baseclasses.py 72.2 KB
Newer Older
1 2 3 4
# -*- coding: utf-8 -*-

import os
import warnings
5
from importlib import util
6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
from collections import OrderedDict

import numpy as np
from matplotlib import pyplot as plt
from osgeo import gdal_array
# custom
from shapely.geometry import Polygon
from shapely.wkt import loads as shply_loads
from six import PY3

# mpl_toolkits.basemap -> imported when GeoArray.show_map() is used
# dill -> imported when dumping GeoArray

try:
    from osgeo import gdal
    from osgeo import gdalnumeric
except ImportError:
    import gdal
    import gdalnumeric
from geopandas import GeoDataFrame, GeoSeries
from pandas import DataFrame
27 28 29 30 31 32 33
from py_tools_ds.convenience.object_oriented import alias_property
from py_tools_ds.geo.coord_calc import get_corner_coordinates
from py_tools_ds.geo.coord_grid import snap_bounds_to_pixGrid
from py_tools_ds.geo.coord_trafo import mapXY2imXY, imXY2mapXY, transform_any_prj, reproject_shapelyGeometry
from py_tools_ds.geo.projection import prj_equal, WKT2EPSG, EPSG2WKT
from py_tools_ds.geo.raster.conversion import raster2polygon
from py_tools_ds.geo.vector.topology \
34
    import get_footprint_polygon, polyVertices_outside_poly, fill_holes_within_poly
35 36
from py_tools_ds.geo.vector.geometry import boxObj
from py_tools_ds.io.raster.gdal import get_GDAL_ds_inmem
37
from py_tools_ds.compatibility.gdal import get_gdal_func
38
from py_tools_ds.numeric.numbers import is_number
39
from py_tools_ds.numeric.array import get_array_tilebounds
40 41 42 43

#  internal imports
from .subsetting import get_array_at_mapPos

44
if PY3:
45
    # noinspection PyCompatibility
46 47 48
    from builtins import TimeoutError, FileNotFoundError
else:
    from py_tools_ds.compatibility.python.exceptions import TimeoutError, FileNotFoundError
49

50
__author__ = 'Daniel Scheffler'
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75


class GeoArray(object):
    def __init__(self, path_or_array, geotransform=None, projection=None, bandnames=None, nodata=None, progress=True,
                 q=False):
        # type: (any, tuple, str, list, float, bool, bool) -> None
        """This class creates a fast Python interface for geodata - either on disk or in memory. It can be instanced with
        a file path or with a numpy array and the corresponding geoinformation. Instances can always be indexed like
        normal numpy arrays, no matter if GeoArray has been instanced from file or from an in-memory array. GeoArray
        provides a wide range of geo-related attributes belonging to the dataset as well as some functions for quickly
        visualizing the data as a map, a simple image or an interactive image.

        :param path_or_array:   a numpy.ndarray or a valid file path
        :param geotransform:    GDAL geotransform of the given array or file on disk
        :param projection:      projection of the given array or file on disk as WKT string
                                (only needed if GeoArray is instanced with an array)
        :param bandnames:       names of the bands within the input array, e.g. ['mask_1bit', 'mask_clouds'],
                                (default: ['B1', 'B2', 'B3', ...])
        :param nodata:          nodata value
        :param progress:        show progress bars (default: True)
        :param q:               quiet mode (default: False)
        """

        # TODO implement compatibility to GDAL VRTs
        if not (isinstance(path_or_array, (str, np.ndarray, GeoArray)) or
76
           issubclass(getattr(path_or_array, '__class__'), GeoArray)):
77
            raise ValueError("%s parameter 'arg' takes only string, np.ndarray or GeoArray(and subclass) instances. "
78
                             "Got %s." % (self.__class__.__name__, type(path_or_array)))
79 80

        if path_or_array is None:
81
            raise ValueError("The %s parameter 'path_or_array' must not be None!" % self.__class__.__name__)
82 83 84 85 86

        if isinstance(path_or_array, str):
            assert ' ' not in path_or_array, "The given path contains whitespaces. This is not supported by GDAL."

            if not os.path.exists(path_or_array):
87
                raise FileNotFoundError(path_or_array)
88

89 90
        if isinstance(path_or_array, GeoArray) or issubclass(getattr(path_or_array, '__class__'), GeoArray):
            self.__dict__ = path_or_array.__dict__.copy()
91
            self._initParams = dict([x for x in locals().items() if x[0] != "self"])
92 93 94 95 96 97
            self.geotransform = geotransform or self.geotransform
            self.projection = projection or self.projection
            self.bandnames = bandnames or list(self.bandnames.values())
            self._nodata = nodata if nodata is not None else self._nodata
            self.progress = False if progress is False else self.progress
            self.q = q if q is not None else self.q
98 99

        else:
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114
            self._initParams = dict([x for x in locals().items() if x[0] != "self"])
            self.arg = path_or_array
            self._arr = path_or_array if isinstance(path_or_array, np.ndarray) else None
            self.filePath = path_or_array if isinstance(path_or_array, str) and path_or_array else None
            self.basename = os.path.splitext(os.path.basename(self.filePath))[0] if not self.is_inmem else 'IN_MEM'
            self.progress = progress
            self.q = q
            self._arr_cache = None  # dict containing key 'pos' and 'arr_cached'
            self._geotransform = None
            self._projection = None
            self._shape = None
            self._dtype = None
            self._nodata = nodata
            self._mask_nodata = None
            self._mask_baddata = None
115 116
            self._footprint_poly = None
            self._gdalDataset_meta_already_set = False
117 118
            self._metadata = None
            self._bandnames = None
119 120

            if bandnames:
121
                self.bandnames = bandnames  # use property in order to validate given value
122
            if geotransform:
123
                self.geotransform = geotransform  # use property in order to validate given value
124
            if projection:
125
                self.projection = projection  # use property in order to validate given value
126 127 128 129 130 131 132 133 134 135

            if self.filePath:
                self.set_gdalDataset_meta()

