Causal discovery in urban data with temporal empirical dynamic modeling: The R package tEDM

Abstract

Urban systems are complex, nonlinear, and adaptive, with interactions that span multiple spatial and temporal scales. Correlation- and regression-based methods often fail to capture their dynamic causal structures. Although causal inference offers a stronger framework, causal discovery remains underexplored in urban research. Empirical dynamic modeling (EDM) reconstructs system dynamics directly from time series without a prespecified model and is well suited to this task, yet it has rarely been applied in urban contexts. To fill this gap, we introduce tEDM, an open-source R package that implements and extends EDM for temporal causal discovery in urban data. With a C++ computational backbone and R integration, tEDM supports heterogeneous, high-frequency, multivariate, and spatially replicated urban time series, and is efficient and usable for large-scale urban analysis. Case studies covering environmental health, carbon-climate dynamics, and epidemic spread demonstrate its applicability. The package is openly available at https://github.com/stscl/tEDM.