Search

In this paper, a new approach for mapping polygenic depressions colonised by vegetation is presented. These spatially periodic vegetation patterns situated in the arid areas of North Africa are known locally as "Dayas". The mapping of these structures is an important component in monitoring their evolution which can be regarded as an indicator of socio-environmental conditions. For this purpose, a method based on satellite image analysis is proposed. First, Landsat Thematic Mapper image inspection is performed to highlight the vegetation spots using the Normalized Difference Vegetation Index (NDVI). Then, a Differential Morphological Profile (DMP) which includes both spectral and morphological information, is built for each NDVI pixel. We adapt the F-Score technique combined with the SVM classifier, to select and classify the most effective DMP images in order to extract the Daya's signature. Using feature selection, overall classification accuracy reached 97.8% with a kappa value of 95.61 but also reduced the dimension of the DMP vector. Dayas' morphological and vegetation characteristics are related to their degree of evolution. Furthermore, we categorise categorise the 170 extracted Dayas into three classes according to their size. On the resulting map, each class corresponds to a specific morphological and vegetation evolution stage. • The established methodology allows Dayas identifying in semi-arid areas. • Differential Morphological Profile is a powerful tool for spots vegetation mapping. • Size-based classification of Dayas enables a chronological distribution. • F-Score combined with SVM for feature selection improves classification accuracies. [ABSTRACT FROM AUTHOR]