Generating 50cm elevation contours from space PhotoSat s s new stereo satellite elevation processing system Gerry Mitchell PhotoSat November 2009
PhotoSat stereo satellite processing history PhotoSat has completed over 100 high definition stereo satellite elevation mapping projects since high resolution stereo satellite photos first became commercially available in 2004. From 2004 to t 2007 we used photogrammetric processes to map the elevations. Since 2007 we have been developing and improving our own stereo satellite geophysical elevation mapping system.
Elevation processing results GeoEye-1 satellite photo Elevation Image Indonesia Open Pit. Stereo GeoEye-1 satellite photo and elevation image processed in September 2009.
Elevation processing results GeoEye-1 satellite photo Elevation Image A GeoEye-1 stereo photo and elevation image of the Indonesia open pit showing a slide area on the NE pit wall
The elevation mapping accuracy Stereo satellite elevation RMSE of 19cm determined by 1,115 surveyed check points. 90% of the survey check point elevations are within 28cm of the mapped stereo satellite elevations. Stereo WorldView-1, 260 km 2 elevation mapping in Chihuahua, Mexico, October 2009.
The elevation mapping accuracy Stereo WorldView-1 1 elevation mapping in Chihuahua, Mexico. The 260 km 2 mapping area was referenced to a single ground survey point. The accuracy was checked using 1,115 survey points.
The elevation mapping accuracy The reference point for the entire 260 km 2 stereo satellite elevation mapping area in Chihuahua Mexico.
The elevation mapping resolution 750m GeoEye-1 satellite photo Elevation Image The one meter high terraces on this hillside in Peru show clearly on the stereo satellite elevation image on the right.
Elevation mapping resolution GeoEye-1 satellite photo Elevation Image These stone fences in Peru show clearly on the stereo satellite elevation mapping. Stone fences are typically less than 1m wide and 1.5m high.
Elevation mapping resolution GeoEye-1 satellite photo draped on the stereo satellite DEM shows a good match between the elevation data and the satellite photo at this bridge.
Elevation mapping resolution GeoEye-1 satellite photo draped on the stereo satellite DEM shows the good match between the satellite photo and the elevation mapping at this clearly defined road cut.
Three key technical components enabling 50cm elevation contouring from space High resolution stereo satellite photos Adaptation of seismic processing systems Graphics Processing Units ( GPUs )
High resolution stereo satellites IKONOS 1m colour 2004 WorldView-1 1 50cm greyscale 2008 GeoEye-1 1 50cm colour 2009 WorldView-2 2 50cm colour 2010
High resolution stereo satellites GeoEye-1 1 50cm colour photo
High resolution stereo satellites Same pass stereo satellite photos. Only about 30 seconds of time elapse between the photos. Each photo is 10 km to 15 km wide. The near uniform look direction to the satellite for every pixel of each photo facilitates the automatic elevation processing.
Adapting seismic processing algorithms and processes Moore et al., 2007 Offshore 3D seismic data cube. In the development our stereo satellite processing system we have made extensive use of processing algorithms and processes developed for 3D seismic processing. 3D seismic cubes such as the one shown above can involve the co registration of 100 or more individual seismic images. A large array of processing techniques for correlating images and attenuating noise and processing artefacts have been developed by the seismic industry over the past 50 years. To our knowledge many of these processes have never before been applied to the processing of stereo satellite data.
Adapting seismic interpretation systems 3D seismic workstation For mapping the ground surface elevations we are employing techniques developed for 3D seismic interpretation workstations and, in some cases, the workstations themselves. Seismic interpreters map the elevations of subsurface rock formations. The slide above shows an underground rock formation surface interpreted in the OpendTect seismic workstation. We are using this technology to map the ground surface elevations.
Graphic Processing Units ( GPUs ) 240 parallel processors 4 gigabytes RAM One teraflop C program compiler The automatic processing of surface elevations at one meter intervals from stereo satellite photos, using seismic processing algorithms, requires the computation of hundreds of millions of forward and inverse 2D Fourier transforms for a 100 km 2 mapping project. GPUs perform numerical processing up to 100 times faster than CPUs, enabling us to automatically produce one meter Digital Surface Models from stereo satellite photos in reasonable times.
Digital photogrammetry With conventional photogrammetric methods elevations are mapped interactively on stereo workstations or mapped automatically with significant interactive editing. Elevation accuracies are typically two to three times the photo pixel size.
Geophysical processing Ground Surface Stereo photo correlation profiles The PhotoSat geophysical processing system automatically calculates vertical profiles of the correlations between the stereo satellite photos. The elevation mapping accuracy is typically ½ of the photo pixel size.
Geophysical processing Cross section of stereo satellite correlation profiles across a hill in Eritrea.
Geophysical processing Cross section of stereo satellite correlation profiles across a hill in Eritrea.
Geophysical mapping Cross section of stereo satellite correlation profiles going from a bare ground upland through a valley with a stand of trees. Dark colours indicate strong correlations. This cross section is displayed in the OpendTect stereo seismic workstation.
Geophysical mapping Two cross sections and a horizontal section of stereo satellite correlation profiles. Dark colours indicate strong correlations.
Geophysical mapping Cross section of stereo satellite correlation profiles in greyscale.
Geophysical mapping Cross section of correlation profiles with the Digital Surface Model (DSM) picked automatically to within about 5cm of elevation.
Geophysical mapping. 3D display of the Digital Surface Model coloured with the correlation amplitudes. The correlations are weaker in the trees than for the bare ground.
Geophysical mapping Digital Surface Model Interpreted bare ground elevations Cross section of correlation profiles showing interactive interpretation of the bare ground elevations. The mapping of bare ground elevations beneath tree canopy and buildings necessarily involves interpretation and interpolation. We are using techniques developed for mapping geological unconformities on 3D seismic data to map the bare ground elevations on cross sections of vertical stereo satellite correlation profiles.
Elevation mapping in steep terrain IKONOS satellite photo IKONOS elevation Image Himalayas IKONOS and Stereo IKONOS elevation image processed in August 2009. The access roads show clearly in the elevation image. The south facing slope is 45 degrees.
Elevation mapping in steep terrain IKONOS satellite photo IKONOS elevation Image Himalayas IKONOS and Stereo IKONOS DEM with a 45 degree slope.
Steep terrain Himalayas IKONOS draped over the Stereo IKONOS elevations.
Steep terrain Himalayas IKONOS draped over the Stereo IKONOS elevations.
Steep terrain Himalayas IKONOS draped over the Stereo IKONOS elevations.
Mine site mapping WorldView-1 greyscale photo of a 10 km x 10 km area over a Athabasca tar sands mine in NE Alberta, Canada
Mine site mapping Stereo WorldView-1 elevation image of a 10 km x 10 km area over a Athabasca tar sands mine in NE Alberta, Canada
Mine site mapping Stereo WorldView-1 elevation image of a 5 km x 2.5 km area over an Athabasca tar sands mine in NE Alberta, Canada
Mine site mapping Stereo WorldView-1 elevation image of a 500m x 250m area over an Athabasca tar sands mine in NE Alberta, Canada
Mine site mapping WorldView-1 satellite photo WorldView-1 elevation Image Stereo WorldView-1 photo and elevation image of a 300m x 300m area over an Athabasca tar sands mine in NE Alberta, Canada