TOC_VEG Processor Overview

 

The proposed algorithm called here TOC_VEG is based on the training of neural networks over a data base simulated using radiative transfer models. It requires top of canopy MERIS reflectances and associated geometry as inputs and allows the derivation of four vegetation biophysical variables:

 

·  fAPAR:  it corresponds to the fraction of photosynthetically active radiation absorbed by the canopy and depends both on the canopy structure and illumination conditions. In order to be consistent with most of the potential users who are dealing with daily integrated fAPAR value as well as with other fAPAR products that are considering the instantaneous fAPAR value at the time of the satellite overpass under clear sky conditions (e.g. MODIS), we decided to use the fAPAR at 10:00 solar time under clear sky conditions. Note also that the fAPAR refers only to the green parts (leaf chlorophyll content higher that 15µg.cm-2) of the canopy.

·  fCover: it corresponds to the gap fraction for nadir direction. It is a canopy intrinsic variable that does not depend on variables such as the geometry of illumination as compared to fAPAR. For this reason, it is a very good candidate for the replacement of classical vegetation indices for the monitoring of green vegetation. Note that similarly to LAI and fAPAR, only the green elements (leaf chlorophyll content higher that 15µg.cm-2) will be considered.

·  LAI: it is defined as half the developed area of green (leaf chlorophyll content higher than 15 µg.cm-2) vegetation elements per unit of horizontal soil Note that LAI of vegetation as estimated from remote sensing will include all the green contributors, i.e. including understory when existing under forest canopies.

·  LAI.Cab: the chlorophyll content can be calculated both at the leaf level and at the canopy level by multiplication of the leaf level chlorophyll content by the leaf area index. As the medium resolution scale of MERIS is generally associated with heterogeneous pixels, the product LAI.Cab more sound than the leaf level chlorophyll content: what would be the chlorophyll content of a pixel with half of very sparse canopy with very high leaf chlorophyll content and half of very dense canopy with very low leaf chlorophyll content? Therefore, the estimation of LAI.Cab has been preferred to that of the leaf chlorophyll content.

 

 

TOC_VEG Algorithm specification

 

 

The proposed algorithm called here TOC_VEG is based on the training of neural network over a data base simulated using radiative transfer models (RTM, Figure 1). The SAIL and PROSPECT models are coupled and used to simulate the reflectance in the 11 MERIS bands considered (490 nm, 510 nm, 560 nm, 620 nm, 665 nm, 681.25 nm, 708.75 nm, 753.75 nm, 778.75 nm, 865 nm, 885 nm). The oxygen and water absorption bands have not been used because they would convey significant uncertainties associated while providing only marginal information on the surface. The two MERIS first bands are also eliminated since they correspond to strong Rayleigh and aerosol scattering and to low canopy reflectances, this induces that top of canopy reflectance in this region are very sensitive to errors in the atmospheric correction. The background optical properties are simulated using a collection of soil, water and snow typical reflectance spectra. A brightness factor is used to provide additional flexibility of the background reflectance. Finally, to account for the medium resolution of MERIS observations, mixed pixels are simulated with variable fractions of pure background and pure vegetation.

 

The simulation of the top of atmosphere reflectance in the 11 MERIS bands requires 14 input variables. They were drawn randomly according to an experimental plan aiming at getting a more evenly populated space of canopy realization. To provide more robust performances of the network, the distributions of each input variable was close to the actual distributions and, when possible. Realistic co-distributions were also used. This was achieved by considering a representative distribution of targets over the earth surface that constrains the observation geometry, as well as possible vegetation amount. A total number of 46 533 cases were simulated. Half of this data set was used for training, one quarter to evaluate hyper-specialization, and the last quarter to quantify the theoretical performances.

 

Back-propagation neural network was trained for estimating the four variables considered. The architecture was optimized, resulting in 2 hidden layers of tangent-sigmoid neurones corresponding to a total around 300 coefficients to adjust, and providing a good ratio (50-100) with the size of the training data base.

 

 

 

Figure 1. Overview of the TOC_VEG algorithm

 

The theoretical performances were evaluated over the test simulated data set. They are close to 0.05 (absolute value) for fAPAR and fCover, and close to 25% (relative value) for LAI and LAI.Cab that shows some loss of sensitivity for the larger values of LAI and LAI.Cab due to saturation effects.

 

 

TOC_VEG processor descriptor

 

 

The TOC_VEG In- and Output Parameters tab

1.      Input product file: Select the input product file by either typing the product path into the text filed or by pressing the file selection button to invoke a standard file dialog.

  1. Output product file: Select the output product file by either typing the product path into the text filed or by pressing the file selection button  to invoke a standard file dialog.
  2. Output product format: Select one of the available output product formats. Currently BEAM-DIMAP, HDF5 and GEOTIFF format are available.
  3. Bitmask: select the flags you want to apply to the MERIS image before applying the TOC-VEG processor