass through all the shapes. If it is not set or set to zero the
default chunk size will be used. Default size will be estimated
based on the GDAL cache buffer size using formula:
``cache_size_bytes/scanline_size_bytes``, so the chunk will not
exceed the cache.
* ``"ALL_TOUCHED=TRUE/FALSE"``: May be set to ``TRUE`` to set all
pixels touched by the line or polygons, not just those whose
center is within the polygon or that are selected by Brezenhams
line algorithm. Defaults to ``FALSE``.
* ``"BURN_VALUE_FROM"``: May be set to "Z" to use the Z values of
the geometries. The value from burn_values or the
attribute field value is added to this before burning. In
default case dfBurnValue is burned as it is (richpsharp:
note, I'm not sure what this means, but copied from formal
docs). This is implemented properly only for points and
lines for now. Polygons will be burned using the Z value
from the first point.
* ``"MERGE_ALG=REPLACE/ADD"``: REPLACE results in overwriting of
value, while ADD adds the new value to the existing
raster, suitable for heatmaps for instance.
Example::
["ATTRIBUTE=npv", "ALL_TOUCHED=TRUE"]
layer_id (str/int): name or index of the layer to rasterize. Defaults
to 0.
where_clause (str): If not None, is an SQL query-like string to filter
which features are used to rasterize, (e.x. where="value=1").
Return:
None
"""
gdal.PushErrorHandler('CPLQuietErrorHandler')
raster = gdal.OpenEx(target_raster_path, gdal.GA_Update | gdal.OF_RASTER)
gdal.PopErrorHandler()
if raster is None:
raise ValueError(
"%s doesn't exist, but needed to rasterize." % target_raster_path)
rasterize_callback = _make_logger_callback(
"pygeoprocessing.rasterize RasterizeLayer %.1f%% complete %s")
if burn_values is None:
burn_values = []
if option_list is None:
option_list = []
if not burn_values and not option_list:
raise ValueError(
"Neither `burn_values` nor `option_list` is set. At least "
"one must have a value.")
if not isinstance(burn_values, (list, tuple)):
raise ValueError(
"`burn_values` is not a list/tuple, the value passed is '%s'",
repr(burn_values))
if not isinstance(option_list, (list, tuple)):
raise ValueError(
"`option_list` is not a list/tuple, the value passed is '%s'",
repr(option_list))
vector = gdal.OpenEx(vector_path, gdal.OF_VECTOR)
layer = vector.GetLayer(layer_id)
if where_clause:
layer.SetAttributeFilter(where_clause)
try:
result = gdal.RasterizeLayer(
raster, [1], layer, burn_values=burn_values,
options=option_list, callback=rasterize_callback)
except Exception:
# something bad happened, but still clean up
# this case came out of a flaky test condition where the raster
# would still be in use by the rasterize layer function
LOGGER.exception('bad error on rasterizelayer')
result = -1
layer = None
vector = None
if result != 0:
# need this __swig_destroy__ because we sometimes encounter a flaky
# test where the path to the raster cannot be cleaned up because
# it is still in use somewhere, likely a bug in gdal.RasterizeLayer
# note it is only invoked if there is a serious error
gdal.Dataset.__swig_destroy__(raster)
raise RuntimeError('Rasterize returned a nonzero exit code.')
raster = None
[docs]def calculate_disjoint_polygon_set(
vector_path, layer_id=0, bounding_box=None,
geometries_may_touch=False):
"""Create a sequence of sets of polygons that don't overlap.
Determining the minimal number of those sets is an np-complete problem so
this is an approximation that builds up sets of maximal subsets.
Args:
vector_path (string): a path to an OGR vector.
layer_id (str/int): name or index of underlying layer in
``vector_path`` to calculate disjoint set. Defaults to 0.
bounding_box (sequence): sequence of floats representing a bounding
box to filter any polygons by. If a feature in ``vector_path``
does not intersect this bounding box it will not be considered
in the disjoint calculation. Coordinates are in the order
[minx, miny, maxx, maxy].
geometries_may_touch=False(bool): If ``True``, geometries in a subset
are allowed to have touching boundaries, but are not allowed to
have intersecting interiors. If ``False`` (the default), no
geometries in a subset may intersect in any way.
Return:
subset_list (sequence): sequence of sets of FIDs from vector_path
"""
vector = gdal.OpenEx(vector_path, gdal.OF_VECTOR)
vector_layer = vector.GetLayer(layer_id)
feature_count = vector_layer.GetFeatureCount()
if feature_count == 0:
raise RuntimeError('Vector must have geometries but does not: %s'
% vector_path)
LOGGER.info("build shapely polygon list")
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
if bounding_box is None:
bounding_box = get_vector_info(vector_path)['bounding_box']
bounding_box = shapely.prepared.prep(shapely.geometry.box(*bounding_box))
shapely_polygon_lookup = {}
for poly_feat in vector_layer:
poly_geom_ref = poly_feat.GetGeometryRef()
if poly_geom_ref is None:
LOGGER.warning(
f'no geometry in {vector_path} FID: {poly_feat.GetFID()}, '
'skipping...')
continue
if poly_geom_ref.IsEmpty():
LOGGER.warning(
f'empty geometry in {vector_path} FID: {poly_feat.GetFID()}, '
'skipping...')
continue
# with GDAL>=3.3.0 ExportToWkb returns a bytearray instead of bytes
shapely_polygon_lookup[poly_feat.GetFID()] = (
shapely.wkb.loads(bytes(poly_geom_ref.ExportToWkb())))
poly_geom_ref = None
poly_feat = None
LOGGER.info("build shapely rtree index")
r_tree_index_stream = [
(poly_fid, poly.bounds, None)
for poly_fid, poly in shapely_polygon_lookup.items()
if bounding_box.intersects(poly)]
if r_tree_index_stream:
poly_rtree_index = rtree.index.Index(r_tree_index_stream)
else:
LOGGER.warning("no polygons intersected the bounding box")
return []
vector_layer = None
vector = None
LOGGER.info(
'poly feature lookup 100.0%% complete on %s',
os.path.basename(vector_path))
LOGGER.info('build poly intersection lookup')
poly_intersect_lookup = collections.defaultdict(set)
for poly_index, (poly_fid, poly_geom) in enumerate(
shapely_polygon_lookup.items()):
timed_logger.info(
"poly intersection lookup approximately %.1f%% complete "
"on %s", 100.0 * float(poly_index+1) / len(
shapely_polygon_lookup), os.path.basename(vector_path))
possible_intersection_set = list(poly_rtree_index.intersection(
poly_geom.bounds))
# no reason to prep the polygon to intersect itself
if len(possible_intersection_set) > 1:
polygon = shapely.prepared.prep(poly_geom)
else:
polygon = poly_geom
for intersect_poly_fid in possible_intersection_set:
# If geometries touch (share 1+ boundary point), then do not count
# it as an intersection.
if geometries_may_touch and polygon.touches(
shapely_polygon_lookup[intersect_poly_fid]):
continue
if intersect_poly_fid == poly_fid or polygon.intersects(
shapely_polygon_lookup[intersect_poly_fid]):
poly_intersect_lookup[poly_fid].add(intersect_poly_fid)
polygon = None
LOGGER.info(
'poly intersection feature lookup 100.0%% complete on %s',
os.path.basename(vector_path))
# Build maximal subsets
subset_list = []
while len(poly_intersect_lookup) > 0:
# sort polygons by increasing number of intersections
intersections_list = [
(len(poly_intersect_set), poly_fid, poly_intersect_set)
for poly_fid, poly_intersect_set in
poly_intersect_lookup.items()]
intersections_list.sort()
# build maximal subset
maximal_set = set()
for _, poly_fid, poly_intersect_set in intersections_list:
timed_logger.info(
"maximal subset build approximately %.1f%% complete "
"on %s", 100.0 * float(
feature_count - len(poly_intersect_lookup)) /
feature_count, os.path.basename(vector_path))
if not poly_intersect_set.intersection(maximal_set):
# no intersection, add poly_fid to the maximal set and remove
# the polygon from the lookup
maximal_set.add(poly_fid)
del poly_intersect_lookup[poly_fid]
# remove all the polygons from intersections once they're computed
for poly_fid, poly_intersect_set in poly_intersect_lookup.items():
poly_intersect_lookup[poly_fid] = (
poly_intersect_set.difference(maximal_set))
subset_list.append(maximal_set)
LOGGER.info(
'maximal subset build 100.0%% complete on %s',
os.path.basename(vector_path))
return subset_list
def _next_regular(base):
"""Find the next regular number greater than or equal to base.
