• No results found

Exploratory spatial data analysis (ESDA), as distinct from GIS, is defined in “Interac- tive techniques and exploratory spatial data analysis” [Ans99]. Here, Anselin is de- scribing interactivity going beyond a regular GIS system, where users can explore and question data, relying on computer graphics for interactive visualisations. He makes the distinction between static techniques, like spatial autocorrelation, and dynamic, in- teractive, techniques, where the user can interact with the data directly in a graphical environment. He goes on to say,

“The main contribution of ESDA with respect to GIS lies therefore in vi- sualising local patterns of spatial association, indicating local non-stationarity and discovering islands of spatial heterogeneity.”

This relates ESDA to GIS, but Chrisman, in his book, “Charting the Unknown. How Computer Mapping at Harvard Became GIS” [Chr04], chronicles the development of GIS from computer graphics. Containing details about the “SYMAP” program from 1963, in the “how SYMAP worked” section, Chrisman says, “SYMAP used a vector model, a collection of objects-points, lines, and areas-in planar coordinate space with thematic values attached to them”. The same is true today, although with the possible exception of bezier patches, point clouds and voxels (all defined in [Fol+93]). Maguire puts forward a definition for GIS in [Mag91], proposing three distinct views which define a GIS system: “map, database and spatial analysis”. In his definition, “each

2.7. Linking GIS and Computer Graphics 53 data set is represented as a map”, which we extend in this thesis to become a data set representing a collection of maps. His definition of “database view” refers to stor- age, which could be cloud-based, while the “spatial analysis view” is the visualisation layer. He goes on to describe variants of GIS software called, “file processing, hybrid and extended designs”. The file processing design pattern stores data sets as separate files, which are linked together later in the processing pipeline. The hybrid approach is similar, but splits attribute data (aspatial) and spatial data. The extended model is sim- ilar to the hybrid, but extends the relational database layer with spatial extensions. Of these three approaches, the idea of holding data in the cloud and joining with geospa- tial data in a different cloud location, while running spatial and aspatial data through different systems would seem to lie somewhere between the file processing and hybrid approaches. This is the approach taken by the MapTube system.

As far as computer graphics goes, this needs to be taken in a web based context, where the computer rendering the data is distinct from the computer that processed the data. In this model, the computer browser does the rendering, which makes for a distributed GIS. Shaowen Wang advocates this type of GIS system, under the defini- tion, “cyber-infrastructure” [Wan10], and in the “PYSAL CyberGIS Toolkit” [Rey+13]. Wang describes this as “software as a service”, offering GIS functionality as a web ser- vice.

Two approaches to rendering maps in a web browser exist: raster or vector. Much depends on the capabilities of the computer browser, but raster involves the server ren- dering the map tiles, while the vector technique sends scaled, clipped and tiled versions of the data to the browser which then has to render the data using WebGL [Khr14]. Ei- ther way, the rendering involves computer graphics algorithms to draw the maps from points, lines or polygons, following the OGC’s SFS standard [Ope10b].

Advances in modern graphics hardware now mean that computers have a graphics processing unit (GPU) in addition to a CPU, which provides specialised hardware de- signed for rendering. To this end, the vertex shader and pixel shader hardware [CR11, Ch3] are used to accelerate the computer graphics. In “A Lightweight CUDA-Based Parallel Map Reprojection Method for Raster Dataset of Continental to Global Extent” [LFB17], Li shows an example of reprojection using the CUDA library designed by NVIDIA to run on their GPU hardware. The authors also point out the CyberGIS link in their introduction:

“To solve these types of computationally intensive problems, previous approaches usually relied on the use of high-end workstations, thus making efficient map reprojections for raster datasets less accessible to scientists who do not possess expensive infrastructural resources such as high-end computer clusters or supercomputers.”

On the client side, the MapBox library (https:// www.mapbox.com/ mapbox-gl-js/ api/) uses WebGL to render maps, while 3D applications in the browser have also been created, for example “OpenWebGlobe” of West et al. [Wes+14]. Rendering using pixel and vertex shaders can have some interesting applications, for example photo- realistic renderings of the Earth, ray tracing and procedural modelling. In “Real-Time Rendering Techniques with Hardware Tessellation” [Nie+16], the authors summarise work on rendering techniques for hardware tessellation, including parametric surfaces and real-time tessellation of surfaces. In the paper they analyse height fields and terrain rendering, which are relevant for 3D work, including the minimisation of splits and cracks caused by the tessellation, a subject which Cozzi and Ring also cover at length in their book on the development of an OpenGlobe system [CR11].