Urban sprawl monitoring is important for developing land management policies at various spatial scales. Segmentation and classification of satellite images allows obtaining polygons of impervious areas regularly over large areas, e.g. as has been implemented for the region Languedoc‐Roussillon in the south of France using 5 m RapidEye images. Starting from the results of this previous study, we aim to: i) evaluate the geometric and thematic accuracy of the impervious polygons (S) using segmentation accuracy metrics, and ii) use these metrics to simulate polygons having the same level of uncertainty. A manual segmentation (M) was used to evaluate the accuracy. After matching the polygons, the distance (d) and azimuth (a) of each vertex of M to the closest segment of the boundary of S was calculated. Spherically correlated random fields of d and a were used to randomly move the vertices of S. Realistic simulations of impervious polygons were obtained.