Methodologies to characterize uncertainties in regional reanalyses
- 1Deutscher Wetterdienst, National Climate Monitoring, Frankfurter Str. 135, 63067 Offenbach, Germany
- 2Swedish Meteorological and Hydrological Institute, Folkborgsvägen 17, 601 76 Norrköping, Sweden
Abstract. When using climate data for various applications, users are confronted with the difficulty to assess the uncertainties of the data. For both in-situ and remote sensing data the issues of representativeness, homogeneity, and coverage have to be considered for the past, and their respective change over time has to be considered for any interpretation of trends. A synthesis of observations can be obtained by employing data assimilation with numerical weather prediction (NWP) models resulting in a meteorological reanalysis. Global reanalyses can be used as boundary conditions for regional reanalyses (RRAs), which run in a limited area (Europe in our case) with higher spatial and temporal resolution, and allow for assimilation of more regionally representative observations. With the spatially highly resolved RRAs, which exhibit smaller scale information, a more realistic representation of extreme events (e.g. of precipitation) compared to global reanalyses is aimed for. In this study, we discuss different methods for quantifying the uncertainty of the RRAs to answer the question to which extent the smaller scale information (or resulting statistics) provided by the RRAs can be relied on. Within the European Union's seventh Framework Programme (EU FP7) project Uncertainties in Ensembles of Regional Re-Analyses (UERRA) ensembles of RRAs (both multi-model and single model ensembles) are produced and their uncertainties are quantified. Here we explore the following methods for characterizing the uncertainties of the RRAs: (A) analyzing the feedback statistics of the assimilation systems, (B) validation against station measurements and (C) grids derived thereof, and (D) against gridded satellite data products. The RRA ensembles (E) provide the opportunity to derive ensemble scores like ensemble spread and other special probabilistic skill scores. Finally, user applications (F) are considered. The various methods are related to user questions they can help to answer.