Comment on hess-2021-445

doi:10.5194/hess-2021-445-rc4

Abstract

We investigate the potential of causal inference methods (CIMs) to reveal hydrological connections from time-series. Four CIMs are selected from two criteria, linear or nonlinear, and bivariate or multivariate. A priori, multivariate and nonlinear CIMs are best suited for revealing hydrological connections because they suit nonlinear processes and deal with confounding factors such as rainfall, evapotranspiration, or seasonality. The four methods are applied to a synthetic case and a real karstic study case. The synthetic experiment indicates that, unlike the other methods, the multivariate nonlinear framework has a low false-positive rate and allows for ruling out a connection between two disconnected reservoirs forced with similar effective precipitation. However, the multivariate nonlinear method appears unstable when it comes to real cases, making the overall meaning of the causal links uncertain. Nevertheless, all CIMs bring valuable insights into the system’s dynamics, making them a cost-effective and recommendable tool for exploring data. Still, causal inference remains attached to subjective choices and operational constraints while building the dataset or constraining the analysis. As a result, the robustness of the conclusions that the CIMs can draw deserves to be questioned, especially with real and imperfect data. Therefore, alongside research perspectives, we encourage a flexible, informed, and limit-aware use of CIMs, without omitting any other approach that aims at the causal understanding of a system.

Highlights

Causal inference methods (CIMs) aim at identifying causal interactions between variables from variables (Spirtes et al, 2000; Pearl, 2009)
We investigate the potential of causal inference methods (CIMs) to reveal hydrological connections from time-series
The four CIMs are applied to both a synthetic and a real case in a karstic study site

Summary

Introduction

Causal inference methods (CIMs) aim at identifying causal interactions between variables from variables (Spirtes et al, 2000; Pearl, 2009). When applied to time-series, these empirical methods are built upon the principle of priority of the cause, which goes back to Hume (Hume, 1748). They infer causation from the expected time-dependencies between causes and effects, i.e., the causes must occur before the effects. They have evolved throughout the 20th century to go beyond the well-known correlation, or cross-correlation, between two time-series (see Runge et al, 2019a, for a broad review). Dependencies between variables can be explained either by a direct causal link or through the common cause principle (Reichenbach, 1956; Runge et al, 2019a).

Methods

Findings

Discussion

Conclusion