def mt_metrics(stack,
out_prefix,
metrics,
rescale_to_datatype=False,
to_power=False,
outlier_removal=False):
with rasterio.open(stack) as src:
# get metadata
meta = src.meta
# update driver and reduced band count
meta.update({'driver': 'GTiff'})
meta.update({'count': 1})
# write all different output files into a dictionary
metric_dict = {}
for metric in metrics:
metric_dict[metric] = rasterio.open(
out_prefix + '.' + metric + '.tif', 'w', **meta)
# scaling factors in case we have to rescale to integer
minimums = {
'avg': -30,
'max': -30,
'min': -30,
'std': 0.00001,
'cov': 0.00001
}
maximums = {'avg': 5, 'max': 5, 'min': 5, 'std': 15, 'cov': 1}
# loop through blocks
for _, window in src.block_windows(1):
# read array with all bands
stack = src.read(range(1, src.count + 1), window=window)
if rescale_to_datatype is True and meta['dtype'] != 'float32':
stack = ras.rescale_to_float(stack, meta['dtype'])
# transform to power
if to_power is True:
stack = ras.convert_to_power(stack)
# outlier removal (only applies if there are more than 5 bands)
if outlier_removal is True and src.count >= 5:
stack = remove_outliers(stack)
# get stats
arr = {}
arr['avg'] = (np.nan_to_num(np.nanmean(stack, axis=0))
if 'avg' in metrics else False)
arr['max'] = (np.nan_to_num(np.nanmax(stack, axis=0))
if 'max' in metrics else False)
arr['min'] = (np.nan_to_num(np.nanmin(stack, axis=0))
if 'min' in metrics else False)
arr['std'] = (np.nan_to_num(np.nanstd(stack, axis=0))
if 'std' in metrics else False)
arr['cov'] = (np.nan_to_num(
stats.variation(stack, axis=0, nan_policy='omit'))
if 'cov' in metrics else False)
# the metrics to be re-turned to dB, in case to_power is True
metrics_to_convert = ['avg', 'min', 'max']
# do the back conversions and write to disk loop
for metric in metrics:
if to_power is True and metric in metrics_to_convert:
arr[metric] = ras.convert_to_db(arr[metric])
if rescale_to_datatype is True and meta['dtype'] != 'float32':
arr[metric] = ras.scale_to_int(arr[metric], meta['dtype'],
minimums[metric],
maximums[metric])
# write to dest
metric_dict[metric].write(np.float32(arr[metric]),
window=window,
indexes=1)
# close the output files
for metric in metrics:
metric_dict[metric].close()
def mt_metrics(stack,
out_prefix,
metrics,
rescale_to_datatype=False,
to_power=False,
outlier_removal=False,
datelist=None):
# from datetime import datetime
with rasterio.open(stack) as src:
harmonics = False
if 'harmonics' in metrics:
logger.debug('INFO: Calculating harmonics')
if not datelist:
logger.debug('WARNING: Harmonics need the datelist. '
'Harmonics will not be calculated')
else:
harmonics = True
metrics.remove('harmonics')
metrics.extend(['amplitude', 'phase', 'residuals'])
if 'percentiles' in metrics:
metrics.remove('percentiles')
metrics.extend(['p95', 'p5'])
# get metadata
meta = src.profile
# update driver and reduced band count
meta.update({'driver': 'GTiff'})
meta.update({'count': 1})
# write all different output files into a dictionary
metric_dict = {}
for metric in metrics:
filename = '{}.{}.tif'.format(out_prefix, metric)
metric_dict[metric] = rasterio.open(filename, 'w', **meta)
# scaling factors in case we have to rescale to integer
minimums = {
'avg': -30,
'max': -30,
'min': -30,
'std': 0.00001,
'cov': 0.00001,
'count': 0
}
