CONTRIBUTION OF REMOTE SENSING TO ONION AGRICULTURAL SYSTEM IMPROVEMENT IN GUIDIMOUNI BASIN (ZINDER REGION)

Abstract

New information and earth observation technologies, remote sensing and the Geographic Information System (T-GIS), have become very effective tools in crop mapping for better management of agricultural plots. The objective of this study is to propose a combination of spectral indexes (SI) based on a time series of sentinel-2 images by comparing the performance of three (3) classification algorithms, namely: Wide Margin Separators (WMS), the Random Forest (RF) and the Decision Trees (DT) in order to produce a map of land cover (LC) in the gardening sites of Guidimouni by analysis of pixel-based images. Five classes of land cover have been retained, namely: Onion, Other plants, Flood zones, Water, and Non-vegetation. 24 optical images were processed to draw the temporal curves of cultures that will be used for processing. 51 classification schemes were tested and evaluated. Thus, the different values of the NDVI time series made it possible to observe three stages linked to the cultural development of onion, namely stage-1 (beginning of germination and the appearance of the first leaves), stage-2 (early bulb formation and strong chlorophyll activity) and stage-3 (bulb thickening and leaf yellowing). On the other hand, the WMS classifiers made it possible to obtain a better mapping of land cover in terms of discrimination between classes by combining the five spectral indexes (NDVI, SAVI, EVI, IV, BI2) with a cartographic precision of 88.89%.