asf_tools v0.4.3 API Reference

Tools developed by ASF for working with SAR data

composite

Create a local-resolution-weighted composite from Sentinel-1 RTC products.

Create a local-resolution-weighted composite from a set of Sentinel-1 RTC products (D. Small, 2012). The local resolution, defined as the inverse of the local contributing (scattering) area, is used to weight each RTC products' contributions to the composite image on a pixel-by-pixel basis. The composite image is created as a Cloud Optimized GeoTIFF (COG). Additionally, a COG specifying the number of rasters contributing to each composite pixel is created.

References

David Small, 2012: https://doi.org/10.1109/IGARSS.2012.6350465

epsg_to_wkt(epsg_code)

Get the WKT representation of a projection from its EPSG code

Parameters:

Name	Type	Description	Default
epsg_code	int	The integer EPSG code	required

Returns:

Name	Type	Description
wkt	str	The WKT representation of the projection

Source code in asf_tools/composite.py

def epsg_to_wkt(epsg_code: int) -> str:
    """Get the WKT representation of a projection from its EPSG code

    Args:
        epsg_code: The integer EPSG code

    Returns:
        wkt: The WKT representation of the projection
    """
    srs = osr.SpatialReference()
    srs.ImportFromEPSG(epsg_code)
    return srs.ExportToWkt()

get_area_raster(raster)

Determine the path of the area raster for a given backscatter raster based on naming conventions for HyP3 RTC products

Parameters:

Name	Type	Description	Default
raster	str	path of the backscatter raster, e.g. S1A_IW_20181102T155531_DVP_RTC30_G_gpuned_5685_VV.tif	required

Returns:

Name	Type	Description
area_raster	str	path of the area raster, e.g. S1A_IW_20181102T155531_DVP_RTC30_G_gpuned_5685_area.tif

Source code in asf_tools/composite.py

def get_area_raster(raster: str) -> str:
    """Determine the path of the area raster for a given backscatter raster based on naming conventions for HyP3 RTC
    products

    Args:
        raster: path of the backscatter raster, e.g. S1A_IW_20181102T155531_DVP_RTC30_G_gpuned_5685_VV.tif

    Returns:
        area_raster: path of the area raster, e.g. S1A_IW_20181102T155531_DVP_RTC30_G_gpuned_5685_area.tif
    """
    return '_'.join(raster.split('_')[:-1] + ['area.tif'])

get_epsg_code(info)

Get the EPSG code from a GDAL Info dictionary

Parameters:

Name	Type	Description	Default
info	dict	The dictionary returned by a gdal.Info call	required

Returns:

Name	Type	Description
epsg_code	int	The integer EPSG code

Source code in asf_tools/composite.py

def get_epsg_code(info: dict) -> int:
    """Get the EPSG code from a GDAL Info dictionary

    Args:
        info: The dictionary returned by a gdal.Info call

    Returns:
        epsg_code: The integer EPSG code
    """
    proj = osr.SpatialReference(info['coordinateSystem']['wkt'])
    epsg_code = int(proj.GetAttrValue('AUTHORITY', 1))
    return epsg_code

get_full_extent(raster_info)

Determine the corner coordinates and geotransform for the full extent of a set of rasters

Parameters:

Name	Type	Description	Default
raster_info	dict	A dictionary of gdal.Info results for the set of rasters	required

Returns:

Name	Type	Description
upper_left		The upper left corner of the extent as a tuple
upper_right		The lower right corner of the extent as a tuple
geotransform		The geotransform of the extent as a list

Source code in asf_tools/composite.py

def get_full_extent(raster_info: dict):
    """Determine the corner coordinates and geotransform for the full extent of a set of rasters

    Args:
        raster_info: A dictionary of gdal.Info results for the set of rasters

    Returns:
        upper_left: The upper left corner of the extent as a tuple
        upper_right: The lower right corner of the extent as a tuple
        geotransform: The geotransform of the extent as a list
    """
    upper_left_corners = [info['cornerCoordinates']['upperLeft'] for info in raster_info.values()]
    lower_right_corners = [info['cornerCoordinates']['lowerRight'] for info in raster_info.values()]

    ulx = min([ul[0] for ul in upper_left_corners])
    uly = max([ul[1] for ul in upper_left_corners])
    lrx = max([lr[0] for lr in lower_right_corners])
    lry = min([lr[1] for lr in lower_right_corners])

    log.debug(f'Full extent raster upper left: ({ulx, uly}); lower right: ({lrx, lry})')

    trans = []
    for info in raster_info.values():
        # Only need info from any one raster
        trans = info['geoTransform']
        break

    trans[0] = ulx
    trans[3] = uly

    return (ulx, uly), (lrx, lry), trans

get_target_epsg_code(codes)

Determine the target UTM EPSG projection for the output composite

Parameters:

Name	Type	Description	Default
codes	List[int]	List of UTM EPSG codes	required

Returns:

Name	Type	Description
target	int	UTM EPSG code

Source code in asf_tools/composite.py

def get_target_epsg_code(codes: List[int]) -> int:
    """Determine the target UTM EPSG projection for the output composite

    Args:
        codes: List of UTM EPSG codes

    Returns:
        target: UTM EPSG code
    """
    # use median east/west UTM zone of all files, regardless of hemisphere
    # UTM EPSG codes for each hemisphere will look like:
    #   North: 326XX
    #   South: 327XX
    valid_codes = list(range(32601, 32661)) + list(range(32701, 32761))
    if bad_codes := set(codes) - set(valid_codes):
        raise ValueError(f'Non UTM EPSG code encountered: {bad_codes}')

    hemispheres = [c // 100 * 100 for c in codes]
    # if even modes, choose lowest (North)
    target_hemisphere = min(multimode(hemispheres))

    zones = sorted([c % 100 for c in codes])
    # if even length, choose fist of median two
    target_zone = zones[(len(zones) - 1) // 2]

    return target_hemisphere + target_zone

make_composite(out_name, rasters, resolution=None)

Creates a local-resolution-weighted composite from Sentinel-1 RTC products

Parameters:

Name	Type	Description	Default
out_name	str	The base name of the output GeoTIFFs	required
rasters	List[str]	A list of file paths of the images to composite	required
resolution	float	The pixel size for the output GeoTIFFs	None

Returns:

Name	Type	Description
out_raster		Path to the created composite backscatter GeoTIFF
out_counts_raster		Path to the created GeoTIFF with counts of scenes contributing to each pixel

Source code in asf_tools/composite.py

def make_composite(out_name: str, rasters: List[str], resolution: float = None):
    """Creates a local-resolution-weighted composite from Sentinel-1 RTC products

    Args:
        out_name: The base name of the output GeoTIFFs
        rasters: A list of file paths of the images to composite
        resolution: The pixel size for the output GeoTIFFs

    Returns:
        out_raster: Path to the created composite backscatter GeoTIFF
        out_counts_raster: Path to the created GeoTIFF with counts of scenes contributing to each pixel
    """
    if not rasters:
        raise ValueError('Must specify at least one raster to composite')

    raster_info = {}
    for raster in rasters:
        raster_info[raster] = gdal.Info(raster, format='json')
        # make sure gdal can read the area raster
        gdal.Info(get_area_raster(raster))

    target_epsg_code = get_target_epsg_code([get_epsg_code(info) for info in raster_info.values()])
    log.debug(f'Composite projection is EPSG:{target_epsg_code}')

    if resolution is None:
        resolution = max([info['geoTransform'][1] for info in raster_info.values()])
    log.debug(f'Composite resolution is {resolution} meters')

    # resample rasters to maximum resolution & common UTM zone
    with TemporaryDirectory(prefix='reprojected_') as temp_dir:
        raster_info = reproject_to_target(raster_info, target_epsg_code=target_epsg_code, target_resolution=resolution,
                                          directory=temp_dir)

        # Get extent of union of all images
        full_ul, full_lr, full_trans = get_full_extent(raster_info)

        nx = int(abs(full_ul[0] - full_lr[0]) // resolution)
        ny = int(abs(full_ul[1] - full_lr[1]) // resolution)

        outputs = np.zeros((ny, nx))
        weights = np.zeros(outputs.shape)
        counts = np.zeros(outputs.shape, dtype=np.int8)

        for raster, info in raster_info.items():
            log.info(f'Processing raster {raster}')
            log.debug(f"Raster upper left: {info['cornerCoordinates']['upperLeft']}; "
                      f"lower right: {info['cornerCoordinates']['lowerRight']}")

            values = read_as_array(raster)

            area_raster = get_area_raster(raster)
            areas = read_as_array(area_raster)

            ulx, uly = info['cornerCoordinates']['upperLeft']
            y_index_start = int((full_ul[1] - uly) // resolution)
            y_index_end = y_index_start + values.shape[0]

            x_index_start = int((ulx - full_ul[0]) // resolution)
            x_index_end = x_index_start + values.shape[1]

            log.debug(
                f'Placing values in output grid at {y_index_start}:{y_index_end} and {x_index_start}:{x_index_end}'
            )

            mask = values == 0
            raster_weights = 1.0 / areas
            raster_weights[mask] = 0

            outputs[y_index_start:y_index_end, x_index_start:x_index_end] += values * raster_weights
            weights[y_index_start:y_index_end, x_index_start:x_index_end] += raster_weights
            counts[y_index_start:y_index_end, x_index_start:x_index_end] += ~mask

    del values, areas, mask, raster_weights

    # Divide by the total weight applied
    outputs /= weights
    del weights

    out_raster = write_cog(f'{out_name}.tif', outputs, full_trans, target_epsg_code, nodata_value=0)
    del outputs

    out_counts_raster = write_cog(f'{out_name}_counts.tif', counts, full_trans, target_epsg_code, dtype=gdal.GDT_Int16)
    del counts

    return out_raster, out_counts_raster

read_as_array(raster, band=1)

