Activities of the GHSL linked with the Copernicus program
S1 GHSL: First Experiment on Global Processing of Sentinel-1 Copernicus Earth Observation Data for Monitoring Human Presence on Earth.
In December 2016 the European Commission’s Joint Research Center (JRC) successfully completed the first experiment of Sentinel 1 global data processing in the frame of the JRC “Global Human Settlement Layer” scientific activities. The S1 GHSL experiment was supported by JRC Earth Observation Data and Processing Platform (JEODPP) developed in the context of the JRC “Big Data Pilot Project”. The JEODPP platform was tasked on use of the Sentinel 1 (S1) Earth Observation data for detailed recognition of all the artificial built-up structures that are visible in the global landmass surface with a spatial resolution of 10x10 meters. At the date, this is the largest known experiment of automatic extraction of such information from S1 data.
Sentinel 1 Earth Observation data are generated by Advanced Synthetic Aperture Radar (ASAR) sensors on board of the twin pair of Sentinel 1 satellites and are provided by the European Copernicus program in a public, open and free data access policy.
Detailed and complete assessment of the global built-up areas are necessary for assessing the human presence on the Earth, collecting evidences for monitoring the implementation of international agreements on sustainable development, disaster risk reduction and climate change.
During the experiment, cutting-edge spatial data mining and analytics technologies have been implemented. More than 2 trillion of Earth Observation data records reporting about the backscattering of electromagnetic energy from each 10x10 square meters portion of the global landmass surface have been analyzed and automatically selected if they were showing an electromagnetic energy pattern associated to roofed built-up structures. The GHSL information production technology is inspired from DNA microarrays data analysis methods used in biomedical informatics for the clustering of gene expressions. By analogy with the genetic association, the new classifier developed at the JRC searched for systematic relationships between more than 100 billion individual sequences of S1 satellite data instances and the “roofed built-up” class abstraction encoded in global reference sets.
The experiment contributes to the scientific work plan of GHSL project of the JRC. Objective of the experiment was to test new technologies for cost-effective, automatic extraction of global thematic information from EO data produced by the Copernicus program. The experiment was enabled by the JRC Earth Observation Data and Processing Platform (JEODPP) that is being developed in the framework of the JRC Big Data Pilot Project. The platform is set-up to answer the emerging needs of the JRC Knowledge Production units following the new challenges posed by Earth Observation entering the big data era. The data streams originating from the Copernicus programme and in particular the Sentinel satellites operated by the European Space Agency will deliver up to 10 TB of image data per day when in full operational capacity. The JEODPP is conceived to timely process large amounts of data using commodity hardware. Novel software technologies for both storage and processing enable the analysis of global coverages of high resolution satellite images in less than one day. It follows that the optimisation of the big data machine learning methods and their parameters can be performed at full scale. The JEODPP benefits from cutting-edge storage technology thanks to a collaboration between the European Organization for Nuclear Research (CERN) and JRC that led to the deployment of a CERN EOS disk-storage instance on the JEODPP.
Several automatic image information extraction models and parameter sets have been tested in the experiment leveraging on the JEODPP capacity to support scientific investigation on large and complex Earth Observation data scenarios. Some of these results can be accessed through the JRC Community Image Data portal.
During 2017, the JRC plans to perform the first assessment of the GHSL production using Sentinel2 multispectral EO data also supported by the JEODPP platform. The first results of the experiments on integrated assessment of S1, S2, and Landsat data in support to the GHSL production are planned during 2018.
The GHSL produces new global spatial information, evidence-based analytics and knowledge describing the human presence on the planet. GHSL operates in an open and free data and methods access policy (open input, open method, open output). The GHSL is supported by the Joint Research Centre (JRC) and the DG for Regional Development (DG REGIO) of the European Commission, together with the international partnership GEO Human Planet Initiative.
According to the preliminary results of the S1 GHSL experiment, S1 EO data is expected to improve the quality of the Global Human Settlement Layer baseline firstly established in late 2014 by processing of 40-years historical records of Landsat EO data. In particular, a reduction of both omission and commission errors in automatic detection of built-up areas is expected by introduction of S1 data in the GHSL information production workflow. Moreover, a better discrimination of water vs land surfaces is expected.
On the 29/10/2020 the Global Human Settlement Layer (GHSL) has publically released a new, free and open global built-up grid from Sentinel-2 Copernicus data. It is the most exhaustive and detailed map of built-up areas at a spatial resolution of 10 meters. It shows humanity’s imprint in more detail and with greater accuracy than ever before.
The key to the success of detecting human settlements is the use of Artificial Intelligence for the processing of global satellite data and the large computing capacities offered by JRC I3 Big Data Analytics project.
By leveraging AI analytical capabilities, Copernicus Earth Observation and High Performance Computing, this dataset sets a path to a new wave of promising research to tackle unprecedented large-scale challenges in urban related applications. It delivers cutting-edge information on human settlements from the megacity to the rural hamlet taking into account also temporary and informal settlements.
The method builds on a deep-learning-based framework for a fully automated extraction of built-up areas at a spatial resolution of 10 m from a global composite of Sentinel-2 imagery recently published in Neural Computing and Applications.