Navigation and service

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and Visualization

Principal Investigator: Principal Investigator: Prof. Dr.-Ing. habil. Xiaoxiang Zhu, Signal Processing in Earth Observation, Technical University of Munich and Remote Sensing Technology Institute, German Aerospace Center (Germany)
HPC Platform: SuperMUC of LRZ


Static 3-D city models are well established for many applications such as architecture, urban planning, navigation, tourism, and disaster management. However, they do not represent the dynamic behavior of the buildings and other infrastructure (e.g. dams, bridges, railway lines). Such temporal change, i.e. 4-D, information is demanded in various aspect of urban administration, especially for the long-term monitoring of building deformation. Very high resolution spaceborne Synthetic Aperture Radar (SAR) Earth observation satellites, like the German TerraSAR-X and TanDEM-X provide for the first time the possibility to derive both shape and deformation parameters of urban infrastructure on a continuous basis.

Therefore, this project aims to generate 4-D city models and their user specific visualizations to reveal not only the 3-D shape of urban infrastructures but also their deformation patterns and motion. The research envisioned in this project will lead to a new kind of city models for monitoring and visualization of the dynamics of urban infrastructure in a very high level of detail. The deformation of different parts of individual buildings will be accessible for different users (geologists, civil engineers, decision makers, etc.) to support city monitoring and management and risk assessment.

With the support of Gauss Centre for Supercomputing (GCS), the project has successfully delivered the world’s first city-scale 4-D model derived from spaceborne SAR sensor in 2012. In addition, in total six million CPU-hours have been dedicated to the generation of 4-D city models of various cities in the duration of the project, including Las Vegas, Berlin, Shanghai, Beijing, Washington D.C., and Paris.


More than half of the world’s population lives in urban areas. For instance, in China there were 83 cities with a population over 1.5 million in 2005 [1]. While this urbanization is expected to increase continuously, with around 135 cities having more than 1.5 million inhabitants in 2025 and around 1 billion people living in China’s cities in 2030 [1], the monitoring of the structural healthiness of urban infrastructure gets increasingly urgent. There are several potential threats which may lead to structural degradation and damage of infrastructure, e.g. erroneous construction, bad building quality, subsidence or uplift due to groundwater level underground construction activities, or natural disasters. For example, Figure 1 shows a collapsed building in Shanghai due to unstable geological condition and poor reconstruction. Such event can be prevented if a continuous monitoring of the local ground deformation was conducted.

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and VisualizationFigure 1.Collapsed apartment building due to unstable geological condition of the ground, and poor construction.
Copyright: GettyImages

Methodology and Challenge

The most competent method for assessing long-term millimeter level deformation over large urban areas is the so called differential SAR tomography (D-TomoSAR). D-TomoSAR is able to reconstruct dense 3-D point cloud as well as the deformation parameters of the monitored area. One can imagine D-TomoSAR as dense GPS measurements covering each temporally coherent pixel (usually more than 50% of all pixels) on the acquired SAR image.

However, for retrieving high precision 3-D position and the deformation parameters, D-TomoSAR needs to solve an inversion problem with a typical dimension of 100×1,000,000 (the forward model matrix) for the data typically used. Such inversion problem is repeated for each pixel in the SAR image with a typical size of 6,000 × 10,000. Thus, D-TomoSAR processing on a city scale is computationally expensive.

Another method employed in this project to improve the resolution of the SAR images is the so called non-local (NL) means filtering. NL means filtering searches similar patches in the SAR image in order to significantly reduced the measurements noise while preserving the spatial resolution. As the similar patches have to be searched within the entire image space, it is extremely expensive. Moreover, the computational complexity increases quadratically with the dimension of the image.

The abovementioned method is not feasible for large area processing without high performance computational (HPC) support.


With the support of Leibniz Supercomputing Centre (LRZ) of GCS, the research team so far is the only team in the world that is able to produce the 3-D reconstruction and deformation in city-scale using D-TomoSAR. This project consumed so far in total 6 million core-hours. For each processing job, a stack of tens to hundreds of images was uploaded and processed. Over 500 cores were usually requested for each job. As the computation is also memory intensive, most of the jobs were processed on the fat-island in LRZ. So far, the following datasets have been processed:

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and Visualization

Some representative results are shown in the following content.

Las Vegas

The following upper subfigure is one of the input TerraSAR-X images of Las Vegas. By applying the D-TomoSAR algorithm on tens of such images, a 3-D point cloud was reconstructed (lower subfigure). This point cloud contains around 10 million points. Most importantly, each point contains not only the 3-D position, but also its deformation information with an accuracy of better than millimeter per year (so-called 4-D).

