Sentinel 2 Atmospheric Correction in Google Earth Engine

Table of Content

Sentinel 2 Atmospheric Correction in Google Earth Engine


Sentinel 2 Atmospheric Correction in Google Earth Engine

Import modules

and initialize Earth Engine

In [1]:
import ee
from Py6S import *
import datetime
import math
import os
import sys
from atmospheric import Atmospheric


time and place

Define the time and place that you are looking for.

In [2]:
date = ee.Date('2017-01-01')
geom = ee.Geometry.Point(-157.816222, 21.297481)
geom = ee.Geometry.Point(12.670733, 41.826685)# ESRIN (ESA Earth Observation Centre)

an image

The following code will grab the first scene that occurs on or after date.

In [3]:
# The first Sentinel 2 image
S2 = ee.Image(

# top of atmosphere reflectance
toa = S2.divide(10000)


In [4]:
info = S2.getInfo()['properties']
scene_date = datetime.datetime.utcfromtimestamp(info['system:time_start']/1000)# i.e. Python uses seconds, EE uses milliseconds
solar_z = info['MEAN_SOLAR_ZENITH_ANGLE']

atmospheric constituents

In [5]:
h2o = Atmospheric.water(geom,date).getInfo()
o3 = Atmospheric.ozone(geom,date).getInfo()
aot = Atmospheric.aerosol(geom,date).getInfo()

target altitude (km)

In [6]:
SRTM = ee.Image('CGIAR/SRTM90_V4')# Shuttle Radar Topography mission covers *most* of the Earth
alt = SRTM.reduceRegion(reducer = ee.Reducer.mean(),geometry = geom.centroid()).get('elevation').getInfo()
km = alt/1000 # i.e. Py6S uses units of kilometers

6S object

The backbone of Py6S is the 6S (i.e. SixS) class. It allows you to define the various input parameters, to run the radiative transfer code and to access the outputs which are required to convert radiance to surface reflectance.

In [7]:
# Instantiate
s = SixS()

# Atmospheric constituents
s.atmos_profile = AtmosProfile.UserWaterAndOzone(h2o,o3)
s.aero_profile = AeroProfile.Continental
s.aot550 = aot

# Earth-Sun-satellite geometry
s.geometry = Geometry.User()
s.geometry.view_z = 0               # always NADIR (I think..)
s.geometry.solar_z = solar_z        # solar zenith angle
s.geometry.month = scene_date.month # month and day used for Earth-Sun distance =     # month and day used for Earth-Sun distance

Spectral Response functions

Py6S uses the Wavelength class to handle the wavelength(s) associated with a given channel (a.k.a. waveband). This might be a single scalar value (e.g. a central wavelength) or, if known, possibly the spectral response function of the waveband. The Sentinel 2 spectral response functions are provided with Py6S (as well as those of a number of missions). For more details please see the docs or the (comment-rich) source code

In [8]:
def spectralResponseFunction(bandname):
    Extract spectral response function for given band name
    bandSelect = {
    return Wavelength(bandSelect[bandname])

TOA Reflectance to Radiance

Sentinel 2 data is provided as top-of-atmosphere reflectance. Lets convert this to at-sensor radiance for the atmospheric correction.*

*You can atmospherically corrected directly from TOA reflectance. However, I suggest radiance for a couple of reasons. Firstly, it is more intuitive. Instead of spherical albedo (which I suspect is more of a mathematical convenience than a physical property) you can use solar irradiance, transmissivity, path radiance, etc. Secondly, Py6S seems to be more geared towards converting from radiance to SR</sup>

In [9]:
def toa_to_rad(bandname):
    Converts top of atmosphere reflectance to at-sensor radiance
    # solar exoatmospheric spectral irradiance
    ESUN = info['SOLAR_IRRADIANCE_'+bandname]
    solar_angle_correction = math.cos(math.radians(solar_z))
    # Earth-Sun distance (from day of year)
    doy = scene_date.timetuple().tm_yday
    d = 1 - 0.01672 * math.cos(0.9856 * (doy-4))#
    # conversion factor
    multiplier = ESUN*solar_angle_correction/(math.pi*d**2)

    # at-sensor radiance
    rad =
    return rad

Radiance to Surface Reflectance

Reflected sunlight can be described as follows (wavelength dependence is implied):

$ L = \tau\rho(E_{dir} + E_{dif})/\pi + L_p$

where L is at-sensor radiance, $\tau$ is transmissivity, $\rho$ is surface reflectance, $E_{dir}$ is direct solar irradiance, $E_{dif}$ is diffuse solar irradiance and $L_p$ is path radiance. There are five unknowns in this equation, 4 atmospheric terms ($\tau$, $E_{dir}$, $E_{dif}$ and $L_p$) and surface reflectance. The 6S radiative transfer code is used to solve for the atmospheric terms, allowing us to solve for surface reflectance.

