Erich Osterberg

Accurate projections of streamflow, which have implications for flooding, water resources, hydropower, and ecosystems, are critical to climate change adaptation and require an understanding of streamflow sensitivity to climate drivers. The northeastern United States has experienced a dramatic increase in extreme precipitation over the past 25 years; however, the effects of these changes, as well as changes in other drivers of streamflow, remain unclear. Here, we use a random forest model forced with a regional climate model to examine historical and future streamflow dynamics of four watersheds across the Northeast. We find that streamflow in the cold season (November–May) is primarily driven by 3‐day rainfall and antecedent wetness (Antecedent Precipitation Index) in three rainfall‐dominant watersheds, and 30‐day rainfall, antecedent wetness, and 30‐day snowmelt in the fourth, more snowmelt‐dominated watershed. In the warm season (June–October), streamflow is driven by antecedent wetness and rainfall in all watersheds. By the end of the century (2070–2099), cold season streamflow depends on the importance placed on snow in the machine learning model, with changes ranging from −7% (with snow) to +40% (without snow) in a single watershed. Simulated future warm season streamflow increases in two watersheds (56% and 193%) due to increased precipitation and antecedent soil wetness, but decreases in the other two watersheds (−6% and −27%) due to reduced precipitation.