October 28, 2014

Level 2 product

Simulation of level 2 products with Formosat-2

Tensift data set (Morocco)
(animation at better resolution)
Muret data set (Southwest France)
(animation at better resolution)
These Images are simulations of Venµs level 2 products, from Formosat-2 data sets. These data are orthorectified, registered, calibrated and converted to top of atmosphere reflectance at level 1, then a cloud screening and an atmospheric correction are performed. The cloud screening algorithm also detects cloud shadows, in these images, both clouds and shadows appear in black. Atmospheric correction for these images is not done using the Venµs algorithm described below. An aeronet sunphotometer is present on both sites to obtain Aerosol Optical Properties.

Description of level 2 product

Product Content

For the level 2 products definition, two basic facts have been taken into account:

  • The Venµs data set will have the following unique features: multi-temporal, high resolution products, with low directional effects. The algorithms to derive bio-physical variables from these data are far from being mature, and it is the aim of Venµs to help develop these algorithms.
  • The Venµs data set is made of 50 different local data sets on 50 sites located around the world, there is no global data set.

For these reasons, it does not seem relevant to develop very complex biophysical inversion algorithms that will have to work on 50 very different sites with different vegetation characteristics and with different applications. At least for the first generation of Venµs products, it seems more relevant to limit the level 2 processing to what will be common to most applications: cloud screening and atmospheric corrections. If some algorithms developed by Venµs users for their sites seem useful for other users and sites, such algorithms might be added to a second generation of Venµs products.

The Venµs level 2 product will provide:

  • a refined cloud mask
  • surface reflectance after atmospheric corrections for all spectral bands (still geolocated)
  • confidence values
  • maybe some vegetation indexes

While level 1 is produced at 5 m resolution (we want to preserve all the information for the Level 1 product), the level 2 geometric resolution is 10 m, because the Venµs Signal to Noise Ratio requirements will only be met at 10 m resollution.

Algorithm description

See below for more information

Cloud Detection

Usually, cloud detection is based on spectral indexes, and relies on spectral bands in the blue where surface reflectances are usually low, and cloud reflectances are relatively high. However, some surfaces have higher reflectances in the blue (deserts, buildings) that prevent the use of very low threshold values. Venus sensor offers two other criteria to detect clouds:

  • the altitude of observed surfaces using Venµs stereoscopic bands (from level 1)
  • the slow variation of ground reflectances

A combination of these 3 criteria will be used to detect the clouds. The use of the criterion on slow variation of ground reflectances has a drawback: level 2 products must be produced sequentially and rely on a few level 1 products.

Atmospheric Correction

For the atmospheric correction, 2 terms are to be corrected

  • absorption
  • scattering


The gaseous absorption correction is done using SMAC model ( Rahman, Dedieu, 1994) and data coming from

  • other satellites (ozone from TOMS)
  • altitude information (pressure to correct oxygen absorption)
  • Venµs images (910 spectral band to correct water vapour absorption)

Rahman, H, et G Dedieu. 1994. SMAC: a simplified method for the atmospheric correction of satellite measurements in the solar spectrum. International journal of remote sensing 15 (1): 123-143.


Scattering comes from either molecules or aerosols. The contribution of molecules depends mainly on the pressure, and thus on the altitude, a Digital Terrain Model will be used as input of the atmospheric correction. The contribution of aerosols is much more difficult to handle, because aerosol optical properties (AOP) are very variable in time and place. Up to now, the only reliable external data source to obtain AOP is to have a sun photometer on the site.

Thanks to Venµs repetivity at constant viewing angles, it is possible to measure directly the AOP from Venµs imagery itself. To do this, we intend to use the following properties:

  • the AOP vary slowly with location, the surface reflectances vary quickly with location
  • the AOP vary quickly with time, the surface reflectances vary slowly with time
  • surface reflectance vary quickly with observation and solar angle, but the observation angles do not vary for Venµs time series, and solar angles vary very slowly.

The principle of the atmospheric correction method is to use the data acquired during a short period of time to obtain the AOP. As a result, a level 2 product will not be obtained from a single level 1 product, but from a sequence of successive level 2 products. Once the AOP are obtained, the atmospheric correction can be done, at full resolution. The atmospheric correction will be based on Look-up tables.

A demonstration of Venµs method for aerosol detection has been done using Formosat2 data. This poster shows the first results.

Slope Correction

The illumination of a given pixel depends on the slope of the terrain. If we do not take this phenomenon into account, the reflectance estimation will be erroneous when the observed surface is not horizontal. For Venµs level 2, we intend to introduce a simple correction of illumination, using a Digital Terrain Model (DTM) as input, and assumiing, as a first order estimation, that the observed surface is lambertian.