Biases in morphological landscape features: challenges for environmental purposes in GIScience and related fields

This is the authors' abstract. We don't add key points for this paper.

During the last two decades, a wide range of geographical tools including the calculation of landscape metrics were transposed to ecological studies to build models for land-use dynamics. Currently, few studies have evaluated the biases which can occur during the rasterization step which could influence the results. The purpose of this study was to evaluate the influence of dataset rasterization on area and perimeter variables, which are frequently used to calculate landscape indices, according to (i) the rasterization cell size and (ii) the shape of geographic features. The Urban Atlas 2006 dataset focused on Bas-Rhin department (France) was used as a vector reference layer. Rasterization was performed for various cell sizes to evaluate the influence of spatial resolution on the errors injected into shape descriptors. Five morphological metrics were calculated for all geographic features. For the first time, a UMAP algorithm was performed to relate the rasterization relative errors at all spatial resolutions with morphological attributes. Results showed that low values of area errors were obtained for cell sizes lower than 5 m (<10%). For higher cell sizes, errors exceeding 10% appeared for linear and low width geographic features. For perimeter, significant errors were observed for cell sizes between 1 and 5 m (>10%) with an overestimation tendency. For cell sizes greater to 10 m, overestimations and underestimations were occurring according to the shape of geographic features. This study showed that sensitivity analyses must be performed before any study carried out on landscape changes estimation to define the best raster cell size as function to the morphological attributes of the geographic features, the predefined error threshold.