def retrieve_data(dataset, pq, pq_masks, path, x, y, overwrite=False, stack=False):
_log.info("Retrieving data from [%s] with pq [%s] and pq mask [%s] to [%s]", dataset.path, pq and pq.path or "", pq and pq_masks or "", path)
if os.path.exists(path) and not overwrite:
_log.error("Output file [%s] exists", path)
raise Exception("Output file [%s] already exists" % path)
data = None
metadata = get_dataset_metadata(dataset)
if pq:
data = get_dataset_data_with_pq(dataset, pq, pq_masks=pq_masks)
else:
data = get_dataset_data(dataset)
_log.debug("data is [%s]", data)
raster_create(path, [data[b] for b in dataset.bands],
metadata.transform, metadata.projection, NDV, gdal.GDT_Int16)
# If we are creating a stack then also add to a file list file...
if stack:
path_file_list = os.path.join(os.path.dirname(path), get_filename_file_list(dataset.satellite, dataset.dataset_type, x, y))
_log.info("Also going to write file list to [%s]", path_file_list)
with open(path_file_list, "ab") as f:
print >>f, path
def create_water_tile(self, tile):
arg = tile.datasets[DatasetType.ARG25]
pqa = tile.datasets[DatasetType.PQ25]
_log.info("ARG tile [%s]", arg)
_log.info("PQ tile [%s]", pqa)
filename = os.path.basename(arg.path)
filename = filename.replace("NBAR", "WETNESS")
filename = filename.replace(".vrt", ".tif")
filename = os.path.join(self.output_directory, filename)
metadata = get_dataset_metadata(arg)
data = get_dataset_data_with_pq(arg, Ls57Arg25Bands, pqa)
# Calculate TCI Wetness
tci = calculate_tassel_cap_index(data,
coefficients=TCI_COEFFICIENTS[arg.satellite][TasselCapIndex.WETNESS])
_log.info("TCI shape is %s | min = %s | max = %s", numpy.shape(tci), tci.min(), tci.max())
raster_create(filename,
[tci],
metadata.transform, metadata.projection, numpy.nan, gdal.GDT_Float32)
def generate_derived_nbar(self, dataset_types, nbar, pqa, pqa_masks, overwrite=False):
for dataset_type in dataset_types:
filename = self.get_output_filename_derived_nbar(nbar, dataset_type)
_log.info("Generating data from [%s] with pq [%s] and pq mask [%s] to [%s]", nbar.path, pqa and pqa.path or "", pqa and pqa_masks or "", filename)
metadata = get_dataset_metadata(nbar)
data = None
if pqa:
data = get_dataset_data_with_pq(nbar, pqa, pq_masks=pqa_masks)
else:
data = get_dataset_data(nbar)
_log.debug("data is [%s]", data)
if dataset_type == DatasetType.NDVI:
ndvi = calculate_ndvi(data[nbar.bands.RED], data[nbar.bands.NEAR_INFRARED])
raster_create(filename, [ndvi], metadata.transform, metadata.projection, NDV, gdal.GDT_Float32)
elif dataset_type == DatasetType.EVI:
evi = calculate_evi(data[nbar.bands.RED], data[nbar.bands.BLUE], data[nbar.bands.NEAR_INFRARED])
raster_create(filename, [evi], metadata.transform, metadata.projection, NDV, gdal.GDT_Float32)
elif dataset_type == DatasetType.NBR:
nbr = calculate_nbr(data[nbar.bands.NEAR_INFRARED], data[nbar.bands.SHORT_WAVE_INFRARED_2])
raster_create(filename, [nbr], metadata.transform, metadata.projection, NDV, gdal.GDT_Float32)
def retrieve_pixel_value(dataset, pq, pq_masks, latitude, longitude, ndv=NDV):
_log.debug("Retrieving pixel value(s) at lat=[%f] lon=[%f] from [%s] with pq [%s] and pq mask [%s]", latitude, longitude, dataset.path, pq and pq.path or "", pq and pq_masks or "")
metadata = get_dataset_metadata(dataset)