Reads data from a raster image into memory

Parameters:

Name	Type	Description	Default
raster	str	The file path to a raster image	required
band	int	The raster band to read	1

Returns:

Name	Type	Description
data	np.array	The raster pixel data as a numpy array

Source code in asf_tools/composite.py

def read_as_array(raster: str, band: int = 1) -> np.array:
    """Reads data from a raster image into memory

    Args:
        raster: The file path to a raster image
        band: The raster band to read

    Returns:
        data: The raster pixel data as a numpy array
    """
    log.debug(f'Reading raster values from {raster}')
    ds = gdal.Open(raster)
    data = ds.GetRasterBand(band).ReadAsArray()
    del ds  # How to close w/ gdal
    return data

reproject_to_target(raster_info, target_epsg_code, target_resolution, directory)

Reprojects a set of raster images to a common projection and resolution

Parameters:

Name	Type	Description	Default
raster_info	dict	A dictionary of gdal.Info results for the set of rasters	required
target_epsg_code	int	The integer EPSG code for the target projection	required
target_resolution	float	The target resolution	required
directory	str	The directory in which to create the reprojected files	required

Returns:

Name	Type	Description
target_raster_info	dict	An updated dictionary of gdal.Info results for the reprojected files

Source code in asf_tools/composite.py

def reproject_to_target(raster_info: dict, target_epsg_code: int, target_resolution: float, directory: str) -> dict:
    """Reprojects a set of raster images to a common projection and resolution

    Args:
        raster_info: A dictionary of gdal.Info results for the set of rasters
        target_epsg_code: The integer EPSG code for the target projection
        target_resolution: The target resolution
        directory: The directory in which to create the reprojected files

    Returns:
        target_raster_info: An updated dictionary of gdal.Info results for the reprojected files
    """
    target_raster_info = {}
    for raster, info in raster_info.items():
        epsg_code = get_epsg_code(info)
        resolution = info['geoTransform'][1]
        if epsg_code != target_epsg_code or resolution != target_resolution:
            log.info(f'Reprojecting {raster}')
            reprojected_raster = os.path.join(directory, os.path.basename(raster))
            gdal.Warp(
                reprojected_raster, raster, dstSRS=f'EPSG:{target_epsg_code}',
                xRes=target_resolution, yRes=target_resolution, targetAlignedPixels=True
            )

            area_raster = get_area_raster(raster)
            log.info(f'Reprojecting {area_raster}')
            reprojected_area_raster = os.path.join(directory, os.path.basename(area_raster))
            gdal.Warp(
                reprojected_area_raster, area_raster, dstSRS=f'EPSG:{target_epsg_code}',
                xRes=target_resolution, yRes=target_resolution, targetAlignedPixels=True
            )

            target_raster_info[reprojected_raster] = gdal.Info(reprojected_raster,  format='json')
        else:
            log.info(f'No need to reproject {raster}')
            target_raster_info[raster] = info

    return target_raster_info

write_cog(file_name, data, transform, epsg_code, dtype=gdal.GDT_Float32, nodata_value=None)

Creates a Cloud Optimized GeoTIFF

Parameters:

Name	Type	Description	Default
file_name	Union[str, Path]	The output file name	required
data	np.ndarray	The raster data	required
transform	List[float]	The geotransform for the output GeoTIFF	required
epsg_code	int	The integer EPSG code for the output GeoTIFF projection	required
dtype		The pixel data type for the output GeoTIFF	gdal.GDT_Float32
nodata_value		The NODATA value for the output Geotiff	None

Returns:

Name	Type	Description
file_name		The output file name

Source code in asf_tools/composite.py

def write_cog(file_name: Union[str, Path], data: np.ndarray, transform: List[float], epsg_code: int,
              dtype=gdal.GDT_Float32, nodata_value=None):
    """Creates a Cloud Optimized GeoTIFF

    Args:
        file_name: The output file name
        data: The raster data
        transform: The geotransform for the output GeoTIFF
        epsg_code: The integer EPSG code for the output GeoTIFF projection
        dtype: The pixel data type for the output GeoTIFF
        nodata_value: The NODATA value for the output Geotiff

    Returns:
        file_name: The output file name
    """
    log.info(f'Creating {file_name}')

    with NamedTemporaryFile() as temp_file:
        driver = gdal.GetDriverByName('GTiff')
        temp_geotiff = driver.Create(temp_file.name, data.shape[1], data.shape[0], 1, dtype)
        temp_geotiff.GetRasterBand(1).WriteArray(data)
        if nodata_value is not None:
            temp_geotiff.GetRasterBand(1).SetNoDataValue(nodata_value)
        temp_geotiff.SetGeoTransform(transform)
        temp_geotiff.SetProjection(epsg_to_wkt(epsg_code))

        driver = gdal.GetDriverByName('COG')
        options = ['COMPRESS=LZW', 'OVERVIEW_RESAMPLING=AVERAGE', 'NUM_THREADS=ALL_CPUS', 'BIGTIFF=YES']
        driver.CreateCopy(str(file_name), temp_geotiff, options=options)

        del temp_geotiff  # How to close w/ gdal
    return file_name

dem

Prepare a Copernicus GLO-30 DEM virtual raster (VRT) covering a given geometry

prepare_dem_vrt(vrt, geometry)

Create a DEM mosaic VRT covering a given geometry

The DEM mosaic is assembled from the Copernicus GLO-30 DEM tiles that intersect the geometry.

Note: asf_tools does not currently support geometries that cross the antimeridian.

Parameters:

Name	Type	Description	Default
vrt	Union[str, Path]	Path for the output VRT file	required
geometry	Union[ogr.Geometry, BaseGeometry]	Geometry in EPSG:4326 (lon/lat) projection for which to prepare a DEM mosaic	required

Source code in asf_tools/dem.py

def prepare_dem_vrt(vrt: Union[str, Path], geometry: Union[ogr.Geometry, BaseGeometry]):
    """Create a DEM mosaic VRT covering a given geometry

    The DEM mosaic is assembled from the Copernicus GLO-30 DEM tiles that intersect the geometry.

    Note: `asf_tools` does not currently support geometries that cross the antimeridian.

    Args:
        vrt: Path for the output VRT file
        geometry: Geometry in EPSG:4326 (lon/lat) projection for which to prepare a DEM mosaic

    """
    with GDALConfigManager(GDAL_DISABLE_READDIR_ON_OPEN='EMPTY_DIR'):
        if isinstance(geometry, BaseGeometry):
            geometry = ogr.CreateGeometryFromWkb(geometry.wkb)

        min_lon, max_lon, _, _ = geometry.GetEnvelope()
        if min_lon < -160. and max_lon > 160.:
            raise ValueError(f'asf_tools does not currently support geometries that cross the antimeridian: {geometry}')

        tile_features = vector.get_features(DEM_GEOJSON)
        if not vector.get_property_values_for_intersecting_features(geometry, tile_features):
            raise ValueError(f'Copernicus GLO-30 DEM does not intersect this geometry: {geometry}')

        dem_file_paths = vector.intersecting_feature_properties(geometry, tile_features, 'file_path')

        gdal.BuildVRT(str(vrt), dem_file_paths)

flood_map

Generate flood depth map from surface water extent map.

Create a flood depth map from a surface water extent map and a HAND image. The HAND image must be pixel-aligned (same extent and size) to the water extent map, and the surface water extent map should be a byte GeoTIFF indicating water (true), not water (false). Flood depth maps are estimated using either a numerical, normalized median absolute deviation, logarithmic or iterative approach.

logstat(data, func=np.nanstd)

Calculate a function in logarithmic scale and return in linear scale. INF values inside the data array are set to nan.

Parameters:

Name	Type	Description	Default
data	np.ndarray	array of data	required
func	Callable	statistical function to calculate in logarithmic scale	np.nanstd

Returns:

Name	Type	Description
statistic	Union[np.ndarray, float]	statistic of data in linear scale

Source code in asf_tools/flood_map.py

def logstat(data: np.ndarray, func: Callable = np.nanstd) -> Union[np.ndarray, float]:
    """ Calculate a function in logarithmic scale and return in linear scale.
        INF values inside the data array are set to nan.

