Simulated errors in deep drainage beneath irrigated settings: Partitioning vegetation, texture and irrigation effects using Monte Carlo

Wednesday, 17 December 2014
Justin Philip Gibson1, John B Gates2 and Paolo Nasta1, (1)University of Nebraska Lincoln, Lincoln, NE, United States, (2)Univ Nebraska Lincoln-Geoscien, Lincoln, NE, United States
Groundwater in irrigated regions is impacted by timing and rates of deep drainage. Because field monitoring of deep drainage is often cost prohibitive, numerical soil water models are frequently the main method of estimation. Unfortunately, few studies have quantified the relative importance of likely error sources. In this study, three potential error sources are considered within a Monte Carlo framework: water retention parameters, rooting depth, and irrigation practice. Error distributions for water retention parameters were determined by 1) laboratory hydraulic measurements and 2) pedotransfer functions. Error distributions for rooting depth were developed from literature values. Three irrigation scheduling regimes were considered: one representing pre-scheduled irrigation ignoring preceding rainfall, one representing pre-scheduled irrigation that was altered based on preceding rainfall, and one representing algorithmic irrigation scheduling informed by profile matric potential sensors. This approach was applied to an experimental site in Nebraska with silt loam soils and irrigated corn for 2002-2012. Results are based on Six Monte-Carlo simulations, each consisting of 1000 Hydrus 1D simulations at daily timesteps, facilitated by parallelization on a 12-node computing cluster.

Results indicate greater sensitivity to irrigation regime than to hydraulic or vegetation parameters (median values for prescheduled irrigation, prescheduled irrigation altered by rainfall, and algorithmic irrigation were 310 ,100, and 110 mm/yr, respectively). Error ranges were up to 700% higher for pedotransfer functions than for laboratory-measured hydraulic functions. Deep drainage was negatively correlated with alpha and maximum root zone depth and, for some scenarios, positively correlated with n. The relative importance of error sources differed amongst the irrigation scenarios because of nonlinearities amongst parameter values, profile wetness, and deep drainage. Compared to pre-scheduled irrigation, the algorithmic irrigation resulted in a drier profile, higher sensitivity to rooting depth, and lower sensitivity to hydraulic parameters. Results highlight that knowledge of irrigation rates and timing are at least as important as biophysical measurements for constraining recharge.