    @property
    def arr(self):
        return self._arr

    @arr.setter
    def arr(self, ndarray):
136 137
        assert isinstance(ndarray, np.ndarray), "'arr' can only be set to a numpy array! Got %s." % type(ndarray)
        # assert ndarray.shape == self.shape, "'arr' can only be set to a numpy array with shape %s. Received %s. " \
138 139 140 141 142
        #                                    "If you need to change the dimensions, create a new instance of %s." \
        #                                    %(self.shape, ndarray.shape, self.__class__.__name__)
        #  THIS would avoid warping like this: geoArr.arr, geoArr.gt, geoArr.prj = warp(...)

        if ndarray.shape != self.shape:
143
            self.flush_cache()  # the cached array is not useful anymore
144 145 146 147 148

        self._arr = ndarray

    @property
    def bandnames(self):
149
        if self._bandnames and len(self._bandnames) == self.bands:
150 151 152 153 154 155 156 157 158 159 160
            return self._bandnames
        else:
            self._bandnames = OrderedDict(('B%s' % band, i) for i, band in enumerate(range(1, self.bands + 1)))
            return self._bandnames

    @bandnames.setter
    def bandnames(self, list_bandnames):
        # type: (list) -> None

        if list_bandnames:
            assert isinstance(list_bandnames, list), "A list must be given when setting the 'bandnames' attribute. " \
161
                                                     "Received %s." % type(list_bandnames)
162 163 164 165 166
            assert len(list_bandnames) == self.bands, \
                'Number of given bandnames does not match number of bands in array.'
            assert len(list(set([type(b) for b in list_bandnames]))) == 1 and type(list_bandnames[0] == 'str'), \
                "'bandnames must be a set of strings. Got other datetypes in there.'"
            bN_dict = OrderedDict((band, i) for i, band in enumerate(list_bandnames))
167 168
            assert len(
                bN_dict) == self.bands, 'Bands must not have the same name. Received band list: %s' % list_bandnames
169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213

            self._bandnames = bN_dict

    @property
    def is_inmem(self):
        """Check if associated image array is completely loaded into memory."""

        return isinstance(self.arr, np.ndarray)

    @property
    def shape(self):
        """Get the array shape of the associated image array."""

        if self.is_inmem:
            return self.arr.shape
        else:
            if self._shape:
                return self._shape
            else:
                self.set_gdalDataset_meta()
                return self._shape

    @property
    def ndim(self):
        """Get the number dimensions of the associated image array."""
        return len(self.shape)

    @property
    def rows(self):
        """Get the number of rows of the associated image array."""

        return self.shape[0]

    @property
    def columns(self):
        """Get the number of columns of the associated image array."""

        return self.shape[1]

    cols = alias_property('columns')

    @property
    def bands(self):
        """Get the number of bands of the associated image array."""

214
        return self.shape[2] if len(self.shape) > 2 else 1
215 216 217 218 219 220 221 222 223 224 225 226 227 228 229

    @property
    def dtype(self):
        """Get the numpy data type of the associated image array."""

        if self._dtype:
            return self._dtype
        elif self.is_inmem:
            return self.arr.dtype
        else:
            self.set_gdalDataset_meta()
            return self._dtype

    @property
    def geotransform(self):
230
        """Get the GDAL GeoTransform of the associated image, e.g., (283500.0, 5.0, 0.0, 4464500.0, 0.0, -5.0)"""
231 232 233 234 235 236 237

        if self._geotransform:
            return self._geotransform
        elif not self.is_inmem:
            self.set_gdalDataset_meta()
            return self._geotransform
        else:
238
            return [0, 1, 0, 0, 0, -1]
239 240 241

    @geotransform.setter
    def geotransform(self, gt):
242 243
        assert isinstance(gt, (list, tuple)) and len(gt) == 6,\
            'geotransform must be a list with 6 numbers. Got %s.' % str(gt)
244

245
        for i in gt:
246
            assert is_number(i), "geotransform must contain only numbers. Got '%s' (type: %s)." % (i, type(i))
247

248
        self._geotransform = gt
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270

    gt = alias_property('geotransform')

    @property
    def xgsd(self):
        """Get the X resolution in units of the given or detected projection."""

        return self.geotransform[1]

    @property
    def ygsd(self):
        """Get the Y resolution in units of the given or detected projection."""

        return abs(self.geotransform[5])

    @property
    def xygrid_specs(self):
        """
        Get the specifications for the X/Y coordinate grid, e.g. [[15,30], [0,30]] for a coordinate with its origin
        at X/Y[15,0] and a GSD of X/Y[15,30].
        """

271
        def get_grid(gt, xgsd, ygsd): return [[gt[0], gt[0] + xgsd], [gt[3], gt[3] - ygsd]]
272 273 274 275 276 277 278 279 280 281 282 283 284
        return get_grid(self.geotransform, self.xgsd, self.ygsd)

    @property
    def projection(self):
        """
        Get the projection of the associated image. Setting the projection is only allowed if GeoArray has been
        instanced from memory or the associated file on disk has no projection.
        """

        if self._projection:
            return self._projection
        elif not self.is_inmem:
            self.set_gdalDataset_meta()
285
            return self._projection  # or "LOCAL_CS[\"MAP\"]"
286
        else:
287
            return ''  # '"LOCAL_CS[\"MAP\"]"
288 289 290 291

    @projection.setter
    def projection(self, prj):
        if self.filePath:
292
            assert self.projection is None or prj_equal(self.projection, prj), \
293
                "Cannot set %s.projection to the given value because it does not match the projection from the file " \
294
                "on disk." % self.__class__.__name__
295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329
        else:
            self._projection = prj

    prj = alias_property('projection')

    @property
    def epsg(self):
        """Get the EPSG code of the projection of the GeoArray."""

        return WKT2EPSG(self.projection)

    @epsg.setter
    def epsg(self, epsg_code):
        self.projection = EPSG2WKT(epsg_code)