Regular numbers are composites of the prime factors 2, 3, and 5.
Also known as 5-smooth numbers or Hamming numbers, these are the optimal
size for inputs to FFTPACK.
This source was taken directly from scipy.signaltools and saves us from
having to access a protected member in a library that could change in
future releases:
https://github.com/scipy/scipy/blob/v0.17.1/scipy/signal/signaltools.py#L211
Args:
base (int): a positive integer to start to find the next Hamming
number.
Return:
The next regular number greater than or equal to ``base``.
"""
if base <= 6:
return base
# Quickly check if it's already a power of 2
if not (base & (base-1)):
return base
match = float('inf') # Anything found will be smaller
p5 = 1
while p5 < base:
p35 = p5
while p35 < base:
# Ceiling integer division, avoiding conversion to float
# (quotient = ceil(base / p35))
quotient = -(-base // p35)
# Quickly find next power of 2 >= quotient
p2 = 2**((quotient - 1).bit_length())
N = p2 * p35
if N == base:
return N
elif N < match:
match = N
p35 *= 3
if p35 == base:
return p35
if p35 < match:
match = p35
p5 *= 5
if p5 == base:
return p5
if p5 < match:
match = p5
return match
[docs]def convolve_2d(
signal_path_band, kernel_path_band, target_path,
ignore_nodata_and_edges=False, mask_nodata=True,
normalize_kernel=False, target_datatype=gdal.GDT_Float64,
target_nodata=None, working_dir=None, set_tol_to_zero=1e-8,
max_timeout=_MAX_TIMEOUT,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Convolve 2D kernel over 2D signal.
Convolves the raster in ``kernel_path_band`` over ``signal_path_band``.
Nodata values are treated as 0.0 during the convolution and masked to
nodata for the output result where ``signal_path`` has nodata.
Note with default values, boundary effects can be seen in the result where
the kernel would hang off the edge of the raster or in regions with
nodata pixels. The function would treat these areas as values with "0.0"
by default thus pulling the total convolution down in these areas. This
is similar to setting ``mode='same'`` in Numpy's ``convolve`` function:
https://numpy.org/doc/stable/reference/generated/numpy.convolve.html
This boundary effect can be avoided by setting
``ignore_nodata_and_edges=True`` which normalizes the target result by
dynamically accounting for the number of valid signal pixels the kernel
overlapped during the convolution step.
Args:
signal_path_band (tuple): a 2 tuple of the form
(filepath to signal raster, band index).
kernel_path_band (tuple): a 2 tuple of the form
(filepath to kernel raster, band index), all pixel values should
be valid -- output is not well defined if the kernel raster has
nodata values. To create a kernel raster, see the documentation
and helper functions available in the
:doc:`pygeoprocessing kernels module <pygeoprocessing.kernels>`.
target_path (string): filepath to target raster that's the convolution
of signal with kernel. Output will be a single band raster of
same size and projection as ``signal_path_band``. Any nodata pixels
that align with ``signal_path_band`` will be set to nodata.
ignore_nodata_and_edges (boolean): If true, any pixels that are equal
to ``signal_path_band``'s nodata value or signal pixels where the
kernel extends beyond the edge of the raster are not included when
averaging the convolution filter. This has the effect of
"spreading" the result as though nodata and edges beyond the
bounds of the raster are 0s. If set to false this tends to "pull"
the signal away from nodata holes or raster edges. Set this value
to ``True`` to avoid distortions signal values near edges for
large integrating kernels. It can be useful to set this value to
``True`` to fill nodata holes through distance weighted averaging.
In this case ``mask_nodata`` must be set to ``False`` so the result
does not mask out these areas which are filled in. When using this
technique be careful of cases where the kernel does not extend
over any areas except nodata holes, in this case the resulting
values in these areas will be nonsensical numbers, perhaps
numerical infinity or NaNs.
normalize_kernel (boolean): If true, the result is divided by the
sum of the kernel.
mask_nodata (boolean): If true, ``target_path`` raster's output is
nodata where ``signal_path_band``'s pixels were nodata. Note that
setting ``ignore_nodata_and_edges`` to ``True`` while setting
``mask_nodata`` to ``False`` can allow for a technique involving
distance weighted averaging to define areas that would otherwise
be nodata. Be careful in cases where the kernel does not
extend over any valid non-nodata area since the result can be
numerical infinity or NaNs.
target_datatype (GDAL type): a GDAL raster type to set the output
raster type to, as well as the type to calculate the convolution
in. Defaults to GDT_Float64. Note signed byte is not
supported.
target_nodata (int/float): nodata value to set on output raster.
If ``target_datatype`` is not gdal.GDT_Float64, this value must
be set. Otherwise defaults to the minimum value of a float32.
raster_creation_options (sequence): an argument list that will be
passed to the GTiff driver for creating ``target_path``. Useful
for blocksizes, compression, and more.
working_dir (string): If not None, indicates where temporary files
should be created during this run.
set_tol_to_zero (float): any value within +- this from 0.0 will get
set to 0.0. This is to handle numerical roundoff errors that
sometimes result in "numerical zero", such as -1.782e-18 that
cannot be tolerated by users of this function. If `None` no
adjustment will be done to output values.
max_timeout (float): maximum amount of time to wait for worker thread
to terminate.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Returns:
``None``
Raises:
ValueError:
if ``ignore_nodata_and_edges`` is ``True`` and ``mask_nodata``
is ``False``.
ValueError:
if ``signal_path_band`` or ``kernel_path_band`` is a row based
blocksize which would result in slow runtimes due to gdal
cache thrashing.
"""
if target_datatype is not gdal.GDT_Float64 and target_nodata is None:
raise ValueError(
"`target_datatype` is set, but `target_nodata` is None. "
"`target_nodata` must be set if `target_datatype` is not "
"`gdal.GDT_Float64`. `target_nodata` is set to None.")
if target_nodata is None:
target_nodata = float(numpy.finfo(numpy.float32).min)
if ignore_nodata_and_edges and not mask_nodata:
LOGGER.debug(
'ignore_nodata_and_edges is True while mask_nodata is False -- '
'this can yield a nonsensical result in areas where the kernel '
'touches only nodata values.')
bad_raster_path_list = []
for raster_id, raster_path_band in [
('signal', signal_path_band), ('kernel', kernel_path_band)]:
if (not _is_raster_path_band_formatted(raster_path_band)):
bad_raster_path_list.append((raster_id, raster_path_band))
if bad_raster_path_list:
raise ValueError(
"Expected raster path band sequences for the following arguments "
f"but instead got: {bad_raster_path_list}")
signal_raster_info = get_raster_info(signal_path_band[0])
kernel_raster_info = get_raster_info(kernel_path_band[0])
for info_dict, raster_path_band in zip(
[signal_raster_info, kernel_raster_info],
[signal_path_band, kernel_path_band]):
if 1 in info_dict['block_size']:
raise ValueError(
f'{raster_path_band} has a row blocksize which can make this '
f'function run very slow, create a square blocksize using '
f'`warp_raster` or `align_and_resize_raster_stack` which '
f'creates square blocksizes by default')
# The nodata value is reset to a different value at the end of this
# function. Here 0 is chosen as a default value since data are
# incrementally added to the raster
new_raster_from_base(
signal_path_band[0], target_path, target_datatype, [0],
raster_driver_creation_tuple=raster_driver_creation_tuple)
n_cols_signal, n_rows_signal = signal_raster_info['raster_size']
n_cols_kernel, n_rows_kernel = kernel_raster_info['raster_size']
s_path_band = signal_path_band
k_path_band = kernel_path_band
s_nodata = signal_raster_info['nodata'][0]
# we need the original signal raster info because we want the output to
# be clipped and NODATA masked to it
signal_raster = gdal.OpenEx(signal_path_band[0], gdal.OF_RASTER)
signal_band = signal_raster.GetRasterBand(signal_path_band[1])
# getting the offset list before it's opened for updating
target_offset_list = list(iterblocks((target_path, 1), offset_only=True))
target_raster = gdal.OpenEx(target_path, gdal.OF_RASTER | gdal.GA_Update)
target_band = target_raster.GetRasterBand(1)
# if we're ignoring nodata, we need to make a parallel convolved signal
# of the nodata mask
if ignore_nodata_and_edges:
raster_file, mask_raster_path = tempfile.mkstemp(
suffix='.tif', prefix='convolved_mask',
dir=os.path.dirname(target_path))
os.close(raster_file)
new_raster_from_base(
signal_path_band[0], mask_raster_path, gdal.GDT_Float64,
[0.0], raster_driver_creation_tuple=raster_driver_creation_tuple)
mask_raster = gdal.OpenEx(
mask_raster_path, gdal.GA_Update | gdal.OF_RASTER)
mask_band = mask_raster.GetRasterBand(1)
LOGGER.info('starting convolve')
timed_logger = TimedLoggingAdapter(_LOGGING_PERIOD)
# calculate the kernel sum for normalization
kernel_nodata = kernel_raster_info['nodata'][0]
kernel_sum = 0.0
for _, kernel_block in iterblocks(kernel_path_band):
if kernel_nodata is not None and ignore_nodata_and_edges:
kernel_block[numpy.isclose(kernel_block, kernel_nodata)] = 0.0
kernel_sum += numpy.sum(kernel_block)
