Author(s): M. Borga; F. Marra

Linked Author(s):

Keywords: Radar rainfall; Flash floods; Parameter uncertainty; Bayesian calibration

Abstract: Current systems for radar-based precipitation estimation consist of several interconnected algorithms, with a large number of (often nonlinear) functions and parameters controlling its performance. The parameter values of the system are often based on limited experimental studies and, more recently, on the implementation of calibration procedures (based on radar-raingauge comparison) aimed at identifying the optimal parameter set. This work suggests, however, that there may be many parameter sets within an algorithm structure that are equally acceptable as predictors of rainfall at the ground, and that these may come from very different regions in the parameter space. A methodology is proposed to take into account and assess the uncertainty arising from errors in parameter selection. The methodology reformulates the algorithm calibration problem into the estimation of posterior probabilities of algorithm responses, thereby avoiding the idea that there is an optimum parameter set. Results are presented for a set of 8 flash flood triggering storms, among the most intense in the last 15 years occurred in Europe. We show that the uncertainty assessment approach is not only practical and relatively simple to implement but can also provide useful information about the scale-dependency of the radar-rainfall uncertainty and about the propagation of radar uncertainty into flood prediction.

DOI: https://doi.org/10.3850/978-981-11-2731-1_368-cd

Year: 2018