    @property
    def box(self):
        mapPoly = get_footprint_polygon(get_corner_coordinates(gt=self.geotransform, cols=self.columns, rows=self.rows))
        return boxObj(gt=self.geotransform, prj=self.projection, mapPoly=mapPoly)

    @property
    def nodata(self):
        """
        Get the nodata value of the GeoArray. If GeoArray has been instanced with a file path the file is checked
        for an existing nodata value. Otherwise (if no value is exlicitly given during object instanciation) the nodata
        value is tried to be automatically detected.
        """

        if self._nodata is not None:
            return self._nodata
        else:
            # try to get nodata value from file
            if not self.is_inmem:
                self.set_gdalDataset_meta()
            if self._nodata is None:
Daniel Scheffler's avatar
Bugfix  
Daniel Scheffler committed
330
                self._nodata = self.find_noDataVal()
331 332 333 334 335 336
                if self._nodata == 'ambiguous':
                    warnings.warn('Nodata value could not be clearly identified. It has been set to None.')
                    self._nodata = None
                else:
                    if self._nodata is not None and not self.q:
                        print("Automatically detected nodata value for %s '%s': %s"
337
                              % (self.__class__.__name__, self.basename, self._nodata))
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352
            return self._nodata

    @nodata.setter
    def nodata(self, value):
        self._nodata = value

    @property
    def mask_nodata(self):
        """
        Get the nodata mask of the associated image array. It is calculated using all image bands.
        """

        if self._mask_nodata is not None:
            return self._mask_nodata
        else:
353
            self.calc_mask_nodata()  # sets self._mask_nodata
354 355 356 357 358 359 360 361 362 363 364
            return self._mask_nodata

    @mask_nodata.setter
    def mask_nodata(self, mask):
        """Set bad data mask.

        :param mask:    Can be a file path, a numpy array or an instance o GeoArray.
        """

        if mask is not None:
            from .masks import NoDataMask
365 366
            geoArr_mask = NoDataMask(mask, progress=self.progress, q=self.q)
            geoArr_mask.gt = geoArr_mask.gt if geoArr_mask.gt not in [None, [0, 1, 0, 0, 0, -1]] else self.gt
367
            geoArr_mask.prj = geoArr_mask.prj if geoArr_mask.prj else self.prj
368
            imName = "the %s '%s'" % (self.__class__.__name__, self.basename)
369 370 371 372

            assert geoArr_mask.bands == 1, \
                'Expected one single band as nodata mask for %s. Got %s bands.' % (self.basename, geoArr_mask.bands)
            assert geoArr_mask.shape[:2] == self.shape[:2], 'The provided nodata mask must have the same number of ' \
373
                                                            'rows and columns as the %s itself.' % imName
374 375
            assert geoArr_mask.gt == self.gt, \
                'The geotransform of the given nodata mask for %s must match the geotransform of the %s itself. ' \
376
                'Got %s.' % (imName, self.__class__.__name__, geoArr_mask.gt)
377 378
            assert not geoArr_mask.prj or prj_equal(geoArr_mask.prj, self.prj), \
                'The projection of the given nodata mask for the %s must match the projection of the %s itself.' \
379
                % (imName, self.__class__.__name__)
380 381

            self._mask_nodata = geoArr_mask
382 383 384 385 386 387
        else:
            del self.mask_nodata

    @mask_nodata.deleter
    def mask_nodata(self):
        self._mask_nodata = None
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406

    @property
    def mask_baddata(self):
        """
        Returns the bad data mask for the associated image array if it has been explicitly previously. It can be set
         by passing a file path, a numpy array or an instance of GeoArray to the setter of this property.
        """

        return self._mask_baddata

    @mask_baddata.setter
    def mask_baddata(self, mask):
        """Set bad data mask.

        :param mask:    Can be a file path, a numpy array or an instance o GeoArray.
        """

        if mask is not None:
            from .masks import BadDataMask
407 408
            geoArr_mask = BadDataMask(mask, progress=self.progress, q=self.q)
            geoArr_mask.gt = geoArr_mask.gt if geoArr_mask.gt not in [None, [0, 1, 0, 0, 0, -1]] else self.gt
409
            geoArr_mask.prj = geoArr_mask.prj if geoArr_mask.prj else self.prj
410
            imName = "the %s '%s'" % (self.__class__.__name__, self.basename)
411 412 413 414

            assert geoArr_mask.bands == 1, \
                'Expected one single band as bad data mask for %s. Got %s bands.' % (self.basename, geoArr_mask.bands)
            assert geoArr_mask.shape[:2] == self.shape[:2], 'The provided bad data mask must have the same number of ' \
415
                                                            'rows and columns as the %s itself.' % imName
416 417
            assert geoArr_mask.gt == self.gt, \
                'The geotransform of the given bad data mask for %s must match the geotransform of the %s itself. ' \
418
                'Got %s.' % (imName, self.__class__.__name__, geoArr_mask.gt)
419 420
            assert prj_equal(geoArr_mask.prj, self.prj), \
                'The projection of the given bad data mask for the %s must match the projection of the %s itself.' \
421
                % (imName, self.__class__.__name__)
422 423

            self._mask_baddata = geoArr_mask
424 425 426 427 428 429
        else:
            del self.mask_baddata

    @mask_baddata.deleter
    def mask_baddata(self):
        self._mask_baddata = None
430 431 432 433 434 435 436 437 438 439

