Knowledge-Based Identification of Urban Green Spaces: Allotments

Abstract. Urban Green Spaces (UGS) play a crucial role in enhancing the quality of life in cities by providing numerous environmental, social, and health benefits. Among these green spaces, allotment gardens stand out as a unique type that contributes to ecological services, preservation of biodiversity, and the overall well-being of urban dwellers. Unfortunately, the significance of allotment gardens as a specific type of UGS is still disregarded and they are not recognized as a separate category in land use / land cover maps or city maps of green spaces. This is mainly due to the mixed use of allotment areas, their small size and absence of tailored identification or mapping workflows. In this research, we address the latter one by proposing an approach that utilizes various semantic characteristics of allotment gardens to create distinctive spatial representations. The semantic characteristics we consider include the presence, density, and height of garden huts, proximity to water bodies and railroads, as well as the presence of pathways within the allotment gardens. Allotments are delineated using a three-step procedure. This involves utilizing a Random Forest machine learning classifier to create maps of the distribution of green spaces, extracting garden huts employing a threshold, and demarcating the area using a density based clustering technique. Furthermore, we repeat the same workflow in a new study area to assess the applicability of the proposed workflow. With the established workflow, we are able to accurately identify 78% of allotments in Augsburg and 88% in Wuerzburg respectively. Our results demonstrate that the proposed workflow can be a useful approach to validate and extend existing land use and land cover data sets while remaining time and cost effective.