        Args:
            data: array of data
            func: statistical function to calculate in logarithmic scale
        Returns:
            statistic: statistic of data in linear scale
    """
    ld = np.log(data)
    ld[np.isinf(ld)] = np.nan
    st = func(ld)
    return np.exp(st)

make_flood_map(out_raster, vv_raster, water_raster, hand_raster, estimator='iterative', water_level_sigma=3.0, known_water_threshold=30.0, iterative_bounds=(0, 15))

Create a flood depth map from a surface water extent map.

WARNING: This functionality is still under active development and the products created using this function are likely to change in the future.

Create a flood depth map from a single surface water extent map and a HAND image. The HAND image must be pixel-aligned to the surface water extent map. The surface water extent map should be a byte GeoTIFF indicating water (true) and not water (false)

Known perennial Global Surface-water data are produced under the Copernicus Programme (Pekel et al., 2016), and are included with surface-water detection maps when generating the flood depth product.

Flood depth maps are estimated using one of the approaches: Iterative: (Default) Basin hopping optimization method matches flooded areas to flood depth estimates given by the HAND layer. This is the most accurate method but also the most time-intensive. Normalized Median Absolute Deviation (nmad): Uses a median operator to estimate the variation to increase robustness in the presence of outliers. Logstat: Calculates the mean and standard deviation of HAND heights in the logarithmic domain to improve robustness for very non-Gaussian data distributions. Numpy: Calculates statistics on a linear scale. Least robust to outliers and non-Gaussian distributions.

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	Flood depth GeoTIFF to create	required
vv_raster	Union[str, Path]	Sentinel-1 RTC GeoTIFF, in power scale, with VV polarization	required
water_raster	Union[str, Path]	Surface water extent GeoTIFF	required
hand_raster	Union[str, Path]	Height Above Nearest Drainage (HAND) GeoTIFF aligned to the surface water extent raster	required
estimator	str	Estimation approach for determining flood depth	'iterative'
water_level_sigma	float	Max water height used in logstat, nmad, and numpy estimations	3.0
known_water_threshold	float	Threshold for extracting the known water area in percent	30.0
iterative_bounds	Tuple[int, int]	Bounds on basin-hopping algorithm used in iterative estimation	(0, 15)

References

Jean-Francios Pekel, Andrew Cottam, Noel Gorelik, Alan S. Belward. 2016. https://doi:10.1038/nature20584

Source code in asf_tools/flood_map.py

def make_flood_map(out_raster: Union[str, Path],  vv_raster: Union[str, Path],
                   water_raster: Union[str, Path], hand_raster: Union[str, Path],
                   estimator: str = 'iterative',
                   water_level_sigma: float = 3.,
                   known_water_threshold: float = 30.,
                   iterative_bounds: Tuple[int, int] = (0, 15)):
    """Create a flood depth map from a surface water extent map.

    WARNING: This functionality is still under active development and the products
    created using this function are likely to change in the future.

    Create a flood depth map from a single surface water extent map and
    a HAND image. The HAND image must be pixel-aligned to the surface water extent map.
    The surface water extent map should be a byte GeoTIFF indicating water (true) and
    not water (false)

    Known perennial Global Surface-water data are produced under the Copernicus Programme (Pekel et al., 2016),
    and are included with surface-water detection maps when generating the flood depth product.

    Flood depth maps are estimated using one of the approaches:
    *Iterative: (Default) Basin hopping optimization method matches flooded areas to flood depth
    estimates given by the HAND layer. This is the most accurate method but also the
    most time-intensive.
    *Normalized Median Absolute Deviation (nmad): Uses a median operator to estimate
    the variation to increase robustness in the presence of outliers.
    *Logstat: Calculates the mean and standard deviation of HAND heights in the logarithmic
    domain to improve robustness for very non-Gaussian data distributions.
    *Numpy: Calculates statistics on a linear scale. Least robust to outliers and non-Gaussian
    distributions.

    Args:
        out_raster: Flood depth GeoTIFF to create
        vv_raster: Sentinel-1 RTC GeoTIFF, in power scale, with VV polarization
        water_raster: Surface water extent GeoTIFF
        hand_raster: Height Above Nearest Drainage (HAND) GeoTIFF aligned to the surface water extent raster
        estimator: Estimation approach for determining flood depth
        water_level_sigma: Max water height used in logstat, nmad, and numpy estimations
        known_water_threshold: Threshold for extracting the known water area in percent
        iterative_bounds: Bounds on basin-hopping algorithm used in iterative estimation

    References:
        Jean-Francios Pekel, Andrew Cottam, Noel Gorelik, Alan S. Belward. 2016. <https://doi:10.1038/nature20584>
    """

    info = gdal.Info(str(water_raster), format='json')
    epsg = get_epsg_code(info)
    geotransform = info['geoTransform']

    hand_array = gdal.Open(str(hand_raster), gdal.GA_ReadOnly).ReadAsArray()

    log.info('Fetching perennial flood data.')
    known_water_mask = get_waterbody(info, threshold=known_water_threshold)
    write_cog(str(out_raster).replace('.tif', f'_{estimator}_PW.tif'), known_water_mask, transform=geotransform,
              epsg_code=epsg, dtype=gdal.GDT_Byte, nodata_value=False)

    water_map = gdal.Open(str(water_raster)).ReadAsArray()
    flood_mask = np.logical_or(water_map, known_water_mask)
    del water_map

    vv_array = read_as_masked_array(vv_raster)
    flood_mask[vv_array.mask] = False
    del vv_array

    labeled_flood_mask, num_labels = ndimage.label(flood_mask)
    object_slices = ndimage.find_objects(labeled_flood_mask)
    log.info(f'Detected {num_labels} water bodies...')

    flood_depth = np.zeros(flood_mask.shape)

    for ll in tqdm(range(1, num_labels)):  # Skip first, largest label.
        slices = object_slices[ll - 1]
        min0, max0 = slices[0].start, slices[0].stop
        min1, max1 = slices[1].start, slices[1].stop

        flood_window = labeled_flood_mask[min0:max0, min1:max1]
        hand_window = hand_array[min0:max0, min1:max1]

        water_height = estimate_flood_depth(ll, hand_window, flood_window, estimator=estimator,
                                            water_level_sigma=water_level_sigma, iterative_bounds=iterative_bounds)

        flood_depth_window = flood_depth[min0:max0, min1:max1]
        flood_depth_window[flood_window == ll] = water_height - hand_window[flood_window == ll]

    flood_depth[flood_depth < 0] = 0

    write_cog(str(out_raster).replace('.tif', f'_{estimator}_WaterDepth.tif'), flood_depth, transform=geotransform,
              epsg_code=epsg, dtype=gdal.GDT_Float64, nodata_value=False)
    write_cog(str(out_raster).replace('.tif', f'_{estimator}_FloodMask.tif'), flood_mask, transform=geotransform,
              epsg_code=epsg, dtype=gdal.GDT_Byte, nodata_value=False)

    flood_mask[known_water_mask] = False
    flood_depth[np.logical_not(flood_mask)] = 0

    write_cog(str(out_raster).replace('.tif', f'_{estimator}_FloodDepth.tif'), flood_depth, transform=geotransform,
              epsg_code=epsg, dtype=gdal.GDT_Float64, nodata_value=False)

hand

calculate_hand_for_basins(out_raster, geometries, dem_file, acc_thresh=100)

Calculate the Height Above Nearest Drainage (HAND) for watershed boundaries (hydrobasins).

For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	HAND GeoTIFF to create	required
geometries	GeometryCollection	watershed boundary (hydrobasin) polygons to calculate HAND over	required
dem_file	Union[str, Path]	DEM raster covering (containing) geometries	required
acc_thresh	Optional[int]	Accumulation threshold for determining the drainage mask. If None, the mean accumulation value is used	100

Source code in asf_tools/hand/calculate.py

def calculate_hand_for_basins(out_raster:  Union[str, Path], geometries: GeometryCollection,
                              dem_file: Union[str, Path], acc_thresh: Optional[int] = 100):
    """Calculate the Height Above Nearest Drainage (HAND) for watershed boundaries (hydrobasins).

    For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

    Args:
        out_raster: HAND GeoTIFF to create
        geometries: watershed boundary (hydrobasin) polygons to calculate HAND over
        dem_file: DEM raster covering (containing) `geometries`
        acc_thresh: Accumulation threshold for determining the drainage mask.
            If `None`, the mean accumulation value is used
    """
    with rasterio.open(dem_file) as src:
        basin_mask, basin_affine_tf, basin_window = rasterio.mask.raster_geometry_mask(
            src, geometries, all_touched=True, crop=True, pad=True, pad_width=1
        )
        basin_array = src.read(1, window=basin_window)

        hand = calculate_hand(basin_array, basin_affine_tf, src.crs, basin_mask, acc_thresh=acc_thresh)

        write_cog(
            out_raster, hand, transform=basin_affine_tf.to_gdal(), epsg_code=src.crs.to_epsg(), nodata_value=np.nan,
        )

make_copernicus_hand(out_raster, vector_file, acc_thresh=100)

Copernicus GLO-30 Height Above Nearest Drainage (HAND)

Make a Height Above Nearest Drainage (HAND) GeoTIFF from the Copernicus GLO-30 DEM covering the watershed boundaries (hydrobasins) defined in a vector file.

