摘要：Abstract. Freshwater resources are of high societal relevance, and understanding theirpast variability is vital to water management in the context of ongoingclimate change. This study introduces a global gridded monthlyreconstruction of runoff covering the period from 1902 to 2014. In situstreamflow observations are used to train a machine learning algorithm thatpredicts monthly runoff rates based on antecedent precipitation andtemperature from an atmospheric reanalysis. The accuracy of thisreconstruction is assessed with cross-validation and compared with anindependent set of discharge observations for large river basins. Thepresented dataset agrees on average better with the streamflow observationsthan an ensemble of 13 state-of-the art global hydrological model runoffsimulations. We estimate a global long-term mean runoff of 38 452 km3 yr−1 in agreement with previous assessments. The temporal coverage ofthe reconstruction offers an unprecedented view on large-scale features ofrunoff variability in regions with limited data coverage, making it anideal candidate for large-scale hydro-climatic process studies, waterresource assessments, and evaluating and refining existing hydrologicalmodels. The paper closes with example applications fostering theunderstanding of global freshwater dynamics, interannual variability,drought propagation and the response of runoff to atmosphericteleconnections. The GRUN dataset is available athttps://doi.org/10.6084/m9.figshare.9228176(Ghiggi et al.,2019).
