Towards model-to-measurement data cubes with rigorous uncertainty propagation

Analytically intractable uncertainty structures occur across the Earth system sciences when large datasets from numerical simulations and measurements are compared. In paleoclimatology, proxy system models map Earth system model (ESM) output onto proxy measurements through a process chain with multiple sources of autocorrelated and non-Gaussian uncertainties. Here, we aim at integrating point measurements and Monte Carlo techniques for uncertainty quantification into a data cube architecture, with paleoclimate proxies as example. Our python package cupsm implements metadata rich objects for paleoclimate simulations and proxy data that permit lazy loading and parallelized data cube operations. Operators mapping between the two objects propagate uncertainties with Monte Carlo methods. A tutorial demonstrates our approach by comparing transient simulations from the Last Glacial Maximum to present-day with a database of sea surface temperature reconstructions from biological and geochemical proxies. cupsm improves interoperability and reusability of data analysis workflows in paleoclimatology. The package is available as open source repository and has been presented at the EGU2024 conference and within NFDI4Earth activities. cupsm has the potential to operationalize the evaluation of paleoclimate simulations from model intercomparison projects. It can also serve as a proof-of-concept for other Earth system science communities in which measurement operators are subject to complex uncertainties.