def heuristic_atmosphere(RT, instrument, x_RT, x_instrument, meas, geom):
'''From a given radiance, estimate atmospheric state with band ratios.
Used to initialize gradient descent inversions.'''
# Identify the latest instrument wavelength calibration (possibly
# state-dependent) and identify channel numbers for the band ratio.
wl, fwhm = instrument.calibration(x_instrument)
b865 = s.argmin(abs(wl-865))
b945 = s.argmin(abs(wl-945))
b1040 = s.argmin(abs(wl-1040))
if not (any(RT.wl > 850) and any(RT.wl < 1050)):
return x_RT
x_new = x_RT.copy()
# Band ratio retrieval of H2O. Depending on the radiative transfer
# model we are using, this state parameter could go by several names.
for h2oname in ['H2OSTR', 'h2o']:
if h2oname not in RT.statevec:
continue
# ignore unused names
if h2oname not in RT.lut_names:
continue
# find the index in the lookup table associated with water vapor
ind_lut = RT.lut_names.index(h2oname)
ind_sv = RT.statevec.index(h2oname)
h2os, ratios = [], []
# We iterate through every possible grid point in the lookup table,
# calculating the band ratio that we would see if this were the
# atmospheric H2O content. It assumes that defaults for all other
# atmospheric parameters (such as aerosol, if it is there).
for h2o in RT.lut_grids[ind_lut]:
# Get Atmospheric terms at high spectral resolution
x_RT_2 = x_RT.copy()
x_RT_2[ind_sv] = h2o
rhoatm_hi, sphalb_hi, transm_hi, transup_hi = RT.get(x_RT_2, geom)
rhoatm = instrument.sample(x_instrument, RT.wl, rhoatm_hi)
transm = instrument.sample(x_instrument, RT.wl, transm_hi)
solar_irr = instrument.sample(x_instrument, RT.wl, RT.solar_irr)
# Assume no surface emission. "Correct" the at-sensor radiance
# using this presumed amount of water vapor, and measure the
# resulting residual (as measured from linear interpolation across
# the absorption feature)
r = (meas*s.pi/(solar_irr*RT.coszen) - rhoatm) / (transm+1e-8)
ratios.append((r[b945]*2.0)/(r[b1040]+r[b865]))
h2os.append(h2o)
# Finally, interpolate to determine the actual water vapor level that
# would optimize the continuum-relative correction
p = interp1d(h2os, ratios)
bounds = (h2os[0]+0.001, h2os[-1]-0.001)
best = min1d(lambda h: abs(1-p(h)), bounds=bounds, method='bounded')
x_new[ind_sv] = best.x
return x_new
def heuristic_atmosphere(self, rdn, geom):
'''From a given radiance, estimate atmospheric state using band ratio
heuristics. Used to initialize gradient descent inversions.'''
x = self.init_val.copy()
# Band ratio retrieval of H2O
for h2oname in ['H2OSTR', 'h2o']:
if (h2oname in self.lut_names) and \
any(self.wl > 850) and any(self.wl < 1050):
ind_lut = self.lut_names.index(h2oname)
ind_sv = self.statevec.index(h2oname)
b865, b945, b1040 = [
s.argmin(abs(self.wl - t)) for t in (865, 945, 1040)
]
h2os, ratios = [], [
] # will be corrected for transmission, path radiance
for h2o in self.lut_grids[ind_lut]:
xnew = x.copy()
xnew[ind_sv] = h2o
rhoatm, sphalb, transm, transup = self.get(xnew, geom)
# assume no surface emission
r = (rdn * s.pi / (self.solar_irr * self.coszen) -
rhoatm) / (transm + 1e-8)
ratios.append((r[b945] * 2.0) / (r[b1040] + r[b865]))
h2os.append(h2o)
p = interp1d(h2os, ratios)
bounds = (h2os[0] + 0.001, h2os[-1] - 0.001)
best = min1d(lambda h: abs(1 - p(h)),
bounds=bounds,
method='bounded')
x[ind_sv] = best.x
Ls_est = s.zeros(rdn.shape)
rfl_est = self.invert_algebraic(x, rdn, Ls_est, geom)
return x, rfl_est
def heuristic_atmosphere(RT: RadiativeTransfer, instrument, x_RT, x_instrument,
meas, geom):
"""From a given radiance, estimate atmospheric state with band ratios.
Used to initialize gradient descent inversions."""
# Identify the latest instrument wavelength calibration (possibly
# state-dependent) and identify channel numbers for the band ratio.
wl, fwhm = instrument.calibration(x_instrument)
b865 = np.argmin(abs(wl - 865))
b945 = np.argmin(abs(wl - 945))
b1040 = np.argmin(abs(wl - 1040))
if not (any(RT.wl > 850) and any(RT.wl < 1050)):
return x_RT
x_new = x_RT.copy()
# Figure out which RT object we are using
# TODO: this is currently very specific to vswir-tir 2-mode, eventually generalize
my_RT = None
for rte in RT.rt_engines:
if rte.treat_as_emissive is False:
my_RT = rte
break
if not my_RT:
raise ValueError('No suiutable RT object for initialization')
# Band ratio retrieval of H2O. Depending on the radiative transfer
# model we are using, this state parameter could go by several names.
for h2oname in ['H2OSTR', 'h2o']:
if h2oname not in RT.statevec_names:
continue
# ignore unused names
if h2oname not in my_RT.lut_names:
continue
# find the index in the lookup table associated with water vapor
ind_lut = my_RT.lut_names.index(h2oname)
ind_sv = RT.statevec_names.index(h2oname)
h2os, ratios = [], []
# We iterate through every possible grid point in the lookup table,
# calculating the band ratio that we would see if this were the
# atmospheric H2O content. It assumes that defaults for all other
# atmospheric parameters (such as aerosol, if it is there).
for h2o in my_RT.lut_grids[ind_lut]:
# Get Atmospheric terms at high spectral resolution
x_RT_2 = x_RT.copy()
x_RT_2[ind_sv] = h2o
rhi = RT.get_shared_rtm_quantities(x_RT_2, geom)
rhoatm = instrument.sample(x_instrument, RT.wl, rhi['rhoatm'])
transm = instrument.sample(x_instrument, RT.wl, rhi['transm'])
sphalb = instrument.sample(x_instrument, RT.wl, rhi['sphalb'])
solar_irr = instrument.sample(x_instrument, RT.wl, RT.solar_irr)
# Assume no surface emission. "Correct" the at-sensor radiance
# using this presumed amount of water vapor, and measure the
# resulting residual (as measured from linear interpolation across
# the absorption feature)
rho = meas * np.pi / (solar_irr * RT.coszen)
r = 1.0 / (transm / (rho - rhoatm) + sphalb)
ratios.append((r[b945] * 2.0) / (r[b1040] + r[b865]))
h2os.append(h2o)
# Finally, interpolate to determine the actual water vapor level that
# would optimize the continuum-relative correction
p = interp1d(h2os, ratios)
bounds = (h2os[0] + 0.001, h2os[-1] - 0.001)
best = min1d(lambda h: abs(1 - p(h)), bounds=bounds, method='bounded')
x_new[ind_sv] = best.x
return x_new