Identification of dominant source of errors in developing streamflow and groundwater projections under near-term climate change

Uncertainties in projecting the changes in hydroclimatic variables (i.e., temperature and precipitation) under climate change partly arises from the inability of global circulation models (GCMs) in explaining the observed changes in hydrologic variables. Apart from the unexplained changes by GCMs, the process of customizing GCM projections to watershed scale through a model chain-spatial downscaling, temporal disaggregation, and hydrologic model-also introduces errors, thereby limiting the ability to explain the observed changes in hydrologic variability. Toward this, we first propose metrics for quantifying the errors arising from different steps in the model chain in explaining the observed changes in hydrologic variables (streamflow and groundwater). The proposed metrics are then evaluated using a detailed retrospective analyses in projecting the changes in streamflow and groundwater attributes in four target basins that span across a diverse hydroclimatic regimes over the U.S. Sunbelt. Our analyses focused on quantifying the dominant sources of errors in projecting the changes in eight hydrologic variables-mean and variability of seasonal streamflow, mean and variability of 3 day peak seasonal streamflow, mean and variability of 7 day low seasonal streamflow, and mean and standard deviation of groundwater depth-over four target basins using an Penn state Integrated Hydrologic Model (PIHM) between the period 1956-1980 and 1981-2005. Retrospective analyses show that small/humid (large/arid) basins show increased (reduced) uncertainty in projecting the changes in hydrologic attributes. Further, changes in error due to GCMs primarily account for the unexplained changes in mean and variability of seasonal streamflow. On the other hand, the changes in error due to temporal disaggregation and hydrologic model account for the inability to explain the observed changes in mean and variability of seasonal extremes. Thus, the proposed metrics provide insights on how the error in explaining the observed changes being propagated through the model under different hydroclimatic regimes.