The authors propose a new calibration algorithm based on a deep neural network learning, and on relations between input and output parameters. They try to formulate a new way to make interval predictions, and compare three approaches. However, as far as the insertion of more uncertainty into the initial parameters, in my opinion the resulting confidence intervals for the forecast in all of these approaches would be wider than in the standard statistical regression model approach. The main idea is to find the inverse function of the regression model that provides the best input parameter setting of the time series when the observational data is given. The results are depicted on a hydrological data set for weather forecasting.