For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	HAND GeoTIFF to create	required
vector_file	Union[str, Path]	Vector file of watershed boundary (hydrobasin) polygons to calculate HAND over	required
acc_thresh	Optional[int]	Accumulation threshold for determining the drainage mask. If None, the mean accumulation value is used	100

Source code in asf_tools/hand/calculate.py

def make_copernicus_hand(out_raster:  Union[str, Path], vector_file: Union[str, Path], acc_thresh: Optional[int] = 100):
    """Copernicus GLO-30 Height Above Nearest Drainage (HAND)

    Make a Height Above Nearest Drainage (HAND) GeoTIFF from the Copernicus GLO-30 DEM
    covering the watershed boundaries (hydrobasins) defined in a vector file.

    For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

    Args:
        out_raster: HAND GeoTIFF to create
        vector_file: Vector file of watershed boundary (hydrobasin) polygons to calculate HAND over
        acc_thresh: Accumulation threshold for determining the drainage mask.
            If `None`, the mean accumulation value is used
    """
    with fiona.open(vector_file) as vds:
        geometries = GeometryCollection([shape(feature['geometry']) for feature in vds])

    with NamedTemporaryFile(suffix='.vrt', delete=False) as dem_vrt:
        prepare_dem_vrt(dem_vrt.name, geometries)
        calculate_hand_for_basins(out_raster, geometries, dem_vrt.name, acc_thresh=acc_thresh)

prepare_hand_vrt(vrt, geometry)

Prepare a HAND mosaic VRT covering a given geometry

Prepare a Height Above Nearest Drainage (HAND) virtual raster (VRT) covering a given geometry. The Height Above Nearest Drainage (HAND) mosaic is assembled from the HAND tiles that intersect the geometry, using a HAND derived from the Copernicus GLO-30 DEM.

Note: asf_tools does not currently support geometries that cross the antimeridian.

Parameters:

Name	Type	Description	Default
vrt	Union[str, Path]	Path for the output VRT file	required
geometry	Union[ogr.Geometry, BaseGeometry]	Geometry in EPSG:4326 (lon/lat) projection for which to prepare a HAND mosaic	required

Source code in asf_tools/hand/prepare.py

def prepare_hand_vrt(vrt: Union[str, Path], geometry: Union[ogr.Geometry, BaseGeometry]):
    """Prepare a HAND mosaic VRT covering a given geometry

    Prepare a Height Above Nearest Drainage (HAND) virtual raster (VRT) covering a given geometry.
    The Height Above Nearest Drainage (HAND) mosaic is assembled from the HAND tiles that intersect
    the geometry, using a HAND derived from the Copernicus GLO-30 DEM.

    Note: `asf_tools` does not currently support geometries that cross the antimeridian.

    Args:
        vrt: Path for the output VRT file
        geometry: Geometry in EPSG:4326 (lon/lat) projection for which to prepare a HAND mosaic

    """
    with GDALConfigManager(GDAL_DISABLE_READDIR_ON_OPEN='EMPTY_DIR'):
        if isinstance(geometry, BaseGeometry):
            geometry = ogr.CreateGeometryFromWkb(geometry.wkb)

        min_lon, max_lon, _, _ = geometry.GetEnvelope()
        if min_lon < -160. and max_lon > 160.:
            raise ValueError(f'asf_tools does not currently support geometries that cross the antimeridian: {geometry}')

        tile_features = vector.get_features(HAND_GEOJSON)
        if not vector.get_property_values_for_intersecting_features(geometry, tile_features):
            raise ValueError(f'Copernicus GLO-30 HAND does not intersect this geometry: {geometry}')

        hand_file_paths = vector.intersecting_feature_properties(geometry, tile_features, 'file_path')

        gdal.BuildVRT(str(vrt), hand_file_paths)

calculate

Calculate Height Above Nearest Drainage (HAND) from the Copernicus GLO-30 DEM

calculate_hand(dem_array, dem_affine, dem_crs, basin_mask, acc_thresh=100)

Calculate the Height Above Nearest Drainage (HAND)

Calculate the Height Above Nearest Drainage (HAND) using pySHEDS library. Because HAND is tied to watershed boundaries (hydrobasins), clipped/cut basins will produce weird edge effects, and incomplete basins should be masked out. For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

This involves: * Filling pits (single-cells lower than their surrounding neighbors) in the Digital Elevation Model (DEM) * Filling depressions (regions of cells lower than their surrounding neighbors) in the Digital Elevation Model (DEM) * Resolving un-drainable flats * Determine the flow direction using the ESRI D8 routing scheme * Determine flow accumulation (number of upstream cells) * Create a drainage mask using the accumulation threshold acc_thresh * Calculating HAND

In the HAND calculation, NaNs inside the basin filled using fill_hand

Parameters:

Name	Type	Description	Default
dem_array		DEM to calculate HAND for	required
dem_crs	rasterio.crs.CRS	DEM Coordinate Reference System (CRS)	required
dem_affine	rasterio.Affine	DEM Affine geotransform	required
basin_mask		Array of booleans indicating wither an element should be masked out (à la Numpy Masked Arrays: https://numpy.org/doc/stable/reference/maskedarray.generic.html#what-is-a-masked-array)	required
acc_thresh	Optional[int]	Accumulation threshold for determining the drainage mask. If None, the mean accumulation value is used	100

Source code in asf_tools/hand/calculate.py

def calculate_hand(dem_array, dem_affine: rasterio.Affine, dem_crs: rasterio.crs.CRS, basin_mask,
                   acc_thresh: Optional[int] = 100):
    """Calculate the Height Above Nearest Drainage (HAND)

     Calculate the Height Above Nearest Drainage (HAND) using pySHEDS library. Because HAND
     is tied to watershed boundaries (hydrobasins), clipped/cut basins will produce weird edge
     effects, and incomplete basins should be masked out. For watershed boundaries,
     see: https://www.hydrosheds.org/page/hydrobasins

     This involves:
        * Filling pits (single-cells lower than their surrounding neighbors)
            in the Digital Elevation Model (DEM)
        * Filling depressions (regions of cells lower than their surrounding neighbors)
            in the Digital Elevation Model (DEM)
        * Resolving un-drainable flats
        * Determine the flow direction using the ESRI D8 routing scheme
        * Determine flow accumulation (number of upstream cells)
        * Create a drainage mask using the accumulation threshold `acc_thresh`
        * Calculating HAND

    In the HAND calculation, NaNs inside the basin filled using `fill_hand`

    Args:
        dem_array: DEM to calculate HAND for
        dem_crs: DEM Coordinate Reference System (CRS)
        dem_affine: DEM Affine geotransform
        basin_mask: Array of booleans indicating wither an element should be masked out (à la Numpy Masked Arrays:
            https://numpy.org/doc/stable/reference/maskedarray.generic.html#what-is-a-masked-array)
        acc_thresh: Accumulation threshold for determining the drainage mask.
            If `None`, the mean accumulation value is used
    """
    nodata_fill_value = np.finfo(float).eps
    with NamedTemporaryFile() as temp_file:
        write_cog(temp_file.name, dem_array,
                  transform=dem_affine.to_gdal(), epsg_code=dem_crs.to_epsg(),
                  # Prevents PySheds from assuming using zero as the nodata value
                  nodata_value=nodata_fill_value)

        # From PySheds; see example usage: http://mattbartos.com/pysheds/
        grid = sGrid.from_raster(str(temp_file.name))
        dem = grid.read_raster(str(temp_file.name))

    log.info('Fill pits in DEM')
    pit_filled_dem = grid.fill_pits(dem)

    log.info('Filling depressions')
    flooded_dem = grid.fill_depressions(pit_filled_dem)
    del pit_filled_dem

    log.info('Resolving flats')
    inflated_dem = grid.resolve_flats(flooded_dem)
    del flooded_dem

    log.info('Obtaining flow direction')
    flow_dir = grid.flowdir(inflated_dem, apply_mask=True)

    log.info('Calculating flow accumulation')
    acc = grid.accumulation(flow_dir)

    if acc_thresh is None:
        acc_thresh = acc.mean()

    log.info(f'Calculating HAND using accumulation threshold of {acc_thresh}')
    hand = grid.compute_hand(flow_dir, inflated_dem, acc > acc_thresh, inplace=False)

    if np.isnan(hand).any():
        log.info('Filling NaNs in the HAND')
        # mask outside of basin with a not-NaN value to prevent NaN-filling outside of basin (optimization)
        hand[basin_mask] = nodata_fill_value
        hand = fill_hand(hand, dem_array)

    # set pixels outside of basin to nodata
    hand[basin_mask] = np.nan

    # TODO: also mask ocean pixels here?

    return hand

calculate_hand_for_basins(out_raster, geometries, dem_file, acc_thresh=100)

Calculate the Height Above Nearest Drainage (HAND) for watershed boundaries (hydrobasins).

