Estimation of riparian irrigated areas using Earth Engine. A case study in the Upper-Comoé watershed, Burkina Faso

Abstract

The implementation of an automatic method for mapping smallholder irrigated plots along river banks in Africa based on remote sensing remains a major challenge because of the diversity of crops grown, the narrowness of the plots, the variability of crop cycles and the similarity of reflectance of irrigated areas, wetlands herbaceous areas and riparian woody vegetation. The purpose of this paper was to design a method enabling an automatic estimate of irrigated croplands on the banks of the Comoé river. To do so, a composite image based on Sentinel-1 and 2 images and spectral indices NDVI, MNDWI, NBR2 were used for supervised classification thanks to Earth Engine platform. A sequence of automatic masks of bare lands, water bodies, infrastructures and the riparian forest was first implemented, before doing a Random Forest classification. Tested with the January 2019 dataset, the proposed method made it possible to distinguish effectively the 'irrigated areas'and 'herbaceous wetlands'classes with an overall accuracy of 98%, a Kappa coefficient of 0.91 and F-scores of 99% for irrigated areas and 92% for herbaceous wetlands. The study showed that it is possible to successfully implement a low-cost and automatic method for monitoring riparian irrigated plots areas along the Comoé river, using the cloud-based Earth Engine platform.