Une méthode simple et rapide pour l'évaluation de statistiques d'occupation du sol à l'aide d'images à très haute résolution acquises par mini-drone

Abstract

Land use monitoring by remote sensing techniques has been developing rapidly, thanks to much easier access, often free of charge, to (very) high-resolution images, and to the development of specific Web applications for land use monitoring. However, access to these applications depends on the existence of a reliable internet connection, which is still lacking in some regions of the world. This study describes a land-use monitoring method based on point-by-point photo-interpretation of very high-resolution images acquired by small drones. The method requires the integration of an application (PINT, for Photo-INTerpretation) into QGIS Open source software. The areas occupied by different land uses are derived from the estimated proportions of the points allocated to each land-use class, based on a systematic grid. To illustrate the advantages of the tool, this study investigated the land-use statistics for two villages in the Greater Garamba Complex of protected areas, in the Democratic Republic of Congo. The results obtained were compared with those of a reference map, on the basis of exhaustive photo-interpretation after segmentation of the images. The differences between the areas estimated by sampling and the reference areas vary from 0.2% to 6.1% for the main land uses (forests and savannas, clearings, fallows, human settlements and crops). Larger differences (17.4% and 13.4%) were recorded for the "isolated trees" class. Implementing the method takes about 1 hour per operator for 60 ha. Using the PINT plugin with drone images appears to be a relevant solution to estimate land-use statistics in Web-isolated regions, for areas of a few to a few dozen km2.