Space-Time Causal Discovery in Climate Science: A Local Stencil Learning Approach

Submitted article (under review)
Authors
Affiliations

J. Jake Nichol

Department of Computer Science, University of New Mexico

Sandia National Laboratories

Baruch College, CUNY

G. Matthew Fricke

Department of Computer Science, University of New Mexico

Melanie E. Moses

Department of Computer Science, University of New Mexico

Santa Fe Institute

Diana Bull

Sandia National Laboratories

Laura P. Swiler

Sandia National Laboratories

Abstract: Causal discovery tools enable scientists to infer meaningful relationships from observational data, spurring advances in fields as diverse as biology, economics, and climate science. Despite these successes, the application of causal discovery to space-time systems remains immensely challenging due to the high-dimensional nature of the data. For example, in climate sciences, modern observational temperature records over the past few decades regularly measure thousands of locations around the globe. To address these challenges, we introduce Causal Space-Time Stencil Learning (CaSTLe), a novel algorithm for discovering causal structures in complex space-time systems. CaSTLe leverages regularities in local dependence to learn governing global dynamics. This local perspective eliminates spurious confounding and drastically reduces sample complexity, making space-time causal discovery practical and effective. These advances enable causal discovery of geophysical phenomena that were previously unapproachable, including non-periodic, transient phenomena such as volcanic eruption plumes. When applied to ever-larger spatial grids, CaSTLe’s performance actually improves because it transforms large grids into informative spatial replicates. We successfully apply CaSTLe to discover the atmospheric dynamics governing the climate response to the 1991 Mount Pinatubo volcanic eruption. We additionally provide extensive validation experiments to demonstrate the effectiveness of CaSTLe over existing causal-discovery frameworks on a range of climate-inspired benchmarks.

Working Copy (ESS Open Archive): 10.22541/essoar.172253117.78663487


Summary: We introduce a new method for learning the dynamics of causal systems, that is, the physical laws that define their behavior. While this problem, causal discovery, is not new, existing tools are ill-suited for large climate datasets. Current state-of-the-art approaches use statistical techniques to search for causal relationships between all aspects of a system, examining billions of possible causal effects. Instead of this ‘needle-in-a-haystack’ search, we incorporate basic physical principles—requiring effects to be ‘local’ and ‘uniform’—to massively simplify the causal discovery problem. We show that our approach can be used to make important climate discoveries by analyzing the 1991 Mt. Pinatubo eruption.

Illustration of the CaSTle Process: Transformation from original domain to reduced coordinates, parent identification in the reduced space, and re-expansion to the original domain.

Application of CaSTLe to the 1991 Mt. Pinatubo Eruption - CaSTLe successfully identifies drivers of stratospheric zonal transport (early) and meridional transport (later). See paper for details.

Citation:

@ARTICLE{Nichol:2024,
  AUTHOR="Jake J. Nichol and Michael Weylandt and G. Matthew Fricke and Melanie E. Moses and Diana L. Bull and Laura P. Swiler",
  TITLE="Space-Time Causal Discovery in Climate Science: A Local Stencil Learning Approach",
  YEAR=2024,
  DOI={10.22541/essoar.172253117.78663487},
  JOURNAL="ESS Open Archive 172253117.78663487"
}