# limit the size of the work queue since a large kernel / signal with small
# block size can have a large memory impact when queuing offset lists.
work_queue = queue.Queue(10)
signal_offset_list = list(iterblocks(s_path_band, offset_only=True))
kernel_offset_list = list(iterblocks(k_path_band, offset_only=True))
n_blocks = len(signal_offset_list) * len(kernel_offset_list)
LOGGER.debug('start fill work queue thread')
def _fill_work_queue():
"""Asynchronously fill the work queue."""
LOGGER.debug('fill work queue')
for signal_offset in signal_offset_list:
for kernel_offset in kernel_offset_list:
work_queue.put((signal_offset, kernel_offset))
work_queue.put(None)
LOGGER.debug('work queue full')
fill_work_queue_worker = threading.Thread(
target=_fill_work_queue)
fill_work_queue_worker.daemon = True
fill_work_queue_worker.start()
# limit the size of the write queue so we don't accidentally load a whole
# array into memory
LOGGER.debug('start worker thread')
write_queue = queue.Queue(10)
worker = threading.Thread(
target=_convolve_2d_worker,
args=(
signal_path_band, kernel_path_band,
ignore_nodata_and_edges, normalize_kernel,
set_tol_to_zero, work_queue, write_queue))
worker.daemon = True
worker.start()
n_blocks_processed = 0
LOGGER.info(f'{n_blocks} sent to workers, wait for worker results')
while True:
# the timeout guards against a worst case scenario where the
# ``_convolve_2d_worker`` has crashed.
write_payload = write_queue.get(timeout=_MAX_TIMEOUT)
if write_payload:
(index_dict, result, mask_result,
left_index_raster, right_index_raster,
top_index_raster, bottom_index_raster,
left_index_result, right_index_result,
top_index_result, bottom_index_result) = write_payload
else:
worker.join(max_timeout)
break
output_array = numpy.empty(
(index_dict['win_ysize'], index_dict['win_xsize']),
dtype=numpy.float32)
# the inital data value in target_band is 0 because that is the
# temporary nodata selected so that manual resetting of initial
# data values weren't necessary. at the end of this function the
# target nodata value is set to `target_nodata`.
current_output = target_band.ReadAsArray(**index_dict)
# read the signal block so we know where the nodata are
potential_nodata_signal_array = signal_band.ReadAsArray(**index_dict)
valid_mask = numpy.ones(
potential_nodata_signal_array.shape, dtype=bool)
# guard against a None nodata value
if s_nodata is not None and mask_nodata:
valid_mask[:] = (
~numpy.isclose(potential_nodata_signal_array, s_nodata))
output_array[:] = target_nodata
output_array[valid_mask] = (
(result[top_index_result:bottom_index_result,
left_index_result:right_index_result])[valid_mask] +
current_output[valid_mask])
target_band.WriteArray(
output_array, xoff=index_dict['xoff'],
yoff=index_dict['yoff'])
if ignore_nodata_and_edges:
# we'll need to save off the mask convolution so we can divide
# it in total later
current_mask = mask_band.ReadAsArray(**index_dict)
output_array[valid_mask] = (
(mask_result[
top_index_result:bottom_index_result,
left_index_result:right_index_result])[valid_mask] +
current_mask[valid_mask])
mask_band.WriteArray(
output_array, xoff=index_dict['xoff'],
yoff=index_dict['yoff'])
n_blocks_processed += 1
timed_logger.info(
"convolution worker approximately %.1f%% complete on %s",
100.0 * float(n_blocks_processed) / (n_blocks),
os.path.basename(target_path))
LOGGER.info(
f"convolution worker 100.0% complete on "
f"{os.path.basename(target_path)}")
target_band.FlushCache()
if ignore_nodata_and_edges:
signal_nodata = get_raster_info(signal_path_band[0])['nodata'][
signal_path_band[1]-1]
LOGGER.info(
"need to normalize result so nodata values are not included")
mask_pixels_processed = 0
mask_band.FlushCache()
for target_offset_data in target_offset_list:
target_block = target_band.ReadAsArray(
**target_offset_data).astype(numpy.float64)
signal_block = signal_band.ReadAsArray(**target_offset_data)
mask_block = mask_band.ReadAsArray(**target_offset_data)
if mask_nodata and signal_nodata is not None:
valid_mask = ~numpy.isclose(signal_block, signal_nodata)
else:
valid_mask = numpy.ones(target_block.shape, dtype=bool)
valid_mask &= (mask_block > 0)
# divide the target_band by the mask_band
target_block[valid_mask] /= mask_block[valid_mask].astype(
numpy.float64)
# scale by kernel sum if necessary since mask division will
# automatically normalize kernel
if not normalize_kernel:
target_block[valid_mask] *= kernel_sum
target_band.WriteArray(
target_block, xoff=target_offset_data['xoff'],
yoff=target_offset_data['yoff'])
mask_pixels_processed += target_block.size
timed_logger.info(
f"""convolution nodata normalizer approximately {
100 * mask_pixels_processed / (n_cols_signal * n_rows_signal)
:.1f}% complete on {os.path.basename(target_path)}""")
mask_raster = None
mask_band = None
os.remove(mask_raster_path)
LOGGER.info(
f"convolution nodata normalize 100.0% complete on "
f"{os.path.basename(target_path)}")
# set the nodata value from 0 to a reasonable value for the result
target_band.SetNoDataValue(target_nodata)
target_band = None
target_raster = None
[docs]def iterblocks(
raster_path_band, largest_block=_LARGEST_ITERBLOCK,
offset_only=False):
"""Iterate across all the memory blocks in the input raster.
Result is a generator of block location information and numpy arrays.
This is especially useful when a single value needs to be derived from the
pixel values in a raster, such as the sum total of all pixel values, or
a sequence of unique raster values. In such cases, ``raster_local_op``
is overkill, since it writes out a raster.
As a generator, this can be combined multiple times with itertools.izip()
to iterate 'simultaneously' over multiple rasters, though the user should
be careful to do so only with prealigned rasters.
Args:
raster_path_band (tuple): a path/band index tuple to indicate
which raster band iterblocks should iterate over.
largest_block (int): Attempts to iterate over raster blocks with
this many elements. Useful in cases where the blocksize is
relatively small, memory is available, and the function call
overhead dominates the iteration. Defaults to 2**20. A value of
anything less than the original blocksize of the raster will
result in blocksizes equal to the original size.
offset_only (boolean): defaults to False, if True ``iterblocks`` only
returns offset dictionary and doesn't read any binary data from
the raster. This can be useful when iterating over writing to
an output.
Yields:
If ``offset_only`` is false, on each iteration, a tuple containing a
dict of block data and a 2-dimensional numpy array are
yielded. The dict of block data has these attributes:
* ``data['xoff']`` - The X offset of the upper-left-hand corner of the
block.
* ``data['yoff']`` - The Y offset of the upper-left-hand corner of the
block.
* ``data['win_xsize']`` - The width of the block.
* ``data['win_ysize']`` - The height of the block.
If ``offset_only`` is True, the function returns only the block offset
data and does not attempt to read binary data from the raster.
