Jean Novais
University of BrasÃlia, Brazil
Title: Spectral mixture model and Sentinel-2A time-series for digital soil mapping in Brazilian Cerrado (savanna)
Biography
Biography: Jean Novais
Abstract
One of the main functions of soil is that of climate regulation on Earth. In-depth knowledge about this natural resource is indispensable for management and conservation actions. One way to know the soil is through Digital Soil Mapping (DSM). Thus, the objective of this work was to evaluate a short Sentinel-2A (S-2A) time-series using topsoil reflectance spectra and multiple endmember spectral mixture analysis (MESMA) for DSM. The study area has about 152 Km2; it is located in Central Brazil and developed over metasedimentary rocks. We collected 19 representative soil samples using toposequence method to physical, chemical and spectroscopic analyzes. So, we proceeded to the classification to the third categorical level of the World Reference Base for Soil Resources (WRB - FAO/UN). The similar spectra soils were grouped in clustering analysis by Euclidean distance. After, we acquired a surface reflectance time-series of 16 S-2A images (Level 2A) from 2015 to 2019 during the dry season. We applied a bare soil mask in all simple images which were used to calculate the pixel median and to produce a synthetic image of reflectance surface. Finally, the soil spectra (endmembers) were applied as inputs on the MESMA algorithm which got to model 98.7% with low global RMSE (0.81%) and high global fraction (60%). MESMA-derived DSM reached a Kappa coefficient of 0.73, indicating a good agreement with the field-verification sites. The short S-2A time-series evaluation showed that has high potential for DSM which tends to improve with data availability over time.