CFE 2019: Start Registration
View Submission - CMStatistics
Title: Challenges and solutions for multiscale global temperature reconstruction Authors:  Finn Lindgren - University of Edinburgh (United Kingdom) [presenting]
Abstract: As part of the EUSTACE project, an ambitious aim of constructing a historical global and daily temperature reconstruction system using recent advances in statistical modelling and computational methodology was envisioned. To approach this ambitious mission, models needed to be developed and investigated, to design a system that could integrate diverse observation systems, station biases, non-Gaussian temperature distributions, and diverse spatial and temporal correlation scales. At the same time, these high-dimensional models required novel computational approaches. We will present some of the challenges and solutions that were developed, with emphasis on the major modelling and computational challenges in taking theoretical models and scaling the associated computational methods from one hundred thousand dimensions up by six orders of magnitude in size, leveraging parallel computing and sparse iterative numerical solvers.