The reduction of information contained in model time series through the useof aggregating statistical performance measures is very high compared to theamount of information that one would like to draw from it for modelidentification and calibration purposes. It has been readily shown that thisloss imposes important limitations on model identification and -diagnosticsand thus constitutes an element of the overall model uncertainty. In thiscontribution we present an approach using a Self-Organizing Map (SOM) tocircumvent the identifiability problem induced by the low discriminatorypower of aggregating performance measures. Instead, a Self-Organizing Map isused to differentiate the spectrum of model realizations, obtained fromMonte-Carlo simulations with a distributed conceptual watershed model, basedon the recognition of different patterns in time series. Further, the SOM isused instead of a classical optimization algorithm to identify those modelrealizations among the Monte-Carlo simulation results that most closelyapproximate the pattern of the measured discharge time series. The resultsare analyzed and compared with the manually calibrated model as well as withthe results of the Shuffled Complex Evolution algorithm (SCE-UA). In ourstudy the latter slightly outperformed the SOM results. The SOM method,however, yields a set of equivalent model parameterizations and thereforealso allows for confining the parameter space to a region that closelyrepresents a measured data set. This particular feature renders the SOMpotentially useful for future model identification applications.
展开▼