For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	HAND GeoTIFF to create	required
geometries	GeometryCollection	watershed boundary (hydrobasin) polygons to calculate HAND over	required
dem_file	Union[str, Path]	DEM raster covering (containing) geometries	required
acc_thresh	Optional[int]	Accumulation threshold for determining the drainage mask. If None, the mean accumulation value is used	100

Source code in asf_tools/hand/calculate.py

def calculate_hand_for_basins(out_raster:  Union[str, Path], geometries: GeometryCollection,
                              dem_file: Union[str, Path], acc_thresh: Optional[int] = 100):
    """Calculate the Height Above Nearest Drainage (HAND) for watershed boundaries (hydrobasins).

    For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

    Args:
        out_raster: HAND GeoTIFF to create
        geometries: watershed boundary (hydrobasin) polygons to calculate HAND over
        dem_file: DEM raster covering (containing) `geometries`
        acc_thresh: Accumulation threshold for determining the drainage mask.
            If `None`, the mean accumulation value is used
    """
    with rasterio.open(dem_file) as src:
        basin_mask, basin_affine_tf, basin_window = rasterio.mask.raster_geometry_mask(
            src, geometries, all_touched=True, crop=True, pad=True, pad_width=1
        )
        basin_array = src.read(1, window=basin_window)

        hand = calculate_hand(basin_array, basin_affine_tf, src.crs, basin_mask, acc_thresh=acc_thresh)

        write_cog(
            out_raster, hand, transform=basin_affine_tf.to_gdal(), epsg_code=src.crs.to_epsg(), nodata_value=np.nan,
        )

fill_hand(hand, dem)

Replace NaNs in a HAND array with values interpolated from their neighbor's HOND

Replace NaNs in a HAND array with values interpolated from their neighbor's HOND (height of nearest drainage) using a 2D Gaussian kernel. Here, HOND is defined as the DEM value less the HAND value. For the kernel, see: https://docs.astropy.org/en/stable/convolution/#using-astropy-s-convolution-to-replace-bad-data

Source code in asf_tools/hand/calculate.py

def fill_hand(hand: np.ndarray, dem: np.ndarray):
    """Replace NaNs in a HAND array with values interpolated from their neighbor's HOND

    Replace NaNs in a HAND array with values interpolated from their neighbor's HOND (height of nearest drainage)
    using a 2D Gaussian kernel. Here, HOND is defined as the DEM value less the HAND value. For the kernel, see:
    https://docs.astropy.org/en/stable/convolution/#using-astropy-s-convolution-to-replace-bad-data
    """
    hond = dem - hand
    hond = fill_nan(hond)

    hand_mask = np.isnan(hand)
    hand[hand_mask] = dem[hand_mask] - hond[hand_mask]
    hand[hand < 0] = 0

    return hand

fill_nan(array)

Replace NaNs with values interpolated from their neighbors

Replace NaNs with values interpolated from their neighbors using a 2D Gaussian kernel, see: https://docs.astropy.org/en/stable/convolution/#using-astropy-s-convolution-to-replace-bad-data

Source code in asf_tools/hand/calculate.py

def fill_nan(array: np.ndarray) -> np.ndarray:
    """Replace NaNs with values interpolated from their neighbors

    Replace NaNs with values interpolated from their neighbors using a 2D Gaussian
    kernel, see: https://docs.astropy.org/en/stable/convolution/#using-astropy-s-convolution-to-replace-bad-data
    """
    kernel = astropy.convolution.Gaussian2DKernel(x_stddev=3)  # kernel x_size=8*stddev
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
        while np.any(np.isnan(array)):
            array = astropy.convolution.interpolate_replace_nans(
                array, kernel, convolve=astropy.convolution.convolve
            )

    return array

make_copernicus_hand(out_raster, vector_file, acc_thresh=100)

Copernicus GLO-30 Height Above Nearest Drainage (HAND)

Make a Height Above Nearest Drainage (HAND) GeoTIFF from the Copernicus GLO-30 DEM covering the watershed boundaries (hydrobasins) defined in a vector file.

For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	HAND GeoTIFF to create	required
vector_file	Union[str, Path]	Vector file of watershed boundary (hydrobasin) polygons to calculate HAND over	required
acc_thresh	Optional[int]	Accumulation threshold for determining the drainage mask. If None, the mean accumulation value is used	100

Source code in asf_tools/hand/calculate.py

def make_copernicus_hand(out_raster:  Union[str, Path], vector_file: Union[str, Path], acc_thresh: Optional[int] = 100):
    """Copernicus GLO-30 Height Above Nearest Drainage (HAND)

    Make a Height Above Nearest Drainage (HAND) GeoTIFF from the Copernicus GLO-30 DEM
    covering the watershed boundaries (hydrobasins) defined in a vector file.

    For watershed boundaries, see: https://www.hydrosheds.org/page/hydrobasins

    Args:
        out_raster: HAND GeoTIFF to create
        vector_file: Vector file of watershed boundary (hydrobasin) polygons to calculate HAND over
        acc_thresh: Accumulation threshold for determining the drainage mask.
            If `None`, the mean accumulation value is used
    """
    with fiona.open(vector_file) as vds:
        geometries = GeometryCollection([shape(feature['geometry']) for feature in vds])

    with NamedTemporaryFile(suffix='.vrt', delete=False) as dem_vrt:
        prepare_dem_vrt(dem_vrt.name, geometries)
        calculate_hand_for_basins(out_raster, geometries, dem_vrt.name, acc_thresh=acc_thresh)

prepare

Prepare a Height Above Nearest Drainage (HAND) virtual raster (VRT) covering a given geometry

prepare_hand_for_raster(hand_raster, source_raster, resampling_method='lanczos')

Create a HAND raster pixel-aligned to a source raster

Parameters:

Name	Type	Description	Default
hand_raster	Union[str, Path]	Path for the output HAND raster	required
source_raster	Union[str, Path]	Path for the source raster	required
resampling_method	str	Name of the resampling method to use. For available methods, see: https://gdal.org/programs/gdalwarp.html#cmdoption-gdalwarp-r	'lanczos'

Source code in asf_tools/hand/prepare.py

def prepare_hand_for_raster(hand_raster: Union[str, Path], source_raster: Union[str, Path],
                            resampling_method: str = 'lanczos'):
    """Create a HAND raster pixel-aligned to a source raster

    Args:
        hand_raster: Path for the output HAND raster
        source_raster: Path for the source raster
        resampling_method: Name of the resampling method to use. For available methods, see:
            https://gdal.org/programs/gdalwarp.html#cmdoption-gdalwarp-r
    """
    info = gdal.Info(str(source_raster), format='json')

    hand_geometry = shape(info['wgs84Extent'])
    hand_bounds = [info['cornerCoordinates']['upperLeft'][0],
                   info['cornerCoordinates']['lowerRight'][1],
                   info['cornerCoordinates']['lowerRight'][0],
                   info['cornerCoordinates']['upperLeft'][1]]

    with NamedTemporaryFile(suffix='.vrt', delete=False) as hand_vrt:
        prepare_hand_vrt(hand_vrt.name, hand_geometry)
        gdal.Warp(str(hand_raster), hand_vrt.name, dstSRS=f'EPSG:{get_epsg_code(info)}',
                  outputBounds=hand_bounds, width=info['size'][0], height=info['size'][1],
                  resampleAlg=Resampling[resampling_method].value)

prepare_hand_vrt(vrt, geometry)

Prepare a HAND mosaic VRT covering a given geometry

Prepare a Height Above Nearest Drainage (HAND) virtual raster (VRT) covering a given geometry. The Height Above Nearest Drainage (HAND) mosaic is assembled from the HAND tiles that intersect the geometry, using a HAND derived from the Copernicus GLO-30 DEM.

Note: asf_tools does not currently support geometries that cross the antimeridian.

Parameters:

Name	Type	Description	Default
vrt	Union[str, Path]	Path for the output VRT file	required
geometry	Union[ogr.Geometry, BaseGeometry]	Geometry in EPSG:4326 (lon/lat) projection for which to prepare a HAND mosaic	required

Source code in asf_tools/hand/prepare.py

def prepare_hand_vrt(vrt: Union[str, Path], geometry: Union[ogr.Geometry, BaseGeometry]):
    """Prepare a HAND mosaic VRT covering a given geometry

    Prepare a Height Above Nearest Drainage (HAND) virtual raster (VRT) covering a given geometry.
    The Height Above Nearest Drainage (HAND) mosaic is assembled from the HAND tiles that intersect
    the geometry, using a HAND derived from the Copernicus GLO-30 DEM.