    @property
    def footprint_poly(self):
        # FIXME should return polygon in image coordinates if no projection is available
        """
        Get the footprint polygon of the associated image array (returns an instance of shapely.geometry.Polygon.
        """

        if self._footprint_poly is None:
            assert self.mask_nodata is not None, 'A nodata mask is needed for calculating the footprint polygon. '
440
            if False in self.mask_nodata[:]:
441 442 443 444
                # do not run raster2polygon if whole image is filled with data
                self._footprint_poly = self.box.mapPoly
            else:
                try:
445 446
                    multipolygon = raster2polygon(self.mask_nodata.astype(np.uint8), self.gt, self.prj, exact=False,
                                                  progress=self.progress, q=self.q, maxfeatCount=10, timeout=3)
447
                    self._footprint_poly = fill_holes_within_poly(multipolygon)
448
                except (RuntimeError, TimeoutError):
449 450 451 452
                    if not self.q:
                        warnings.warn("\nCalculation of footprint polygon failed for %s '%s'. Using outer bounds. One "
                                      "reason could be that the nodata value appears within the actual image (not only "
                                      "as fill value). To avoid this use another nodata value. Current nodata value is "
453
                                      "%s." % (self.__class__.__name__, self.basename, self.nodata))
454 455 456
                    self._footprint_poly = self.box.mapPoly

            # validation
457 458 459 460
            assert not polyVertices_outside_poly(self._footprint_poly, self.box.mapPoly), \
                "Computing footprint polygon for %s '%s' failed. The resulting polygon is partly or completely " \
                "outside of the image bounds." % (self.__class__.__name__, self.basename)
            # assert self._footprint_poly
461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
            # for XY in self.corner_coord:
            #    assert self.GeoArray.box.mapPoly.contains(Point(XY)) or self.GeoArray.box.mapPoly.touches(Point(XY)), \
            #        "The corner position '%s' is outside of the %s." % (XY, self.imName)

        return self._footprint_poly

    @footprint_poly.setter
    def footprint_poly(self, poly):
        if isinstance(poly, Polygon):
            self._footprint_poly = poly
        elif isinstance(poly, str):
            self._footprint_poly = shply_loads(poly)
        else:
            raise ValueError("'footprint_poly' can only be set from a shapely polygon or a WKT string.")

    @property
    def metadata(self):
        """
        Returns a GeoDataFrame containing all available metadata (read from file if available).
        Use 'metadata[band_index].to_dict()' to get a metadata dictionary for a specific band.
        Use 'metadata.loc[row_name].to_dict()' to get all metadata values of the same key for all bands as dictionary.
        Use 'metadata.loc[row_name, band_index] = value' to set a new value.

        :return:  geopandas.GeoDataFrame
        """

        if self._metadata is not None:
            return self._metadata
        else:
            default = GeoDataFrame(columns=range(self.bands))
491
            # for bn,idx in self.bandnames.items():
492 493 494 495 496 497 498 499 500 501
            #    default.loc['band_index',bn] = idx
            self._metadata = default
            if not self.is_inmem:
                self.set_gdalDataset_meta()
                return self._metadata
            else:
                return self._metadata

    @metadata.setter
    def metadata(self, GDF):
502
        assert isinstance(GDF, (GeoDataFrame, DataFrame)) and len(GDF.columns) == self.bands, \
503
            "%s.metadata can only be set with an instance of geopandas.GeoDataFrame of which the column number " \
504
            "corresponds to the band number of %s." % (self.__class__.__name__, self.__class__.__name__)
505 506 507 508 509
        self._metadata = GDF

    meta = alias_property('metadata')

    def __getitem__(self, given):
510
        if isinstance(given, (int, float, slice)) and self.ndim == 3:
511 512 513 514 515 516 517 518 519 520 521 522 523
            # handle 'given' as index for 3rd (bands) dimension
            if self.is_inmem:
                return self.arr[:, :, given]
            else:
                return self.from_path(self.arg, [given])

        elif isinstance(given, str):
            # behave like a dictionary and return the corresponding band
            if self.bandnames:
                if given not in self.bandnames:
                    raise ValueError("'%s' is not a known band. Known bands are: %s"
                                     % (given, ', '.join(list(self.bandnames.keys()))))
                if self.is_inmem:
524
                    return self.arr if self.ndim == 2 else self.arr[:, :, self.bandnames[given]]
525 526 527 528
                else:
                    return self.from_path(self.arg, [self.bandnames[given]])
            else:
                raise ValueError('String indices are only supported if %s has been instanced with bandnames given.'
529
                                 % self.__class__.__name__)
530 531 532 533 534 535 536

        elif isinstance(given, (tuple, list)):
            # handle requests like geoArr[[1,2],[3,4]  -> not implemented in from_path if array is not in mem
            types = [type(i) for i in given]
            if list in types or tuple in types:
                self.to_mem()

537
            if len(given) == 3:
538 539

                # handle strings in the 3rd dim of 'given' -> convert them to a band index
540
                if isinstance(given[2], str):
541 542 543 544 545 546 547 548
                    if self.bandnames:
                        if given[2] not in self.bandnames:
                            raise ValueError("'%s' is not a known band. Known bands are: %s"
                                             % (given[2], ', '.join(list(self.bandnames.keys()))))

                        band_idx = self.bandnames[given[2]]
                        # NOTE: the string in the 3rd is ignored if ndim==2 and band_idx==0
                        if self.is_inmem:
549
                            return self.arr if (self.ndim == 2 and band_idx == 0) else self.arr[:, :, band_idx]
550
                        else:
551 552
                            getitem_params = \
                                given[:2] if (self.ndim == 2 and band_idx == 0) else given[:2] + (band_idx,)
553 554 555 556 557 558 559
                            return self.from_path(self.arg, getitem_params)
                    else:
                        raise ValueError(
                            'String indices are only supported if %s has been instanced with bandnames given.'
                            % self.__class__.__name__)

                # in case a third dim is requested from 2D-array -> ignore 3rd dim if 3rd dim is 0
560
                elif self.ndim == 2 and given[2] == 0:
561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584
                    if self.is_inmem:
                        return self.arr[given[:2]]
                    else:
                        return self.from_path(self.arg, given[:2])

        # if nothing has been returned until here -> behave like a numpy array
        if self.is_inmem:
            return self.arr[given]
        else:
            getitem_params = [given] if isinstance(given, slice) else given
            return self.from_path(self.arg, getitem_params)

    def __setitem__(self, idx, array2set):
        """Overwrites the pixel values of GeoArray.arr with the given array.

