Screening a precipitation stable isotope database for inconsistencies prior to hydrological applications – examples from the Austrian Network for Isotopes in Precipitation

The ratio between the heavy and light stable isotopes in precipitation (δ_p) is an effective tool in answering questions in hydrology, climatology, biogeochemistry and other disciplines, but only if spatiotemporally sufficient data is available provided by precipitation monitoring networks. However, when gathered into large databases this can contain errors that can severely impact research outcomes. The present study aims to systematically identify and propose, for the first time, a screening procedure and possible adequate solution(s) to database errors detected in precipitation stable isotope monitoring networks in a reproducible way. The proposed approach is a distance-based outlier detection variant heavily relying on empirical inspection of spatially clustered δ_p time series. The core of the methodology consisted of screening the (i) δ¹⁸O vs. δ²H cross plot and (ii) δ_p station time-series, and comparing them to their neighbors by organizing the δ_p monitoring stations into spatial domains. Potential errors were categorized into (i) point anomalies (isolated erroneous data points) and (ii) interval anomalies (sustained errors over time). The approach is demonstrated on the Austrian Network for Isotopes in Precipitation, a data base that collects data on a monthly basis since 1972 with more than 70 active stations at its peak in 2014. In this sense, it is a crucial backbone for understanding hydrological processes in Central Europe. At 10 stations only point anomalies were found, at six stations only interval anomalies (Achenkirch, Bad Bleiberg, Hütten, Lahn, Salzburg, Schoppernau), and at five (Apetlon, Podersdorf, Saalfelden, Villacher Alps, Weyregg) both kind of anomalies were detected. By addressing these errors case-by-case the reliability of a precipitation isotope database for hydrological and climatological research could be enhanced.