Learning General Causal Structures with Hidden Dynamic Process for Climate Analysis

Understanding climate dynamics requires going beyond correlations in observational data to uncover their underlying causal process. Latent drivers, such as atmospheric processes, play a critical role in temporal dynamics, while direct causal influences also exist among geographically proximate observed variables. Traditional Causal Representation Learning (CRL) typically focuses on latent factors but overlooks such observable-to-observable causal relations, limiting its applicability to climate analysis. In this paper, we introduce a unified framework that jointly uncovers (i) causal relations among observed variables and (ii) latent driving forces together with their interactions. We establish conditions under which both the hidden dynamic processes and the causal structure among observed variables are simultaneously identifiable from time-series data. Remarkably, our guarantees hold even in the nonparametric setting, leveraging contextual information to recover latent variables and observable relations. Building on these insights, we propose CaDRe (Causal Discovery and Representation learning), a time-series generative model with structural constraints that integrates CRL and causal discovery. Experiments on synthetic datasets validate our theoretical results. On real-world climate datasets, CaDRe not only delivers competitive forecasting accuracy but also recovers visualized causal graphs aligned with domain expertise, thereby offering interpretable insights into climate systems.

Climate is driven by an invisible web of cause-and-effect: hidden atmospheric forces shape what we measure, and neighboring places affect each other through shared weather. Most AI tools that study climate either ignore the hidden drivers or assume the observed places are independent, so they reveal correlations but not the actual causes behind heatwaves, droughts, or floods. We built a method that uncovers both pieces at once: the hidden climate drivers and the cause-and-effect links between observed places, using only the time-stamped records climate scientists already collect. We proved this joint recovery is mathematically possible even when the underlying processes are messy and far from textbook assumptions, and turned the proof into a practical AI model called CaDRe. On standard real-world climate datasets, CaDRe forecasts as accurately as the leading AI models while also producing diagrams that match known phenomena such as westward Pacific winds. Climate scientists can therefore go beyond "what is correlated" to "what causes what," moving us toward more transparent and trustworthy AI for science.