V okviru 121. doktorskega seminarja Grajeno okolje bo imel Bruno Morales iz Univerze Leibniz v Hannoverju predavanje z naslovom:
Regionalization of the precipitation model’s parameters
Doktorski seminar bo potekal v četrtek, 6. septembra 2018, ob 14.30 uri v Predavalnici H-10 UL FGG (Hajdrihova 28).
Vljudno vabljeni!
ABSTRACT
Long and continuous series of rainfall data in a high temporal resolution are relevant for several purposes, namely for quantifying rainfall erosivity, hydrological applications involving urban or small, steep rural catchments, engineering design of flash flood control structures, etc. However, in Germany records with these characteristics are usually short and only available for some locations. Due to this issue, it is the aim to generate synthetic time series of precipitation with a high temporal resolution for all Germany on a grid with a spatial resolution of 1km x 1km. This research is part of an ongoing project “SYNOPSE II” designed to develop a regionalisation of a precipitation model for the whole country.
For this purpose, a stochastic precipitation model developed during the first stage of the project (SYNOPSE I) is applied, which is based on an alternating renewal process. The model is divided into an external and an internal structure and defined by rainfall characteristics which are described by probability distribution for each specific location. The regionalisation of the parameters of the probability distributions is performed by estimating the L-moments describing the rainfall characteristics. Different regionalization methods are considered, which are Nearest Neighbors, Multiple Linear Regression, Ordinary Kriging, External Drift Kriging and Residual Kriging. In order to estimate the parameters for locations without measurements, some of the regionalisation methods use as input site descriptors available for all grids based on long-term observations, which include spatial, topographic and hydrometeorological information.
In the study area, there are 803 stations available with a 5 minutes rainfall observation well distributed all over Germany. From the total number of stations, 45 stations are assigned for the validation process. The verification of the results is based on a statistical comparison between the observed and simulated precipitation time series and additionally, indirectly based on a small fictional drainage sewer system model concerning overflow and flooding simulations. The selection of the best regionalisation method is based on split sampling. For that purpose, a simulation of 100 synthetic samples of events is generated for each location in order to compare statistical characteristics between the observed and simulated time series. The ability to reproduce extreme events is additionally considered as a validation for the regionalization method.
Keywords: Continuous simulation, stochastic precipitation, high temporal resolution, regionalization, kriging.