In this cumulative dissertation, different features and methods are presented to assess and process multi-sensor derived radar data for climatological analysis. The overall objectives were to appraise the limitations of an hourly radar-based quantitative precipitation estimate (QPE) product and to develop and apply reasonable approaches to process these data. Hence the spatial and temporal limitations of radar-derived precipitation rates are discussed in the context of climatological applications, and two types of climatologies are obtained, first a climatology of daily precipitation fields and second a long term precipitation climatology. These relate to questions concerning the methodologies rather than climatological significance or assessment of precipitation and its role in the water balance. Current radar data availability limits such a hydro-climatic analysis. The thesis consists of three peer-reviewed publications. All investigations in this thesis are based on the RADOLAN rw-product of the German Weather Service (DWD) for an extended study region including the Free State of Saxony, Germany, for the period from April 2004 to November 2011. The first publication is dedicated to the classification of daily precipitation fields by unsupervised neural networks. In the presented work, the quality of the radar-derived precipitation rates is analysed by a temporal comparison between recording and non-recording gauges and the corresponding pixels of the RADOLAN rw-product on hourly and daily bases. The analysis shows that a temporal aggregation of the original product should be limited to a temporal scale up to 24 h because of the processing algorithms and the reappearance of previously suppressed errors. Nevertheless, an unsupervised neural network was successfully used for the classification of daily patterns. The derived daily precipitation classes and corresponding precipitation patterns could be assigned to properties of the associated weather patterns and seasonal dependencies. Hence, it could be shown that the classified patterns not only occurred by chance but by statistically proven properties of the atmosphere and of the season. The second publication is primarily concerned with two tasks: first, the pixel-wise fitting of mixture distributions on the bases of the obtained patterns from the first publication, and second, the analysis of spatial consistency of the radar-derived precipitation data set. The fitted parametric distribution functions were analysed in terms of Akaike\'s information criterion and the Kolmogorov-Smirnov test. These benchmarks showed, that the performances are best for mixture distributions derived by an initial classification by an unsupervised neural network and cluster analysis, and by gamma distributions. These results underline the significance of the derived precipitation classes obtained in the first publication. Furthermore, the Kolmogorov-Smirnov test indicates that independent of the distribution function, the radar-derived daily precipitation rates under the assumption of the deployed parametric distribution function has the best or most natural order of precipitation rates at spatial scales from 2 to 4 km for daily precipitation fields. Thus, it is recommended to use the original radar product at these scales rather than at 1 km resolution for daily precipitation sums. In the last publication, the focus shifts from daily to long-term precipitation climatology. The work introduces a rapid and simple approach for processing radar-derived precipitation rates for long-term climatologies. The method could successfully be applied to the radar-derived precipitation rates by excluding or correcting the errors that reappear due to temporal aggregation. Despite the fact that the approach is empirical, the introduced parameters could almost be objectively derived by means of simulation and optimisation. This could be achieved by utilising the reasonable relationship between elevation and precipitation rates for longer periods. Finally, the obtained results are compared to two independently derived precipitation data sets. The comparison shows good agreement of the precipitation fields and illustrates a reasonable application of the introduced procedure. The presented results support the application of the approach for precipitation aggregates of, at least, annual or longer periods. However the derivation of climatologies led to satisfactory results at the respective temporal scales, though the influence of radar-specific errors can only be minimized to a certain degree. Further studies have to prove if an application independent processing of radar-derived precipitation rates leads to higher qualities and validities of the derived data in time and space.

In this cumulative dissertation, different features and methods are presented to assess and process multi-sensor derived radar data for climatological analysis. The overall objectives were to appraise the limitations of an hourly radar-based quantitative precipitation estimate (QPE) product and to develop and apply reasonable approaches to process these data. Hence the spatial and temporal limitations of radar-derived precipitation rates are discussed in the context of climatological applications, and two types of climatologies are obtained, first a climatology of daily precipitation fields and second a long term precipitation climatology. These relate to questions concerning the methodologies rather than climatological significance or assessment of precipitation and its role in the water balance. Current radar data availability limits such a hydro-climatic analysis. The thesis consists of three peer-reviewed publications. All investigations in this thesis are based on the RADOLAN rw-product of the German Weather Service (DWD) for an extended study region including the Free State of Saxony, Germany, for the period from April 2004 to November 2011. The first publication is dedicated to the classification of daily precipitation fields by unsupervised neural networks. In the presented work, the quality of the radar-derived precipitation rates is analysed by a temporal comparison between recording and non-recording gauges and the corresponding pixels of the RADOLAN rw-product on hourly and daily bases. The analysis shows that a temporal aggregation of the original product should be limited to a temporal scale up to 24 h because of the processing algorithms and the reappearance of previously suppressed errors. Nevertheless, an unsupervised neural network was successfully used for the classification of daily patterns. The derived daily precipitation classes and corresponding precipitation patterns could be assigned to properties of the associated weather patterns and seasonal dependencies. Hence, it could be shown that the classified patterns not only occurred by chance but by statistically proven properties of the atmosphere and of the season. The second publication is primarily concerned with two tasks: first, the pixel-wise fitting of mixture distributions on the bases of the obtained patterns from the first publication, and second, the analysis of spatial consistency of the radar-derived precipitation data set. The fitted parametric distribution functions were analysed in terms of Akaike\'s information criterion and the Kolmogorov-Smirnov test. These benchmarks showed, that the performances are best for mixture distributions derived by an initial classification by an unsupervised neural network and cluster analysis, and by gamma distributions. These results underline the significance of the derived precipitation classes obtained in the first publication. Furthermore, the Kolmogorov-Smirnov test indicates that independent of the distribution function, the radar-derived daily precipitation rates under the assumption of the deployed parametric distribution function has the best or most natural order of precipitation rates at spatial scales from 2 to 4 km for daily precipitation fields. Thus, it is recommended to use the original radar product at these scales rather than at 1 km resolution for daily precipitation sums. In the last publication, the focus shifts from daily to long-term precipitation climatology. The work introduces a rapid and simple approach for processing radar-derived precipitation rates for long-term climatologies. The method could successfully be applied to the radar-derived precipitation rates by excluding or correcting the errors that reappear due to temporal aggregation. Despite the fact that the approach is empirical, the introduced parameters could almost be objectively derived by means of simulation and optimisation. This could be achieved by utilising the reasonable relationship between elevation and precipitation rates for longer periods. Finally, the obtained results are compared to two independently derived precipitation data sets. The comparison shows good agreement of the precipitation fields and illustrates a reasonable application of the introduced procedure. The presented results support the application of the approach for precipitation aggregates of, at least, annual or longer periods. However the derivation of climatologies led to satisfactory results at the respective temporal scales, though the influence of radar-specific errors can only be minimized to a certain degree. Further studies have to prove if an application independent processing of radar-derived precipitation rates leads to higher qualities and validities of the derived data in time and space.