        :param idx:         <int, list, slice> the index position to overwrite
        :param array2set:   <np.ndarray> array to be set. Must be compatible to the given index position.
        :return:
        """

        if self.is_inmem:
            self.arr[idx] = array2set
        else:
            raise NotImplementedError('Item assignment for %s instances that are not in memory is not yet supported.'
585
                                      % self.__class__.__name__)
586 587 588

    def __getattr__(self, attr):
        # check if the requested attribute can not be present because GeoArray has been instanced with an array
589 590
        if attr not in self.__dir__() and not self.is_inmem and attr in ['shape', 'dtype', 'geotransform',
                                                                         'projection']:
591 592
            self.set_gdalDataset_meta()

593 594 595
        if attr in self.__dir__():  # __dir__() includes also methods and properties
            return self.__getattribute__(attr)  # __getattribute__ avoids infinite loop
        elif hasattr(np.array([]), attr):
596 597
            return self[:].__getattribute__(attr)
        else:
598
            raise AttributeError("%s object has no attribute '%s'." % (self.__class__.__name__, attr))
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625

    def __getstate__(self):
        """Defines how the attributes of GMS object are pickled."""

        # clean array cache in order to avoid cache pickling
        self.flush_cache()

        return self.__dict__

    def __setstate__(self, state):
        """Defines how the attributes of GMS object are unpickled.
        NOTE: This method has been implemented because otherwise pickled and unpickled instances show recursion errors
        within __getattr__ when requesting any attribute.
        """

        self.__dict__ = state

    def calc_mask_nodata(self, fromBand=None, overwrite=False):
        """Calculates a no data mask with (values: 0=nodata; 1=data)

        :param fromBand:  <int> index of the band to be used (if None, all bands are used)
        :param overwrite: <bool> whether to overwrite existing nodata mask that has already been calculated
        :return:
        """

        if self._mask_nodata is None or overwrite:
            assert self.ndim in [2, 3], "Only 2D or 3D arrays are supported. Got a %sD array." % self.ndim
626
            arr = self[:, :, fromBand] if self.ndim == 3 and fromBand is not None else self[:]
627 628 629

            if self.nodata is None:
                self.mask_nodata = np.ones((self.rows, self.cols), np.bool)
630 631 632 633
            elif np.isnan(self.nodata):
                self.mask_nodata = \
                    np.invert(np.isnan(arr)) if arr.ndim == 2 else \
                    np.all(np.invert(np.isnan(arr)), axis=2)
634
            else:
635
                self.mask_nodata = \
636 637
                    np.ma.masked_not_equal(arr, self.nodata).mask if arr.ndim == 2 else \
                    np.all(np.ma.masked_not_equal(arr, self.nodata).mask, axis=2)
638

639 640 641 642 643 644 645 646
    def find_noDataVal(self, bandIdx=0, sz=3):
        """Tries to derive no data value from homogenious corner pixels within 3x3 windows (by default).
        :param bandIdx:
        :param sz: window size in which corner pixels are analysed
        """
        wins = [self[0:sz, 0:sz, bandIdx], self[0:sz, -sz:, bandIdx],
                self[-sz:, -sz:, bandIdx], self[-sz:, 0:sz, bandIdx]]  # UL, UR, LR, LL

647 648
        means, stds = [np.mean(win) for win in wins], [np.std(win) for win in wins]
        possVals = [mean for mean, std in zip(means, stds) if std == 0 or np.isnan(std)]
649 650 651 652
        # possVals==[]: all corners are filled with data; np.std(possVals)==0: noDataVal clearly identified

        if possVals:
            if np.std(possVals) != 0:
653 654 655 656 657 658
                if np.isnan(np.std(possVals)):
                    # at least one of the possible values is np.nan
                    nodata = np.nan
                else:
                    # different possible nodata values have been found in the image corner
                    nodata = 'ambiguous'
659 660 661 662 663
            else:
                if len(possVals) <= 2:
                    # each window in each corner
                    warnings.warn("\nAutomatic nodata value detection returned the value %s for GeoArray '%s' but this "
                                  "seems to be unreliable (occurs in only %s). To avoid automatic detection, just pass "
664 665 666
                                  "the correct nodata value."
                                  % (possVals[0], self.basename, ('2 image corners' if len(possVals) == 2 else
                                                                  '1 image corner')))
667
                nodata = possVals[0]
668
        else:
669 670 671
            nodata = None

        return nodata
672

673 674 675 676 677 678 679 680 681 682
    def set_gdalDataset_meta(self):
        """Retrieves GDAL metadata from file. This function is only executed once to avoid overwriting of user defined
         attributes, that are defined after object instanciation.

        :return:
        """

        if not self._gdalDataset_meta_already_set:
            assert self.filePath
            ds = gdal.Open(self.filePath)
683 684 685
            if not ds:
                raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())

686
            # set private class variables (in order to avoid recursion error)
687 688
            self._shape = tuple([ds.RasterYSize, ds.RasterXSize] + ([ds.RasterCount] if ds.RasterCount > 1 else []))
            self._dtype = gdal_array.GDALTypeCodeToNumericTypeCode(ds.GetRasterBand(1).DataType)
689
            self._geotransform = list(ds.GetGeoTransform())
690 691

            # for some reason GDAL reads arbitrary geotransforms as (0, 1, 0, 0, 0, 1) instead of (0, 1, 0, 0, 0, -1)
692
            self._geotransform[5] = -abs(self._geotransform[5])  # => force ygsd to be negative
693

694 695
            # temp conversion to EPSG needed because GDAL seems to modify WKT string when writing file to disk
            # (e.g. using gdal_merge) -> conversion to EPSG and back undos that
696 697 698 699
            try:
                self._projection = EPSG2WKT(WKT2EPSG(ds.GetProjection()))
            except:
                self._projection = None
700

701 702 703 704
            if 'nodata' not in self._initParams or self._initParams['nodata'] is None:
                band = ds.GetRasterBand(1)
                # FIXME this does not support different nodata values within the same file
                self._nodata = band.GetNoDataValue()
705

