On the interpolation of precipitation data over complex terrain

The knowledge of areal precipitation is crucial for many scientific applications, e.g. in hydrology, environmental and agricultural meteorology, climatology as well as for practical applications, e.g. land-use management, risk assessment. Thus, many different interpolation methods have been developed within the last century.

The motivation for this paper stems from the need of operational high resolution real-time analyses of areal precipitation over complex terrain including data quality control. Aspects, which have to be dealt with, are: a) treatment of observational and measurement errors, b) anisotropic structures in the field, forced by complex topography, c) performance of the analysis scheme to changing station density, d) sensitivity of the results to different station distributions and e) the analysis method should be stable and cheap.

The Vienna Enhanced Resolution Analysis Scheme (VERA) comprises most of the mentioned aspects. It includes a sophisticated data quality procedure and the so called fingerprint technique, which recognises patterns of topographic impact in the data and allows to introduce anisotropic structures. The system runs operationally in a stable mode since several years and takes about the order of minutes on a commercial PC to calculate fields on a 20km grid for the Alpine region and to produce the graphics for several meteorological parameters.

This paper focuses on the performance of the VERA system on changing station density and distribution. Two pilot case studies of the project RISK-Advanced Weather forecasting system to Advice on Risk Events and management (RISK-AWARE) have been selected in the Upper Danube Catchment area. Both cases have in common that they caused severe widespread flooding in the area.

The investigation has been performed within three domains. The areas of the domains differ by a factor of 10 approximately. An inverse distance weighting (IDW) approach has been used to compare the VERA-fields against the results of a common interpolation scheme. Beside the routinely and in real-time available SYNOP data, the precipitation data from the dense hydrological network have been utilized on a post event basis. Efficiency and root mean square error have been used as statistical measures to characterize the quality of the results. The mean areal precipitation can be well estimated for the two cases in view from the routine available SYNOP network in the largest domain (≈170.000 km²). Additional stations from the hydrological network increase the variance in the field but do not change the mean value substantially. This finding changes for smaller domains when the mean areal precipitation increases by a factor of two by using the dense hydrological network. The two interpolation methods used show quite similar results. It seems that the nature of precipitation (large-scale versus convective) has the strongest impact on the quality of the results. For example, the efficiency drops from 0.84 (for large scale event) to about 0.47 (for convective event).