    Note: `asf_tools` does not currently support geometries that cross the antimeridian.

    Args:
        vrt: Path for the output VRT file
        geometry: Geometry in EPSG:4326 (lon/lat) projection for which to prepare a HAND mosaic

    """
    with GDALConfigManager(GDAL_DISABLE_READDIR_ON_OPEN='EMPTY_DIR'):
        if isinstance(geometry, BaseGeometry):
            geometry = ogr.CreateGeometryFromWkb(geometry.wkb)

        min_lon, max_lon, _, _ = geometry.GetEnvelope()
        if min_lon < -160. and max_lon > 160.:
            raise ValueError(f'asf_tools does not currently support geometries that cross the antimeridian: {geometry}')

        tile_features = vector.get_features(HAND_GEOJSON)
        if not vector.get_property_values_for_intersecting_features(geometry, tile_features):
            raise ValueError(f'Copernicus GLO-30 HAND does not intersect this geometry: {geometry}')

        hand_file_paths = vector.intersecting_feature_properties(geometry, tile_features, 'file_path')

        gdal.BuildVRT(str(vrt), hand_file_paths)

raster

convert_scale(array, in_scale, out_scale)

Convert calibrated raster scale between db, amplitude and power

Source code in asf_tools/raster.py

def convert_scale(array: Union[np.ndarray, np.ma.MaskedArray], in_scale: Literal['db', 'amplitude', 'power'],
                  out_scale: Literal['db', 'amplitude', 'power']) -> Union[np.ndarray, np.ma.MaskedArray]:
    """Convert calibrated raster scale between db, amplitude and power"""
    if in_scale == out_scale:
        warnings.warn(f'Nothing to do! {in_scale} is same as {out_scale}.')
        return array

    log10 = np.ma.log10 if isinstance(array, np.ma.MaskedArray) else np.log10

    if in_scale == 'db':
        if out_scale == 'power':
            return 10 ** (array / 10)
        if out_scale == 'amplitude':
            return 10 ** (array / 20)

    if in_scale == 'amplitude':
        if out_scale == 'power':
            return array ** 2
        if out_scale == 'db':
            return 10 * log10(array ** 2)

    if in_scale == 'power':
        if out_scale == 'amplitude':
            return np.sqrt(array)
        if out_scale == 'db':
            return 10 * log10(array)

    raise ValueError(f'Cannot convert raster of scale {in_scale} to {out_scale}')

read_as_masked_array(raster, band=1)

Reads data from a raster image into memory, masking invalid and NoData values

Parameters:

Name	Type	Description	Default
raster	Union[str, Path]	The file path to a raster image	required
band	int	The raster band to read	1

Returns:

Name	Type	Description
data	np.ma.MaskedArray	The raster pixel data as a numpy MaskedArray

Source code in asf_tools/raster.py

def read_as_masked_array(raster: Union[str, Path], band: int = 1) -> np.ma.MaskedArray:
    """Reads data from a raster image into memory, masking invalid and NoData values

    Args:
        raster: The file path to a raster image
        band: The raster band to read

    Returns:
        data: The raster pixel data as a numpy MaskedArray
    """
    log.debug(f'Reading raster values from {raster}')
    ds = gdal.Open(str(raster))
    band = ds.GetRasterBand(band)
    data = np.ma.masked_invalid(band.ReadAsArray())
    nodata = band.GetNoDataValue()
    if nodata is not None:
        return np.ma.masked_values(data, nodata)
    return data

threshold

expectation_maximization_threshold(tile, number_of_classes=3)

Water threshold Calculation using a multi-mode Expectation Maximization Approach

Thresholding works best when backscatter tiles are provided on a decibel scale to get Gaussian distribution that is scaled to a range of 0-255, and performed on a small tile that is likely to have a transition between water and not water.

Parameters:

Name	Type	Description	Default
tile	np.ndarray	array of backscatter values for a tile from an RTC raster	required
number_of_classes	int	classify the tile into this many classes. Typically, three classes capture: (1) urban and bright slopes, (2) average brightness farmland, and (3) water, as is often seen in the US Midwest.	3

Returns:

Name	Type	Description
threshold	float	threshold value that can be used to create a water extent map

Source code in asf_tools/threshold.py

def expectation_maximization_threshold(tile: np.ndarray, number_of_classes: int = 3) -> float:
    """Water threshold Calculation using a multi-mode Expectation Maximization Approach

    Thresholding works best when backscatter tiles are provided on a decibel scale
    to get Gaussian distribution that is scaled to a range of 0-255, and performed
    on a small tile that is likely to have a transition between water and not water.


    Args:
        tile: array of backscatter values for a tile from an RTC raster
        number_of_classes: classify the tile into this many classes. Typically, three
            classes capture: (1) urban and bright slopes, (2) average brightness farmland,
            and (3) water, as is often seen in the US Midwest.

    Returns:
        threshold: threshold value that can be used to create a water extent map
    """

    image_copy = tile.copy()
    image_copy2 = np.ma.filled(tile.astype(float), np.nan)  # needed for valid posterior_lookup keys
    image = tile.flatten()
    minimum = np.amin(image)
    image = image - minimum + 1
    maximum = np.amax(image)

    histogram = _make_histogram(image)
    nonzero_indices = np.nonzero(histogram)[0]
    histogram = histogram[nonzero_indices]
    histogram = histogram.flatten()
    class_means = (
            (np.arange(number_of_classes) + 1) * maximum /
            (number_of_classes + 1)
    )
    class_variances = np.ones(number_of_classes) * maximum
    class_proportions = np.ones(number_of_classes) * 1 / number_of_classes
    sml = np.mean(np.diff(nonzero_indices)) / 1000
    iteration = 0
    while True:
        class_likelihood = _make_distribution(
            class_means, class_variances, class_proportions, nonzero_indices
        )
        sum_likelihood = np.sum(class_likelihood, 1) + np.finfo(
            class_likelihood[0][0]).eps
        log_likelihood = np.sum(histogram * np.log(sum_likelihood))
        for j in range(0, number_of_classes):
            class_posterior_probability = (
                    histogram * class_likelihood[:, j] / sum_likelihood
            )
            class_proportions[j] = np.sum(class_posterior_probability)
            class_means[j] = (
                    np.sum(nonzero_indices * class_posterior_probability)
                    / class_proportions[j]
            )
            vr = (nonzero_indices - class_means[j])
            class_variances[j] = (
                    np.sum(vr * vr * class_posterior_probability)
                    / class_proportions[j] + sml
            )
            del class_posterior_probability, vr
        class_proportions += 1e-3
        class_proportions /= np.sum(class_proportions)
        class_likelihood = _make_distribution(
            class_means, class_variances, class_proportions, nonzero_indices
        )
        sum_likelihood = np.sum(class_likelihood, 1) + np.finfo(
            class_likelihood[0, 0]).eps
        del class_likelihood
        new_log_likelihood = np.sum(histogram * np.log(sum_likelihood))
        del sum_likelihood
        if (new_log_likelihood - log_likelihood) < 0.000001:
            break
        iteration += 1
    del log_likelihood, new_log_likelihood
    class_means = class_means + minimum - 1
    s = image_copy.shape
    posterior = np.zeros((s[0], s[1], number_of_classes))
    posterior_lookup = dict()
    for i in range(0, s[0]):
        for j in range(0, s[1]):
            pixel_val = image_copy2[i, j]
            if pixel_val in posterior_lookup:
                for n in range(0, number_of_classes):
                    posterior[i, j, n] = posterior_lookup[pixel_val][n]
            else:
                posterior_lookup.update({pixel_val: [0] * number_of_classes})
                for n in range(0, number_of_classes):
                    x = _make_distribution(
                        class_means[n], class_variances[n], class_proportions[n],
                        image_copy[i, j]
                    )
                    posterior[i, j, n] = x * class_proportions[n]
                    posterior_lookup[pixel_val][n] = posterior[i, j, n]

    sorti = np.argsort(class_means)
    xvec = np.arange(class_means[sorti[0]], class_means[sorti[1]], step=.05)
    x1 = _make_distribution(class_means[sorti[0]], class_variances[sorti[0]], class_proportions[sorti[0]], xvec)
    x2 = _make_distribution(class_means[sorti[1]], class_variances[sorti[1]], class_proportions[sorti[1]], xvec)
    dx = np.abs(x1 - x2)

    return xvec[np.argmin(dx)]

tile

tile_array(array, tile_shape=(200, 200), pad_value=None)

Tile a 2D numpy array

Turn a 2D numpy array like

array = [[0, 0, 1, 1], ... [0, 0, 1, 1], ... [2, 2, 3, 3], ... [2, 2, 3, 3]] array.shape (4, 4)

into a tiled array like

tiles = tiled_array(array, 2, 2) print(tiles) [[[0, 0], [0, 0]], [[1, 1], [1, 1]], [[2, 2], [2, 2]], [[3, 3], [3, 3]]] tiles.shape (4, 2, 2)

Parameters:

