Generating precipitation time series using simulated annealing

Long, high time resolution precipitation time series are often needed in hydrology. In most cases available measurements are not sufficient, because of a coarse time resolution and/or because observations are taken at differing locations. Often generated time series are used. These series are usually obtained from stochastic precipitation models that reproduce properties of the observed series. The purpose of this paper is to present a different methodology, one in which precipitation series can be generated by directly using their properties. The method uses simulated annealing, based on the Metropolis-Hastings algorithm. An objective function including all desired properties is formulated. The method generates series with the desired properties. The advantage Of this formulation is that properties of the precipitation that are important for the target application can directly be incorporated. The method can also be applied for generating precipitation series in a changed climate. For this purpose, changes of the statistical properties of the series have to be assessed.