"""
if not _is_raster_path_band_formatted(raster_path_band):
raise ValueError(
"`raster_path_band` not formatted as expected. Expects "
"(path, band_index), received %s" % repr(raster_path_band))
raster = gdal.OpenEx(raster_path_band[0], gdal.OF_RASTER)
if raster is None:
raise ValueError(
"Raster at %s could not be opened." % raster_path_band[0])
band = raster.GetRasterBand(raster_path_band[1])
block = band.GetBlockSize()
cols_per_block = block[0]
rows_per_block = block[1]
n_cols = raster.RasterXSize
n_rows = raster.RasterYSize
block_area = cols_per_block * rows_per_block
# try to make block wider
if int(largest_block / block_area) > 0:
width_factor = int(largest_block / block_area)
cols_per_block *= width_factor
if cols_per_block > n_cols:
cols_per_block = n_cols
block_area = cols_per_block * rows_per_block
# try to make block taller
if int(largest_block / block_area) > 0:
height_factor = int(largest_block / block_area)
rows_per_block *= height_factor
if rows_per_block > n_rows:
rows_per_block = n_rows
n_col_blocks = int(math.ceil(n_cols / float(cols_per_block)))
n_row_blocks = int(math.ceil(n_rows / float(rows_per_block)))
for row_block_index in range(n_row_blocks):
row_offset = row_block_index * rows_per_block
row_block_width = n_rows - row_offset
if row_block_width > rows_per_block:
row_block_width = rows_per_block
for col_block_index in range(n_col_blocks):
col_offset = col_block_index * cols_per_block
col_block_width = n_cols - col_offset
if col_block_width > cols_per_block:
col_block_width = cols_per_block
offset_dict = {
'xoff': col_offset,
'yoff': row_offset,
'win_xsize': col_block_width,
'win_ysize': row_block_width,
}
if offset_only:
yield offset_dict
else:
yield (offset_dict, band.ReadAsArray(**offset_dict))
band = None
raster = None
[docs]def mask_raster(
base_raster_path_band, mask_vector_path, target_mask_raster_path,
mask_layer_id=0, target_mask_value=None, working_dir=None,
all_touched=False, where_clause=None,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Mask a raster band with a given vector.
Args:
base_raster_path_band (tuple): a (path, band number) tuple indicating
the data to mask.
mask_vector_path (path): path to a vector that will be used to mask
anything outside of the polygon that overlaps with
``base_raster_path_band`` to ``target_mask_value`` if defined or
else ``base_raster_path_band``'s nodata value.
target_mask_raster_path (str): path to desired target raster that
is a copy of ``base_raster_path_band`` except any pixels that do
not intersect with ``mask_vector_path`` are set to
``target_mask_value`` or ``base_raster_path_band``'s nodata value
if ``target_mask_value`` is None.
mask_layer_id (str/int): an index or name to identify the mask
geometry layer in ``mask_vector_path``, default is 0.
target_mask_value (numeric): If not None, this value is written to
any pixel in ``base_raster_path_band`` that does not intersect
with ``mask_vector_path``. Otherwise the nodata value of
``base_raster_path_band`` is used.
working_dir (str): this is a path to a directory that can be used to
hold temporary files required to complete this operation.
all_touched (bool): if False, a pixel is only masked if its centroid
intersects with the mask. If True a pixel is masked if any point
of the pixel intersects the polygon mask.
where_clause (str): (optional) if not None, it is an SQL compatible
where clause that can be used to filter the features that are used
to mask the base raster.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to a GTiff driver tuple
defined at geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
"""
with tempfile.NamedTemporaryFile(
prefix='mask_raster', delete=False, suffix='.tif',
dir=working_dir) as mask_raster_file:
mask_raster_path = mask_raster_file.name
new_raster_from_base(
base_raster_path_band[0], mask_raster_path, gdal.GDT_Byte, [255],
fill_value_list=[0],
raster_driver_creation_tuple=raster_driver_creation_tuple)
base_raster_info = get_raster_info(base_raster_path_band[0])
rasterize(
mask_vector_path, mask_raster_path, burn_values=[1],
layer_id=mask_layer_id,
option_list=[('ALL_TOUCHED=%s' % all_touched).upper()],
where_clause=where_clause)
base_nodata = base_raster_info['nodata'][base_raster_path_band[1]-1]
if target_mask_value is None:
mask_value = base_nodata
if mask_value is None:
LOGGER.warning(
"No mask value was passed and target nodata is undefined, "
"defaulting to 0 as the target mask value.")
mask_value = 0
else:
mask_value = target_mask_value
def mask_op(base_array, mask_array):
result = numpy.copy(base_array)
result[mask_array == 0] = mask_value
return result
raster_calculator(
[base_raster_path_band, (mask_raster_path, 1)], mask_op,
target_mask_raster_path, base_raster_info['datatype'], base_nodata,
raster_driver_creation_tuple=raster_driver_creation_tuple)
os.remove(mask_raster_path)
def _gdal_to_numpy_type(gdal_type, metadata):
"""Calculate the equivalent numpy datatype from a GDAL type and metadata.
Args:
gdal_type: GDAL.GDT_* data type code
metadata: mapping or list of strings to check for the existence of
the 'PIXELTYPE=SIGNEDBYTE' flag
Returns:
numpy.dtype that is the equivalent of the input gdal type
Raises:
ValueError if an unsupported data type is entered
"""
if (GDAL_VERSION < (3, 7, 0) and gdal_type == gdal.GDT_Byte and
(('PIXELTYPE=SIGNEDBYTE' in metadata) or
('PIXELTYPE' in metadata and metadata['PIXELTYPE'] == 'SIGNEDBYTE'))):
return numpy.int8
numpy_type = gdal_array.GDALTypeCodeToNumericTypeCode(gdal_type)
if numpy_type is None:
raise ValueError(f"Unsupported DataType: {gdal_type}")
return numpy_type
def _numpy_to_gdal_type(numpy_type):
"""Calculate the equivalent GDAL type and metadata from a numpy type.
Args:
numpy_type: numpy data type
Returns:
(gdal type, metadata) tuple. gdal type is a gdal.GDT_* type code.
metadata is an empty list in most cases, or ['PIXELTYPE=SIGNEDBYTE']
if needed to indicate a signed byte type.
Raises:
ValueError if an unsupported data type is entered
"""
numpy_dtype = numpy.dtype(numpy_type)
if GDAL_VERSION < (3, 7, 0) and numpy_dtype == numpy.dtype(numpy.int8):
return gdal.GDT_Byte, ['PIXELTYPE=SIGNEDBYTE']
gdal_type = gdal_array.NumericTypeCodeToGDALTypeCode(numpy_dtype)
if gdal_type is None:
raise ValueError(f"Unsupported DataType: {numpy_type}")
return gdal_type, []
[docs]def merge_bounding_box_list(bounding_box_list, bounding_box_mode):
"""Create a single bounding box by union or intersection of the list.
Args:
bounding_box_list (sequence): a sequence of bounding box coordinates
in the order [minx, miny, maxx, maxy].
mode (string): either ``'union'`` or ``'intersection'`` for the
corresponding reduction mode.
Return:
A four tuple bounding box that is the union or intersection of the
input bounding boxes.
Raises:
ValueError
if the bounding boxes in ``bounding_box_list`` do not
intersect if the ``bounding_box_mode`` is 'intersection'.
"""
def _merge_bounding_boxes(bb1, bb2, mode):
"""Merge two bounding boxes through union or intersection.
Args:
bb1, bb2 (sequence): sequence of float representing bounding box
in the form bb=[minx,miny,maxx,maxy]
mode (string); one of 'union' or 'intersection'
Return:
Reduced bounding box of bb1/bb2 depending on mode.
"""
def _less_than_or_equal(x_val, y_val):
return x_val if x_val <= y_val else y_val
def _greater_than(x_val, y_val):
return x_val if x_val > y_val else y_val
if mode == "union":
comparison_ops = [
_less_than_or_equal, _less_than_or_equal,
_greater_than, _greater_than]
if mode == "intersection":
comparison_ops = [
_greater_than, _greater_than,
_less_than_or_equal, _less_than_or_equal]
bb_out = [op(x, y) for op, x, y in zip(comparison_ops, bb1, bb2)]
return bb_out
result_bb = functools.reduce(
functools.partial(_merge_bounding_boxes, mode=bounding_box_mode),
bounding_box_list)
if result_bb[0] > result_bb[2] or result_bb[1] > result_bb[3]:
raise ValueError(
"Bounding boxes do not intersect. Base list: %s mode: %s "
" result: %s" % (bounding_box_list, bounding_box_mode, result_bb))
return result_bb
[docs]def get_gis_type(path):
"""Calculate the GIS type of the file located at ``path``.