Name	Type	Description	Default
array	Union[np.ndarray, np.ma.MaskedArray]	2D array to tile	required
tile_shape	Tuple[int, int]	the shape of each tile	(200, 200)
pad_value	float	right-bottom pad a with pad as needed so a is evenly divisible into tiles	None

Returns:

Type	Description
Union[np.ndarray, np.ma.MaskedArray]	the tiled array

Source code in asf_tools/tile.py

def tile_array(array: Union[np.ndarray, np.ma.MaskedArray], tile_shape: Tuple[int, int] = (200, 200),
               pad_value: float = None) -> Union[np.ndarray, np.ma.MaskedArray]:
    """Tile a 2D numpy array

    Turn a 2D numpy array like:
        >>> array = [[0, 0, 1, 1],
        ...          [0, 0, 1, 1],
        ...          [2, 2, 3, 3],
        ...          [2, 2, 3, 3]]
        >>> array.shape
        (4, 4)

    into a tiled array like:
        >>> tiles = tiled_array(array, 2, 2)
        >>> print(tiles)
        [[[0, 0],
          [0, 0]],
         [[1, 1],
          [1, 1]],
         [[2, 2],
          [2, 2]],
         [[3, 3],
          [3, 3]]]
        >>> tiles.shape
        (4, 2, 2)

    Args:
        array: 2D array to tile
        tile_shape: the shape of each tile
        pad_value: right-bottom pad `a` with `pad` as needed so `a` is evenly divisible into tiles

    Returns:
        the tiled array
    """
    array_rows, array_columns = array.shape
    tile_rows, tile_columns = tile_shape

    # CREDIT: https://twitter.com/LizzUltee/status/1379508448262512641
    rpad = -array_rows % tile_rows
    cpad = -array_columns % tile_columns

    if (rpad or cpad) and pad_value is None:
        raise ValueError(f'Cannot evenly tile a {array.shape} array into ({tile_rows},{tile_columns}) tiles')

    if rpad or cpad:
        padded_array = np.pad(array, ((0, rpad), (0, cpad)), constant_values=pad_value)
        if isinstance(array, np.ma.MaskedArray):
            mask = np.pad(array.mask, ((0, rpad), (0, cpad)), constant_values=True)
            padded_array = np.ma.MaskedArray(padded_array, mask=mask)
    else:
        padded_array = array

    tile_list = []
    for rows in np.vsplit(padded_array, range(tile_rows, array_rows, tile_rows)):
        tile_list.extend(np.hsplit(rows, range(tile_columns, array_columns, tile_columns)))

    dstack = np.ma.dstack if isinstance(array, np.ma.MaskedArray) else np.dstack
    tiled = np.moveaxis(dstack(tile_list), -1, 0)

    return tiled

untile_array(tiled_array, array_shape)

Untile a tiled array into a 2D numpy array

This is the reverse of tile_array and will turn a tiled array like: >>> tiled_array = [[[0,0], ... [0,0]], ... [[1,1], ... [1,1]], ... [[2,2], ... [2,2]], ... [[3,3], ... [3,3]]] >>> tiled_array.shape (4, 2, 2)

into a 2D array like

array = untile_array(tiled_array) print(array) [[0, 0, 1, 1], [0, 0, 1, 1], [2, 2, 3, 3], [2, 2, 3, 3]] array.shape (4, 4)

Parameters:

Name	Type	Description	Default
tiled_array	Union[np.ndarray, np.ma.MaskedArray]	a tiled array	required
array_shape	Tuple[int, int]	shape to untile the array to. If array_shape's size is smaller than the tiled array, untile_array will subset the tiled array assuming bottom right padding was added when tiling.	required

Returns:

Type	Description
Union[np.ndarray, np.ma.MaskedArray]	the untiled array

Source code in asf_tools/tile.py

def untile_array(tiled_array: Union[np.ndarray, np.ma.MaskedArray], array_shape: Tuple[int, int]) \
        -> Union[np.ndarray, np.ma.MaskedArray]:
    """Untile a tiled array into a 2D numpy array

    This is the reverse of `tile_array` and will turn a tiled array like:
        >>> tiled_array = [[[0,0],
        ...                 [0,0]],
        ...                [[1,1],
        ...                 [1,1]],
        ...                [[2,2],
        ...                 [2,2]],
        ...                [[3,3],
        ...                 [3,3]]]
        >>> tiled_array.shape
        (4, 2, 2)

    into a 2D array like:
        >>> array = untile_array(tiled_array)
        >>> print(array)
        [[0, 0, 1, 1],
         [0, 0, 1, 1],
         [2, 2, 3, 3],
         [2, 2, 3, 3]]
        >>> array.shape
        (4, 4)

    Args:
        tiled_array: a tiled array
        array_shape: shape to untile the array to. If array_shape's size is smaller
            than the tiled array, `untile_array` will subset the tiled array assuming
            bottom right padding was added when tiling.

    Returns:
        the untiled array
    """
    _, tile_rows, tile_columns = tiled_array.shape
    array_rows, array_columns = array_shape

    untiled_rows = int(np.ceil(array_rows / tile_rows))
    untiled_columns = int(np.ceil(array_columns / tile_columns))

    untiled = np.zeros((untiled_rows*tile_rows, untiled_columns*tile_columns), dtype=tiled_array.dtype)

    if (array_size := array_rows * array_columns) > tiled_array.size:
        raise ValueError(
            f'array_shape {array_shape} will result in an array bigger than the tiled array:'
            f' {array_size} > {tiled_array.size}'
        )

    for ii in range(untiled_rows):
        for jj in range(untiled_columns):
            untiled[ii*tile_rows:(ii+1)*tile_rows, jj*tile_columns:(jj+1)*tile_columns] = \
                tiled_array[ii * untiled_columns + jj, :, :]

    if isinstance(tiled_array, np.ma.MaskedArray):
        untiled_mask = untile_array(tiled_array.mask, untiled.shape)
        untiled = np.ma.MaskedArray(untiled, mask=untiled_mask)

    return untiled[:array_rows, :array_columns]

util

GDALConfigManager

Context manager for setting GDAL config options temporarily

Source code in asf_tools/util.py

class GDALConfigManager:
    """Context manager for setting GDAL config options temporarily"""
    def __init__(self, **options):
        """
        Args:
            **options: GDAL Config `option=value` keyword arguments.
        """
        self.options = options.copy()
        self._previous_options = {}

    def __enter__(self):
        for key in self.options:
            self._previous_options[key] = gdal.GetConfigOption(key)

        for key, value in self.options.items():
            gdal.SetConfigOption(key, value)

    def __exit__(self, exc_type, exc_val, exc_tb):
        for key, value in self._previous_options.items():
            gdal.SetConfigOption(key, value)

__init__(**options)

Parameters:

Name	Type	Description	Default
**options		GDAL Config option=value keyword arguments.	{}

Source code in asf_tools/util.py

def __init__(self, **options):
    """
    Args:
        **options: GDAL Config `option=value` keyword arguments.
    """
    self.options = options.copy()
    self._previous_options = {}

water_map

Generate surface water maps from Sentinel-1 RTC products

Create a surface water extent map from a dual-pol Sentinel-1 RTC product and a HAND image. The HAND image must be pixel-aligned (same extent and size) to the RTC images. The water extent maps are created using an adaptive Expectation Maximization thresholding approach and refined using Fuzzy Logic.

make_water_map(out_raster, vv_raster, vh_raster, hand_raster=None, tile_shape=(100, 100), max_vv_threshold=-15.5, max_vh_threshold=-23.0, hand_threshold=15.0, hand_fraction=0.8, membership_threshold=0.45)

Creates a surface water extent map from a Sentinel-1 RTC product

Create a surface water extent map from a dual-pol Sentinel-1 RTC product and a HAND image. The HAND image must be pixel-aligned (same extent and size) to the RTC images. The water extent maps are created using an adaptive Expectation Maximization thresholding approach and refined with Fuzzy Logic.

The input images are broken into a set of corresponding tiles with a shape of tile_shape, and a set of tiles are selected from the VH RTC image that contain water boundaries to determine an appropriate water threshold. Candidate tiles must meet these criteria: * hand_fraction of pixels within a tile must have HAND pixel values lower than hand_threshold * The median backscatter value for the tile must be lower than an average tiles' backscatter values * The tile must have a high variance -- high variance is considered initially to be a variance in the 95th percentile of the tile variances, but progressively relaxed to the 5th percentile if there not at least 5 candidate tiles.

The 5 VH tiles with the highest variance are selected for thresholding and a water threshold value is determined using an Expectation Maximization approach. If there were not enough candidate tiles or the threshold is too high, max_vh_threshold and/or max_vv_threshold will be used instead.

From the initial threshold-based water extent maps, Fuzzy Logic is used to remove spurious false detections and improve the water extent map quality. The fuzzy logic uses these indicators for the presence of water: * radar cross section in a pixel relative to the determined detection threshold * the height above nearest drainage (HAND) * the surface slope, which is derived from the HAND data * the size of the detected water feature

For each indicator, a Z-shaped activation function is used to determine pixel membership. The membership maps are combined to form the final water extent map. Pixels classified as water pixels will: * have non-zero membership in all of the indicators, and * have an average membership above the membership_threshold value.

