Mapping past land cover on Poitiers in 1993 at very high resolution using GEOBIA approach and open data

Abstract

The land cover data presented here is a reconstruction of the past landscape (1993) at Very High Resolution (VHR) for the city of Poitiers, France. This reconstruction is based on multiple sources of images and data. We combined the strengths of both mono-temporal and multi-temporal classifications. Orthophotos were created at a spatial resolution of 0.5 m using aerial raw images from the French National Geographic Institute (IGN), taken during two aerial missions in July and August 1993. These orthophotos were merged at a spatial resolution of 5 m to conduct a first object-based classification using Landsat-5 TM images. The goal was to identify croplands, grasslands, coniferous and deciduous forests, urban areas, water bodies, and shadows. This learning-based classification employed a dataset consisting of 1371 polygons and demonstrated strong classification performances, achieving an overall accuracy of 86.31% and a kappa index of 0.832. On the other hand, mono-temporal classifications at a 0.5 m spatial resolution were carried out on each orthophoto to extract trees and herbaceous vegetation, especially in urban contexts. As mono-temporal classifications contained less information, we used a larger number of polygons for the learning step: 3849 and 5173 polygons for the northern and southern classifications, respectively. The segmentation step performed better in urban areas compared to rural areas. Consequently, the performance of classifications was evaluated separately for both contexts. Urban areas exhibited excellent performances, achieving kappa indices of 0.897 and 0.881 for the northern and southern classifications, respectively, whereas only tree vegetation was accurately detected in rural areas. To compensate for the lack of information such as buildings, railways, or roads, we modified the BD Topo^Ⓡ dataset from IGN. This land cover map provides highly detailed information, facilitating the understanding of urban sprawl and changes in urban and rural vegetation surrounding the city of Poitiers. Due to these reasons, this freely accessible map can be utilized by researchers, land managers, and private companies for addressing urban and ecological challenges.

Keywords: Aerial images; GEOBIA; Land cover map; Landsat-5; Poitiers; Remote sensing.