Args:
path (str): path to a file on disk or network.
Raises:
ValueError
if ``path`` is not a file or cannot be opened as a
``gdal.OF_RASTER`` or ``gdal.OF_VECTOR``.
Return:
A bitwise OR of all GIS types that PyGeoprocessing models, currently
this is
``pygeoprocessing.RASTER_TYPE``, or ``pygeoprocessing.VECTOR_TYPE``.
"""
from pygeoprocessing import UNKNOWN_TYPE
gis_type = UNKNOWN_TYPE
gis_raster = gdal.OpenEx(path, gdal.OF_RASTER)
if gis_raster is not None:
from pygeoprocessing import RASTER_TYPE
gis_type |= RASTER_TYPE
gis_raster = None
gis_vector = gdal.OpenEx(path, gdal.OF_VECTOR)
if gis_vector is not None:
from pygeoprocessing import VECTOR_TYPE
gis_type |= VECTOR_TYPE
gis_vector = None
if gis_type == UNKNOWN_TYPE:
raise ValueError(
f"Could not open {path} as a gdal.OF_RASTER or gdal.OF_VECTOR.")
return gis_type
def _make_logger_callback(message):
"""Build a timed logger callback that prints ``message`` replaced.
Args:
message (string): a string that expects 2 placement %% variables,
first for % complete from ``df_complete``, second from
``p_progress_arg[0]``.
Return:
Function with signature:
logger_callback(df_complete, psz_message, p_progress_arg)
"""
def logger_callback(df_complete, _, p_progress_arg):
"""Argument names come from the GDAL API for callbacks."""
try:
current_time = time.time()
if ((current_time - logger_callback.last_time) > 5.0 or
(df_complete == 1.0 and
logger_callback.total_time >= 5.0)):
# In some multiprocess applications I was encountering a
# ``p_progress_arg`` of None. This is unexpected and I suspect
# was an issue for some kind of GDAL race condition. So I'm
# guarding against it here and reporting an appropriate log
# if it occurs.
if p_progress_arg:
LOGGER.info(message, df_complete * 100, p_progress_arg[0])
else:
LOGGER.info(message, df_complete * 100, '')
logger_callback.last_time = current_time
logger_callback.total_time += current_time
except AttributeError:
logger_callback.last_time = time.time()
logger_callback.total_time = 0.0
except Exception:
LOGGER.exception("Unhandled error occurred while logging "
"progress. df_complete: %s, p_progress_arg: %s",
df_complete, p_progress_arg)
return logger_callback
def _is_raster_path_band_formatted(raster_path_band):
"""Return true if raster path band is a (str, int) tuple/list."""
if not isinstance(raster_path_band, (list, tuple)):
return False
elif len(raster_path_band) != 2:
return False
elif not isinstance(raster_path_band[0], str):
return False
elif not isinstance(raster_path_band[1], int):
return False
else:
return True
def _convolve_2d_worker(
signal_path_band, kernel_path_band,
ignore_nodata, normalize_kernel, set_tol_to_zero,
work_queue, write_queue):
"""Worker function to be used by ``convolve_2d``.
Args:
signal_path_band (tuple): a 2 tuple of the form
(filepath to signal raster, band index).
kernel_path_band (tuple): a 2 tuple of the form
(filepath to kernel raster, band index).
ignore_nodata (boolean): If true, any pixels that are equal to
``signal_path_band``'s nodata value are not included when
averaging the convolution filter.
normalize_kernel (boolean): If true, the result is divided by the
sum of the kernel.
set_tol_to_zero (float): Value to test close to to determine if values
are zero, and if so, set to zero.
work_queue (Queue): will contain (signal_offset, kernel_offset)
tuples that can be used to read raster blocks directly using
GDAL ReadAsArray(**offset). Indicates the block to operate on.
write_queue (Queue): mechanism to pass result back to the writer
contains a (index_dict, result, mask_result,
left_index_raster, right_index_raster,
top_index_raster, bottom_index_raster,
left_index_result, right_index_result,
top_index_result, bottom_index_result) tuple that's used
for writing and masking.
Return:
None
"""
signal_raster = gdal.OpenEx(signal_path_band[0], gdal.OF_RASTER)
kernel_raster = gdal.OpenEx(kernel_path_band[0], gdal.OF_RASTER)
signal_band = signal_raster.GetRasterBand(signal_path_band[1])
kernel_band = kernel_raster.GetRasterBand(kernel_path_band[1])
signal_raster_info = get_raster_info(signal_path_band[0])
kernel_raster_info = get_raster_info(kernel_path_band[0])
n_cols_signal, n_rows_signal = signal_raster_info['raster_size']
n_cols_kernel, n_rows_kernel = kernel_raster_info['raster_size']
signal_nodata = signal_raster_info['nodata'][0]
kernel_nodata = kernel_raster_info['nodata'][0]
mask_result = None # in case no mask is needed, variable is still defined
# calculate the kernel sum for normalization
kernel_sum = 0.0
for _, kernel_block in iterblocks(kernel_path_band):
if kernel_nodata is not None and ignore_nodata:
kernel_block[numpy.isclose(kernel_block, kernel_nodata)] = 0.0
kernel_sum += numpy.sum(kernel_block)
while True:
payload = work_queue.get()
if payload is None:
break
signal_offset, kernel_offset = payload
# ensure signal and kernel are internally float64 precision
# irrespective of their base type
signal_block = signal_band.ReadAsArray(**signal_offset).astype(
numpy.float64)
kernel_block = kernel_band.ReadAsArray(**kernel_offset).astype(
numpy.float64)
# don't ever convolve the nodata value
if signal_nodata is not None:
signal_nodata_mask = numpy.isclose(signal_block, signal_nodata)
signal_block[signal_nodata_mask] = 0.0
if not ignore_nodata:
signal_nodata_mask[:] = 0
else:
signal_nodata_mask = numpy.zeros(
signal_block.shape, dtype=bool)
left_index_raster = (
signal_offset['xoff'] - n_cols_kernel // 2 +
kernel_offset['xoff'])
right_index_raster = (
signal_offset['xoff'] - n_cols_kernel // 2 +
kernel_offset['xoff'] + signal_offset['win_xsize'] +
kernel_offset['win_xsize'] - 1)
top_index_raster = (
signal_offset['yoff'] - n_rows_kernel // 2 +
kernel_offset['yoff'])
bottom_index_raster = (
signal_offset['yoff'] - n_rows_kernel // 2 +
kernel_offset['yoff'] + signal_offset['win_ysize'] +
kernel_offset['win_ysize'] - 1)
# it's possible that the piece of the integrating kernel
# doesn't affect the final result, if so we should skip
if (right_index_raster < 0 or
bottom_index_raster < 0 or
left_index_raster > n_cols_signal or
top_index_raster > n_rows_signal):
continue
if kernel_nodata is not None:
kernel_block[numpy.isclose(kernel_block, kernel_nodata)] = 0.0
if normalize_kernel:
kernel_block /= kernel_sum
# determine the output convolve shape
shape = (
numpy.array(signal_block.shape) +
numpy.array(kernel_block.shape) - 1)
# add zero padding so FFT is fast
fshape = [_next_regular(int(d)) for d in shape]
signal_fft = numpy.fft.rfftn(signal_block, fshape)
kernel_fft = numpy.fft.rfftn(kernel_block, fshape)
