Similarity of catchment dynamics based on the interaction between streamflow and forcing time series: Use of a transfer entropy signature

The transfer of the hydrological information between catchments is founded on the definition of hydrological similarity, which is in turn strictly connected to the features to be regionalised. In order to characterise the catchment behaviour in the streamflow generation processes, the similarity should reflect also the interaction between meteorological forcings and river streamflow time series. While previous hydrological research has identified basins with similar meteorological forcings (i.e. similarity of climate) or with similar streamflow time-series (i.e. similarity of runoff response), the present work proposes, for the first time, to quantify the interaction between the entire time-series of different forcing data and streamflow observations, to be considered as a novel hydrological signature and used as a catchment similarity metric. In particular, the present study proposes the use of a multi-variate entropy-based measure, the so-called transfer entropy, a time-asymmetric quantity which analyses the interaction between different signals. The concept of transfer entropy is applied for identifying the dominant hydrological processes occurring in a catchment, measuring the transfer of information from different meteorological forcings over the catchment to the corresponding observed time series of daily streamflow at the basin outlet. The resulting transfer entropy values are then used as signatures to characterise the main catchment dynamics, and a classification of the basins region is obtained assuming that similar values of transfer entropy correspond to hydrologically similar basins. The methodology is tested on a densely-gauged set of more than 200 catchments across Austria and the outcomes of the approach are evaluated against a set of morpho-climatic catchment attributes and typical streamflow signatures. Despite the limitations, the method is able to distinguish the predominant or partial role of snow melt and evapotranspiration across the dataset, helping to assess differences in catchment response time and to highlight the role of very high orographic precipitation in catchments with a dominant snow regime. The study demonstrates the potential of transfer entropy as complementary to consolidated streamflow signatures for assessing hydrological similarity and for quantifying the connection between different catchment processes.