Stochastic generation of missing data: methods and applications
The last years have seen ever-increasing remote sensing capabilities and improved numerical models that feed our understanding of Earth surface processes. However, it appears that all such data have intrinsic limitations: any acquisition procedure, no matter how sophisticated, is limited by sensor constraints (e.g., coverage, resolution, frequency), and numerical models are challenged for predicting the state of the environment under a changing climate. Addressing these limitations calls for increasing the data harvesting capability, which is often not possible.

This talk will provide a survey of models and algorithms that palliate this lack of exhaustive measurements for applications in hydrology and climate science. In particular, geostatistical tools can be used to stochastically generate unmeasured data about a studied process, which ideally should be statistically indistinguishable from the truth. This is enabled by new multiple-point geostatistical approaches that extract training information from analogues. An important aspect is that large data requirements are accompanied by large computational costs, which need to be addressed with efficient algorithms and cloud computing. Recent simulation algorithms will be presented, along with 1D and 2D geoscience applications.
Date: 14 November 2022, 11:30 (Monday, 6th week, Michaelmas 2022)
Venue: Dyson Perrins Building, off South Parks Road OX1 3QY
Venue Details: Diversity Room
Speaker: Gregoire Mariethoz (Université de Lausanne)
Organising department: School of Geography and the Environment
Booking required?: Not required
Cost: Free
Audience: Members of the University only
Editors: Chris White, Donna Palfreman