M-CaStLe: Discovering Local Causal Structures in Multivariate Space-Time Gridded Data

Paper: M-CaStLe: Uncovering Local Causal Structures in Multivariate Space-Time Gridded Data Authors: J. Jake Nichol, Michael Weylandt, G. Matthew Fricke, Jhayron Perez-Carrasquilla arXiv: 2605.00398 | 2026-04-29

1. The Classic Problem: Correlation Is Not Causation

Imagine analyzing weather data:

• Temperature rises in region A → rainfall increases in region B
• The two are highly correlated

But: Does A cause B? Does B cause A? Or does some C influence both?

Limitations of traditional methods:

• They only find correlations
• They cannot determine causal direction
• They struggle in complex spatiotemporal systems

2. Challenges of Spatiotemporal Causal Discovery

High-dimensional gridded data:

• Many spatial grid points
• Multiple variables per point
• Relatively few temporal observations
• Curse of dimensionality

Limitations of existing methods:

• CaStLe (the predecessor): handles only univariate cases; multivariate scenarios are out of reach

What a new method needs:

• Multivariate handling
• Exploiting spatiotemporal locality
• Leveraging stationarity assumptions

3. M-CaStLe: Multivariate Causal Space-Time Stencil Learning

The paper proposes M-CaStLe (Multivariate Causal Space-Time Stencil Learning).

Core idea:

> Exploit spatiotemporal locality and stationarity to uncover local causal structures in high-dimensional gridded data.

Technical approach:

1. Local causal stencils — instead of finding global causal relationships, find local "stencil" patterns, e.g., "a rise near A leads to rainfall near B" as a local mode. 2. Space-time stencils — consider spatial neighborhoods and time lags, e.g., A at time t influences B at time t+1, but only in nearby regions. 3. Multivariate extension — handle multiple variables simultaneously (temperature, humidity, wind speed, etc.) and discover causal networks among them. 4. Scalability — leverage locality and stationarity; no global model needed; suitable for large-scale data.

A detective analogy: instead of asking "who influenced the whole world," ask "who influenced whom in this neighborhood." Local causality is more reliable and more verifiable.

4. Why Local Beats Global Causality

Problems with global causality:

• Curse of dimensionality: global model parameters explode; computation becomes infeasible; samples are insufficient
• Heterogeneity: different regions have different causal structures; a unified global model is inaccurate

Advantages of local causality:

• Tractable: local dimensionality is low; computation is feasible; samples are relatively sufficient
• More accurate: accounts for spatial heterogeneity; different regions get different causal structures; closer to reality
• Verifiable: local causality can be validated locally using neighboring data; high reliability

5. A Feynman-Style Judgment: Understanding the Local Is the Basis of Understanding the Global

Feynman said:

> "If you can't explain it simply, you can't explain it at all."

In causal discovery:

> "The global causal structure may be too complex to discover directly. But local causal structures are simpler and more reliable. M-CaStLe's wisdom is to start locally and gradually build global understanding."

This reflects the essence of the scientific method: from simple to complex, from local to global, from concrete to abstract.

6. Takeaways

If you work with spatiotemporal data or causal inference, ask yourself:

1. Does my data exhibit locality and stationarity? 2. Is global causality too complex to estimate? 3. Can local causal stencils provide more reliable insights? 4. How can causal networks among multiple variables be discovered?

M-CaStLe reminds us: in complex systems, causal discovery doesn't have to happen in one step. Starting locally and expanding gradually may be the more reliable path.

When causal structures in spatiotemporal data are revealed layer by layer, we understand not only "what causes what" but also "where it causes what." On the map of causal science, local discovery is the building block toward global understanding.

In the fog of correlation, causality is the lighthouse — and local causality is the brightest lighthouse of all.