Bridging Prediction and Attribution: Identifying Forward and Backward Causal Influence Ranges Using Assimilative Causal Inference
Marios Andreou, Nan Chen · 2025
Read Bridging Prediction and Attribution: Identifying Forward and Backward Causal Influence Ranges Using Assimilative Causal Inference by Marios Andreou, Nan Chen (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-21889.txt.
Introduction
Marios Andreou and Nan Chen extend a method called assimilative causal inference, which blends real observations with a working model of a system to trace causes from effects. Knowing that two things are causally linked is only half the answer; the pair also asks when an influence began and how long it lasts.
They define two influence ranges for each moment in time. The forward range measures how long a cause keeps mattering after it acts, and the backward range reaches back to find when the trigger of an observed effect first emerged. Clean mathematical formulas and objective measures for both ranges remove the need for arbitrary cutoffs. The framework ties causal detection to prediction and attribution, separating what can be forecast from what can be explained afterward.
Catalog record
| CREATOR | Marios Andreou, Nan Chen |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 101 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-21889
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Bridging Prediction and Attribution: Identifying Forward and Backward Causal Influence Ranges Using Assimilative Causal Inference Thanks: Submitted to SIAM/ASA Journal on Uncertainty Quantification on October 21, 2025. Marios Andreou, Nan Chen Abstract Causal inference identifies cause-and-effect relationships between variables. While traditional approaches rely on data to reveal causal links, a recently developed method, assimilative causal inference (ACI), integrates observations with dynamical models. It utilizes Bayesian data assimilation to trace causes back from observed effects by quantifying the reduction in uncertainty. ACI advances the detection of instantaneous causal relationships and the intermittent reversal of causal roles over time. Beyond identifying causal connections, an equally important challenge is determining the associated causal influence range (CIR), indicating when causal influences emerged and for how long they persist. In this paper, ACI is employed to develop mathematically rigorous formulations of both forward and backward CIRs at each time. The forward CIR quantifies the temporal impact of a cause, while the backward CIR traces the onset of triggers for an observed effect, thus characterizing causal predictability and attribution of outcomes at each transient phase, respectively. Objective and robust metrics for both CIRs are introduced, eliminating the need for empirical thresholds. Computationally efficient approximation algorithms to compute CIRs are developed, which facilitate the use of closed-form expressions for a broad class of nonlinear dynamical systems. Numerical simulations demonstrate how this forward and backward CIR framework provides new possibilities for probing complex dynamical systems. It advances the study of bifurcation-driven and noise-induced tipping points in Earth systems, investigates the impact from resolving the interfering variables when determining the influence ranges, and elucidates atmospheric blocking mechanisms in the equatorial region. These results have direct implications for science, policy, and decision-making. keywords causal inference, causal influence range, data assimilation, Bayesian inverse problems, uncertainty quantification Funding. This work was funded by the Office of Naval Research (N00014-24-1-2244) and the Army Research Office (W911NF-23-1-0118). † † runningheads: Bridging Prediction and Attribution / Marios Andreou and Nan Chen MSC 62F15, 62D20, 62M20, 93E11, 93E14, 60H10 1 Introduction Causal inference identifies cause-and-effect relationships between state variables [ 64 , 62 , 78 ] . It is widely used across various scientific and societal areas. Beyond improving the understanding of the underlying dynamics, causal inference additionally plays a crucial role with assisting in the discovery of the governing equations, evaluating interventions, and informing decision-making [ 72 , 16 , 8 , 82 , 71 , 69 , 92 , 80 , 94 ] . Most of the traditional causal inference methods utilize multivariate time series data. Methods falling into this category include direct time series analysis (Granger causality [ 32 ] , transfer entropy [ 79 , 9 ] , causation entropy [ 84 , 85 ] ), state space reconstruction methods (topological causality [ 34 ] , convergent cross mapping [ 83 , 61 ] ) and statistical learning of structural causal models (interventional data methods [ 81 , 12 ] , mixture causal discovery [ 91 ] ). In these approaches, the identified average-in-time causal relationship provides an overall indication of the causal