TITLE:
Complementarity of Sentinel-1 and Sentinel-2 Data for Mapping Agricultural Areas in Senegal
AUTHORS:
Gayane Faye, Fama Mbengue, Lacina Coulibaly, Mamadou Adama Sarr, Modou Mbaye, Amath Tall, Dome Tine, Omar Marigo, Mouhamadou Moustapha Mbacke Ndour
KEYWORDS:
Agricultural Areas, Remote Sensing, Sentinel-1, Sentinel-2, Senegal
JOURNAL NAME:
Advances in Remote Sensing,
Vol.9 No.3,
September
30,
2020
ABSTRACT: The
small size of agricultural plots is the main difficulty for crops mapping with
remote sensing data in the Sahelian region of Africa. The study aims to combine
Sentinel-1 (radar) and Sentinel-2 (Optic) data to discriminate millet, maize
and peanut crops. Training plots were used in order to analyse temporal
variation of the three crops’ signals. The
NDVI (Normalized Difference Vegetation Index) was able to differentiate crops
only at the end of the rainy season (October). The
optical data as well as the radar ones could not easily discriminate the three
crops during the growing season, because in that period vegetation cover is
low, and soil contribution to the signals (due to roughness and moisture) was
more important than that of real vegetation. However, the ratio of VH/VV (VH:
incident signal in vertical polarization and reflected signal in horizontal
polarization; VV: incident signal in vertical polarization and reflected signal
in horizontal polarization) gave a difference between millet and the two other
crops at the beginning cultural season (July 11). Difference appears from the
second third of September when the harvest of cereals crops (millet and maize)
began. From middle of October, the peanut signal dropped sharply thus
facilitating the differentiation of peanut from the two other crops. This
analysis led to the identification of data that have could be used to
discriminate these crops (useful data). Classification of the combined useful
data gave an overall high accuracy of 82%, with 96%, 61% and 65% for peanut,
maize and millet, respectively. The non-agricultural areas (water, natural
vegetation, habit, bare soil) were well classified with an accuracy greater
than 90%.