The goal of this project is to achieve more reliable predictions of hydrological signatures (peaks, flow duration curves etc.) in natural catchments using improved rainfall-runoff models and inference algorithms.
It is well known that rainfall- and runoff-time series obey remarkably universal scaling laws. This means that the long-range features of these time series might not depend on many details of the underlying small scale processes and thus be amenable to relatively simple stochastic modelling, which we want to exploit to arrive at more reliable runoff forecasts, in particular of extreme events.
Calibrating stochastic models to measured data and quantifying the ensuing uncertainty is a very computation-intense task, which we’ll tackle with our newly developed inference algorithms. They will yield more reliable estimates for parameter uncertainty and thus increase the reliability of our predictions.
A simple way of calibrating stochastic models is to compare only certain signatures of measured and simulated hydrographs. This is also interesting for un-gauged catchments, where we can infer certain signatures from the climate and geology. In hydrology, scaling laws have hardly been considered in the context of signatures so far. We expect the exponents of these laws to contain important information about the underlying catchment.