maximums = {
'avg': 5,
'max': 5,
'min': 5,
'std': 15,
'cov': 1,
'count': 64000
}
if harmonics:
# construct independent variables
dates, sines, cosines = [], [], []
two_pi = np.multiply(2, np.pi)
for date in sorted(datelist):
delta = difference_in_years(
datetime.strptime('700101', "%y%m%d"),
datetime.strptime(date, "%y%m%d"))
dates.append(delta)
sines.append(np.sin(np.multiply(two_pi, delta - 0.5)))
cosines.append(np.cos(np.multiply(two_pi, delta - 0.5)))
arr = np.array([dates, cosines, sines])
# loop through blocks
for _, window in src.block_windows(1):
# read array with all bands
stack = src.read(range(1, src.count + 1), window=window)
if rescale_to_datatype is True and meta['dtype'] != 'float32':
stack = ras.rescale_to_float(stack, meta['dtype'])
# transform to power
if to_power is True:
stack = ras.convert_to_power(stack)
# outlier removal (only applies if there are more than 5 bands)
if outlier_removal is True and src.count >= 5:
stack = remove_outliers(stack)
# get stats
np.seterr(divide='ignore', invalid='ignore')
arr = {}
arr['p95'], arr['p5'] = (np.nan_to_num(
nan_percentile(stack, [95, 5])) if 'p95' in metrics else
(False, False))
arr['median'] = (np.nan_to_num(np.nanmedian(stack, axis=0))
if 'median' in metrics else False)
arr['avg'] = (np.nan_to_num(np.nanmean(stack, axis=0))
if 'avg' in metrics else False)
arr['max'] = (np.nan_to_num(np.nanmax(stack, axis=0))
if 'max' in metrics else False)
arr['min'] = (np.nan_to_num(np.nanmin(stack, axis=0))
if 'min' in metrics else False)
arr['std'] = (np.nan_to_num(np.nanstd(stack, axis=0))
if 'std' in metrics else False)
arr['cov'] = (np.nan_to_num(
stats.variation(stack, axis=0, nan_policy='omit'))
if 'cov' in metrics else False)
arr['count'] = (np.nan_to_num(np.count_nonzero(stack, axis=0))
if 'count' in metrics else False)
if harmonics:
stack_size = (stack.shape[1], stack.shape[2])
if to_power is True:
y = ras.convert_to_db(stack).reshape(stack.shape[0], -1)
else:
y = stack.reshape(stack.shape[0], -1)
arr, residuals, _, _ = np.linalg.lstsq(arr.T, y)
arr['amplitude'] = np.hypot(arr[1], arr[2]).reshape(stack_size)
arr['phase'] = np.arctan2(arr[2], arr[1]).reshape(stack_size)
arr['residuals'] = np.sqrt(np.divide(
residuals, stack.shape[0])).reshape(stack_size)
# the metrics to be re-turned to dB, in case to_power is True
metrics_to_convert = ['avg', 'min', 'max', 'p95', 'p5', 'median']
# do the back conversions and write to disk loop
for metric in metrics:
if to_power is True and metric in metrics_to_convert:
arr[metric] = ras.convert_to_db(arr[metric])
if rescale_to_datatype is True and meta['dtype'] != 'float32':
arr[metric] = ras.scale_to_int(arr[metric], meta['dtype'],
minimums[metric],
maximums[metric])
# write to dest
metric_dict[metric].write(np.float32(arr[metric]),
window=window,
indexes=1)
metric_dict[metric].update_tags(
1,
BAND_NAME='{}_{}'.format(os.path.basename(out_prefix),
metric))
metric_dict[metric].set_band_description(
1, '{}_{}'.format(os.path.basename(out_prefix), metric))
# close the output files
for metric in metrics:
# close rio opening
metric_dict[metric].close()
dirname = os.path.dirname(out_prefix)
check_file = opj(dirname,
'.{}.processed'.format(os.path.basename(out_prefix)))
with open(str(check_file), 'w') as file:
file.write('passed all tests \n')