The European Sentinel-1, Sentinel-2 and Sentinel-3 missions will generate a daily data volume in excess of 20 TB by the end of 2017. In view of this large quantity, new processing procedures are called for - like those used by the TimeScan processor successfully implemented by the U-TEP consortium. Basically, the procedure distils a single information product from a multitude of satellite images acquired over a lengthy time period. Using the already available “big data” of the U.S. Landsat mission(s) as testbed, over 450,000 images collected by the Landsat-8 satellite from 2013 to 2015 were processed to produce the now published “TimeScan-Landsat 2015” product. Some 500 terabytes of separate images, condensed to a 20th of their original size, can be analysed in the form of a single, global, cloud-free data set. This is not simply a satellite image mosaic. Instead, the information content of the multispectral images was condensed as statistical ranges (e.g. minimal, maximal and mean) covering the entire recording period from 2013 to 2015 and taking into account the various spectral channels. Analysis of the land surface is possible based on these indices for such aspects as the state of human settlements (built-up areas), vegetation, or water cover.
The procedure is designed to help end users exploit information from masses of data that until now were too unwieldy for them to handle. The individual satellite images are no longer transferred to users, as formerly, but processed on large computer clusters, ideally where the data were first acquired. This eliminates distribution of immense quantities of data to numerous individual users, who no longer need to have their own computer infrastructure to analyse them. On the contrary, now only the analysed end product, whose size is only a fraction of the original amount of input data, is sent to the user as an intermediate layer for further value-adding.
To test the procedure, the TimeScan processor was transferred to the Super Computing Center IT4Innovations in Ostrava-Poruba, Czech Republic and used to calculate the so-far unique global TimeScan Landsat 2015 product. The new data set has a spatial resolution of 30 m per pixel. Including all the intermediate products, that means that over 1.5 petabytes of data had to be processed. The Landsat mission has recorded over four million individual scenes over the last four decades, making it an ideal resource for testing the processing of mass data, like what the Sentinel programme will be supplying in the near future.
The TimeScan Landsat processor contains a module that uses an automated procedure to download the required Landsat scenes from different databases. A cloud mask is provided and, if desired, atmospheric correction can be undertaken for the scenes. The next step is to calculate selected spectral indices. By taking into account all the validated initial values, such as the times without cloud cover, the spectral behaviour is determined for the selected time period and used to determine the actual land cover type. Compared with the mass of original input data, a much smaller volume of data has to be processed and analysed if the TimeScan layer is used as input.
The new TimeScan Landsat 2015 data set is being used in the context of the U-TEP project to generate various settlement-related geoinformation products such as up-to-date settlement masks or maps showing the urban green and imperviousness. This data is supposed to support scientists and decision makers in public planning and environmental offices and in development banks. It will also contribute to improved understanding of the worldwide phenomenon of urbanisation. The TimeScan data sets now available and those that will become available based on Sentinel data have great potential for a wide range of applications beyond the urban context. These include research questions related to land cover and land use mapping, agriculture, forestry, monitoring polar and coastal regions, risk management, disaster prevention and resource management.
Currently, the TimeScan Landsat 2015 is just accessible for visualisation, but as soon as the validation of the data set is finished the layer will [be provided to the interested users on request. Moreover, the TimeScan processor is currently being used at the DLR, IT4Innovation and Brockmann Consult processing centres to create comparable products based on Sentinel-1 and Sentinel-2.
The U-TEP team will continue to add new valuable products and services required for urban management and monitoring. To get more information, keep on checking the U-TEP product and service portfolio envisaged for the first implementation phase of our project as well as the U-TEP App Store and stay tuned!