x, y = latlon_to_xy(latitude, longitude, metadata.transform)
_log.debug("Retrieving value at x=[%d] y=[%d]", x, y)
data = None
if pq:
data = get_dataset_data_with_pq(dataset, pq, x=x, y=y, x_size=1, y_size=1, pq_masks=pq_masks, ndv=ndv)
else:
data = get_dataset_data(dataset, x=x, y=y, x_size=1, y_size=1)
_log.debug("data is [%s]", data)
return data
def test_retrieve_data_ls5_arg_with_pqa(config=None):
filename = "LS5_TM_NBAR_WITH_PQA_{x:03d}_{y:04d}_{date}.{x_offset:04d}_{y_offset:04d}.{x_size:04d}x{y_size:04d}.tif".format(x=CELL_X, y=CELL_Y, date=DATE, x_offset=X_OFFSET, y_offset=Y_OFFSET, x_size=X_SIZE, y_size=Y_SIZE)
tiles = list_tiles_as_list(x=[CELL_X], y=[CELL_Y],
acq_min=ACQ_LS5, acq_max=ACQ_LS5,
satellites=[Satellite.LS5],
dataset_types=[ARG_DATASET_TYPE, PQ_DATASET_TYPE],
config=config)
assert len(tiles) == 1
tile = tiles[0]
assert ARG_DATASET_TYPE in tile.datasets
dataset = tile.datasets[ARG_DATASET_TYPE]
assert PQ_DATASET_TYPE in tile.datasets
pqa = tile.datasets[PQ_DATASET_TYPE]
data = get_dataset_data_with_pq(dataset=dataset, dataset_pqa=pqa, x=X_OFFSET, y=Y_OFFSET, x_size=X_SIZE, y_size=Y_SIZE)
assert(data)
_log.info("data is [%s]\n%s", numpy.shape(data), data)
ndv = get_dataset_ndv(dataset)
assert(is_ndv(ndv, ARG_NDV))
data_type = get_dataset_datatype(dataset)
assert(data_type == ARG_DATA_TYPE)
metadata = generate_dataset_metadata(x=CELL_X, y=CELL_Y, acq_dt=ACQ_LS5,
dataset=dataset, bands=None,
mask_pqa_apply=False, mask_pqa_mask=None,
mask_wofs_apply=False, mask_wofs_mask=None)
raster_create_geotiff(filename, [data[b] for b in dataset.bands], CELL_GEO_TRANSFORM, CELL_PROJECTION, ndv, data_type,
dataset_metadata=metadata, band_ids=[b.name for b in dataset.bands])
assert filecmp.cmp(filename, get_test_data_path(filename))
def run(self):
self.parse_arguments()
config = Config()
_log.debug(config.to_str())
path = self.get_output_filename(self.dataset_type)
_log.info("Output file is [%s]", path)
if os.path.exists(path):
if self.overwrite:
_log.info("Removing existing output file [%s]", path)
os.remove(path)
else:
_log.error("Output file [%s] exists", path)
raise Exception("Output file [%s] already exists" % path)
# TODO
bands = get_bands(self.dataset_type, self.satellites[0])
# TODO once WOFS is in the cube
tiles = list_tiles_as_list(x=[self.x], y=[self.y], acq_min=self.acq_min, acq_max=self.acq_max,
satellites=[satellite for satellite in self.satellites],
dataset_types=[self.dataset_type],
database=config.get_db_database(),
user=config.get_db_username(),
password=config.get_db_password(),
host=config.get_db_host(), port=config.get_db_port())
raster = None
metadata = None
# TODO - PQ is UNIT16 (others are INT16) and so -999 NDV doesn't work
ndv = self.dataset_type == DatasetType.PQ25 and UINT16_MAX or NDV
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
import itertools
for x, y in itertools.product(range(0, 4000, self.chunk_size_x), range(0, 4000, self.chunk_size_y)):
_log.info("About to read data chunk ({xmin:4d},{ymin:4d}) to ({xmax:4d},{ymax:4d})".format(xmin=x, ymin=y, xmax=x+self.chunk_size_x-1, ymax=y+self.chunk_size_y-1))
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
stack = dict()
for tile in tiles:
if self.list_only:
_log.info("Would summarise dataset [%s]", tile.datasets[self.dataset_type].path)
continue
pqa = None
