Mapping of tank silt application using Sentinel-2 images over the Berambadi catchment (India)

Mapping soil properties is becoming more and challenging due to the increase in anthropogenic modification of landscape, calling for new methods identify these changes. A striking example modifications widespread practice South India applying large quantities silt from dry river dams (or “tanks”) agricultural fields. Whereas several studies have demonstrated interest tank fertility, no assessment actual extent this age-old traditional exists. Over South-Indian pedological context, characterized by an application black-colored red-colored soils such as Ferralsols. The objective work was evaluate usefulness Sentinel-2 images mapping applications, hypothesizing that observed changes surface color can be a proxy application. We used data collected cultivated watershed including 217 samples terms Munsell color. two acquired on February April 2017. over each image classified into types (“Black” “Red” soils). change 2017 “Black” attributed Soil were analyzed accounting possible moisture proposed methodology based well-balanced Calibration created initial imbalanced dataset thanks Synthetic Minority Over-sampling Technique (SMOTE) methodology, coupled Cost-Sensitive Classification And Regression Trees (Cost-Sensitive CART) algorithm. To estimate uncertainties i) two-class classification at date ii) “Black”, bootstrap procedure providing fifty classifications image. results showed 1) CART method allowed classify with correct overall accuracy both images, 2) identified 202 fields 3) not related between dates. With availability past LANDSAT satellite imageries, study may open way toward simple accurate delivering so possibly quantify retroactively farmer practice.