GENERATING RAINFALL-RUNOFF DATA COLLECTION FOR CALIBRATION OF MACHINE LEARNING DRIVEN MODELS

Machine learning driven models provide a useful alternative for analytic modeling software in many domains. Simulating the rainfall-runoff process (i.e. transforming the fallen precipitations into the runoff in the corresponding outlet) is no exception and there are a lot of machine learning alternatives such as case-based reasoning, artificial neural networks etc. To facilitate their proper function, it is necessary to correctly set up the algorithm parameters or to provide a meaningful training data collection. However, in some areas, where the floods are not very frequent, it can be almost impossible to obtain the required combination of input measured precipitations amount and the corresponding measured output discharge in the outlet. In such case, the utilization of analytic modeling software (such as HEC-HMS, MIKESHE, HBV etc.) can be very helpful. This paper describes in detail our procedure for generating desirable data collection using such software including distorting of inputs and concatenation of partial results. It also clarifies the usage of verified rainfall-runoff model (the Floreon(+) system) and selection of studied area (Odra catchment).