_log.debug("Reading dataset [%s]", tile.datasets[self.dataset_type].path)
if not metadata:
metadata = get_dataset_metadata(tile.datasets[self.dataset_type])
# Apply PQA if specified
if self.apply_pqa_filter:
data = get_dataset_data_with_pq(tile.datasets[self.dataset_type], tile.datasets[DatasetType.PQ25], bands=bands, x=x, y=y, x_size=self.chunk_size_x, y_size=self.chunk_size_y, pq_masks=self.pqa_mask, ndv=ndv)
else:
data = get_dataset_data(tile.datasets[self.dataset_type], bands=bands, x=x, y=y, x_size=self.chunk_size_x, y_size=self.chunk_size_y)
for band in bands:
if band in stack:
stack[band].append(data[band])
else:
stack[band] = [data[band]]
_log.debug("data[%s] has shape [%s] and MB [%s]", band.name, numpy.shape(data[band]), data[band].nbytes/1000/1000)
_log.debug("stack[%s] has [%s] elements", band.name, len(stack[band]))
# Apply summary method
_log.info("Finished reading {count} datasets for chunk ({xmin:4d},{ymin:4d}) to ({xmax:4d},{ymax:4d}) - about to summarise them".format(count=len(tiles), xmin=x, ymin=y, xmax=x+self.chunk_size_x-1, ymax=y+self.chunk_size_y-1))
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
masked_stack = dict()
for band in bands:
masked_stack[band] = numpy.ma.masked_equal(stack[band], ndv)
_log.debug("masked_stack[%s] is %s", band.name, masked_stack[band])
_log.debug("masked stack[%s] has shape [%s] and MB [%s]", band.name, numpy.shape(masked_stack[band]), masked_stack[band].nbytes/1000/1000)
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
if self.summary_method == TimeSeriesSummaryMethod.MIN:
masked_summary = numpy.min(masked_stack[band], axis=0)
elif self.summary_method == TimeSeriesSummaryMethod.MAX:
masked_summary = numpy.max(masked_stack[band], axis=0)
elif self.summary_method == TimeSeriesSummaryMethod.MEAN:
masked_summary = numpy.mean(masked_stack[band], axis=0)
elif self.summary_method == TimeSeriesSummaryMethod.MEDIAN:
masked_summary = numpy.median(masked_stack[band], axis=0)
# aka 50th percentile
elif self.summary_method == TimeSeriesSummaryMethod.MEDIAN_NON_INTERPOLATED:
masked_sorted = numpy.ma.sort(masked_stack[band], axis=0)
masked_percentile_index = numpy.ma.floor(numpy.ma.count(masked_sorted, axis=0) * 0.95).astype(numpy.int16)
masked_summary = numpy.ma.choose(masked_percentile_index, masked_sorted)
elif self.summary_method == TimeSeriesSummaryMethod.COUNT:
# TODO Need to artificially create masked array here since it is being expected/filled below!!!
masked_summary = numpy.ma.masked_equal(masked_stack[band].count(axis=0), ndv)
elif self.summary_method == TimeSeriesSummaryMethod.SUM:
masked_summary = numpy.sum(masked_stack[band], axis=0)
elif self.summary_method == TimeSeriesSummaryMethod.STANDARD_DEVIATION:
masked_summary = numpy.std(masked_stack[band], axis=0)
elif self.summary_method == TimeSeriesSummaryMethod.VARIANCE:
masked_summary = numpy.var(masked_stack[band], axis=0)
# currently 95th percentile
elif self.summary_method == TimeSeriesSummaryMethod.PERCENTILE:
masked_sorted = numpy.ma.sort(masked_stack[band], axis=0)
masked_percentile_index = numpy.ma.floor(numpy.ma.count(masked_sorted, axis=0) * 0.95).astype(numpy.int16)
masked_summary = numpy.ma.choose(masked_percentile_index, masked_sorted)
elif self.summary_method == TimeSeriesSummaryMethod.YOUNGEST_PIXEL:
