Automating Code Generation for Point Evaluation in Geoscience Models

Abstract:

Firedrake is a code generation system which enables straightforward and scalable model development using finite element methods via high level abstractions of mathematical concepts. Several software packages build on Firedrake for applications in the geophysical sciences: the ice sheet model Icepack, the atmospheric model Gusto, and the ocean model Thetis. In the past, assimilating sparse point data (as opposed to gridded data) into models discretised using the finite element method has been challenging. We present progress towards evaluating, manipulating, assimilating and creating equation systems with large numbers of point measurements in such models. Examples of point data sets include measurements of ice sheet elevation from satellite altimetry and measurements of ocean salinity and temperature from drifting buoys. Our framework is compatible with the automated adjoint system pyadjoint, facilitating automated code generation for inverse problems with point data. This work is part of a broader project aiming to use Domain Specific Languages (DSLs) to automate diagnostics of geoscientific models with very large output data sets.