# this variable determines the output slice that doesn't include
# the padded array region made for fast FFTs.
fslice = tuple([slice(0, int(sz)) for sz in shape])
# classic FFT convolution
result = numpy.fft.irfftn(signal_fft * kernel_fft, fshape)[fslice]
# nix any roundoff error
if set_tol_to_zero is not None:
result[numpy.isclose(result, set_tol_to_zero)] = 0.0
# if we're ignoring nodata, we need to make a convolution of the
# nodata mask too
if ignore_nodata:
mask_fft = numpy.fft.rfftn(
numpy.where(signal_nodata_mask, 0.0, 1.0), fshape)
mask_result = numpy.fft.irfftn(
mask_fft * kernel_fft, fshape)[fslice]
left_index_result = 0
right_index_result = result.shape[1]
top_index_result = 0
bottom_index_result = result.shape[0]
# we might abut the edge of the raster, clip if so
if left_index_raster < 0:
left_index_result = -left_index_raster
left_index_raster = 0
if top_index_raster < 0:
top_index_result = -top_index_raster
top_index_raster = 0
if right_index_raster > n_cols_signal:
right_index_result -= right_index_raster - n_cols_signal
right_index_raster = n_cols_signal
if bottom_index_raster > n_rows_signal:
bottom_index_result -= (
bottom_index_raster - n_rows_signal)
bottom_index_raster = n_rows_signal
# Add result to current output to account for overlapping edges
index_dict = {
'xoff': left_index_raster,
'yoff': top_index_raster,
'win_xsize': right_index_raster-left_index_raster,
'win_ysize': bottom_index_raster-top_index_raster
}
write_queue.put(
(index_dict, result, mask_result,
left_index_raster, right_index_raster,
top_index_raster, bottom_index_raster,
left_index_result, right_index_result,
top_index_result, bottom_index_result))
# Indicates worker has terminated
write_queue.put(None)
def _assert_is_valid_pixel_size(target_pixel_size):
"""Return true if ``target_pixel_size`` is a valid 2 element sequence.
Raises ValueError if not a two element list/tuple and/or the values in
the sequence are not numerical.
"""
def _is_number(x):
"""Return true if x is a number."""
try:
if isinstance(x, str):
return False
float(x)
return True
except (ValueError, TypeError):
return False
if not isinstance(target_pixel_size, (list, tuple)):
raise ValueError(
"target_pixel_size is not a tuple, its value was '%s'",
repr(target_pixel_size))
if (len(target_pixel_size) != 2 or
not all([_is_number(x) for x in target_pixel_size])):
raise ValueError(
"Invalid value for `target_pixel_size`, expected two numerical "
"elements, got: %s", repr(target_pixel_size))
return True
[docs]def shapely_geometry_to_vector(
shapely_geometry_list, target_vector_path, projection_wkt,
vector_format, fields=None, attribute_list=None,
ogr_geom_type=ogr.wkbPolygon):
"""Convert list of geometry to vector on disk.
Args:
shapely_geometry_list (list): a list of Shapely objects.
target_vector_path (str): path to target vector.
projection_wkt (str): WKT for target vector.
vector_format (str): GDAL driver name for target vector.
fields (dict): a python dictionary mapping string fieldname
to OGR Fieldtypes, if None no fields are added
attribute_list (list of dicts): a list of python dictionary mapping
fieldname to field value for each geometry in
`shapely_geometry_list`, if None, no attributes are created.
ogr_geom_type (ogr geometry enumerated type): sets the target layer
geometry type. Defaults to wkbPolygon.
Return:
None
"""
if fields is None:
fields = {}
if attribute_list is None:
attribute_list = [{} for _ in range(len(shapely_geometry_list))]
num_geoms = len(shapely_geometry_list)
num_attrs = len(attribute_list)
if num_geoms != num_attrs:
raise ValueError(
f"Geometry count ({num_geoms}) and attribute count "
f"({num_attrs}) do not match.")
vector_driver = ogr.GetDriverByName(vector_format)
target_vector = vector_driver.CreateDataSource(target_vector_path)
layer_name = os.path.basename(os.path.splitext(target_vector_path)[0])
projection = osr.SpatialReference()
projection.ImportFromWkt(projection_wkt)
target_layer = target_vector.CreateLayer(
layer_name, srs=projection, geom_type=ogr_geom_type)
for field_name, field_type in fields.items():
target_layer.CreateField(ogr.FieldDefn(field_name, field_type))
layer_defn = target_layer.GetLayerDefn()
for field_index, (shapely_feature, feature_attributes) in enumerate(
zip(shapely_geometry_list, attribute_list)):
new_feature = ogr.Feature(layer_defn)
new_geometry = ogr.CreateGeometryFromWkb(shapely_feature.wkb)
new_feature.SetGeometry(new_geometry)
if set(fields.keys()) != set(feature_attributes.keys()):
raise ValueError(
f"The fields and attributes for feature {field_index} "
f"do not match. Field definitions={fields}. "
f"Feature attributes={feature_attributes}.")
for field_name, field_value in feature_attributes.items():
new_feature.SetField(field_name, field_value)
target_layer.CreateFeature(new_feature)
target_layer = None
target_vector = None
[docs]def numpy_array_to_raster(
base_array, target_nodata, pixel_size, origin, projection_wkt,
target_path,
raster_driver_creation_tuple=DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS):
"""Create a single band raster of size ``base_array.shape``.
The GDAL datatype of the target raster is determined by the numpy dtype of
``base_array``.
Note:
The ``origin`` and ``pixel_size`` parameters must both be defined
properly as 2-tuples of floats, or else must both be set to ``None``.
A ``ValueError`` will be raised otherwise.
Args:
base_array (numpy.array): a 2d numpy array.
target_nodata (numeric): nodata value of target array, can be None.
pixel_size (tuple): square dimensions (in ``(x, y)``) of pixel. Can be
None to indicate no stated pixel size.
origin (tuple/list): x/y coordinate of the raster origin. Can be None
to indicate no stated origin.
projection_wkt (str): target projection in wkt. Can be None to
indicate no projection/SRS.
target_path (str): path to raster to create that will be of the
same type of base_array with contents of base_array.
raster_driver_creation_tuple (tuple): a tuple containing a GDAL driver
name string as the first element and a GDAL creation options
tuple/list as the second. Defaults to
geoprocessing.DEFAULT_GTIFF_CREATION_TUPLE_OPTIONS.
Return:
None
"""
driver_name, creation_options = raster_driver_creation_tuple
raster_driver = gdal.GetDriverByName(driver_name)
ny, nx = base_array.shape
gdal_type, type_creation_options = _numpy_to_gdal_type(base_array.dtype)
new_raster = raster_driver.Create(
target_path, nx, ny, 1, gdal_type,
options=list(creation_options) + type_creation_options)
if projection_wkt is not None:
new_raster.SetProjection(projection_wkt)
if origin is not None and pixel_size is not None:
new_raster.SetGeoTransform(
[origin[0], pixel_size[0], 0, origin[1], 0, pixel_size[1]])
elif origin is not None or pixel_size is not None:
raise ValueError(
"Origin and pixel size must both be defined or both be None")
new_band = new_raster.GetRasterBand(1)
if target_nodata is not None:
new_band.SetNoDataValue(target_nodata)
new_band.WriteArray(base_array)
new_band = None
new_raster = None
[docs]def raster_to_numpy_array(raster_path, band_id=1):
"""Read the entire contents of the raster band to a numpy array.
Args:
raster_path (str): path to raster.
band_id (int): band in the raster to read.
Return:
numpy array contents of `band_id` in raster.
"""
raster = gdal.OpenEx(raster_path, gdal.OF_RASTER)
band = raster.GetRasterBand(band_id)
array = band.ReadAsArray()
band = None
raster = None
return array
[docs]def stitch_rasters(
base_raster_path_band_list,
resample_method_list,
target_stitch_raster_path_band,
overlap_algorithm='etch',
area_weight_m2_to_wgs84=False,
osr_axis_mapping_strategy=DEFAULT_OSR_AXIS_MAPPING_STRATEGY):
"""Stitch the raster in the base list into the existing target.
Args:
base_raster_path_band_list (sequence): sequence of raster path/band
tuples to stitch into target.
resample_method_list (sequence): a sequence of resampling methods
which one to one map each path in ``base_raster_path_band_list``
during resizing. Each element must be one of,
'rms | mode | sum | q1 | near | q3 | average | cubicspline |
bilinear | max | med | min | cubic | lanczos'
target_stitch_raster_path_band (tuple): raster path/band tuple to an
existing raster, values in ``base_raster_path_band_list`` will
be stitched into this raster/band in the order they are in the
list. The nodata value for the target band must be defined and
will be written over with values from the base raster. Nodata
values in the base rasters will not be written into the target.
