A one-step similarity approach for the regionalization of hydrological model parameters based on Self-Organizing Maps

A novel one-step similarity approach for the regionalization of hydrological parameters has been developed. The method automatically associates model parameters (MP) with catchment descriptors (CD), using techniques based on Self-Organizing Maps (SOM-B). The combination of the one-step approach with the automatic association reduces the problem of equifinality and avoids subjective predefinition of relationships between CDs and MPs. Further positive properties of the SOM-B are the feasibility to use discrete variables as CDs and the ability to capture interdependencies between MPs. The SOM-B was validated against a known one-step regionalization method which is based on linear transfer functions (LT-B). Furthermore the influence of different factors like combination CDs, or size of the SOM, on the regionalization performance of the SOM-B was analyzed. The hydrological model used for the evaluations was a modified version of the HBV-IWS, which simulates continuous hourly discharge series. The investigations were carried out in 41 catchments of the Aller-Leine river basin in Northern Germany. For the validation a spatial split-sampling in 12 donor and 29 validation catchments, as well as a temporal split-sampling were applied. The results showed that: (I) MPs regionalized with the SOM-B approach led to similar or slightly better model performance than with the LT-B method, (II) some hydrologically plausible relationships between MPs and CDs were adjusted automatically without any predefinition by the SOM-B, (III) the relationships between CDs and MPs were more transparent using the SOM-B approach, (IV) the sensitivity of different factors of the SOM-B on the performance was smaller than expected.