Newer
Older
# from netCDF4 import Dataset
import xarray as xr
import numpy as np
import ancillary_data as anc
_DTR = np.pi/180.
_RTD = 180./np.pi
def read_data(sensor: str, l1b_filename: str, geo_filename: str):
data = xr.open_dataset(l1b_filename, group='observation_data', decode_cf=False)
in_data = xr.Dataset()
if sensor.lower() == 'viirs':
for band in list(data.variables):
if 'reflectance' in data[band].long_name:
if hasattr(data[band], 'VCST_scale_factor'):
scale_factor = data[band].VCST_scale_factor * data[band].bias_correction
else:
scale_factor = data[band].radiance_scale_factor
in_data[band] = (('number_of_lines', 'number_of_pixels'),
data[band].values * scale_factor)
elif 'radiance' in data[band].long_name:
in_data[band] = (('number_of_lines', 'number_of_pixels'),
data[f'{band}_brightness_temperature_lut'].values[data[band].values])
else:
pass
data = xr.open_dataset(geo_filename, group='geolocation_data')
in_data = in_data.merge(data)
relazi = relative_azimuth_angle(data.sensor_azimuth.values, data.solar_azimuth.values)
sunglint = sun_glint_angle(data.sensor_zenith.values, data.solar_zenith.values, relazi)
scatt_angle = scattering_angle(data.solar_zenith.values, data.sensor_zenith.values, relazi)
in_data['relative_azimuth'] = (('number_of_lines', 'number_of_pixels'), relazi)
in_data['sunglint_angle'] = (('number_of_lines', 'number_of_pixels'), sunglint)
in_data['scattering_angle'] = (('number_of_lines', 'number_of_pixels'), scatt_angle)
return in_data
def relative_azimuth_angle(sensor_azimuth: float, solar_azimuth: float) -> float:
rel_azimuth = np.abs(180. - np.abs(sensor_azimuth - solar_azimuth))
return rel_azimuth
def sun_glint_angle(sensor_zenith: float, solar_zenith: float, rel_azimuth: float) -> float:
cossna = (np.sin(sensor_zenith*_DTR) * np.sin(solar_zenith*_DTR) * np.cos(rel_azimuth*_DTR) +
np.cos(sensor_zenith*_DTR) * np.cos(solar_zenith*_DTR))
cossna[cossna > 1] = 1
sunglint_angle = np.arccos(cossna) * _RTD
return sunglint_angle
def scattering_angle(solar_zenith, sensor_zenith, relative_azimuth):
cos_scatt_angle = -1. * (np.cos(solar_zenith*_DTR) * np.cos(sensor_zenith*_DTR) -
np.sin(solar_zenith*_DTR) * np.sin(sensor_zenith*_DTR) *
np.cos(relative_azimuth*_DTR))
scatt_angle = np.arccos(cos_scatt_angle) * _RTD
return scatt_angle
def correct_reflectances():
pass
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
def read_ancillary_data(viirs_data):
# Ancillary files temporarily defined here. Eventually we will find a better way to pass these
start_time = '2014-08-01 15:48:00.000'
anc_dir = '/ships19/hercules/pveglio/neige_data/snpp_test_input/ancillary/2014_08_01_213'
sst_file = 'oisst.20140730'
ndvi_file = 'NDVI.FM.c004.v2.0.WS.00-04.209.hdf'
geos_cnst = 'GEOS.fp.asm.const_2d_asm_Nx.00000000_0000.V01.nc4'
geos_lnd = 'GEOS.fpit.asm.tavg1_2d_lnd_Nx.GEOS5124.20140801_1530.V01.nc4'
geos_ocn = 'GEOS.fpit.asm.tavg1_2d_ocn_Nx.GEOS5124.20140801_1530.V01.nc4'
geos1 = 'GEOS.fpit.asm.inst3_2d_asm_Nx.GEOS5124.20140801_1500.V01.nc4'
geos2 = 'GEOS.fpit.asm.inst3_2d_asm_Nx.GEOS5124.20140801_1800.V01.nc4'
sst = np.empty((3232*3200, ), dtype=np.float32)
ndvi = np.empty((3232*3200, ), dtype=np.float32)
eco = np.empty((3232*3200, ), dtype=np.ubyte)
geos_data = {'tpw': np.empty((3232*3200, ), dtype=np.float32),
'snowfr': np.empty((3232*3200, ), dtype=np.float32),
'icefr': np.empty((3232*3200, ), dtype=np.float32),
'ocnfr': np.empty((3232*3200, ), dtype=np.float32),
'landicefr': np.empty((3232*3200, ), dtype=np.float32),
'sfct': np.empty((3232*3200, ), dtype=np.float32),
}
sst = anc.py_get_Reynolds_SST(viirs_data.latitude.values.reshape((3232*3200, )),
viirs_data.longitude.values.reshape((3232*3200, )),
anc_dir, sst_file, sst)
ndvi = anc.py_get_NDVI_background(viirs_data.latitude.values.reshape((3232*3200, )),
viirs_data.longitude.values.reshape((3232*3200, )),
anc_dir, ndvi_file, ndvi)
eco = anc.py_get_Olson_eco(viirs_data.latitude.values.reshape((3232*3200, )),
viirs_data.longitude.values.reshape((3232*3200, )),
anc_dir, eco)
geos_data = anc.py_get_GEOS(viirs_data.latitude.values.reshape((3232*3200, )),
viirs_data.longitude.values.reshape((3232*3200, )),
start_time, anc_dir, geos1, geos2, geos_lnd, geos_ocn, geos_cnst, geos_data)
data = viirs_data
data['sst'] = (['number_of_lines', 'number_of_pixels'], sst)
data['ndvi'] = (['number_of_lines', 'number_of_pixels'], ndvi)
data['eco'] = (['number_of_lines', 'number_of_pixels'], eco)
data['geos_tpw'] = (['number_of_lines', 'number_of_pixels'], geos_data['tpw'])
data['geos_snowfr'] = (['number_of_lines', 'number_of_pixels'], geos_data['snowfr'])
data['geos_icefr'] = (['number_of_lines', 'number_of_pixels'], geos_data['icefr'])
data['geos_ocnfr'] = (['number_of_lines', 'number_of_pixels'], geos_data['ocnfr'])
data['geos_landicefr'] = (['number_of_lines', 'number_of_pixels'], geos_data['landicefr'])
data['geos_sfct'] = (['number_of_lines', 'number_of_pixels'], geos_data['sfct'])
return data