If the pixel size or projection are different between base and
target the base is warped to the target's cell size and target
with the interpolation method provided. If any part of the
base raster lies outside of the target, that part of the base
is ignored. A warning is logged if the entire base raster is
outside of the target bounds.
overlap_algorithm (str): this value indicates which algorithm to use
when a raster is stitched on non-nodata values in the target
stitch raster. It can be one of the following:
- 'etch': write a value to the target raster only if the target
raster pixel is nodata. If the target pixel is non-nodata
ignore any additional values to write on that pixel.
- 'replace': write a value to the target raster irrespective
of the value of the target raster
- 'add': add the value to be written to the target raster to
any existing value that is there. If the existing value
is nodata, treat it as 0.0.
area_weight_m2_to_wgs84 (bool): If ``True`` the stitched raster will
be converted to a per-area value before reprojection to wgs84,
then multiplied by the m^2 area per pixel in the wgs84 coordinate
space. This is useful when the quantity being stitched is a total
quantity per pixel rather than a per unit area density. Note
this assumes input rasters are in a projected space of meters,
if they are not the stitched output will be nonsensical.
osr_axis_mapping_strategy (int): OSR axis mapping strategy for
``SpatialReference`` objects. Defaults to
``geoprocessing.DEFAULT_OSR_AXIS_MAPPING_STRATEGY``. This
parameter should not be changed unless you know what you are
doing.
Return:
None.
"""
valid_overlap_algorithms = ['etch', 'replace', 'add']
if overlap_algorithm not in valid_overlap_algorithms:
raise ValueError(
f'overlap algorithm {overlap_algorithm} is not one of '
f'{valid_overlap_algorithms}')
if not _is_raster_path_band_formatted(target_stitch_raster_path_band):
raise ValueError(
f'Expected raster path/band tuple for '
f'target_stitch_raster_path_band but got '
f'"{target_stitch_raster_path_band}"')
if len(base_raster_path_band_list) != len(resample_method_list):
raise ValueError(
f'Expected same number of elements in '
f'`base_raster_path_band_list` as `resample_method_list` but '
f'got {len(base_raster_path_band_list)} != '
f'{len(resample_method_list)} respectively')
if not os.path.exists(target_stitch_raster_path_band[0]):
raise ValueError(
f'Target stitch raster does not exist: '
f'"{target_stitch_raster_path_band[0]}"')
gis_type = get_gis_type(target_stitch_raster_path_band[0])
from pygeoprocessing import RASTER_TYPE
if gis_type != RASTER_TYPE:
raise ValueError(
f'Target stitch raster is not a raster. '
f'Location: "{target_stitch_raster_path_band[0]}" '
f'GIS type: {gis_type}')
target_raster_info = get_raster_info(target_stitch_raster_path_band[0])
if target_stitch_raster_path_band[1] > len(target_raster_info['nodata']):
raise ValueError(
f'target_stitch_raster_path_band refers to a band that exceeds '
f'the number of bands in the raster:\n'
f'target_stitch_raster_path_band[1]: '
f'{target_stitch_raster_path_band[1]} '
f'n bands: {len(target_raster_info["nodata"])}')
target_nodata = target_raster_info['nodata'][
target_stitch_raster_path_band[1]-1]
if target_nodata is None:
raise ValueError(
f'target stitch raster at "{target_stitch_raster_path_band[0]} "'
f'nodata value is `None`, expected non-`None` value')
target_raster = gdal.OpenEx(
target_stitch_raster_path_band[0], gdal.OF_RASTER | gdal.GA_Update)
target_band = target_raster.GetRasterBand(
target_stitch_raster_path_band[1])
target_inv_gt = gdal.InvGeoTransform(target_raster_info['geotransform'])
target_raster_x_size, target_raster_y_size = target_raster_info[
'raster_size']
for (raster_path, raster_band_id), resample_method in zip(
base_raster_path_band_list, resample_method_list):
LOGGER.info(
f'stitching {(raster_path, raster_band_id)} into '
f'{target_stitch_raster_path_band}')
raster_info = get_raster_info(raster_path)
projected_raster_bounding_box = transform_bounding_box(
raster_info['bounding_box'],
raster_info['projection_wkt'],
target_raster_info['projection_wkt'])
try:
# merge the bounding boxes only to see if they don't intersect
_ = merge_bounding_box_list(
[projected_raster_bounding_box,
target_raster_info['bounding_box']], 'intersection')
except ValueError:
LOGGER.warning(
f'the raster at "{raster_path}"" does not intersect the '
f'stitch raster at "{target_stitch_raster_path_band[0]}", '
f'skipping...')
continue
# use this to determine if we need to warp and delete if we did at
# the end
if (raster_info['projection_wkt'] ==
target_raster_info['projection_wkt'] and
raster_info['pixel_size'] ==
target_raster_info['pixel_size']):
warped_raster = False
base_stitch_raster_path = raster_path
else:
workspace_dir = tempfile.mkdtemp(
dir=os.path.dirname(target_stitch_raster_path_band[0]),
prefix='stitch_rasters_workspace')
base_stitch_raster_path = os.path.join(
workspace_dir, os.path.basename(raster_path))
warp_raster(
raster_path, target_raster_info['pixel_size'],
base_stitch_raster_path, resample_method,
target_projection_wkt=target_raster_info['projection_wkt'],
working_dir=workspace_dir,
osr_axis_mapping_strategy=osr_axis_mapping_strategy)
warped_raster = True
if warped_raster and area_weight_m2_to_wgs84:
# determine base area per pixel currently and area per pixel
# once it is projected to wgs84 pixel sizes
base_pixel_area_m2 = abs(numpy.prod(raster_info['pixel_size']))
base_stitch_raster_info = get_raster_info(
base_stitch_raster_path)
_, lat_min, _, lat_max = base_stitch_raster_info['bounding_box']
n_rows = base_stitch_raster_info['raster_size'][1]
# this column is a longitude invariant latitude variant pixel
# area for scaling area dependent values
m2_area_per_lat = _create_latitude_m2_area_column(
lat_min, lat_max, n_rows)
def _mult_op(base_array, base_nodata, scale, datatype):
"""Scale non-nodata by scale."""
result = base_array.astype(datatype)
if base_nodata is not None:
valid_mask = ~numpy.isclose(base_array, base_nodata)
else:
valid_mask = numpy.ones(
base_array.shape, dtype=bool)
result[valid_mask] = result[valid_mask] * scale[valid_mask]
return result
base_stitch_nodata = base_stitch_raster_info['nodata'][0]
scaled_raster_path = os.path.join(
workspace_dir,
f'scaled_{os.path.basename(base_stitch_raster_path)}')
gdal_type = _gdal_to_numpy_type(
target_band.DataType,
target_band.GetMetadata('IMAGE_STRUCTURE'))
# multiply the pixels in the resampled raster by the ratio of
# the pixel area in the wgs84 units divided by the area of the
# original pixel
raster_calculator(
[(base_stitch_raster_path, 1), (base_stitch_nodata, 'raw'),
m2_area_per_lat/base_pixel_area_m2,
(gdal_type, 'raw')], _mult_op,
scaled_raster_path,
target_raster_info['datatype'], base_stitch_nodata)
# swap the result to base stitch so the rest of the function
# operates on the area scaled raster
os.remove(base_stitch_raster_path)
base_stitch_raster_path = scaled_raster_path
base_raster = gdal.OpenEx(base_stitch_raster_path, gdal.OF_RASTER)
base_gt = base_raster.GetGeoTransform()
base_band = base_raster.GetRasterBand(raster_band_id)
base_nodata = base_band.GetNoDataValue()
# Get the target upper left xoff/yoff w/r/t the stitch raster 0,0
# coordinates
target_to_base_xoff, target_to_base_yoff = [
int(_) for _ in gdal.ApplyGeoTransform(
target_inv_gt, *gdal.ApplyGeoTransform(base_gt, 0, 0))]
for offset_dict in iterblocks(
(base_stitch_raster_path, raster_band_id), offset_only=True):
_offset_vars = {}
overlap = True
for (target_to_base_off, off_val,
target_off_id, off_clip_id, win_size_id, raster_size) in [
(target_to_base_xoff, offset_dict['xoff'],
'target_xoff', 'xoff_clip', 'win_xsize',
target_raster_x_size),
(target_to_base_yoff, offset_dict['yoff'],
'target_yoff', 'yoff_clip', 'win_ysize',
target_raster_y_size)]:
_offset_vars[target_off_id] = (target_to_base_off+off_val)
if _offset_vars[target_off_id] >= raster_size:
overlap = False
break
# how far to move right to get in the target raster
_offset_vars[off_clip_id] = 0
if _offset_vars[target_off_id] < 0:
_offset_vars[off_clip_id] = -_offset_vars[target_off_id]
_offset_vars[win_size_id] = offset_dict[win_size_id]
if _offset_vars[off_clip_id] >= _offset_vars[win_size_id]:
# its too far left for the whole window
overlap = False
break
# make the _offset_vars[win_size_id] smaller if it shifts
# off the target window
if (_offset_vars[off_clip_id] + _offset_vars[target_off_id] +
_offset_vars[win_size_id] >= raster_size):
_offset_vars[win_size_id] -= (
_offset_vars[off_clip_id] +
_offset_vars[target_off_id] +
_offset_vars[win_size_id] - raster_size)
if not overlap:
continue
target_array = target_band.ReadAsArray(
xoff=_offset_vars['target_xoff']+_offset_vars['xoff_clip'],
yoff=_offset_vars['target_yoff']+_offset_vars['yoff_clip'],
win_xsize=_offset_vars['win_xsize'],
win_ysize=_offset_vars['win_ysize'])
target_nodata_mask = numpy.isclose(target_array, target_nodata)
base_array = base_band.ReadAsArray(
xoff=offset_dict['xoff']+_offset_vars['xoff_clip'],
yoff=offset_dict['yoff']+_offset_vars['yoff_clip'],
win_xsize=_offset_vars['win_xsize'],
win_ysize=_offset_vars['win_ysize'])
if base_nodata is not None:
base_nodata_mask = numpy.isclose(base_array, base_nodata)
else:
base_nodata_mask = numpy.zeros(
base_array.shape, dtype=bool)
if overlap_algorithm == 'etch':
# place values only where target is nodata
valid_mask = ~base_nodata_mask & target_nodata_mask
target_array[valid_mask] = base_array[valid_mask]
elif overlap_algorithm == 'replace':
# write valid values into the target -- disregard any
# existing values in the target
valid_mask = ~base_nodata_mask
target_array[valid_mask] = base_array[valid_mask]
elif overlap_algorithm == 'add':
