Exploiting big data for greenhouse gas emissions estimation using INLA

Greenhouse gases are accumulating in the atmosphere but our current understanding of their sources and sinks is poor. Robust, transparent evaluation of global greenhouse gas emissions is now recognised as a major international research priority (e.g. Peters et al., “Towards real-time verification of CO2 emissions”, Nature Climate Change, 13th November 2017). Policy makers need to understand emissions at national scales so that they can enact appropriate regulation and ensure that current protocols are being followed; industry representatives need to know the impact that their activities have on the global climate; scientists hope to the improve their understanding of the global carbon cycle, which will allow for more accurate predictions of future climate change.

The Earth’s surface is far too large to measure its gas emissions directly. The Atmospheric Chemistry Research Group has therefore pioneered the use of statistical methods to indirectly infer spatio-temporal emissions based on a limited number of concentration measurements and atmospheric models. Instruments located on towers at a handful of locations around the world generally provide these measurements, along with few auxiliary datasets, such as from aircraft campaigns and limited satellite observations. This limited availability of measurement data is about to change. The successful launch of the TROPOMI instrument on board the Sentinel 5 Precursor satellite in October 2017 will provide unprecedented data volumes on atmospheric methane in the atmosphere. The new measurements have the potential to provide us with a wealth of information to improve emission estimates but the currently used statistical techniques will become computationally prohibitive with such large data volumes.

Colleagues in the School of Geographical Sciences and School of Mathematics have faced a similar big data problem when researching spatio-temporal sea-level changes. They have been able take advantage of similar volumes of satellite data using two recent complementary advances in spatio-temporal statistics: the integrated nested Laplace approximation (INLA) for Bayesian inference on hierarchical models and the stochastic partial differential equation (SPDE) approach for approximating spatial Gaussian processes. Using the knowledge and experience of an interdisciplinary team of scientists we will apply this method to infer emissions of greenhouse gases.