The usual way to store and provide data for virtual globe applications is to use special encodings that are tailored for digital elevation models (DEM) on the one hand and objects such as buildings, labels, and other objects on the other hand.
Most applications, e.g. NASA WorldWind, use image or raster formats for encoding elevation data, e.g. GeoTIFF or ASCII raster files. In images, the elevation values are stored in the three available bands. On the client side these rasters can be very quickly processed and transformed into triangle meshes for rendering. Although this approach can be very efficient in terms of bandwidth requirements and parsing speed, it involves some drawbacks limiting its use for highly detailed surface models. 1) Overhangs, caves and vertical structures cannot be modeled correctly. 2) The triangulation schema is fixed and cannot adapt to morphological structures. 3) Flat areas are filled up with unnecessary vertices. Storing DEM in a free geometry structure using e.g. indexed triangle meshes allows using more complex triangulation algorithms (Figure 71). In the project OSM-3D for instance, an algorithm taking into account local surface curvature was used replacing the standard Delaunay algorithm. Moreover, land use areas and the road network were integrated into the triangle mesh as described by Schilling at al. (2007). Since such computations are quite time consuming, they should not be done on request by the live system; all terrain data is preferably processed offline before publishing the data, or at least cached, very much like a map tile cache. In the system as described by Lanig et al. (2011) complex surface models were generated using terrestrial lasers canning (Figure 72), which also contained cavities.
Non-tangible or non-physical geographic features include labels for populated places, countries, mountain peaks, lakes, streams, and so on, as well as icons or 3D symbols representing points of interest or features with specific attributes. Such data is usually stored as set of points containing just a tuple of longitude/latitude values and optionally an elevation value. If the latter is missing, some formats such as KML allow specifying that objects are “sticking on the ground”, i.e. elevation is retrieved from the DEM. The great benefit of transmitting raw point data is that custom symbols tailored for a specific application can be implemented which make use of screen overlay techniques avoiding overlapping labels and other complications. Similarly, linear features representing borders, bus lines, and the like can be transmitted and processed in a similar fashion as raw data. In the scene graph based approach, symbols are applied on the server and the complete symbol definition is encoded in 3D formats and streamed to the client. This assumes that suitable node types are available for describing text components, billboards, switches, user interactions and the like, which is the case with X3D.
Figure 71: Advanced DEM processing. A: fixed triangulation scheme. B: free triangulation using mesh fairing algorithm. C: reduced mesh by 50 percent. D: forest areas integrated into mesh. CIAT-CSI SRTM.
Models of buildings and other manmade structures are usually created by 3D artists or CAD professionals using modeling software which operates within a local Cartesian coordinate system. By defining a translation vector shifting the local model to the correct reference point on the earth surface, models can be quickly published. Again, KML/KMZ is a well known example containing a reference point defined in longitude/latitude, rotation angles, and a reference to a local model encoded in COLLADA. This already complies with the scene graph concept, except that in the presented framework, X3D is used to encode buildings. The Geospatial Component is part of the X3D normative specification and provides a GeoLocation node, being equivalent to the KML reference point. Arnaud (2007) explains the differences between COLLADA and X3D.
Figure 72: Surface models with overhanging parts and caves. Site in Sichuan, China which has been captured using terrestrial laser scanning. The lying Buddha has been carved into the stone and is part of the terrain model. Project Sutras 3D (Auer et al. 2011).
Figure 73 shows the data flows of both, raw data approach and scene graph based approach. Although additional feature classes may be possible, only the 3 main feature classes are discussed
here: DEM, point data and buildings. As can be seen from the server/client configurations, most logic has been moved to the server in the scene graph based approach. Especially mapping images on the DEM and applying symbols to point and line data must be performed by the server. This implies that the client does not necessarily include any symbol repositories and switching between different symbols or image maps requires additional server requests. However, different symbols may be promoted as styles by the server to choose from. The great advantage of this approach is that the server interface can be unified and that additional feature types, such as atmospheric layers, geological strata, sensor data, or any other feature type which was not predefined can be easily added without changing the interface. All layer data is homogenously transmitted as scene graph models and can be processed by the same chain in the client.
The server interface is compliant to the Web 3D Service (W3DS) 0.4.1 specification, a candidate OGC standard for interactive 3D portrayal (Schilling & Kolbe 2011). A first attempt to use the W3DS on a global scale was done by Misund et al. (2005). The W3DS requires that at least the X3D format must be supported, however additional encodings and formats may be used that reduce bandwidth requirements and parsing time. For example, GZIP compression on the HTML layer and binary encoding of X3D can greatly increase transmission and parsing performance. The task of the client is reduced to process scene graph data and, in the case of a Virtual Globe application, to correctly implement LOD and loading schemas.
Figure 73: Comparison of raw data approach and scene graph based approach for setting up 3D geospatial visualization systems