# TODO the fact that this is band at a time might be problematic. We really should be considering
# all bands at once (that is what the landsat_mosaic logic did). If PQA is being applied then
# it's probably all good but if not then we might get odd results....
masked_summary = empty_array(shape=(self.chunk_size_x, self.chunk_size_x), dtype=numpy.int16, ndv=ndv)
# Note the reversed as the stack is created oldest first
for d in reversed(stack[band]):
masked_summary = numpy.where(masked_summary == ndv, d, masked_summary)
# If the summary doesn't contain an no data values then we can stop
if not numpy.any(masked_summary == ndv):
break
# TODO Need to artificially create masked array here since it is being expected/filled below!!!
masked_summary = numpy.ma.masked_equal(masked_summary, ndv)
elif self.summary_method == TimeSeriesSummaryMethod.OLDEST_PIXEL:
# TODO the fact that this is band at a time might be problematic. We really should be considering
# all bands at once (that is what the landsat_mosaic logic did). If PQA is being applied then
# it's probably all good but if not then we might get odd results....
masked_summary = empty_array(shape=(self.chunk_size_x, self.chunk_size_x), dtype=numpy.int16, ndv=ndv)
# Note the NOT reversed as the stack is created oldest first
for d in stack[band]:
masked_summary = numpy.where(masked_summary == ndv, d, masked_summary)
# If the summary doesn't contain an no data values then we can stop
if not numpy.any(masked_summary == ndv):
break
# TODO Need to artificially create masked array here since it is being expected/filled below!!!
masked_summary = numpy.ma.masked_equal(masked_summary, ndv)
masked_stack[band] = None
_log.debug("NONE-ing masked stack[%s]", band.name)
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
_log.debug("masked summary is [%s]", masked_summary)
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
# Create the output file
if not os.path.exists(path):
_log.info("Creating raster [%s]", path)
driver = gdal.GetDriverByName("GTiff")
assert driver
raster = driver.Create(path, metadata.shape[0], metadata.shape[1], len(bands), gdal.GDT_Int16)
assert raster
raster.SetGeoTransform(metadata.transform)
raster.SetProjection(metadata.projection)
for b in bands:
raster.GetRasterBand(b.value).SetNoDataValue(ndv)
_log.info("Writing band [%s] data to raster [%s]", band.name, path)
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
raster.GetRasterBand(band.value).WriteArray(masked_summary.filled(ndv), xoff=x, yoff=y)
raster.GetRasterBand(band.value).ComputeStatistics(True)
raster.FlushCache()
masked_summary = None
_log.debug("NONE-ing the masked summary")
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
stack = None
_log.debug("Just NONE-ed the stack")
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
raster = None
_log.debug("Just NONE'd the raster")
_log.debug("Current MAX RSS usage is [%d] MB", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
_log.info("Memory usage was [%d MB]", resource.getrusage(resource.RUSAGE_SELF).ru_maxrss / 1024)
_log.info("CPU time used [%s]", timedelta(seconds=int(resource.getrusage(resource.RUSAGE_SELF).ru_utime)))
def doit(self):
shape = (4000, 4000)
no_data_value = NDV
best_pixel_data = dict()
# TODO
if Satellite.LS8.value in self.satellites:
bands = Ls8Arg25Bands
else:
bands = Ls57Arg25Bands
for band in bands:
best_pixel_data[band] = empty_array(shape=shape, dtype=numpy.int16, ndv=no_data_value)
best_pixel_satellite = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
# best_pixel_epoch = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
best_pixel_date = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
current_satellite = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
# current_epoch = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
current_date = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
metadata = None
SATELLITE_DATA_VALUES = {Satellite.LS5: 5, Satellite.LS7: 7, Satellite.LS8: 8}
for tile in self.get_tiles(sort=SortType.DESC):
# Get ARG25 dataset
dataset = tile.datasets[DatasetType.ARG25]
_log.info("Processing ARG tile [%s]", dataset.path)
if not metadata:
metadata = get_dataset_metadata(dataset)
band_data = None
if self.apply_pq_filter:
band_data = get_dataset_data_with_pq(dataset, tile.datasets[DatasetType.PQ25])
else:
band_data = get_dataset_data(dataset)
