A Causal Inference Framework for Catchment Boundary Delineation

Over three decades ago, James Dooge emphasized the fundamental importance of closing the catchment water balance in hydrology. Yet, accurately closing the water balance within the true catchment boundary—a problem that appears simple in theory—remains unresolved. Conventional delineation methods based on surface topography often misrepresent the hydrologic system due to subsurface complexities such as karst conduits, fault lines, and slow cross-boundary groundwater flows. While delineation based on the water table may be conceptually sound, accurate, high-resolution water table depth data remain largely unavailable. In this paper, we propose a causality-based paradigm to address this challenge. We begin by positing that precipitation at a location contributes to (or triggers) streamflow if and only if that location lies within the true catchment boundary. We then explain why the purportedly “causal” methods usually utilized in hydrology, such as transfer entropy and mutual information, cannot be applied to such a problem. In turn, we introduce the theory of causal effect estimation as a potential remedy to delineate a catchment boundary from rainfall-runoff data. Finally, we explore where and under what circumstances such causal reasoning can solve hydrologists’ catchment delineation problem by virtually experimenting on a large sample of catchments in Germany.