Finally, the VV and VH water masks will be combined to include all water pixels from both masks, and the combined water map will be written to out_raster.

Parameters:

Name	Type	Description	Default
out_raster	Union[str, Path]	Water map GeoTIFF to create	required
vv_raster	Union[str, Path]	Sentinel-1 RTC GeoTIFF, in power scale, with VV polarization	required
vh_raster	Union[str, Path]	Sentinel-1 RTC GeoTIFF, in power scale, with VH polarization	required
hand_raster	Optional[Union[str, Path]]	Height Above Nearest Drainage (HAND) GeoTIFF aligned to the RTC rasters	None
tile_shape	Tuple[int, int]	shape (height, width) in pixels to tile the image to	(100, 100)
max_vv_threshold	float	Maximum threshold value to use for vv_raster in decibels (db)	-15.5
max_vh_threshold	float	Maximum threshold value to use for vh_raster in decibels (db)	-23.0
hand_threshold	float	The maximum height above nearest drainage in meters to consider a pixel valid	15.0
hand_fraction	float	The minimum fraction of valid HAND pixels required in a tile for thresholding	0.8
membership_threshold	float	The average membership to the fuzzy indicators required for a water pixel	0.45

Source code in asf_tools/water_map.py

def make_water_map(out_raster: Union[str, Path], vv_raster: Union[str, Path], vh_raster: Union[str, Path],
                   hand_raster: Optional[Union[str, Path]] = None, tile_shape: Tuple[int, int] = (100, 100),
                   max_vv_threshold: float = -15.5, max_vh_threshold: float = -23.0,
                   hand_threshold: float = 15., hand_fraction: float = 0.8, membership_threshold: float = 0.45):
    """Creates a surface water extent map from a Sentinel-1 RTC product

    Create a surface water extent map from a dual-pol Sentinel-1 RTC product and
    a HAND image. The HAND image must be pixel-aligned (same extent and size) to
    the RTC images. The water extent maps are created using an adaptive Expectation
    Maximization thresholding approach and refined with Fuzzy Logic.

    The input images are broken into a set of corresponding tiles with a shape of
    `tile_shape`, and a set of tiles are selected from the VH RTC
    image that contain water boundaries to determine an appropriate water threshold.
     Candidate tiles must meet these criteria:
    * `hand_fraction` of pixels within a tile must have HAND pixel values lower
      than `hand_threshold`
    * The median backscatter value for the tile must be lower than an average tiles'
      backscatter values
    * The tile must have a high variance -- high variance is considered initially to
      be a variance in the 95th percentile of the tile variances, but progressively
      relaxed to the 5th percentile if there not at least 5 candidate tiles.

    The 5 VH tiles with the highest variance are selected for thresholding and a
    water threshold value is determined using an Expectation Maximization approach.
    If there were not enough candidate tiles or the threshold is too high,
    `max_vh_threshold` and/or `max_vv_threshold` will be used instead.

    From the initial threshold-based water extent maps, Fuzzy Logic is used to remove
    spurious false detections and improve the water extent map quality. The fuzzy logic
    uses these indicators for the presence of water:
    * radar cross section in a pixel relative to the determined detection threshold
    * the height above nearest drainage (HAND)
    * the surface slope, which is derived from the HAND data
    * the size of the detected water feature

    For each indicator, a Z-shaped activation function is used to determine pixel membership.
    The membership maps are combined to form the final water extent map. Pixels classified
    as water pixels will:
    * have non-zero membership in all of the indicators, and
    * have an average membership above the `membership_threshold` value.

    Finally, the VV and VH water masks will be combined to include all water pixels
    from both masks, and the combined water map will be written to `out_raster`.

    Args:
        out_raster: Water map GeoTIFF to create
        vv_raster: Sentinel-1 RTC GeoTIFF, in power scale, with VV polarization
        vh_raster: Sentinel-1 RTC GeoTIFF, in power scale, with VH polarization
        hand_raster: Height Above Nearest Drainage (HAND) GeoTIFF aligned to the RTC rasters
        tile_shape: shape (height, width) in pixels to tile the image to
        max_vv_threshold: Maximum threshold value to use for `vv_raster` in decibels (db)
        max_vh_threshold:  Maximum threshold value to use for `vh_raster` in decibels (db)
        hand_threshold: The maximum height above nearest drainage in meters to consider
            a pixel valid
        hand_fraction: The minimum fraction of valid HAND pixels required in a tile for
            thresholding
        membership_threshold: The average membership to the fuzzy indicators required for a water pixel
    """
    if tile_shape[0] % 2 or tile_shape[1] % 2:
        raise ValueError(f'tile_shape {tile_shape} requires even values.')

    info = gdal.Info(str(vh_raster), format='json')

    out_transform = info['geoTransform']
    out_epsg = get_epsg_code(info)

    if hand_raster is None:
        hand_raster = str(out_raster).replace('.tif', '_HAND.tif')
        log.info(f'Extracting HAND data to: {hand_raster}')
        prepare_hand_for_raster(hand_raster, vh_raster)

    log.info(f'Determining HAND memberships from {hand_raster}')
    hand_array = read_as_masked_array(hand_raster)
    hand_tiles = tile_array(hand_array, tile_shape=tile_shape, pad_value=np.nan)

    hand_candidates = select_hand_tiles(hand_tiles, hand_threshold, hand_fraction)
    log.debug(f'Selected HAND tile candidates {hand_candidates}')

    selected_tiles = None
    water_extent_maps = []
    for max_db_threshold, raster, pol in ((max_vh_threshold, vh_raster, 'VH'), (max_vv_threshold, vv_raster, 'VV')):
        log.info(f'Creating initial {pol} water extent map from {raster}')
        array = read_as_masked_array(raster)
        tiles = tile_array(array, tile_shape=tile_shape, pad_value=0.)
        # Masking less than zero only necessary for old HyP3/GAMMA products which sometimes returned negative powers
        tiles = np.ma.masked_less_equal(tiles, 0.)
        if selected_tiles is None:
            selected_tiles = select_backscatter_tiles(tiles, hand_candidates)
            log.info(f'Selected tiles {selected_tiles} from {raster}')

        with np.testing.suppress_warnings() as sup:
            sup.filter(RuntimeWarning)  # invalid value and divide by zero encountered in log10
            tiles = np.log10(tiles) + 30.  # linear power scale -> Gaussian scale optimized for thresholding
        max_gaussian_threshold = max_db_threshold / 10. + 30.  # db -> Gaussian scale optimized for thresholding
        if selected_tiles.size:
            scaling = 256 / (np.mean(tiles) + 3 * np.std(tiles))
            gaussian_threshold = determine_em_threshold(tiles[selected_tiles, :, :], scaling)
            threshold_db = 10. * (gaussian_threshold - 30.)
            log.info(f'Threshold determined to be {threshold_db} db')
            if gaussian_threshold > max_gaussian_threshold:
                log.warning(f'Threshold too high! Using maximum threshold {max_db_threshold} db')
                gaussian_threshold = max_gaussian_threshold
        else:
            log.warning(f'Tile selection did not converge! using default threshold {max_db_threshold} db')
            gaussian_threshold = max_gaussian_threshold

        gaussian_array = untile_array(tiles, array.shape)
        water_map = np.ma.masked_less_equal(gaussian_array, gaussian_threshold).mask
        water_map &= ~array.mask

        write_cog(str(out_raster).replace('.tif', f'_{pol}_initial.tif'), water_map, transform=out_transform,
                  epsg_code=out_epsg, dtype=gdal.GDT_Byte, nodata_value=False)

        log.info(f'Refining initial {pol} water extent map using Fuzzy Logic')
        array = np.ma.masked_where(~water_map, array)
        gaussian_lower_limit = np.log10(np.ma.median(array)) + 30.

        water_map = fuzzy_refinement(
            water_map, gaussian_array, hand_array, pixel_size=out_transform[1],
            gaussian_thresholds=(gaussian_lower_limit, gaussian_threshold), membership_threshold=membership_threshold
        )
        water_map &= ~array.mask

        write_cog(str(out_raster).replace('.tif', f'_{pol}_fuzzy.tif'), water_map, transform=out_transform,
                  epsg_code=out_epsg, dtype=gdal.GDT_Byte, nodata_value=False)

        water_extent_maps.append(water_map)

    log.info('Combining Fuzzy VH and VV extent map')
    combined_water_map = np.logical_or(*water_extent_maps)

    combined_segments = measure.label(combined_water_map, connectivity=2)
    combined_water_map = remove_small_segments(combined_segments)

    write_cog(out_raster, combined_water_map, transform=out_transform,
              epsg_code=out_epsg, dtype=gdal.GDT_Byte, nodata_value=False)