# Create the provenance datasets
# NOTE: need to do this BEFORE selecting the pixel since it is actually using the fact that the
# selected pixel currently doesn't have a value
# NOTE: band values are propagated "as a job lot" so can just check any band
# TODO better way than just saying....RED....?
band = bands.RED
# Satellite
current_satellite.fill(SATELLITE_DATA_VALUES[dataset.satellite])
best_pixel_satellite = numpy.where(best_pixel_data[band] == no_data_value, current_satellite, best_pixel_satellite)
# # Epoch dataset
#
# current_epoch.fill(calendar.timegm(tile.end_datetime.timetuple()))
# best_pixel_epoch = numpy.where(best_pixel_data[band] == no_data_value, current_epoch, best_pixel_epoch)
# Date dataset (20150101)
current_date.fill(tile.end_datetime.year * 10000 + tile.end_datetime.month * 100 + tile.end_datetime.day)
best_pixel_date = numpy.where(best_pixel_data[band] == no_data_value, current_date, best_pixel_date)
for band in bands:
data = band_data[band]
# _log.debug("data = \n%s", data)
# Replace any NO DATA best pixels with data pixels
# TODO should I explicitly do the AND data is not NO DATA VALUE?
best_pixel_data[band] = numpy.where(best_pixel_data[band] == no_data_value, data, best_pixel_data[band])
# _log.debug("best pixel = \n%s", best_pixel_data[band])
still_no_data = numpy.any(numpy.array([best_pixel_data[b] for b in bands]) == no_data_value)
# _log.debug("still no data pixels = %s", still_no_data)
if not still_no_data:
break
# Now want to mask out values in the provenance datasets if we haven't actually got a value
# TODO better way than just saying....RED....?
band = bands.RED
mask = numpy.ma.masked_equal(best_pixel_data[band], NDV).mask
best_pixel_satellite = numpy.ma.array(best_pixel_satellite, mask=mask).filled(NDV)
# best_pixel_epoch = numpy.ma.array(best_pixel_epoch, mask=mask).fill(NDV)
best_pixel_date = numpy.ma.array(best_pixel_date, mask=mask).filled(NDV)
# Composite NBAR dataset
raster_create(self.get_output_path("NBAR"), [best_pixel_data[b] for b in bands],
metadata.transform, metadata.projection, NDV, gdal.GDT_Int16)
# Provenance (satellite) dataset
raster_create(self.get_output_path("SAT"),
[best_pixel_satellite],
metadata.transform, metadata.projection, no_data_value,
gdal.GDT_Int16)
# # Provenance (epoch) dataset
#
# raster_create(self.get_output_path("EPOCH"),
# [best_pixel_epoch],
# metadata.transform, metadata.projection, no_data_value,
# gdal.GDT_Int32)
# Provenance (day of month) dataset
raster_create(self.get_output_path("DATE"),
[best_pixel_date],
metadata.transform, metadata.projection, no_data_value,
gdal.GDT_Int32)
def doit(self):
_log.debug("Bare Soil Cell Task - doit()")
shape = (4000, 4000)
no_data_value = NDV
best_pixel_fc = dict()
for band in Fc25Bands:
best_pixel_fc[band] = empty_array(shape=shape, dtype=numpy.int16, ndv=INT16_MIN)
best_pixel_nbar = dict()
for band in Ls57Arg25Bands:
best_pixel_nbar[band] = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
best_pixel_satellite = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
best_pixel_year = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
best_pixel_month = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
best_pixel_epoch = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
current_satellite = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
current_year = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
current_month = empty_array(shape=shape, dtype=numpy.int16, ndv=NDV)
current_epoch = empty_array(shape=shape, dtype=numpy.int32, ndv=NDV)
SATELLITE_DATA_VALUES = {Satellite.LS5: 5, Satellite.LS7: 7, Satellite.LS8: 8}
metadata_nbar = None
metadata_fc = None
for tile in self.get_tiles():
# Get the PQ mask
pq = tile.datasets[DatasetType.PQ25]
data_pq = get_dataset_data(pq, [Pq25Bands.PQ])[Pq25Bands.PQ]
