Sacha W Ruzzante, Tom Gleeson

Excessively low stream flows harm ecosystems and societies, so two key goals of low-flow hydrology are to understand their drivers and to predict their severity and frequency. We show that linear regressions can accomplish both goals across diverse catchments. We analyse 230 unregulated moderate to high relief catchments across rainfall-dominated, hybrid, snowmelt-dominated, and glacial regimes in British Columbia, Canada, with drainage areas spanning 5 orders of magnitude from 0.5 to 55,000 km2. Summer low flows are decreasing in rainfall-dominated and hybrid catchments but have been stable in catchments that remain snowmelt or glacial-dominated. However, we find that since 1950 approximately one third of snowmelt-dominated catchments have transitioned to a hybrid rain-snow regime. The declines in rainfall-dominated and hybrid catchments are dominantly driven by summer precipitation and temperature, and only weakly influenced by winter storage. We apply this understanding to create regression models that predict the minimum summer flow using monthly temperature and precipitation data. These models outperform distributed process-based models for every common goodness-of-fit metric; the performance improvement is mostly a result of abandoning the requirement to simulate all parts of the annual hydrograph. Using these regression models we reconstruct streamflow droughts and low flow anomalies from 1901-2022. We reproduce recent drying trends in rainfall-dominated and hybrid catchments, but also show that present conditions are comparable to those seen one hundred years ago. However, anomalously low flows last century were caused by large precipitation deficits while current declines are driven by rising summer temperatures despite near-normal precipitation.