# add values to the target and treat target nodata as 0.
valid_mask = ~base_nodata_mask
masked_target_array = target_array[valid_mask]
target_array_nodata_mask = numpy.isclose(
masked_target_array, target_nodata)
target_array[valid_mask] = (
base_array[valid_mask] +
numpy.where(
target_array_nodata_mask, 0, masked_target_array))
else:
raise RuntimeError(
f'overlap_algorithm {overlap_algorithm} was not defined '
f'but also not detected earlier -- this should never '
f'happen')
target_band.WriteArray(
target_array,
xoff=_offset_vars['target_xoff']+_offset_vars['xoff_clip'],
yoff=_offset_vars['target_yoff']+_offset_vars['yoff_clip'])
base_raster = None
base_band = None
if warped_raster:
shutil.rmtree(workspace_dir)
target_raster = None
target_band = None
[docs]def build_overviews(
raster_path, internal=False, resample_method='near',
overwrite=False, levels='auto'):
"""Build overviews for a raster dataset.
Args:
raster_path (str): A path to a raster on disk for which overviews
should be built.
internal=False (bool): Whether to modify the raster when building
overviews. In GeoTiffs, this builds internal overviews when
``internal=True``, and external overviews when ``internal=False``.
resample_method='near' (str): The resample method to use when
building overviews. Must be a valid resampling method for
``gdal.GDALDataset.BuildOverviews``, one of
'rms | mode | sum | q1 | near | q3 | average | cubicspline |
bilinear | max | med | min | cubic | lanczos'.
overwrite=False (bool): Whether to overwrite existing overviews, if
any exist.
levels='auto' (sequence): A sequence of integer overview levels. If
``'auto'``, overview levels will be determined by using factors of
2 until the overview's x and y dimensions are both less than 256.
Example:
Generate overviews, regardless of whether overviews already exist
for the raster, letting the function determine the levels of overviews
to generate::
build_overviews(raster_path)
Generate overviews for 4 levels, at 1/2, 1/4, 1/8 and 1/16 the
resolution::
build_overviews(raster_path, levels=[2, 4, 8, 16])
Returns:
``None``
"""
def overviews_progress(*args, **kwargs):
pct_complete, name, other = args
percent = round(pct_complete * 100, 2)
if time.time() - overviews_progress.last_progress_report > 5.0:
LOGGER.info(f"Overviews progress: {percent}%")
overviews_progress.last_progress_report = time.time()
overviews_progress.last_progress_report = time.time()
open_flags = gdal.OF_RASTER
if internal:
open_flags |= gdal.GA_Update
LOGGER.info(f"Building internal overviews on {raster_path}")
else:
LOGGER.info("Building external overviews.")
raster = gdal.OpenEx(raster_path, open_flags)
overview_count = 0
for band_index in range(1, raster.RasterCount + 1):
band = raster.GetRasterBand(band_index)
overview_count += band.GetOverviewCount()
if overview_count > 0:
if overwrite:
LOGGER.info(f"Clearing existing overviews from {raster_path}")
result = raster.BuildOverviews(
resampling=resample_method,
overviewlist=[],
callback=overviews_progress
)
LOGGER.info(f"Overviews cleared from {raster_path}")
else:
raise ValueError(
f"Raster already has overviews. Use "
"overwrite=True to override this and regenerate overviews on "
f"{raster_path}")
# This loop and limiting factor borrowed from gdaladdo.cpp.
# Create overviews so long as the overviews are at least 256 pixels in
# either x or y dimensions.
if levels == 'auto':
overview_scales = []
factor = 2
limiting_factor = 256
while (math.ceil(raster.RasterXSize / factor) > limiting_factor or
math.ceil(raster.RasterYSize / factor) > limiting_factor):
overview_scales.append(factor)
factor *= 2
else:
overview_scales = [int(level) for level in levels]
LOGGER.debug(f"Using overviews {overview_scales}")
result = raster.BuildOverviews(
resampling=resample_method,
overviewlist=overview_scales,
callback=overviews_progress
)
LOGGER.info(f"Overviews completed for {raster_path}")
if result: # Result will be nonzero on error.
raise RuntimeError(
f"Building overviews failed or was interrupted for {raster_path}")
def _m2_area_of_wg84_pixel(pixel_size, center_lat):
"""Calculate m^2 area of a square wgs84 pixel.
Adapted from: https://gis.stackexchange.com/a/127327/2397
Args:
pixel_size (float): length of side of a square pixel in degrees.
center_lat (float): latitude of the center of the pixel. Note this
value +/- half the `pixel-size` must not exceed 90/-90 degrees
latitude or an invalid area will be calculated.
Returns:
Area of square pixel of side length `pixel_size` centered at
`center_lat` in m^2.
"""
a = 6378137 # meters
b = 6356752.3142 # meters
e = math.sqrt(1 - (b/a)**2)
area_list = []
for f in [center_lat+pixel_size/2, center_lat-pixel_size/2]:
zm = 1 - e*math.sin(math.radians(f))
zp = 1 + e*math.sin(math.radians(f))
area_list.append(
math.pi * b**2 * (
math.log(zp/zm) / (2*e) +
math.sin(math.radians(f)) / (zp*zm)))
return abs(pixel_size / 360. * (area_list[0] - area_list[1]))
def _create_latitude_m2_area_column(lat_min, lat_max, n_pixels):
"""Create a (n, 1) sized numpy array with m^2 areas in each element.
Creates a per pixel m^2 area array that varies with changes in latitude.
This array can be used to scale values by area when converting to or
from a WGS84 projection to a projected one.
Args:
lat_max (float): maximum latitude in the bound
lat_min (float): minimum latitude in the bound
n_pixels (int): number of pixels to create for the column. The
size of the target square pixels are (lat_max-lat_min)/n_pixels
degrees per side.
Return:
A (n, 1) sized numpy array whose elements are the m^2 areas in each
element estimated by the latitude value at the center of each pixel.
"""
pixel_size = (lat_max - lat_min) / n_pixels
center_lat_array = numpy.linspace(
lat_min+pixel_size/2, lat_max-pixel_size/2, n_pixels)
area_array = numpy.array([
_m2_area_of_wg84_pixel(pixel_size, lat)
for lat in reversed(center_lat_array)]).reshape((n_pixels, 1))
return area_array