GeoComputation in the Big Data era

IWGIS-2014, June 21-22, 2014, Beijing
9th International Workshop of Geographical Information Science
GeoComputation in the Big Data era
Qingfeng Guan
National Engineering Research Center of Geographic Information System & Faculty of Information Engineering,
China University of Geosciences
[email protected] Wuhan, Hubei 430074, China
The advancements in geospatial data acquiring technologies (e.g., high-resolution and hyperspectral
remote sensing, LiDAR, social media, and location-aware mobile devices) have led to the arrival of the Big
Data era in geospatial science. Spatial Big Data provides a variety of opportunities for geospatial scientists
and professionals to retrieve and generate more information and knowledge (e.g., spatial patterns, spatial
relationships, and driving mechanism of spatio-temporal dynamics) and solve complex geospatial problems
through spatial analysis, data mining, and simulation. On the other hand, Spatial Big Data imposes a series
of challenges to GIScience. Spatial Big Data is characterized by the same traits of Big Data in the general
sense, i.e., big volume, high updating velocity, large variety, and varying veracity (a.k.a. 4V’s), thus
requires innovative data management and analytical methods and technologies. GeoComputation, focusing
on geospatial computing approaches, provides promising means to utilizing Spatial Big Data to solve
complex geospatial problems. Two methodological components of GeoComputation are especially useful
in handling Spatial Big Data: Computational Intelligence (CI) and High-performance Computing (HPC).
With the capabilities of self-learning, self-adapting, and self-organizing, CI methods, such as Artificial
Neural Networks (ANN), Genetic Algorithm (GA), Simulated Annealing (SA), and Support Vector
Machine (SVM), focus on adaptive mechanisms to enable or facilitate intelligent behaviors in complex and
changing environments. CI methods have been proven to be able to deal with complex, redundant or
incomplete, and noisy data, and to solve complex geospatial problems (including spatio-temporal modeling
and spatial optimization), to which traditional statistical methods may not be applicable.
HPC, especially parallel computing, can not only speed up the computation and reduce the computing
time by utilizing powerful computing resources, but also makes it feasible/possible to handle vast amount
of data and conduct super large-scale and complex computation and simulation, which are previously
infeasible, or even impossible, using desktop/individual computing. The recent development in HPC
technologies, e.g., Graphics Processing Units (GPUs), Many Integrated Core (MIC), Cloud Computing,
and Cyberinfrastructure, has greatly stimulated the adoption of HPC in a wide range of geospatial
computing, such as terrain analysis, geostatistics, and spatio-temporal simulation.
Fewer efforts, however, have been made to combine CI and HPC in GeoComputation studies and
applications. We believe that effectively integrating the self-learning, self-adapting, and self-organizing
intelligence of CI and the high-throughput, on-demand, and collaborative computing of the emerging HPC
technologies will be one of the most important and productive research directions of GeoComputation in
the Big Data era. A high-performance land-use and land-cover change (LUCC) model, pLandDym, has
been under development. Some preliminary results will be presented as a showcase of combining CI and
HPC in solving complex geospatial problems.