Print this Page

Topical Session 5: Dimensionality Reduction and Local Methods for Big Spatial and Space-time Data

Conveners: D. Hristopulos (Technical University of Crete, Greece), D. Allard (Applied Mathematics and Informatics Division, INRA, France)

Big data is starting to make an impact in the geosciences given the abundance of remote sensing and earth-based observations related to climate and environmental processes, as well as the existence of large databases pertaining to mining applications. This data explosion underscores the need for algorithms that can handle large spatial and space-time data sets. Most current methods of data analysis, however, have not been designed to cope with large data. Therefore, new methods are needed which can achieve favorable scaling of computational resources with increasing data size.
This session will comprise contributions focusing on methodologies that can lead to improved scaling of the computational resources with the size of the data as well as the analysis of interesting big spatial or space-time data sets by means of existing methods. Examples include local approximations, sparse constructions, dimensionality reduction techniques, and parallel algorithms. Contributions using local approximations may involve methods such as maximum composite likelihood and maximum pseudo-likelihood and covariance tapering. Dimensionality reduction contributions will comprise methods such as fixed rank kriging, polynomial chaos, and Karhunen-Loève expansions. Sparse constructions will involve, among other topics, Markov random fields and extensions to irregularly spaced data by means of various approaches. The methodological tools that will be portrayed in this session could have their roots in traditional spatial statistics or in machine learning.

Permanent link to this article: