River flow changes on timescales ranging from minutes to millennia. These variations influence fundamental functions of ecosystems, including biogeochemical fluxes, aquatic habitat, and human society. Efforts to describe temporal variation in river flow—i.e. flow regime—have resulted in hundreds of unique descriptors, complicating interpretation and identification of global drivers of overall flow regime. In this study, we used three analytical approaches to investigate three related questions: 1. how interrelated are flow regime metrics, 2. what catchment characteristics are most associated with flow regime at different timescales globally, and 3. what hydrological processes could explain these associations? To answer these questions, we analyzed a new global database of river discharge from 3,685 stations with coverage from 1987 to 2016. We calculated and condensed 189 traditional flow metrics via principal components analysis (PCA). We then used wavelet analysis to perform a frequency decomposition of each time series, allowing comparison with the flow metrics and characterization of variation in flow at different timescales across sites. Finally, we used three machine learning algorithms to relate flow regime to catchment properties, including climate, land-use, and ecosystem characteristics. For both the PCA and wavelet analysis, just a few catchment properties (catchment size, precipitation, and temperature) were sufficient to predict most aspects of flow regime across sites. The wavelet analysis revealed that variability in flow at short timescales was negatively correlated with variability at long timescales. We propose a hydrological framework that integrates these dynamics across daily to decadal timescales, which we call the Budyko-Darcy hypothesis.