Exploring the capabilities of very high spatial resolution satellite imagery to detect sparse entomophilous flowers located in grass strips : a case study with Pleiades imagery

Details

Supervisors

Defourny, Pierre ; Radoux, Julien

Faculty

Faculté des bioingénieurs

Degree label

Master [120] : bioingénieur en sciences et technologies de l'environnement, à finalité spécialisée

Abstract

Nowadays, the urbanisation and the intensification of agriculture are putting pollinators under pressure. The reduction in floral resources may be considered the most worrying threat to which pollinators are subjected. Mapping flowers on a large scale will allow to visualise landscape connectivity and take measures in the most critical areas in order to limit the pollinator decline. In this context, the aim of this master thesis was to extend the exploration of detection possibilities based on satellite images to the case of grass strips. A Pleiades image with a spatial resolution of two metres was analysed, based on data such as flower counts in grass strips on the Marbaix farm, and high spatial resolution drone images. Experiments requiring increasing detection capabilities were then carried out. Firstly, it was assessed in the field that flowering trees were visible on the Pleiades images. The next experiments were based on the results obtained by applying vegetation indices to the Pleiades image. Firstly, a statistical test was used to assess the spectral separability of flowering areas from non-flowering vegetated areas. The link between Pleiades pixel and field counts was then estimated, before carrying out the test specifically for certain species. Finally, the index that gave the best results was applied to the whole Pleiades scene to check the robustness of the relationship obtained. This master thesis showed that Pleiades images have a real potential for detecting such flower covers. The best results were obtained for flowers of the genus Papaver spp, with a Pearson’s coefficient of determination of 86% obtained with the RGBVI2 index. However, this index does not distinguish vegetation from non-vegetated elements.