Improving Hydrologic Predictions of Distributed Watershed Model via Uncertainty Quantification of Evapotranspiration Methods

Improving Hydrologic Predictions of Distributed Watershed Model via Uncertainty Quantification of Evapotranspiration Methods

Successful initialization and accurate estimation of evapotranspiration (ET) in the coastal plain landscapes are crucial for the prediction of hydrologic variables including streamflow, surficial aquifer lost and infiltration. The aim of this study is to examine the ability of Soil and Water Assessment Tool (SWAT) to accurately represent the characterization of three potential ET methods (Priestley-Taylor (P-T), Penman–Monteith (P-M) and Hargreaves (HG)) using the Sequential Uncertainty FItting (SUFI-2) algorithm during 2003-2005 and 2006-2007 as calibration and validation intervals. The study area was the Waccamaw River watershed, a low-gradient coastal plain watershed in the southeastern US. The results indicated that in estimating ET for a coastal plain landscape, P-T method bracketed more than 75% of daily streamflow during calibration period while both P-M and HG bracketed 57% and 69% of measured streamflow during calibration period, respectively. Model daily performance using P-T method was “very good” (calibration NSE = 0.77; validation NSE=0.90) but only “satisfactory” (P-M calibration NSE = 0.55; HG calibration NSE =0.61) to “good” (P-M validation NSE=0.75; HG validation NSE=0.70) in P-M and HG methods. The prediction mean square error (MSE) for P-T method was comparably low (57.88 and 325.68) compared to P-M (68.34 and 635.95) and HG (69.99 and 551.99) methods at upstream and downstream outlets, respectively. This result suggests that radiation based ET method performed significant results in forested wetland dominated ecosystem with wet and humid surfaces. Based on the water balance analysis, only about 21.2% of flow loss was consumed via stream evaporation and floodplains evapotranspiration, indicating that 78.8% of the loss within the entire study area represented land ET and shallow aquifer recharge. Furthermore, uncertainty quantification revealed that low flows are sensitive to the changes in ET process in dry period and at the beginning of the wet season, but insensitive at the end of the wet season due to nonlinear control of coastal plain soil on water movement. In particular, under conditions of so-called “deep uncertainty” in the coastal plain landscapes, uncertainty quantification of ET methods can lead to the identification of optimal land and water management strategies in the southeastern ecosystems.