706
            # read global domain metadata
707 708
            # TODO check to specifically use the 'ENVI' metadata domain ds.GetMetadata('ENVI')
            global_meta = ds.GetMetadata()
709

710 711
            # read band domain metadata
            for b in range(self.bands):
712 713
                band = ds.GetRasterBand(b + 1)
                meta_gs = GeoSeries(band.GetMetadata())
714

715 716 717 718 719 720
                # add band names if available
                if 'Band_%s' % str(b + 1) in global_meta.keys():
                    meta_gs['band_name'] = global_meta['Band_%s' % str(b + 1)]

                # TODO add the remaining global metadata

721 722 723 724 725
                # avoid double-call of set_gdalDataset_meta by setting self._metadata to default value
                self._metadata = \
                    self._metadata if self._metadata is not None else GeoDataFrame(columns=range(self.bands))

                # fill metadata
726
                self.metadata[b] = meta_gs
727
                del band
728

729
            del ds
730 731 732 733 734 735 736 737 738 739 740 741 742 743

        self._gdalDataset_meta_already_set = True

    def from_path(self, path, getitem_params=None):
        # type: (str, list) -> np.ndarray
        """Read a GDAL compatible raster image from disk, with respect to the given image position.
        NOTE: If the requested array position is already in cache, it is returned from there.

        :param path:            <str> the file path of the image to read
        :param getitem_params:  <list> a list of slices in the form [row_slice, col_slice, band_slice]
        :return out_arr:        <np.ndarray> the output array
        """

        ds = gdal.Open(path)
744 745 746
        if not ds:
            raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())

747
        R, C, B = ds.RasterYSize, ds.RasterXSize, ds.RasterCount
748
        del ds
749

750
        # convert getitem_params to subset area to be read ##
751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776
        rS, rE, cS, cE, bS, bE, bL = [None] * 7

        # populate rS, rE, cS, cE, bS, bE, bL
        if getitem_params:
            if len(getitem_params) >= 2:
                givenR, givenC = getitem_params[:2]
                if isinstance(givenR, slice):
                    rS = givenR.start
                    rE = givenR.stop - 1 if givenR.stop is not None else None
                elif isinstance(givenR, int):
                    rS = givenR
                    rE = givenR
                if isinstance(givenC, slice):
                    cS = givenC.start
                    cE = givenC.stop - 1 if givenC.stop is not None else None
                elif isinstance(givenC, int):
                    cS = givenC
                    cE = givenC
            if len(getitem_params) in [1, 3]:
                givenB = getitem_params[2] if len(getitem_params) == 3 else getitem_params[0]
                if isinstance(givenB, slice):
                    bS = givenB.start
                    bE = givenB.stop - 1 if givenB.stop is not None else None
                elif isinstance(givenB, int):
                    bS = givenB
                    bE = givenB
777
                elif isinstance(givenB, (tuple, list)):
778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
                    typesInGivenB = [type(i) for i in givenB]
                    assert len(list(set(typesInGivenB))) == 1, \
                        'Mixed data types within the list of bands are not supported.'
                    if isinstance(givenB[0], int):
                        bL = list(givenB)
                    elif isinstance(givenB[0], str):
                        bL = [self.bandnames[i] for i in givenB]
                elif type(givenB) in [str]:
                    bL = [self.bandnames[givenB]]

        # set defaults for not given values
        rS = rS if rS is not None else 0
        rE = rE if rE is not None else R - 1
        cS = cS if cS is not None else 0
        cE = cE if cE is not None else C - 1
        bS = bS if bS is not None else 0
        bE = bE if bE is not None else B - 1
        bL = list(range(bS, bE + 1)) if not bL else bL

        # convert negative to positive ones
        rS = rS if rS >= 0 else self.rows + rS
        rE = rE if rE >= 0 else self.rows + rE
        cS = cS if cS >= 0 else self.columns + cS
        cE = cE if cE >= 0 else self.columns + cE
        bS = bS if bS >= 0 else self.bands + bS
        bE = bE if bE >= 0 else self.bands + bE
804
        bL = [b if b >= 0 else (self.bands + b) for b in bL]
805 806

        # validate subset area bounds to be read
807 808 809 810 811 812 813
        def msg(v, idx, sz):
            # FIXME numpy raises that error ONLY for the 2nd axis
            return '%s is out of bounds for axis %s with size %s' % (v, idx, sz)

        for val, axIdx, axSize in zip([rS, rE, cS, cE, bS, bE], [0, 0, 1, 1, 2, 2], [R, R, C, C, B, B]):
            if not 0 <= val <= axSize - 1:
                raise ValueError(msg(val, axIdx, axSize))
814 815

        # summarize requested array position in arr_pos
816
        # NOTE: # bandlist must be string because truth value of an array with more than one element is ambiguous
817 818 819
        arr_pos = dict(rS=rS, rE=rE, cS=cS, cE=cE, bS=bS, bE=bE, bL=bL)

        # check if the requested array position is already in cache -> if yes, return it from there
820
        if self._arr_cache is not None and self._arr_cache['pos'] == arr_pos:
821
            out_arr = self._arr_cache['arr_cached']
822 823 824 825 826 827 828

        else:
            # TODO insert a multiprocessing.Lock here in order to prevent IO bottlenecks?
            # read subset area from disk
            if bL == list(range(0, B)):
                tempArr = gdalnumeric.LoadFile(path, cS, rS, cE - cS + 1, rE - rS + 1)
                out_arr = np.swapaxes(np.swapaxes(tempArr, 0, 2), 0, 1) if B > 1 else tempArr
829 830
                if out_arr is None:
                    raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
831 832 833 834
            else:
                ds = gdal.Open(path)
                if len(bL) == 1:
                    band = ds.GetRasterBand(bL[0] + 1)
835
                    out_arr = band.ReadAsArray(cS, rS, cE - cS + 1, rE - rS + 1)
836 837
                    if out_arr is None:
                        raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
838
                    del band
839 840 841 842 843
                else:
                    out_arr = np.empty((rE - rS + 1, cE - cS + 1, len(bL)))
                    for i, bIdx in enumerate(bL):
                        band = ds.GetRasterBand(bIdx + 1)
                        out_arr[:, :, i] = band.ReadAsArray(cS, rS, cE - cS + 1, rE - rS + 1)
844 845
                        if out_arr is None:
                            raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())
846
                        del band
847

