Our approach

The project will pioneer a powerful and generalised approach to the classical signal processing problem of “source separation”. Our approach is based on a Bayesian Hierarchical Model (BHM) – shown schematically below – which contains three ‘layers’ of information: (i) an observational layer that utilises all available direct observations, (ii) a process layer that describes the relationship between the physical processes and the observations; and (iii) a parameter layer that contains prior information about the unknown parameters in the other two layers.

At its core is the concept that each observational dataset has markedly different spatio-temporal characteristics and patterns of error (or smoothness in time and space). These differences, if accounted for correctly, mean that a statistically rigorous combination of the datasets can produce robust ‘separation’ of the signal between the five physical processes that influence sea level. When the observations are not uniquely controlled by a single process, as is the case here, the use of prior information greatly facilitates source separation.

This approach has already been developed and tested over Antarctica in the NERC-funded RATES (Resolving Antarctic ice mass TrEndsS) project, using a subset of observational datasets. GlobalMass extends this work to include more direct observations and prior information, and by simultaneously solving the problem at a global scale.

In this project, this approach will be applied to the global sea level budget. Sea level budget determines change in sea level, which can be approximated as follows:

This equation can be further expanded to recognise the latent geophysical processes involved:

The GlobalMass project will investigate each of these components, as well as establishing a statistical framework that will allow the vast amount of observations and prior knowledge that exists about them to be combined and resolved. These components define the five work packages (WPs): Statistical framework (WP1), Solid-Earth (WP2), Oceans (WP3), Land ice (WP4) and Hydrology (WP5).

All five work packages are strongly inter-linked. WPs 2-5 will make use of complementary data sets, similar processing and all possess common intellectual themes, challenges and tasks. They will be characterised by a similar approach: i) compilation of the observational time series, ii) defining their error covariance and spatio-temporal properties, iii) development and integration of the priors from both simulators and observations and iv) integration into the statistical framework. WP1 will develop and provide this framework.

Next page: Statistical framework (WP1)