Integration of spatial continuity into the multidimensional structure of a data warehouse – raster SOLAP

JEAN-PAUL KASPRZYK | 2015

Summary

Technological advances in recent decades have created a massive acquisition of digital data whose volume grows exponentially. To efficiently extract the information they contain, powerful tools have been developed to collect, store and analyze these data. These tools are gathered in a discipline called “business intelligence”. Among them, data warehouses are responsible for archiving data by structuring them in a multidimensional way (time, space or others). They are called data hypercubes or data cubes when they are limited to three dimensions. Hypercubes can supply OLAP (On Line Analytical Processing) systems that aim at quickly synthesizing information in interactive tables and charts for decision-makers from various fields: marketing, environment, criminology, etc. Thus, users can navigate into hypercubes using OLAP operations such as slicing on dimension members (e.g. data aggregation for the month of January in the time dimension), or drilling into hierarchies (e.g. switching from the “year” level to the “month” level in the time dimension). When OLAP is coupled with spatial analysis techniques supplied by geographic information systems (GIS), a map interface then improves the exploration of data: OLAP operations can be applied to dimensions defined in the geographical space (spatial drilling or spatial slicing). This kind of tool is called SOLAP (Spatial OLAP). SOLAP tools currently available on the market all suffer from the same deficiency: they are unable to represent spatial dimensions (X, Y) in a continuous way. This representation is nevertheless essential for the management of spatially continuous phenomena (temperature, pollution, etc.) but also for visualizing spatially discrete events (product sales, crimes, etc.) while minimizing the Modifiable Areal Unit Problem (MAUP). This kind of visualization is used especially by the police to predict the location of future crimes through hotspot maps which are generated by the Kernel Density Estimation (KDE) method. In the field of GIS, raster data (as opposed to vector data) enable effective representation of spatial continuity through digital georeferenced grids. Whereas current SOLAP tools only consider vector data, our research uses the raster model to integrate spatial continuity into the multidimensional structure of a data warehouse feeding a SOLAP ("raster SOLAP"). Despite its underutilization in the SOLAP literature, the raster model has many similarities with a particular kind of data cube: the MOLAP cube (Multidimensional OLAP). Like a satellite image (raster) representing the two planimetric spatial dimensions and one "spectral band" dimension, a MOLAP cube is a three-dimensional array whose cells’ coordinates (similar to raster pixels) enable an efficient indexation of dimensions’ members (describing the analyzed facts). In a first original model that we call "raster cube" we define the bases for a three-dimensional raster SOLAP, starting from the definition of a MOLAP cube. Unlike vector SOLAP - where spatiality is attached to a semantic dimension through pointers to geometries - our model directly integrates spatial dimensions (X, Y) in the multidimensional structure of the data warehouse. With this original feature, any geographical entity (country, building, road, etc.) can be imported on the fly as a member in the analysis of the user, which is hardly possible with conventional vector SOLAP tools. An extension of this SOLAP model, called "raster hypercube", is then developed by entrusting the management of extra non-spatial dimensions to a relational database management system (Relational OLAP or ROLAP). The raster hypercube is then populated by KDE raster fields representing crime densities, which are defined in a continuous space (raster dimensions) through time and crime types (ROLAP dimensions). Our model is able to combine the production of hotspot maps at different scales of analysis with SOLAP navigation operations: slicing on spatial or non-spatial members, and drilling into the hierarchy of spatial or non-spatial dimensions. Our raster hypercube model is validated by an operating prototype which is based on open source tools only. Several datasets are integrated through KDE fields, including crime data from London and Seattle. At the end of our work, the results of a comparative study between raster SOLAP and vector SOLAP demonstrate that hybrid vector/raster SOLAP architectures present the same interest for spatial data as hybrid ROLAP/MOLAP architectures do for purely semantic data (management of detailed hypercubes).

Comments

The SOLAP raster model developed in this thesis was implemented and integrated into the OSCAR tool (“Outil SIG et Cartographique pour l’Analyse de Risque”). OSCAR is currently used by the emergency services of Brussels (SIAMU) and the Province of Liège in order to optimize the strategic distribution of their resources across the territory. Such an optimized distribution guarantees quick interventions by emergency services while avoiding their saturation.

Integration of spatial continuity into the multidimensional structure of a data warehouse – raster SOLAP

Summary

Comments

Links