Data-Driven Interpretation of High-Resolution Stream Tracer Data

Compared with other hydrologic disciplines, tracer hydrology remains relatively data scarce, with most tracer series not exceeding a few hundred data points. Nevertheless, the availability and temporal resolution of tracer data is increasing, with opportunities to develop new models that rely less on a-priori assumptions and more on the data themselves.

Here, we introduce a novel data-driven approach for interpreting conservative tracer measurements in streamflow through the lens of transit time distributions (TTDs). We build on concepts from traditional TTD modelling and integrate them with tools from statistical learning. The proposed model is designed to infer time-variable TTDs by leveraging hydrologic and tracer data, but it can also incorporate additional information that may help characterize the catchment state, as e.g. time series of soil moisture, snow cover or plant status.

As transit times cannot be measured directly in any real-world catchment, we test and validate the model using virtual benchmark datasets designed to reflect realistic flow and transport dynamics. Results suggest that both long-term monitoring campaigns (e.g. multiple years of fortnightly sampling) and high-frequency in-situ measurements (e.g. 1-2 years of subdaily sampling) may provide sufficient information for data-driven interpretations of TTDs. While more applications and testing are needed, these early developments highlight the potential of data-driven methods for advancing our understanding of flow and transport processes in catchments.