Large-Area Characterization of Urban Morphology—Mapping of Built-Up Height and Density Using TanDEM-X and Sentinel-2 Data

Abstract

In this paper, we establish a novel multistep procedure for morphologic characterization of built environments in terms of built-up height and density. Thereby, we rely on elevation measurements from the TanDEM-X mission (TDM) and multispectral Sentinel-2 imagery. These earth observation systems feature a notable tradeoff between a fairly high spatial resolution and large-area coverage and, thus, allow for spatially continuous analysis of built environments around the globe. To this purpose, we follow an automated workflow which foresees the distinction of "built-up" and "non built-up" areas by relying on the so-called Global Urban Footprint processor (step 1). This information is deployed within a tailored filtering procedure for the TDM digital surface model data to extract elevation information for built-up areas (step 2). Subsequently, the intra-urban land cover is mapped under consideration of Sentinel-2 imagery and serves as basis to compute built-up heights and densities (step 3). These two measures are finally combined for a morphologic characterization of the built environment on an ordinal scale of measurement. Empirical validation efforts are provided based on comparative analysis with respect to more than 3.2 million individual building geometries and affiliated height measurements from cadastral data sources. The data sets cover the settlement areas of the capital cities and other major cities in Germany, England and the Netherlands. The experimental results underline the capability for a morphologic characterization of built environments with viable accuracies.

Multistrategy ensemble regression for mapping of built-up density and height with Sentinel-2 data

“…The spatial processing units constrain analyses explicitly to settlement areas. Subsequently, a training set is compiled by either relying on an automatic workflow which estimates built-up density and height based on areas where both TDM elevation data and Sentinel-2 imagery are available (Geiß et al, 2017a(Geiß et al, , 2019. In addition, we evaluate the incorporation of geospatial vector data from cadastral sources for this purpose (details can be found in the experimental setup section).…”

“…The first strategy builds upon the TDM and Sentinel-2 data to map built-up density and height with an automatic workflow (Geiß et al, 2017a(Geiß et al, , 2019. First, it foresees the distinction of "built-up" and "non built-up" areas by relying on the so-called GUF processor (Esch et al, 2012(Esch et al, , 2017.…”

“…Sentinel-2 imagery was already deployed to estimate the degree of imperviousness (Lefebvre et al, 2016;Xu et al, 2018). However, a joint derivation of built-up density and height was just recently proposed by Geiß et al (2017Geiß et al ( , 2019 based on an integrative analysis of TDM and Sentinel-2 data. As such, these data sets allow for a unique mapping of urban morphology around the globe for large areas.…”

“…With it, we aim to estimate the target variables for areas where only Sentinel-2 data is available. Consequently, the regression estimation problem is regarded as finding the mapping between an incoming vector ∈ ℝ , i.e., ubiquitously available features computed from Sentinel-2 imagery, and an observable output ∈ ℝ (i.e., built-up density and height) from a given pool of labeled training samples = , , , i.e., automatically derived according to the workflow proposed in Geiß et al (2017Geiß et al ( , 2019 over spatially limited areas where data from both TDM and Sentinel-2 are available. In addition, we evaluate the incorporation of geospatial vector data, i.e., level of detail 1 (LoD-1) building geometries from cadastral sources, which also enable the computation of built-up density and height and, thus, allow compiling a training set.…”

Towards a Large-Scale 3D Modeling of the Built Environment—Joint Analysis of TanDEM-X, Sentinel-2 and Open Street Map Data

“…The same method was later employed by Clinton et al [26] to conduct a worldwide estimation of built-up heights from the 30 m data of the AW3D30 DSM. Geiß et al [27] used the TDX-DEM elevation data in combination with Sentinel-2 imagery to estimate the built-up height and density for selected urban areas. Recently, Li et al [28] presented an approach for the mapping and analysis of 3D building structures based on random forest models applied to a large collection of earth observation data to produce maps of 3D building structures for China, Europe and the United States.…”