Fractional Vegetation Cover Data Extraction from Sentinel-2 Imagery Using ESA SNAP

Vegetation cover data is one of the most important biophysical parameters that can be effectively and efficiently extracted from Remote Sensing Satellite Imagery. A wide range of vegetation cover extraction methods have been developed since years ago, ranging from indirect regression using NDVI data and field measurements, greenness index conversion, and many others. These methods can give vegetation cover information in raster format that can be used for many kinds of applications. For example, you can incorporate vegetation cover data as additional input for biomass estimation, soil erosion modeling, vegetation cover multitemporal analysis, et cetera. 

Vegetation cover is best estimated using low to medium-resolution satellite imagery. On this one, these days we have so many options of satellite images, and most importantly, they are mostly free to use and free to download. LANDSAT, ALOS AVNIR, SPOT 5/6/7, SENTINEL-2, MODIS, ASTER just name a few of them. Specifically for LANDSAT-8 and SENTINEL-2, ESA has been developed a specific module to extract vegetation cover data along with other useful information like Leaf Area Index, Canopy Water Content, chlorophyll content, and a few more. This module has been implemented in ESA SNAP software that you can get and download for free from this LINK. 

And if you are curious about how to do it, I already made a video tutorial below that demonstrates in a step-by-step way how to get fractional vegetation cover, using Sentinel-2 imagery as an example. Mind that this workflow also can be applied to LANDSAT-8 Imagery. 

last but not least, there is some consideration about using the method I describe in the video for successful Vegetation cover data extraction. Here it is:

1. Don’t use Sentinel-2 L1C. Just use L2A because L2A data is already atmospherically corrected (bottom of atmosphere reflectance), atmospheric corrected data will give more consistent FVC results. 

Get imagery with no clouds cover. Cloud cover will cause the algorithm not to perform as intended and lead to over-estimation of FVC. 

2. The S2 SNAP Toolbox biophysical variable retrieval algorithm is based on specific radiative transfer models associated with strong assumptions, particularly regarding canopy architecture (turbid medium model). All the variables derived from such algorithms should be seen as effective, i.e. the variables that would correspond to the measured satellite signal reflected by a canopy verifying all the assumptions made through the radiative transfer models. Depending on the variable, this may lead to differences with ground values that may be accessed from field measurements. 

3. Just like other Remote Sensing based FVC extraction algorithms, these algorithms assumed vegetation homogeneity, so implementation for strong indication of heterogeneity vegetated areas could lead to uncertainty of FVC values