mask_pq = get_pq_mask(data_pq)
# Get NBAR dataset
nbar = tile.datasets[DatasetType.ARG25]
_log.info("Processing NBAR tile [%s]", nbar.path)
if not metadata_nbar:
metadata_nbar = get_dataset_metadata(nbar)
data_nbar = get_dataset_data_with_pq(nbar, Ls57Arg25Bands, tile.datasets[DatasetType.PQ25])
# Get the NDVI mask
red = data_nbar[Ls57Arg25Bands.RED]
nir = data_nbar[Ls57Arg25Bands.NEAR_INFRARED]
ndvi_data = calculate_ndvi(red, nir)
ndvi_data = numpy.ma.masked_equal(ndvi_data, NDV)
ndvi_data = numpy.ma.masked_outside(ndvi_data, 0, 0.3, copy=False)
mask_ndvi = ndvi_data.mask
# Get FC25 dataset
fc = tile.datasets[DatasetType.FC25]
_log.info("Processing FC tile [%s]", fc.path)
if not metadata_fc:
metadata_fc = get_dataset_metadata(fc)
_log.debug("metadata fc is %s", metadata_fc)
data_fc = get_dataset_data(fc, Fc25Bands)
data_bare_soil = data_fc[Fc25Bands.BS]
data_bare_soil = numpy.ma.masked_equal(data_bare_soil, -999)
data_bare_soil = numpy.ma.masked_outside(data_bare_soil, 0, 8000)
data_bare_soil.mask = (data_bare_soil.mask | mask_pq | mask_ndvi)
data_bare_soil = data_bare_soil.filled(NDV)
# Compare the bare soil value from this dataset to the current "best" value
best_pixel_fc[Fc25Bands.BS] = numpy.fmax(best_pixel_fc[Fc25Bands.BS], data_bare_soil)
# Now update the other best pixel datasets/bands to grab the pixels we just selected
for band in Ls57Arg25Bands:
best_pixel_nbar[band] = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
data_nbar[band],
best_pixel_nbar[band])
for band in [Fc25Bands.PV, Fc25Bands.NPV, Fc25Bands.ERROR]:
best_pixel_fc[band] = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
data_fc[band],
best_pixel_fc[band])
# And now the other "provenance" data
current_satellite.fill(SATELLITE_DATA_VALUES[fc.satellite])
best_pixel_satellite = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
current_satellite,
best_pixel_satellite)
current_year.fill(tile.end_datetime_year)
best_pixel_year = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
current_year,
best_pixel_year)
current_month.fill(tile.end_datetime_month)
best_pixel_month = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
current_month,
best_pixel_month)
current_epoch.fill(calendar.timegm(tile.end_datetime.timetuple()))
best_pixel_epoch = propagate_using_selected_pixel(best_pixel_fc[Fc25Bands.BS],
data_bare_soil,
current_epoch,
best_pixel_epoch)
# Create the output datasets
# FC composite
raster_create(self.get_dataset_filename("FC"),
[best_pixel_fc[b] for b in Fc25Bands],
metadata_fc.transform, metadata_fc.projection, metadata_fc.bands[Fc25Bands.BS].no_data_value,
metadata_fc.bands[Fc25Bands.BS].data_type)
# NBAR composite
raster_create(self.get_dataset_filename("NBAR"),
[best_pixel_nbar[b] for b in Ls57Arg25Bands],
metadata_nbar.transform, metadata_nbar.projection,
metadata_nbar.bands[Ls57Arg25Bands.BLUE].no_data_value,
metadata_nbar.bands[Ls57Arg25Bands.BLUE].data_type)
# "Provenance" composites
raster_create(self.get_dataset_filename("SAT"),
[best_pixel_satellite],
metadata_nbar.transform, metadata_nbar.projection, no_data_value,
gdal.GDT_Int16)
raster_create(self.get_dataset_filename("YEAR"),
[best_pixel_year],
metadata_nbar.transform, metadata_nbar.projection, no_data_value,
gdal.GDT_Int16)
raster_create(self.get_dataset_filename("MONTH"),
[best_pixel_month],
metadata_nbar.transform, metadata_nbar.projection, no_data_value,
gdal.GDT_Int16)
raster_create(self.get_dataset_filename("EPOCH"),
[best_pixel_epoch],
metadata_nbar.transform, metadata_nbar.projection, no_data_value,
gdal.GDT_Int32)
def zonal_stats(dataset, rasterised_fname, dataset_type):
"""
Computes the Observed Count, Min, Max, Sum and Sum of Squares for
the segments defined by the rasterised image. The stats are
derived from the `dataset` defined by the `dataset_type`.