848
                del ds
849 850

            # only set self.arr if the whole cube has been read (in order to avoid sudden shape changes)
851
            if out_arr.shape == self.shape:
852 853 854 855 856
                self.arr = out_arr

            # write _arr_cache
            self._arr_cache = dict(pos=arr_pos, arr_cached=out_arr)

857 858
        return out_arr  # TODO implement check of returned datatype (e.g. NoDataMask should always return np.bool
        # TODO -> would be np.int8 if an int8 file is read from disk
859 860 861 862 863 864 865 866

    def save(self, out_path, fmt='ENVI', creationOptions=None):
        # type: (str, str, list) -> None
        """Write the raster data to disk.

        :param out_path:        <str> output path
        :param fmt:             <str> the output format / GDAL driver code to be used for output creation, e.g. 'ENVI'
                                Refer to http://www.gdal.org/formats_list.html to get a full list of supported formats.
867 868
        :param creationOptions: <list> GDAL creation options,
                                e.g., ["QUALITY=80", "REVERSIBLE=YES", "WRITE_METADATA=YES"]
869 870 871
        """

        if not self.q:
872 873
            print('Writing GeoArray of size %s to %s.' % (self.shape, out_path))
        assert self.ndim in [2, 3], 'Only 2D- or 3D arrays are supported.'
874 875 876 877 878 879 880 881 882 883

        driver = gdal.GetDriverByName(fmt)
        if driver is None:
            raise Exception("'%s' is not a supported GDAL driver. Refer to www.gdal.org/formats_list.html for full "
                            "list of GDAL driver codes." % fmt)

        if not os.path.isdir(os.path.dirname(out_path)):
            os.makedirs(os.path.dirname(out_path))

        if self.is_inmem:
884 885
            ds = get_GDAL_ds_inmem(self.arr, self.geotransform, self.projection,
                                   self.nodata)  # expects rows,columns,bands
886 887 888

            # set metadata
            if not self.metadata.empty:
889 890 891
                global_meta = {}

                # set band domain metadata
892
                for bidx in range(self.bands):
893
                    band = ds.GetRasterBand(bidx + 1)
894
                    meta2write = self.metadata[bidx].to_dict()
895
                    meta2write = dict((k, v) for k, v in meta2write.items() if v is not np.nan)
896 897

                    if 'band_name' in meta2write:
898
                        global_meta['Band_%s' % str(bidx + 1)] = meta2write['band_name']
899 900
                        del meta2write['band_name']

901
                    band.SetMetadata(meta2write)
902
                    del band
903

904 905 906
                # set global domain metadata
                ds.SetMetadata(global_meta)

907 908 909 910 911
                # get ENVI metadata domain
                # ds_orig = gdal.Open(self.filePath)
                # envi_meta_domain = ds_orig.GetMetadata('ENVI')
                # ds.SetMetadata(envi_meta_domain, 'ENVI')
                # ds_orig = None
912

913 914
            driver.CreateCopy(out_path, ds, options=creationOptions if creationOptions else [])

915 916 917 918
            # rows, columns, bands => bands, rows, columns
            # out_arr = self.arr if self.ndim == 2 else np.swapaxes(np.swapaxes(self.arr, 0, 2), 1, 2)
            # gdalnumeric.SaveArray(out_arr, out_path, format=fmt, prototype=ds) # expects bands,rows,columns
            del ds
919 920 921

        else:
            src_ds = gdal.Open(self.filePath)
922 923 924
            if not src_ds:
                raise Exception('Error reading file:  ' + gdal.GetLastErrorMsg())

925 926
            gdal_Translate = get_gdal_func('Translate')
            gdal_Translate(out_path, src_ds, format=fmt, creationOptions=creationOptions)
927
            del src_ds
928 929 930 931 932 933 934 935 936

        if not os.path.exists(out_path):
            raise Exception(gdal.GetLastErrorMsg())

    def dump(self, out_path):
        # type: (str) -> None
        """Serialize the whole object instance to disk using dill."""

        import dill
937 938
        with open(out_path, 'wb') as outF:
            dill.dump(self, outF)
939 940 941 942 943 944 945 946 947

    def _get_plottable_image(self, xlim=None, ylim=None, band=None, boundsMap=None, boundsMapPrj=None, res_factor=None,
                             nodataVal=None, out_prj=None):
        # handle limits
        if boundsMap:
            boundsMapPrj = boundsMapPrj if boundsMapPrj else self.prj
            image2plot, gt, prj = self.get_mapPos(boundsMap, boundsMapPrj, band2get=band,
                                                  fillVal=nodataVal if nodataVal is not None else self.nodata)
        else:
948 949
            cS, cE = xlim if isinstance(xlim, (tuple, list)) else (0, self.columns)
            rS, rE = ylim if isinstance(ylim, (tuple, list)) else (0, self.rows)
950 951

            image2plot = self[rS:rE, cS:cE, band] if band is not None else self[rS:rE, cS:cE]
952
            gt, prj = self.geotransform, self.projection
953