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and VisualizationFigure 2. Upper: TerraSAR-X high resolution spotlight image of Las Vegas, and lower: 3D point cloud of Las Vegas reconstructed using our algorithm. Color represents the height [2], [3].
Copyright: Signal Processing in Earth Observation, TUM

Figure 3 shows an example of the precise deformation discovered in Las Vegas. Since July 2009, the Las Vegas Convention Center is undergoing a pronounced subsidence. The color of the figure shows the estimated linear deformation velocity in mm/year.

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and VisualizationFigure 3. Deformation estimates of an area around the Las Vegas Convention Center: linear deformation velocity (unit: mm/y)[4].
Copyright: Signal Processing in Earth Observation, TUM


By fusing two point clouds reconstructed using SAR images acquired from different viewing angles, one can obtain a complete coverage over an entire city. Figure 4 is the example of Berlin. As always, each point is associated with its movement information. The combined point cloud contains about 40 million points. The number of points exceeds 100 million, if all six reconstructed point clouds of Berlin are combined.

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and VisualizationFigure 4. Fusion of two reconstructed 3D point clouds. The combined point cloud contains over 40 million points [5], [6].
Copyright: Signal Processing in Earth Observation, TUM

NL Means Filter

Figure 5 is a comparison of the standard 12m TanDEM-X 3-D digital elevation model (DEM) and the resolution-enhanced DEM by NL means filtering [7]. The NL means filtered DEM revealed much more detailed structure buried in the noise.

4D City − Space-time Urban Infrastructure Mapping by Multi-sensor Fusion and VisualizationFigure 5. Jülich city area: Optical image ©Google (left), the standard 12m TanDEM-X DEM (middle) and improved nonlocal TanDEM-X DEM with 6m resolution (right) [7].
Copyright: Signal Processing in Earth Observation, TUM

Conclusion and Outlook

With the HPC support of GSC, this project delivered the world’s first 4-D city model derived from spaceborne SAR data. Such 4-D model is essential in continuously monitoring of urban areas. As the data volume of earth observation mission exponentially increase, e.g. the Copernicus programme, HPC will surely be an essential edge in the future research. Such edge also supported us of winning an ambitious European Research Council starting grant project So2Sat: Big Data for 4-D Global Mapping – 1016 Bytes from Social Media to EO Satellites (, whose aim is to provide global 4-D urban models.


[1] J. Woetzel et al., “Preparing for China’s Urban Billion,” McKinsey Global Institute, 2009.

[2] X. Zhu, Very High Resolution Tomographic SAR Inversion for Urban Infrastructure Monitoring: A Sparse and Nonlinear Tour, vol. 666. Deutsche Geodätische Kommission, 2011.

[3] X. Zhu, Y. Wang, S. Gernhardt, and R. Bamler, “Tomo-GENESIS: DLR’s Tomographic SAR Processing System,” in Urban Remote Sensing Event (JURSE), 2013 Joint, 2013, pp. 159–162.

[4] X. X. Zhu and R. Bamler, “Let’s Do the Time Warp: Multicomponent Nonlinear Motion Estimation in Differential SAR Tomography,” IEEE Geosci. Remote Sens. Lett., vol. 8, no. 4, pp. 735–739, 2011.

[5] Y. Wang, X. Zhu, and R. Bamler, “An Efficient Tomographic Inversion Approach for Urban Mapping Using Meter Resolution SAR Image Stacks,” IEEE Geosci. Remote Sens. Lett., vol. 11, no. 7, pp. 1250–1254, 2014.

[6] Y. Wang and X. Zhu, “Automatic Feature-based Geometric Fusion of Multi-view TomoSAR Point Clouds in Urban Area,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 8, no. 3, pp. 953–965, 2015.

[7] X. X. Zhu, R. Bamler, M. Lachaise, F. Adam, Y. Shi, and M. Eineder, “Improving TanDEM-X DEMs by Non-local InSAR Filtering,” in EUSAR 2014; 10th European Conference on Synthetic Aperture Radar; Proceedings of, 2014, pp. 1–4.

Scientific Contact:

Prof. Dr. -Ing. habil. Xiaoxiang Zhu
Signal Processing in Earth Observation
Technische Universität München
Arcisstr. 21, D-80333 München (Germany)
e-mail: xiaoxiang.zhu[at]

August 2017