Atmospheric correction in the visible spectrumΒΆ

_images/truecolor_composite_20170518_nrk_reception_cropped_thumb.png

In particular at the shorter wavelengths around 400-600 nm, which include the region used e.g. for ocean color products or true color imagery, in particular the Rayleigh scattering due to atmospheric molecules or atoms, and Mie scattering and absorption of aerosols, becomes significant. As this atmospheric scattering and absorption is obscuring the retrieval of surface parameters and since it is strongly dependent on observation geometry, it is custom to try to correct or subtract this unwanted signal from the data before performing the geophysical retrieval or generating useful and nice looking imagery.

In order to correct for this atmospheric effect we have simulated the solar reflectance under various sun-satellite viewing conditions for a set of different standard atmospheres, assuming a black surface, using a radiative transfer model. For a given atmosphere the reflectance is dependent on wavelength, solar-zenith angle, satellite zenith angle, and the relative sun-satellite azimuth difference angle:

\sigma = \sigma({\theta}_0, \theta, \phi, \lambda)

The relative sun-satellite azimuth difference angle is defined as illustrated in the figure below:

Definition of the sun-satellite viewing geometry

The method is described in detail in a scientific paper.

To apply the atmospheric correction for a given satellite sensor band, the procedure involves the following three steps:

  • Find the effective wavelength of the band
  • Derive the atmospheric absorption and scattering contribution
  • Subtract that from the observations

As the Rayleigh scattering, which is the dominating part we are correcting for under normal situations (when there is no excessive pollution or aerosols in the line of sight) is proportional to \frac{1}{{\lambda}^4} the effective wavelength is derived by convolution of the spectral response with \frac{1}{{\lambda}^4}.

To get the atmospheric contribution for an arbitrary band, first the solar zenith, satellite zenith and the sun-satellite azimuth difference angles should be loaded.

For the VIIRS M2 band the interface looks like this:

>>> from pyspectral.rayleigh import Rayleigh
>>> viirs = Rayleigh('Suomi-NPP', 'viirs')
>>> import numpy as np
>>> sunz = np.array([[32., 40.], [31., 41.]])
>>> satz = np.array([[45., 20.], [46., 21.]])
>>> ssadiff = np.array([[110, 170], [120, 180]])
>>> refl_cor_m2 = viirs.get_reflectance(sunz, satz, ssadiff, 'M2')
>>> print(refl_cor_m2)
[[ 10.45746088   9.69434733]
 [ 10.35336108   9.74561515]]

This Rayleigh (including Mie scattering and aborption by aerosols) contribution should then of course be subtracted from the data. Optionally the red band can be provided as the fifth argument, which will provide a more gentle scaling in cases of high reflectances (above 20%):

>>> redband =  np.array([[23., 19.], [24., 18.]])
>>> refl_cor_m2 = viirs.get_reflectance(sunz, satz, ssadiff, 'M2', redband)
>>> print(refl_cor_m2)
[[ 10.06530609   9.69434733]
 [  9.83569303   9.74561515]]

In case you want to bypass the reading of the sensor response functions or you have a sensor for which there are no RSR data available in PySpectral it is still possible to derive an atmospheric correction for that band. All what is needed is the effective wavelength of the band, given in micrometers (:math: mu m). This wavelength is normally calculated by PySPectral from the RSR data when passing the name of the band as above.

>>> from pyspectral.rayleigh import Rayleigh
>>> viirs = Rayleigh('UFO', 'ufo')
>>> import numpy as np
>>> sunz = np.array([[32., 40.], [31., 41.]])
>>> satz = np.array([[45., 20.], [46., 21.]])
>>> ssadiff = np.array([[110, 170], [120, 180]])
>>> refl_cor_m2 = viirs.get_reflectance(sunz, satz, ssadiff, 0.45, redband)
[[ 9.55453743  9.20336915]
 [ 9.33705413  9.25222498]]

You may choose any name for the platform and sensor as you like, as long as it does not match a platform/sensor pair for which RSR data exist in PySpectral.

At the moment we have done simulations for a set of standard atmospheres in two different configurations, one only considering Rayleigh scattering, and one also accounting for Aerosols. On default we use the simulations with the marine-clean aerosol distribution and the US-standard atmosphere, but it is possible to specify if you want another setup, e.g.:

>>> from pyspectral.rayleigh import Rayleigh
>>> viirs = Rayleigh('Suomi-NPP', 'viirs', atmosphere='midlatitude summer', rural_aerosol=True)
>>> refl_cor_m2 = viirs.get_reflectance(sunz, satz, ssadiff, 'M2', redband)
[[ 10.01281363   9.65488615]
 [  9.78070046   9.70335278]]

At high solar zenith angles the assumptions used in the simulations begin to break down, which can lead to unrealistic correction values. In particular, for true color imagery this often results in the red channel being too bright compared to the green and / or blue channels. We have implemented an optional scaling method to minimise this overcorrection at high solar zeniths, reduce_rayleigh:

>>> from pyspectral.rayleigh import Rayleigh
>>> import numpy as np
>>> viirs = Rayleigh('Suomi-NPP', 'viirs', atmosphere='midlatitude summer', rural_aerosol=True)
>>> sunz = np.array([[32., 40.], [80., 88.]])
>>> satz = np.array([[45., 20.], [46., 21.]])
>>> ssadiff = np.array([[110, 170], [120, 180]])
>>> refl_cor_m2 = viirs.get_reflectance(sunz, satz, ssadiff, 0.45, redband)
[[ 10.40291763 9.654881]
 [ 30.9275331  39.41288558]]
>>> reduced_refl_cor_m2 = viirs.reduce_rayleigh_highzenith(sunz, refl_cor_m2, 70., 90., 1.)
[[ 10.40291763 9.654881],
 [ 15.46376655 3.94128856]]

These reduced atmospheric correction (primarily due to increased Rayleigh scattering in the clear atmosphere) values can then be used to correct the original satellite reflectance data to produce more visually pleasing imagery, also at low sun elevations. Due to the nature of this reduction method they should not be used for any scientific analysis.