def get_kk(fhead):
mete = readxml('%smetadata.xml'%fhead)
vZa = np.array([i+ np.zeros((23,23)) for i in mete['mVz']])
vAa = np.array([i+ np.zeros((23,23)) for i in mete['mVa']])
ks = []
for i in [2,3,4,8,11,12]:
if i ==8:
i -= 1
kk = kernels.Kernels(vZa[i] ,mete['SAG_Z'],mete['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
kk = kernels.Kernels(vZa[i+1] ,mete['SAG_Z'],mete['SAG_A']-vAa[i+1],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
else:
i -= 1
kk = kernels.Kernels(vZa[i] ,mete['SAG_Z'],mete['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
return ks
def get_kk(fhead):
mete = readxml('%smetadata.xml' % fhead)
vZa = np.nanmean(mete['VIAG_Z'].reshape(13,
len(mete['VIAG_Z']) / 13, 23, 23),
axis=1)
vAa = np.nanmean(mete['VIAG_A'].reshape(13,
len(mete['VIAG_Z']) / 13, 23, 23),
axis=1)
ks = []
for i in [2, 3, 4, 8, 11, 12]:
if i == 8:
i -= 1
kk = kernels.Kernels(vZa[i] ,mete['SAG_Z'],mete['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
kk = kernels.Kernels(vZa[i+1] ,mete['SAG_Z'],mete['SAG_A']-vAa[i+1],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
else:
i -= 1
kk = kernels.Kernels(vZa[i] ,mete['SAG_Z'],mete['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
return ks
def create_observation_operator(self, metadata, x_forecast, band):
"""The linear kernel models..."""
K = kernels.Kernels(metadata.vza[metadata.mask],
metadata.sza[metadata.mask],
metadata.raa[metadata.mask],
LiType="Sparse",
doIntegrals=False,
normalise=1,
RecipFlag=True,
RossHS=False,
MODISSPARSE=True,
RossType="Thick")
good_obs = metadata.mask.sum() # size of H_matrix
all_obs = metadata.mask.ravel().shape[0] # All obs
zz = np.zeros(all_obs)
ross = zz * 0.
li = zz * 0.
ross[metadata.mask.flatten()] = K.Ross
li[metadata.mask.flatten()] = K.Li
data = np.c_[np.r_[np.ones(all_obs), zz, zz], np.r_[zz, ross, zz],
np.r_[zz, zz, li]].T
offsets = [0, all_obs, 2 * all_obs]
H_matrix = sp.dia_matrix((data, offsets),
shape=(all_obs, self.n_params * all_obs),
dtype=np.float32)
# The following arrangement is all isotropics, all volumetrics and then
# all geometrics. For the non-linear model, it might be better to have
# all data per grid cell, and not all clobbered together.
return H_matrix.tocsr()
def get_kk(angles):
vza ,sza,raa = angles
kk = kernels.Kernels(vza,sza,raa,\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Sparse',RossType='Thick')
return kk
def get_kk(fhead):
mete = readxml('%smetadata.xml' % fhead)
vZa = np.array([i + np.zeros((23, 23)) for i in mete['mVz']])
vAa = np.array([i + np.zeros((23, 23)) for i in mete['mVa']])
sza = mete['mSz'][0]
vza = mete['mVz'][3]
rel_a = (mete['mSa'] - mete['mVa'])[3]
kk = kernels.Kernels(vza ,sza,rel_a,\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
return kk
def predict_boa_refl_MCD43(self, the_date, the_band):
# find modis matching dataset
print "FIX band_equivalences!!!!!!!!!!"
band_equivalences = [[[1, 1.]], [[2, 1.]], [[3, 1.]], [[4, 1.]],
[[5, 1.]], [[6, 1.]], [[7, 1.]]]
sza, saa, vza, vaa = self.get_l1c_angles(the_date)
raa = vaa - saa
K = kernels.Kernels(vza,
sza,
raa,
LiType='Sparse',
doIntegrals=False,
normalise=1,
RecipFlag=True,
RossHS=False,
MODISSPARSE=True,
RossType='Thick')
kk = np.array([1., K.Ross[0], K.Li[0]])
k = min(self.modis_filelist["MCD43A1"].keys(),
key=lambda x: abs(x - the_date))
pred_boa_refl = np.zeros((2400, 2400))
for band, w in band_equivalences[the_band]:
g = gdal.Open(
'HDF4_EOS:EOS_GRID:"%s":MOD_Grid_BRDF:BRDF_Albedo_Parameters_Band%d'
% (self.modis_filelist["MCD43A1"][k], band))
kernel_weights = g.ReadAsArray() / 10000.
g = gdal.Open(
'HDF4_EOS:EOS_GRID:"%s":MOD_Grid_BRDF:BRDF_Albedo_Band_Quality_Band%d'
% (self.modis_filelist["MCD43A2"][k], band))
qa = g.ReadAsArray()
print "Only best quality MCD43 samples selected"
kernel_weights = np.where(qa == 0, kernel_weights, np.nan)
pred_boa_refl += w * np.sum(kernel_weights * kk[:, None, None],
axis=0)
return pred_boa_refl
return a
co_bands = readfile([2, 3, 4, 8, 13, 11, 12],
'data/50SMG20164100',
bounds=None)
vZa = np.nanmean(co_bands['VIAG_Z'].reshape(13, 12, 23, 23), axis=1)
vAa = np.nanmean(co_bands['VIAG_A'].reshape(13, 12, 23, 23), axis=1)
ks = []
for i in [2, 3, 4, 8, 11, 12]:
if i == 8:
i -= 1
kk = kernels.Kernels(vZa[i] , co_bands['SAG_Z'],co_bands['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
kk = kernels.Kernels(vZa[i+1] , co_bands['SAG_Z'],co_bands['SAG_A']-vAa[i+1],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)
else:
i -= 1
kk = kernels.Kernels(vZa[i] , co_bands['SAG_Z'],co_bands['SAG_A']-vAa[i],\
RossHS=False,MODISSPARSE=True,\
RecipFlag=True,normalise=1,\
doIntegrals=False,LiType='Dense',RossType='Thick')
ks.append(kk)