Exploring the controls of catchment similarity for the transfer of rainfall-runoff model parameters: a comparative study in different large-sample datasets around the globe

The question of what makes two catchments hydrologically similar is of fundamental importance for the understanding of catchment hydrology and for transferring hydrological information from gauged to ungauged catchments. In the regionalisation of rainfall-runoff model parameters, the definition of a similarity measure for identifying the donors basins as a function of catchment characteristics is an essential step for the most consolidated techniques. As demonstrated by a very large number of studies in the literature, conducted all around the world, the main controls of catchment similarity may change subtsantially across different hydroclimatic regions. The recent availability of large-sample catchment dataset for rainfall-runoff studies in different hydroclimatic regions across the globe allows scientists to conduct comparative experiments for enhancing our knowledge about the factors that shape hydrological processes, including catchment similarity and regionalisation. The aim of this study is to test how hydroclimatic characteristics in different regions of the world influence the main factors that control catchment similarity when regionalising rainfall-runoff model parameters, using a homogenised modeling protocol. Two conceptually different bucket-type rainfall-runoff models are calibrated on more CAMELS-type large samples of catchments all around the world, characterised by different hydroclimates and data availability (i.e. streamgauge density). For each regional sample and for each model, one of the most consolidated parameter regionalisation approaches, based on the choice of a set of “most similar” donor catchments and on the transfer of the entire sets of model parameters from the donors to the target catchment, is applied in jack-knife cross-validation. Naturally, in such approach the choice of the donors (and therefore the regionalised model parameters) strictly depends on the catchment descriptors used to define the similarity measure between target and gauged basins. Assuming that the higher is the similarity of the donors to the target catchment, the better is the model performance, the idea of the work is to assess which catchment features better represent similarity for the transfer of model parameters in each of the regional samples. In particular, it is interesting to analyse if and how such features change across different hydroclimates. In order to reach such goal, the regionalisation technique is implemented by including different typologies and combinations of climatic and/or morphological characteristics when defining similarity, therefore obtaining different donors and different regionalised model performances. The findings achieved in the different large samples are compared, mainly focusing on how the set of basin descriptors bringing to the best model performances varies across the different hydroclimatic regions.