The newest generation of satellites can provide us with incredibly precise data about the surface of the Earth. Professor Xiaoxiang Zhu has been using high performance computers to decipher this data and is using it to provide support to some of the poorest urban areas in the world.
Earth observation is the gathering of information about the physical, chemical, and biological systems of the planet via remote-sensing technologies, supplemented by Earth-surveying techniques, which encompasses the collection, analysis, and presentation of data. Earth observation is used to monitor and assess the status of and changes in natural and built environments.
In recent years, Earth observation has become increasingly technologically sophisticated. It has also become more important due to the dramatic impact that modern human civilisation is having on the world, along with the opportunities such observation provides to improve social and economic well-being.
Xiaoxiang Zhu is a professor of signal processing in Earth observation at the German Aerospace Center and Technical University of Munich, Germany. Zhu and her team develop explorative algorithms to improve information retrieval from remote sensing data, in particular those from the current and next generation of Earth observation missions. One of the outstanding achievement of her research work has been to use satellite imagery and supercomputing to predict the risks of structural degradation and damage to city buildings. Furthermore, she and her team are developing the first ever global urban models in 3D and 4D.
“We have satellites orbiting the Earth taking measurements of the Earth’s surface. For example, through the ESA’s Copernicus programme, 15 terabytes of data are acquired every day by its Sentinel satellite fleet and made open and freely accessible. As Earth observation computer scientists, what my team and I do is to turn this kind of measured data into geo-information that is useful to different user groups such as scientists, politicians, and people from industry.”
Figure caption: First impression of the global 3D urban LOD1 models: Munich city reconstructed using only five TanDEM-X acquisitions. The building footprints have been reconstructed from optical data, the building heights (color) from the TomoSAR point cloud. (Image: Xiaoxiang Zhu, DLR/TUM)
Global processing of petabytes of geospatial data requires big computing power. Working at a resolution of approximately ten meters means high-performance computing is absolutely essential. Since 2012, Zhu has been granted over 46 million core hours on the SuperMUC computer at the Leibniz Supercomputing Centre.
One of the main drivers of Zhu’s work has been the ever increasing global urban population. More people now live in cities than ever before, and the difference will only become more extreme in the future. In many developing countries, especially those with large informal settlements and slums, urban growth is happening so quickly that there is often inadequate data to help provide basic infrastructure like healthcare and clean water for everyone. “This kind of urban growth is not controlled, and unfortunately this means that the quality of life for many people in these areas is not good,” explains Zhu. “Our work will hopefully improve the understanding of urban areas on a global scale by providing better information about them.”
At present, the greatest repository of data on urban development across the globe is known as the Global Urban Footprint, which was created using over 180 000 images together with additional data such as digital terrain models. The result is a map which, besides possessing a strange beauty that has been likened to Chinese ink drawings, documents practically all of humanity’s physical presence on the surface of the earth. However, the information it provides is limited to a simple binary representation of where there are settlements and where there are not.
Zhu is now trying to provide a much more detailed picture of the global urban population and urban development, using global radar satellite data to ascertain the 3D height of buildings and optical data to work out their shapes. Furthermore, by combining all of these data sources, the semantic details of an area can be filled in, such as the classification of a building as residential or commercial.
Figure caption: The ERC Project So2Sat in a nutshell. To provide a better understanding of global urbanisation, Xiaoxiang Zhu and her team combine Big Earth Data from satellites and social media to derive a global and consistent 3D/4D spatial dataset on urban morphology.
“A detailed study of census data compared to actual numbers in Mumbai slums found that there was a discrepancy of up to 400%”
Of course, working out all of this using satellites alone would be impossible, and so Zhu and her colleagues have also been using social media data to help them with their task. “We have social media pictures, which are usually taken from a street view perspective, where you can see building facades, which tells you a lot more about the function of a building than the top,” says Zhu. “As well as this, we can also analyse the tweets coming from a specific building to work out its function and also how many people are living or working there.
“We are also interested in combining our 3D building layers with estimated building semantics to deliver the first ever transparent estimation of high resolution population density on a global scale. Census data is often inaccurate when it comes to slum areas. A team led by my collaborator Dr Hannes Taubenböck carried out a detailed study of census data compared to actual numbers in Mumbai slums and found that there was a discrepancy of up to 400 per cent. In particular, the numbers of small children are often underestimated.”
Using HPC, Zhu and her team can now also measure temporal changes down to the centimetre or even millimetre scale to observe whether buildings have been disturbed by uplift or subsidence. The huge global datasets being created are a giant leap forwards for urban geography research.
Resources awarded by PRACE: Xiaoxiang Zhu was awarded 46 000 000 core hours on SuperMUC hosted by GCS at LRZ, Germany