Time-Varying Causal Treatment for Quantifying the Causal Effect of Short-Term Variations on Arctic Sea Ice Dynamics
Akila Sampath, Vandana P. Janeja, Jianwu Wang · 2026
Read Time-Varying Causal Treatment for Quantifying the Causal Effect of Short-Term Variations on Arctic Sea Ice Dynamics by Akila Sampath, Vandana P. Janeja, Jianwu Wang (2026) 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-2601-17647.txt.
Introduction
Akila Sampath, Vandana P. Janeja, and Jianwu Wang quantify causal links in Arctic sea ice change. Ice melt and freshwater spread interact in complex ways. Their effects show as regional shifts in sea surface height. The team uses that height as a proxy for ice dynamics. They apply time-varying causal treatment methods. Those methods track effects that change over time.
Causal proof is stronger than correlation for climate choices. It can show which feedbacks drive polar change. Polar change drives global sea-level rise. Sharper causal maps help planners weigh risks to coasts. The work brings modern causal tools to a hard climate case. Time-varying effects fit a system that never stands still.
Catalog record
| CREATOR | Akila Sampath, Vandana P. Janeja, Jianwu Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-17647
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Time-Varying Causal Treatment for Quantifying the Causal Effect of Short-Term Variations on Arctic Sea Ice Dynamics Akila Sampath, Vandana P. Janeja, Jianwu Wang Abstract Quantifying the causal relationship between ice melt and freshwater distribution is critical, as these complex interactions manifest as regional fluctuations in sea surface height (SSH). Leveraging SSH as a proxy for sea ice dynamics enables improved understanding of the feedback mechanisms driving polar climate change and global sea-level rise. However, conventional deep learning models often struggle with reliable treatment effect estimation in spatiotemporal settings due to unobserved confounders and the absence of physical constraints. To address these challenges, we propose the Knowledge-Guided Causal Model Variational Autoencoder (KGCM-VAE) to quantify causal mechanisms between sea ice thickness and SSH. The proposed framework integrates a velocity modulation scheme in which smoothed velocity signals are dynamically amplified via a sigmoid function governed by SSH transitions to generate physically grounded causal treatments. In addition, the model incorporates Maximum Mean Discrepancy (MMD) to balance treated and control covariate distributions in the latent space, along with a causal adjacency-constrained decoder to ensure alignment with established physical structures. Experimental results on both synthetic and real-world Arctic datasets demonstrate that KGCM-VAE achieves superior PEHE compared to state-of-the-art benchmarks. Ablation studies further confirm the effectiveness of the approach, showing that the joint application of MMD and causal adjacency constraints yields a 1.88% reduction in estimation error. 1 Introduction Earth system data exhibit large variability, stemming from climate model simulations and real-world observational data, leading to the increased use of machine learning (ML) and statistical methods to find patterns in the respective systems [ 1 ] . However, despite their strong predictive ability, current ML and deep learning methods have limitations in understanding the underlying cause-and-effect relationships between variables [ 1 , 2 ] , which is crucial for process understanding. Causal inference combines domain knowledge, machine learning models, and observational or interventional data [ 3 , 2 ] to identify a system’s underlying causal structure and quantify its effects. In this study, our objective is to move beyond correlation and estimate the total causal effect of sea surface height (SSH) on sea ice thickness. This chosen real-world scenario represents a highly impactful contribution because the two phenomena are critically linked to global climate change. Changes in SSH are a key factor in monitoring sea level rise, and the complex, non-direct influence of SSH dynamics on sea ice thickness through mechanisms like ocean circulation and freshwater distribution is a significant, yet difficult-to-quantify, area of climate research. Quantifying this non-direct causal effect is crucial for refining climate models and enhancing predictability. For this purpose, we propose a novel, specialized VAE-based architecture: the Knowledge-Guided Causal Modeling Variational Autoencoder (KGCM-VAE). This unique framework is designed to leverage the flexibility of VAEs and causal modeling to provide robust estimates of the total causal effect in such complex, indirect systems. To address this complexity, we first rely on a qualitative literature review [ 4 , 5 , 6 ] to inform the construction of a preliminary causal (time series) graph. This graph helps identify observed and unobserved confounding variables and potential connections between Arctic climate drivers and sea ice thickness. While Arctic climate drivers are known to influence sea ice thickness, our causal effect estimation is designed to quantify this influence precisely, especially when considering a graph of other observed and unobserved variables. Deeper understanding of a system requires combining causal inference with predictive machine learning, as it reveals insights that go beyond mere correlations [ 7 ] . Traditional predictive machine learning has demonstrated remarkable success in discerning patterns and generating predictions from data. However, these models primarily uncover correlations, often lacking a deep, theoretical understanding of the system’s true causal mechanisms. Judea Pearl argues that incorporating causal understanding into machine learning is essential for advancing beyond correlation-based predictions, empowering models to perform more reliably in new settings, improving their ability to handle unseen scenarios and cross-domain shifts [ 2 ] . Answering fundamental causal questions depends on accurately modeling how a treatment variable influences an outcome’s distribution [ 8 , 9 ] . However, inferring these relationships from observational data is highly challenging, primarily due to unmeasured confounders [ 10 , 11 ] . These hidden factors simultaneously affect both the observed treatment and the outcome, creating spurious correlations that obscure true causal pathways. Challenges such as fairness, generalization to unseen data, and domain adaptation often stem from these underlying causal complexities, which causal inference [ 10 , 11 ] aims to mitigate by learning robust data representations that facilitate the estimation of true causal effects. Climate research often relies on simulated models to discover causal dynamics, but trust in these outcomes is determined by the realism of the simulated physics compared to reality. Consequently, there’s a critical demand for deriving causal connections directly from real-world time series [ 12 , 1 , 13 ] , enabling data-driven insights that are grounded in observed phenomena rather than solely in model-based assumptions. Therefore, we aim to address a pivotal problem in both AI and causality by identifying abstract causal variables from concrete, low-level observations [ 14 ] . The main contributions in this paper are as follows: 1. We present the KGCM-VAE, a novel architecture that successfully integrates Maximum Mean Discrepancy (MMD)-based latent deconfounding within a recurrent VAE framework specifically designed for causal modeling. 2. Our method discovers the underlying structural dependencies (the adjacency matrix) among the variables directly from the data, providing a scientific advantage over models that require pre-defined or fixed relationships. 3. We demonstrate our KGCM-VAE provides highly reliable, unbiased estimates of time-varying causal effects, moving beyond correlation to accurately quantify treatment impact in complex sequential systems. 2 Related Work Causal inference approaches are typically divided into parametric and non-parametric methods, depending on the assumptions made about the structure of the relationships among variables [ 15 ] . In non-parametric approaches such as nearest-neighbor matching, propensity score matching, and propensity score re-weighting, the context-intervention-outcome relationship is not modeled explicitly [ 15 ] . Parametric methods such as linear and logistic regression, as well as structured models like random forests and regression trees, build explicit mappings to model the relationship between covariates, treatment, and outcomes [ 16 , 17 ] . Doubly robust approaches combine models for both the outcome and the treatment assignment, allowing for consistent estimation even if only one of the models is correctly specified [ 18 ] . However, these methods are efficient only when the treatment assignment probability is known. Machine learning methods like tree based methods are increasingly being adapted for causal effect inference, notably for individual-level treatment effects [ 17 , 19 ] . In addition to tree-based models, researchers have shown that high dimensional regression techniques like Lasso can be adapted to reliably estimate treatment effects, attaining convergence rates comparable to those in semi-parametric settings [ 20 , 19 ] . Representation learning offers a compelling approach to counterfactual inference by learning embeddings that help address the challenge of unobserved potential outcomes [ 27 , 22 ] . Although the SITE (Similarity-Preserved Individual Treatment Effect) framework leverages deep representation learning for estimating individual treatment effects, it generally lacks mechanisms to capture