$ \rho = \pi(L - L_p) / \tau(E_{dir} + E_{dif}) $

In [10]:
def surface_reflectance(bandname):
    Calculate surface reflectance from at-sensor radiance given waveband name
    # run 6S for this waveband
    s.wavelength = spectralResponseFunction(bandname)
    # extract 6S outputs
    Edir = s.outputs.direct_solar_irradiance             #direct solar irradiance
    Edif = s.outputs.diffuse_solar_irradiance            #diffuse solar irradiance
    Lp   = s.outputs.atmospheric_intrinsic_radiance      #path radiance
    absorb  = s.outputs.trans['global_gas'].upward       #absorption transmissivity
    scatter = s.outputs.trans['total_scattering'].upward #scattering transmissivity
    tau2 = absorb*scatter                                #total transmissivity
    # radiance to surface reflectance
    rad = toa_to_rad(bandname)
    ref = rad.subtract(Lp).multiply(math.pi).divide(tau2*(Edir+Edif))
    return ref

Atmospheric Correction

In [11]:
# surface reflectance rgb
b = surface_reflectance('B2')
g = surface_reflectance('B3')
r = surface_reflectance('B4')
ref = r.addBands(g).addBands(b)

# # all wavebands
# output ='QA60')
# for band in ['B1','B2','B3','B4','B5','B6','B7','B8','B8A','B9','B10','B11','B12']:
#     print(band)
#     output = output.addBands(surface_reflectance(band))

Display results

In [12]:
from IPython.display import display, Image

region = geom.buffer(5000).bounds().getInfo()['coordinates']
channels = ['B4','B3','B2']

original = Image({

corrected = Image({

display(original, corrected)

Export to Asset

In [ ]:
# # set some properties for export
# dateString = scene_date.strftime("%Y-%m-%d")
# ref = ref.set({'satellite':'Sentinel 2',
#               'fileID':info['system:index'],
#               'date':dateString,
#               'aerosol_optical_thickness':aot,
#               'water_vapour':h2o,
#               'ozone':o3})
In [ ]:
# define YOUR assetID 
# in my case it was something like this..
# assetID = 'users/samsammurphy/shared/sentinel2/6S/ESRIN_'+dateString
In [ ]:
# # export
# export = ee.batch.Export.image.toAsset(\
#     image=ref,
#     description='sentinel2_atmcorr_export',
#     assetId = assetID,
#     region = region,
#     scale = 30)

# # uncomment to run the export
# export.start() 

0 thoughts on “Sentinel 2 Atmospheric Correction in Google Earth Engine”

    1. I’m not sure if it can be used for landsat8 or not, I tried to find out that someone said this:

      GEE doesn’t have a specific algorithm to atmospherically correct scenes, just Cloud Score in Landsat case. Instead to trying to correct images, use Surface Reflectance products:

      USGS Landsat 8 Surface Reflectance Tier 1

      This dataset is the atmospherically corrected surface reflectance from the Landsat 8 OLI/TIRS sensors. These images contain 5 visible and near-infrared (VNIR) bands and 2 short-wave infrared (SWIR) bands processed to orthorectified surface reflectance, and two thermal infrared (TIR) bands processed to orthorectified brightness temperature

      These data have been atmospherically corrected using LaSRC and includes a cloud, shadow, water and snow mask produced using CFMASK, as well as a per-pixel saturation mask.

      Strips of collected data are packaged into overlapping “scenes” covering approximately 170km x 183km using a standardized reference grid.


      var imageCollection = ee.ImageCollection(“LANDSAT/LC08/C01/T2_SR”);

  1. hello there,
    in [11] i get this error:ExecutionError: 6S executable not found.
    and for display result i get this :NameError: name ‘ref’ is not defined.

    what should I do?!

Leave a Reply

Your email address will not be published. Required fields are marked *