:param dataset:
A class of type `Dataset`.
:param rasterised_fname:
A string containing the full file pathname of an image
containing the rasterised features. These features will be
interpreted as segments.
:param dataset_type:
A class of type `DatasetType`.
:return:
A `pandas.DataFrame` containing the statistics for each segment
and for each raster band contained with the `dataset_type`.
"""
# Initialiase a blank dataframe
headings = ["SID", "Timestamp", "Band", "Observed_Count", "Min", "Max",
"Sum", "Sum_of_Squares"]
df = pandas.DataFrame(columns=headings, dtype=numpy.float)
# Read the rasterised image
with rasterio.open(rasterised_fname) as src:
img = src.read(1)
# Initialise the segment visitor
seg_vis = Segments(img)
# Do we have any data to analyse???
if seg_vis.n_segments == 0:
return df
# We need to get the PQ data and the DatasetType of interest
pq_ds = dataset.datasets[DatasetType.PQ25]
ds = dataset.datasets[dataset_type]
timestamp = dataset.start_datetime
bands = ds.bands
no_data = -999
# TODO have a user choice at the config level to determine which PQ flags
# to apply
# pq_flags = [PqaMask.PQ_MASK_CLEAR] # The default will msk everything
pq_flags = [PqaMask.PQ_MASK_SATURATION,
PqaMask.PQ_MASK_CONTIGUITY,
PqaMask.PQ_MASK_CLOUD] # cloud and cloud shadow
for band in bands:
# When the api has a release of get_pq_mask this will have to do
# It'll re-compute the PQ every time which is not ideal
# Otherwise go back to eotools???
ds_data = (get_dataset_data_with_pq(ds, pq_ds, bands=[band],
ndv=no_data)[band]).astype('float')
# Set no-data to NaN
ds_data[ds_data == no_data] = numpy.nan
# Loop over each segment and get the data.
# In other instances we may just need the locations
for seg_id in seg_vis.ids:
data = seg_vis.data(ds_data, segment_id=seg_id)
# dimensions of the data which will be 1D
dim = data.shape
# Returns are 1D arrays, so check if we have an empty array
if dim[0] == 0:
continue # Empty bin, (no data), skipping
# Compute the stats
count = numpy.sum(numpy.isfinite(data))
sum_ = numpy.nansum(data)
sum_sq = numpy.nansum(data**2)
min_ = numpy.nanmin(data)
max_ = numpy.nanmax(data)
format_dict = {"SID": seg_id,
"Timestamp": timestamp,
"Band": band.name,
"Observed_Count": count,
"Min": min_,
"Max": max_,
"Sum": sum_,
"Sum_of_Squares": sum_sq}
# Append the stat to the data frame
df = df.append(format_dict, ignore_index=True)
return df
def classifier(arg25_dataset, pq25_dataset):
"""
Runs the classifier designed by SF.
