Image processing and remote sensing data

Given its good spatial resolution, we selected multispectral S2 imagery for this study. Ten S2 bands were used, i.e., four bands at 10 m (B2-490 nm, B3-560 nm, B4-665 nm, B8-842 nm) and six bands at 20 m (B5-705 nm, B6-740 nm, B7-783 nm, B8a-865 nm, B11-1610 nm, B12-2190 nm) (ESA 2015). There are three bands in the visible spectrum, a band to the near infrared, four bands located in the red edge and two bands of SWIR. S2 has a temporal resolution of 5 days with both satellites operating (S2-A, S2-B), or 10 days referring to a single satellite. The S2 images used are Level-1C processing products (ESA 2015) (see Table ​Table1),1), downloaded from Global Visualization Viewer (GloVis) web service of the United States Geological Survey (USGS 2017).

Sentinel-2 images used

UTM

Zone 17S

Datum GS84

^a Cloud cover (CC) percentage

Level-1C products include radiometric and geometric corrections, and also orthorectification and spatial registration in a global reference system using a Digital Elevation Model (DEM) to project the image in cartographic coordinates (ESA 2015), leading to Top Of Atmosphere (TOA) reflectances. To obtain per-pixel Bottom Of Atmosphere (BOA) reflectances, i.e. processing Level 2A, the Sen2Cor Toolbox version 2.4 through the Sentinel Applications Platform (SNAP) software version 5.0 was used (see Fig. 2a). The complex study area is typically characterized by a high percent of cloud cover, which demands for a special treatment for encapsulation and removal of clouds. The clouds encapsulation was realized using the additional Cloud Mask products combined with manual identifying and removing of clouds. Finally, a BOA mosaic of the herbaceous páramo ecosystem was obtained with ArcGis 10.2 software (see Fig. 2b).

a Atmospheric correction and creation of clouds mask based on additional products from S2, b Process to eliminate clouds and creation of final mosaic

This study uses the S2 spectral indices NDVI, SAVI, WDRVI, EVI2, NDWI, VARIg, NDSI, BI, NDMI, NBR, NBR2 and adds a physiologically-based green leaf area index (LAIgreen index) known as the Sentinel-2 LAIgreen Index (SeLI) (Pasqualotto et al. 2019) (see Table ​Table2).2). The SeLI index has the potential to be used in a unified algorithm for LAIgreen estimation or to identify bare areas, wherefore it can provide relevant information as a SOC storage indicator.

S2-Spectral indices evaluated in the regression model

SAVI=NIR-RNIR+R+L1+L

L=0.15

WDRVI=aNIR-RaNIR+R

a=0.05

Punctual spectral information (i.e., individual S2-bands, spectral indices) is extracted from the final mosaic corresponding to the SOC sampling databases (493 SOC data points in the 0–30 cm profile and 464 SOC data points in the 30–60 cm profile); this database is used for training the regression model for SOC prediction. It contains spectral information of the input variables to be evaluated as predictors (S2-bands), spectral indices, meteorological and soil data, and also biophysical data). In the same way, the final mosaic is used to create an extensive geo-database of points, one point per pixel in the study area. The database contains information on the found SOC predictors, and it is used to obtain the SOC prediction values using the trained and optimized regression model.