Crisis DSM Generation to Support IDP Camp Management
Gstaiger, Veronika; D' Angelo, Pablo; Schneiderhan, Tobias; Krauss, Thomas
The Center for Satellite Based Crisis Information (ZKI) at the German Aerospace Center (DLR) provides satellite-based maps in the frame of natural and environmental disasters, humanitarian aid, and civil security worldwide. During the last years, information extraction methods have been developed in several research projects in order to carry out analyses in a rapid and precise manner as well as to develop operational and application-oriented products for a wide spectrum of use. After being a key protagonist in setting up the European Emergency Response Service (GIO-EMS), now the German Federal Ministry of the Interior (BMI) established a framework with DLR-ZKI which enables national authorities and other authorized users to order products of the ZKI, beginning from January 2013. Benefiters of this framework agreement are national level political decision makers, situation centers, as well as rescue and aid organizations.
In this paper the authors present an example how scientific results, in this case an ad hoc generated crisis DSM (Digital Surface Model) based on three regular WorldView-1 and -2 scenes, can be implemented in order to meet user requirements during an emergency request, based on a practical example of the first mapping request under the national mandate. Experiences from value adding for the International Charter and projects like SAFER have shown, that ad hoc available digital elevation models (DEMs) with very high resolution of a small crisis affected area are of high interest for the emergency management. Application fields are especially volcano eruptions or flood situations, where information about elevation can be used in water or lava flow models to identify areas at risk.
The first request within the new framework came from the German Federal Agency for Technical Relief (THW) in January 2013 and was an opportunity to test current scientific developments in the operational service finally leading to a full integration. The background of the request was the continuing conflict in Syria and the build-up of camps for internally displaced persons (IDPs) in Jordan. The demand covered a simple derivation of the camp structure, but also a detailed elevation information since THW is managing the construction of a drainage system in the camp. Apart from the logistical challenge of organizing hundreds of arrivals and a continuing growth of the camps, camp Al Zaatari had to additionally face damages and flooded areas after a heavy storm with rainfalls, which shows the importance of elevation details. Information was needed to evaluate the development, damage and the selection of the camp's location with regard to flood risk and also to optimally coordinate the work of the emergency response teams. The need for elevation information was covered by providing products derived from an ASTER DEM. Unfortunately the ASTER DEM was acquired before 2011 and construction activities for the camp had seriously changed the surface. Apart from the dated information the resolution of 30 m was also not sufficient for such a small-scale application (see figure 1). For these reasons a method for generating digital surface models from optical satellite data was tested with three archive scenes of WoldView-1 and -2. The generation of the DSM is based on DLR’s processing system CATENA and can be used to fully automatically produce DSMs from high resolution satellite images. The process first finds high quality tie points between all images using scale-invariant feature transform and local least squares image matching. Optionally, other information such as Ground Control Points, reference ortho images and reference DEMs can be used to ensure an appropriate absolute orientation, if available. A bundle block adjustment based on Rational Polynomial Coefficients (RPC) ensures a good relative orientation as basis for accurate stereo matching. The Semi-Global Matching algorithm is then used to match all possible stereo pairs. Semi-Global Matching performs pixel wise matching with a local smoothness prior, but also allows local height jumps, and can thus reconstruct fine surface details. The Census cost function was used to compare each pixel with all possible matches in the second image. Census is a very robust with respect to radiometric differences, as it is invariant to monotonous gray value changes. As no along track stereo pair was available for the camp area, three scenes of WorldView-1 and -2 (2013-01-05, 2012-12-19 and 2012-12-12) were used for matching. The use of three images provides additional redundancy and allows an reliable detection and removal of matching outliers. All possible image pairs are matched independently, and only height values where all pairs agreed have been accepted. The remaining holes were filled using hierarchical B-spline interpolation. The process worked very well for unchanged, arid areas around the camp (see figure 2). Areas that changed between acquisitions of the images, such as travelled dirt tracks and newly constructed tents could not be reconstructed. Along-track stereo imagery would be required for a complete reconstruction. As no ground control was available, the absolute accuracy of the DSM is estimated to 6 m. Experience with other datasets has shown that accuracies of 1-2m LE90 can be archived, depending on the type of terrain. The whole processing chain is highly automated and uses grid computing for faster processing of large images. For the small camp area, the block adjustment and DSM generation required less than 3 hours on a quad core CPU.
The used method provides a fast, robust and fully automated way to derive commonly needed surface information, if there is no sufficient reference data available. It was possible to derive valuable information for a precise and updated evaluation of the topology and its influence on the hydrological system, fully meeting the user requirements. But it was also a chance to connect the requirements of an operational rapid mapping service during an emergency to scientific research topics and to identify new fields of application. This paper will provide insights to the applied technique and will show a comparison between the results and the available ASTER DEM.
Figure 1: Camp Al Zaatari on ASTER DEM (left) and WorldView DSM (right)
Figure 2: Camp Al Zaatari and mobile homes on WoldView DSM (A) and WoldView imagery of January 3, 2013 (B)