"""
# Get the metadata
md = get_dataset_metadata(arg25_dataset)
cols, rows = md.shape
# Read the data and mask pixels via the PQ dataset
data = get_dataset_data_with_pq(arg25_dataset, pq25_dataset)
# Get the wetness coefficients and calculate
coef = TCI_COEFFICIENTS[arg25_dataset.satellite][TasselCapIndex.WETNESS]
wetness = calculate_tassel_cap_index(data, coef)
# NDVI
ndvi = calculate_ndvi(data[arg25_dataset.bands.RED],
data[arg25_dataset.bands.NEAR_INFRARED],
output_ndv=numpy.nan)
# Dump the reflectance data, the classifier only needs tc_wetness and ndvi
del data
# Allocate the result
classified = numpy.zeros((rows, cols), dtype='uint8')
# Water
r1 = wetness > 0
classified[r1] = 1
_tmp = ~r1
#r2 = _tmp & ((wetness >= -250) & (wetness < 0))
r2 = (wetness >= -250) & (wetness < 0)
r3 = ndvi <= 0.3
#_tmp2 = _tmp & r2 & ~r3
_tmp2 = _tmp & r2
# non-veg
classified[_tmp2 & r3] = 2
_tmp3 = _tmp2 & ~r3
r4 = ndvi <= 0.45
# saltmarsh
classified[_tmp3 & r4] = 3
_tmp2 = _tmp3 & ~r4
r5 = ndvi <= 0.6
# mangrove/saltmarsh
classified[_tmp2 & r5] = 4
# mangrove
classified[_tmp2 & ~r5] = 5
# finished rhs of r2
_tmp2 = _tmp & ~r2
r6 = wetness < -750
r7 = ndvi >= 0.3
_tmp3 = _tmp2 & r6
# saltmarsh
classified[_tmp3 & r7] = 3
# non-veg
classified[_tmp3 & ~r7] = 2
r8 = ndvi <= 0.3
_tmp3 = _tmp2 & ~r6
# non-veg
classified[_tmp3 & r8] = 2
r9 = ndvi <= 0.45
_tmp2 = _tmp3 & ~r8
# saltmarsh
classified[_tmp2 & r9] = 3
r10 = ndvi <= 0.6
_tmp3 = _tmp2 & ~r9
# mangrove-saltmarsh
classified[_tmp3 & r10] = 4
# mangrove
classified[_tmp3 & ~r10] = 5
# set any nulls
valid = numpy.isfinite(ndvi)
classified[~valid] = 0
return classified
def classifier(arg25_dataset, pq25_dataset):
"""
Runs the classifier designed by SF.
"""
# Get the metadata
md = get_dataset_metadata(arg25_dataset)
cols, rows = md.shape
# Read the data and mask pixels via the PQ dataset
data = get_dataset_data_with_pq(arg25_dataset, pq25_dataset)
# Get the wetness coefficients and calculate
coef = TCI_COEFFICIENTS[arg25_dataset.satellite][TasselCapIndex.WETNESS]
wetness = calculate_tassel_cap_index(data, coef)
# NDVI
ndvi = calculate_ndvi(data[arg25_dataset.bands.RED],
data[arg25_dataset.bands.NEAR_INFRARED],
output_ndv=numpy.nan)
# Dump the reflectance data, the classifier only needs tc_wetness and ndvi
del data
# Allocate the result
classified = numpy.zeros((rows,cols), dtype='uint8')
# Water
r1 = wetness > 0
classified[r1] = 1
_tmp = ~r1
#r2 = _tmp & ((wetness >= -250) & (wetness < 0))
r2 = (wetness >= -250) & (wetness < 0)
r3 = ndvi <= 0.3
#_tmp2 = _tmp & r2 & ~r3
_tmp2 = _tmp & r2
# non-veg
classified[_tmp2 & r3] = 2
_tmp3 = _tmp2 & ~r3
r4 = ndvi <= 0.45
# saltmarsh
classified[_tmp3 & r4] = 3
_tmp2 = _tmp3 & ~r4
r5 = ndvi <= 0.6
# mangrove/saltmarsh
classified[_tmp2 & r5] = 4
# mangrove
classified[_tmp2 & ~r5] = 5
# finished rhs of r2
_tmp2 = _tmp & ~r2
r6 = wetness < -750
r7 = ndvi >= 0.3
_tmp3 = _tmp2 & r6
# saltmarsh
classified[_tmp3 & r7] = 3
# non-veg
classified[_tmp3 & ~r7] = 2
r8 = ndvi <= 0.3
_tmp3 = _tmp2 & ~r6
# non-veg
classified[_tmp3 & r8] = 2
r9 = ndvi <= 0.45
_tmp2 = _tmp3 & ~r8
# saltmarsh
classified[_tmp2 & r9] = 3
r10 = ndvi <= 0.6
_tmp3 = _tmp2 & ~r9
# mangrove-saltmarsh
classified[_tmp3 & r10] = 4
# mangrove
classified[_tmp3 & ~r10] = 5
# set any nulls
valid = numpy.isfinite(ndvi)
classified[~valid] = 0
return classified
