Advances in shot noise modeling of daily streamflows

Univariate shot noise models for streamflow generation at short time scales are examined in detail, to reconsider the verification of the basic hypotheses behind the models, the problem of objectively evaluating their performances, and the importance of model parsimony. The classical approach to model estimation is shown to produce some inconsistencies in the inverse evaluation of the model input, in particular regarding the assumed independence and Poissonianity of the pulses; an alternative procedure for pulses identification is proposed, which enables the mentioned hypotheses to be respected. To evaluate model performances, two indices are proposed, respectively based on the comparison of real and generated flow duration curves (I1) and annual maxima statistics (I2). A method for explicitly accounting for the dependence of I1 and I2 on the number of model parameters is described. An application to seven daily streamflow time series in northern Italy demonstrates the validity of the proposed procedure for the identification of the input and the usefulness of the performance indices in discerning among competing models.