CombiPrecip [1] is the real-time, multi-sensor scheme used for precipitation estimation in Switzerland. It is the outcome of a three-year development within MeteoSwiss and has already operated successfully for one year. The application combines weather radar precipitation grid estimates and raingauge precipitation point measurements into a single rainfall map. Co-existence of raingauge and radar precipitation estimates has naturally, since decades, generated expectations for effective combination schemes. Although, the problem has been discussed in the literature, there are plenty of subtleties in both theoretical and practical level, which are far from obvious at first glance. This presentation aims at communicating the experience we acquired during the development of CombiPrecip especially on such subtleties. This could be valuable for institutes which would like to develop or improve their own radar-raingauge combination applications. Several questions are addressed.(a) What combination techniques are optimal for an operational combination scheme?(b) How can one merge efficiently two types of observations with widely different domain, resolution, sampling volume, and error scheme (such as those produced by radars and raingauges)?(c) How should the combination be properly extrapolated to regions not monitored by raingauges (for example outside the country border or over oceans)?(d) How should convective precipitation be treated, given that in such cases raingauges often suffer from a larger representativeness error?(e) How is it possible to produce very high temporal resolution (eg 5-minutes) combination maps?

CombiPrecip is based on a geostatistical scheme, but its modeling differs from other similar designs in that it relies not only on spatial but also on temporal data. Its input is hourly-aggregated, so discrepancies between raingauge and radar are not very sizeable, while its output relies on a disaggregation scheme to produce high temporal resolution rainfall fields (5-min), potentially valuable for hydrological forecasting. It incorporates an algorithm for providing rainfall estimates in regions far away from the available raingauge network. It also incorporates an efficient technique for improving results during convective precipitation. CombiPrecip has already been employed in postprocessing eight years of hourly data while currently operates automatically in real-time. It typically requires only a few minutes to produce its output. Cross-validation skill scores for a rich range of explored events suggest that there is a significant improvement over the radar-only precipitation fields. The most pronounced result is that CombiPrecip images are typically mean-field bias-free.