Evaluation of the spatial distribution of a stochastically reconstructed ensemble of precipitation fields using RMWSPy

Precipitation is one of the most essential variables within the hydrological system, and accordingly one of the main drivers for terrestrial hydrological processes. The quality of many hydrological applications such as climate prediction, water resource management, and flood forecasting, depends on the correct reproduction of its spatiotemporal distribution. Not only are there a variety of methods for reconstructing precipitation maps, but the reconstruction can also be based on different observation types or its combinations. In our approach we use rain gauge observations and path-averaged rain rates, derived from Commercial Microwave Link (CML) attenuations, as observations. Using these two observation types we apply Random Mixing Whittaker-Shannon (RMWSPy) to stochastically simulate precipitation fields. 

The algorithm generates precipitation fields as a linear combination of unconditional spatial random fields, where the spatial dependence structure is described by copulas. The weights of the linear combination are optimized in such a way that the precipitation observations themselves as well as their spatial structure are reproduced. Using RMWSPy allows to simulate a precipitation field ensemble of any size, where each ensemble member is in concordance with the underlying observations. 

Here, we apply RMWSPy to the whole of Germany and various catchments of different sizes, covering cases with different amount of available observations and different orographic complexity. The resulting ensembles of precipitation fields are evaluated regarding the quality of the reproduced spatial distribution of the precipitation and its pattern. We show that the reconstructed precipitation fields reproduce the observed spatiotemporal distribution in a quality that is comparable to the gauge-adjusted radar product RADOLAN-RW provided by the German Weather Service (DWD).