954
        transOpt = ['SRC_METHOD=NO_GEOTRANSFORM'] if tuple(gt) == (0, 1, 0, 0, 0, -1) else None
955
        xdim, ydim = None, None
956
        nodataVal = nodataVal if nodataVal is not None else self.nodata
957 958

        if res_factor != 1. and image2plot.shape[0] * image2plot.shape[1] > 1e6:  # shape > 1000*1000
959 960 961 962
            # sample image down / normalize
            xdim, ydim = \
                (self.columns * res_factor, self.rows * res_factor) if res_factor else \
                tuple((np.array([self.columns, self.rows]) / (np.array([self.columns, self.rows]).max() / 1000)))
963 964 965
            xdim, ydim = int(xdim), int(ydim)

        if xdim or ydim or out_prj:
966
            from py_tools_ds.geo.raster.reproject import warp_ndarray
967 968 969 970 971
            image2plot, gt, prj = warp_ndarray(image2plot, self.geotransform, self.projection,
                                               out_XYdims=(xdim, ydim), in_nodata=nodataVal, out_nodata=nodataVal,
                                               transformerOptions=transOpt, out_prj=out_prj, q=True)
            if transOpt and 'NO_GEOTRANSFORM' in ','.join(transOpt):
                image2plot = np.flipud(image2plot)
972 973
                gt = list(gt)
                gt[3] = 0
974 975 976 977 978 979 980

            if xdim or ydim:
                print('Note: array has been downsampled to %s x %s for faster visualization.' % (xdim, ydim))

        return image2plot, gt, prj

    def show(self, xlim=None, ylim=None, band=None, boundsMap=None, boundsMapPrj=None, figsize=None,
981 982
             interpolation='none', vmin=None, vmax=None, pmin=2, pmax=98, cmap=None, nodataVal=None,
             res_factor=None, interactive=False):
983 984 985 986 987 988 989 990 991 992
        """Plots the desired array position into a figure.

        :param xlim:            [start_column, end_column]
        :param ylim:            [start_row, end_row]
        :param band:            the band index of the band to be plotted (if None and interactive==True all bands are
                                shown, otherwise the first band is chosen)
        :param boundsMap:       xmin, ymin, xmax, ymax
        :param boundsMapPrj:
        :param figsize:
        :param interpolation:
993 994 995 996
        :param vmin:            darkest pixel value to be included in stretching
        :param vmax:            brightest pixel value to be included in stretching
        :param pmin:            percentage to be used for excluding the darkest pixels from stretching (default: 2)
        :param pmax:            percentage to be used for excluding the brightest pixels from stretching (default: 98)
997 998
        :param cmap:
        :param nodataVal:
Daniel Scheffler's avatar
Daniel Scheffler committed
999 1000
        :param res_factor:      <float> resolution factor for downsampling of the image to be plotted in order to save
                                plotting time and memory (default=None -> downsampling is performed to 1000x1000)
1001 1002 1003 1004 1005 1006 1007 1008
        :param interactive:     <bool> activates interactive plotting based on 'holoviews' library.
                                NOTE: this deactivates the magic '% matplotlib inline' in Jupyter Notebook
        :return:
        """

        band = (band if band is not None else 0) if not interactive else band

        # get image to plot
1009
        nodataVal = nodataVal if nodataVal is not None else self.nodata
Daniel Scheffler's avatar
Daniel Scheffler committed
1010 1011 1012
        image2plot, gt, prj = \
            self._get_plottable_image(xlim, ylim, band, boundsMap=boundsMap, boundsMapPrj=boundsMapPrj,
                                      res_factor=res_factor, nodataVal=nodataVal)
1013 1014

        # set color palette
1015 1016
        palette = cmap if cmap else plt.cm.gray
        if nodataVal is not None and np.std(image2plot) != 0:  # do not show nodata
1017
            image2plot = np.ma.masked_equal(image2plot, nodataVal)
1018
            vmin_auto, vmax_auto = \
1019
                np.nanpercentile(image2plot.compressed(), pmin), np.nanpercentile(image2plot.compressed(), pmax)
1020 1021
            palette.set_bad('aqua', 0)
        else:
1022
            vmin_auto, vmax_auto = np.nanpercentile(image2plot, pmin), np.nanpercentile(image2plot, pmax)
1023 1024 1025 1026

        vmin = vmin if vmin is not None else vmin_auto
        vmax = vmax if vmax is not None else vmax_auto

1027
        palette.set_over('1')
1028 1029
        palette.set_under('0')

1030 1031 1032 1033 1034 1035
        # check availability of holoviews
        if not util.find_spec('holoviews'):
            warnings.warn("Interactive mode requires holoviews. Install it by running, e.g., "
                          "'conda install --yes -c ioam bokeh holoviews'. Using non-interactive mode.")
            interactive = False

1036
        if interactive and image2plot.ndim == 3:
1037 1038 1039 1040 1041 1042 1043
            import holoviews as hv
            from skimage.exposure import rescale_intensity
            hv.notebook_extension('matplotlib')

            cS, cE = xlim if isinstance(xlim, (tuple, list)) else (0, self.columns - 1)
            rS, rE = ylim if isinstance(ylim, (tuple, list)) else (0, self.rows - 1)

1044
            image2plot = np.array(rescale_intensity(image2plot, in_range=(vmin, vmax)))
1045

1046 1047 1048 1049 1050 1051 1052
            def get_hv_image(b):
                # FIXME ylabels have the wrong order
                return hv.Image(image2plot[:, :, b] if b is not None else image2plot,
                                bounds=(cS, rS, cE, rE))(
                    style={'cmap': 'gray'}, plot={'fig_inches': 4 if figsize is None else figsize, 'show_grid': True})

            # hvIm = hv.Image(image2plot)(style={'cmap': 'gray'}, figure_inches=figsize)
1053 1054 1055 1056 1057 1058 1059
            hmap = hv.HoloMap([(band, get_hv_image(band)) for band in range(image2plot.shape[2])], kdims=['band'])

            return hmap

        else:
            if interactive:
                warnings.warn('Currently there is no interactive mode for single-band arrays. '
1060
                              'Switching to standard matplotlib figure..')  # TODO implement zoomable fig
1061 1062 1063 1064 1065

            # show image
            plt.figure(figsize=figsize)
            rows, cols = image2plot.shape[:2]
            plt.imshow(image2plot, palette, interpolation=interpolation, extent=(0, cols, rows, 0),
1066
                       vmin=vmin, vmax=vmax, )  # compressed excludes nodata values
1067 1068 1069 1070 1071 1072