Propagating Geometric Uncertainties from Image Segmentation to Spatial Indicators of Urban Sprawl

International audience; Urban sprawl is a crucial issue for governments at all administrative levels. Monitoring urban sprawl requires spatial indicators at varying spatial scales. These indicators integrate information about impervious areas, population census, socioeconomic context, agricultural activities and soil quality. All the input data, however, suffer from uncertainties which affect the accuracy of the spatial indicator maps. In this contribution, we aim to: 1) evaluate the geometric and thematic uncertainties of the impervious areas extracted from remote sensing images, 2) propagate them to the spatial indicators using simulation, and 3) produce uncertainty maps for two spatial indicators: the area consumption per inhabitant, and the dispersion of impervious areas. In the region Languedoc-Roussillon, which covers 27,376 km 2 in the south of France, a methodology exists to obtain impervious polygons by segmentation and classification (S dataset) of 5 m resolution RapidEye imagery. Reference impervious polygons (R dataset) were digitalized on the RapidEye images for 75 randomly selected municipalities. The segmentation accuracy was evaluated using the intersection area of S and R, and the boundary distance distribution signature (BDDS) (Huang and Dom 1995). The BDDS is the histogram of the distance between each vertex of S to the closest vertex of R, calculated using only the polygons of S having more than 50% area overlap with R. In a Monte-Carlo set up, 1,000 simulations of impervious polygons were generated using Gaussian random fields of polar coordinates of the polygon vertices. The corresponding 1,000 spatial indicator maps allowed mapping the uncertainties and calculating confidence intervals for each indicator of urban sprawl.