erzonsky, Matthew, Department of Civil and Environmental Engineering, The Pennsylvania State University, 208 ECORE Building, 556 White Course Drive, University Park, PA 16802, mpb5796@psu.edu; Smykalov, Valerie, Department of Civil and Environmental Engineering, The Pennsylvania State University, 208 ECORE Building, 556 White Course Drive, University Park, PA 16802, vds5105@psu.edu; Sadayappan, Kayalvizhi, Department of Civil and Environmental Engineering, The Pennsylvania State University 208 ECORE Building, 556 White Course Drive University Park, PA 16802, kayal@psu.edu; Seibert, Jan, Department of Geography, University of Zurich, Winterthurerstrasse 190, Zurich, Switzerland 8057, jan.seibert@geo.uzh.ch; Li, Li, Department of Civil and Environmental Engineering Pennsylvania State University, 208 ECORE Building, 556 White Course Drive, University Park, PA 16802, lxl35@psu.edu.
Climate change has caused varied long-term changes in river discharge. Although precipitation is often considered as primarily driving discharge, some rivers have shown dwindling discharge despite increasing precipitation, whereas others have shown increasing discharge albeit declining precipitation. It is however not clear how widespread such diverging trends are, and what are the primary drivers for diverging trends. Here we quantify the temporal trends of precipitation and river discharge over 1980 – 2014 in 671 sites, and identify the drivers of the trends in the continental US using the CAMELS-US dataset. The relative fractions of surface flow, shallow subsurface flow and deep subsurface flow based on their depth of generation to total stream discharge were also estimated using the HBV (Hydrologiska Byråns Vattenbalansavdelning) hydrology model. Long-term trends in total discharge and its flow fractions were calculated with a Theil-Sen regression. Results show that many sites have different trend directions in discharge and precipitation, with diverging trends observed in 28% of the 671 sites. Surface, shallow subsurface and deep subsurface flow fractions are increasing at 48%, 56% and 39% of sites and decreasing at 37%, 43% and 61% of sites respectively. Catchment attributes that best explained these trends were identified using a random forest model. Evapotranspiration emerged as the primary driver for trends of discharge and deep subsurface flow. However, precipitation is the primary control over surface and shallow subsurface flow. These results indicate that the discrepancies between discharge and precipitation trends are largely driven by simultaneously changing evapotranspiration and precipitation.
Hydrological Modeling, Machine Learning