link. However, many complex dynamical systems exhibit strong intermittency and nonlinearity, which means the cause-and-effect relationships between state variables may cease or shift repeatedly over time. Therefore, identifying the instantaneous causal link is crucial for understanding transient behaviors such as extreme events, bifurcations, and regime switches. There exist model-based strategies that track time-dependent relationships, such as those utilizing functional operators (information flow [ 44 ] , Koopman causality [ 73 ] ) and linear response theory (fluctuation-dissipation theorem [ 26 ] , Green’s function [ 50 ] ). However, these methods are usually computationally intractable even when validating the hypothesis of a single possible causal scenario, while some of them are only able to provide temporal results for a very short time. Assimilative causal inference (ACI) [ 6 ] , an approach which aims to address the time evolution of causal relationships, has recently been developed. It integrates data and models for causal inference through Bayesian data assimilation, formulating the study of causality as an inverse problem of uncertainty quantification, where causes are traced backward from observed effects. Key advantages of ACI include the online tracking of cause-and-effect roles which can intermittently swap, the accommodation of short and incomplete datasets that contain observations only from potential effects, and computational efficiency in high-dimensional settings by leveraging established techniques from data assimilation. Although the discovery of instantaneous causal links is a critical first step, a comprehensive causal analysis must also quantify the associated causal influence range (CIR) that defines the temporal domain of significant causal impact. This concept has two distinct interpretations: A forward (predictive) CIR, describing the future period over which a current cause will remain influential, and a backward (attributable) CIR, describing the past period where causal progenitors responsible for a currently observed effect emerged: (A) The forward CIR quantifiably answers the operational question: “For how long will this causal effect persist?”. It assesses the remaining duration of an ongoing event or, following an early warning signal, infers the likelihood of an extreme event’s onset. By providing a forecast of causal persistence, the forward CIR facilitates proactive preparedness for preventing catastrophic events, such as droughts, tsunamis, or cyclones. (B) The backward CIR addresses the question: “When did the causal precursors for this observed event emerge?”. It pinpoints the onset of a damaging event and reveals its underlying driving mechanisms. By supporting root-cause analysis through causal attribution, the backward CIR guides targeted strategies for disaster preparedness and risk reduction. For instance, it can identify which factors most strongly influence long-term changes in Earth system processes, thereby helping to prioritize the most effective adaptation measures. This work builds upon the ACI framework to develop mathematically rigorous metrics for forward and backward CIRs. These formulations provide the first general framework for quantifying the temporal reach of a causal link in turbulent systems, both in the predictive (future) and attributable (past) directions. A key advantage of this method is that it utilizes Bayesian data assimilation in ACI to yield computable expressions for CIRs. These expressions are applicable to complex dynamical systems, including those with diverse autocorrelation structures and chaotic dynamics. Critically, these CIR measures are objectively defined without relying on empirical thresholds. Furthermore, they admit computationally efficient algorithms with closed-form approximations for a broad class of turbulent systems, making asymptotic analysis and practical estimation possible even in high-dimensional applications. The remainder of this paper is organized as follows. Section 2 provides a brief overview of ACI fundamentals and develops the theory for forward and backward CIRs, including derivations of computationally efficient approximations. We also introduce a broad class of nonlinear dynamical systems that enables the analytical evaluation of CIR metrics without requiring ensemble approximations. Section 3 applies the CIR framework to: (a) Distinguishing bifurcation-driven from noise-induced tipping points in Earth system models, (b) clarifying prediction and attribution in multiscale atmospheric models by resolving interfering dynamical components, and (c) identifying atmospheric blocking mechanisms in an equatorial circulation model with flow-wave interactions and seasonal feedbacks. Concluding remarks and discussion are presented in Section 4 , with additional mathematical derivations in the Appendix. 2 Methodology 2.1 Notations Throughout this work, boldface variables are used for multidimensional