Groundwater Modeling for the Management of Agricultural Impacts on Surface Water Quality

While significant efforts have been made to reduce agricultural nitrogen export to the Chesapeake Bay, loads from many agriculturally-intensive areas continue to increase due to the discharge of legacy nitrates from surficial aquifers beneath agricultural systems. In order to improve understanding of the relationship between land surface management improvements and water quality response, the Chesapeake Bay Program (CBP) has endorsed the use of targeted small watershed studies as a means of better understanding the impact of agricultural best management practices (BMPs) on surface water quality. This dissertation supports the CBP objectives for small watershed studies by characterizing groundwater flow and nitrate transport and removal in one of the targeted watersheds. The dissertation uses an agriculturally-intensive catchment on the Maryland Eastern Shore to examine a series of questions related to understanding and managing nitrogen in agricultural systems, with particular focus on (i) modeling nitrogen transport subject to groundwater lag times and (ii) calibrating the simulation tools used for that modeling. The collective purpose of the four studies included in the dissertation is to improve our ability to calibrate and use numerical groundwater simulation tools, with the aim of better modeling the impact of agricultural intensification and de-intensification on water quality, and thus support management by improving our ability to interpret signs of decline and improvement in receiving streams.
The first study compares the range and character of different catchment-scale simulation models that result when an automated calibration routine is driven by various combinations of spatially- and temporally-distributed CFC, SF6, 3H, and 3He information. While researchers commonly use groundwater-age information to calibrate subsurface flow and transport models, rarely is there a rich enough age-tracer dataset to investigate the question of what number, spatial distribution, and measurement uncertainty of calibration targets are needed in order to adequately characterize a site. In this investigation 238 environmental-tracer measurements were used individually and in combination to estimate recharge rates, hydraulic conductivities, and effective porosity for a three-dimensional groundwater flow and transport model of our study site. The various combinations of spatially- and temporally-distributed CFC, SF6, and 3H information used to drive the automated inverse modeling routine resulted in a range of catchment-scale simulation models and associated parameter uncertainty bounds. The study demonstrates that while tracer data can provide necessary supplemental information for the calibration of flow and transport models, the use of data from a single tracer or from a small tracer set may be insufficient to fully interpret the information content of the tracers.
In the second study we use the calibrated groundwater flow and transport model to resolve the key components of the nitrogen budget for the targeted watershed. While subsurface nitrate transport and catchment removal processes have been widely investigated, there have been few fully distributed, three-dimensional modeling studies of nitrate transport and removal in catchments with nitrogen removal rates that are highly spatially-variable, as is the case with our study site. We link the re-constructed time-variable land surface loadings to time-variable stream responses in two subcatchments that have similar land use histories but highly disparate nitrate export signatures, and we estimate the impact of soil denitrification and in-stream nitrogen removal as well as the potential influence of retarded nitrate transport. We show that in spite of spatial and temporal uncertainty in loading, multiple calibration scenarios agree that in-stream nitrate removal efficiencies vary significantly between the two sub-catchments, with one stream removing 60-70% of incoming nitrogen loads and the contrasting stream removing only 15-30%.
While we use a steady state representation of the flow system in the descriptions of environmental tracer and nitrate transport in the first two studies, in some situations variability in base-flow age may impact in-stream solute concentrations. In the third study, we examine the impact of time-variable hydrologic forcing – such as that due to seasonal changes in precipitation and evapotranspiration – on the age of base-flow discharge. We develop a method for simulating the transient delivery of base-flow age from subsurface to receiving stream as a function of seasonal changes in hydrology and aquifer storage, and we apply the method to a variety of synthetic two-dimensional (2D) aquifers as well as to the study site. We found that the timing of maximum base-flow age relative to the timing of minimum base-flow discharge varied with both the hydraulic conductivity field and the annually averaged recharge, which determines the system mean age. The two assumptions of (i) an aquifer in which ages are vertically well-mixed and (ii) an aquifer in which ages are strongly stratified provide two end-members for estimation of how the base-flow age might respond to seasonal changes in recharge and base-flow, and the simulations in this study found that the change that occurs in real systems is somewhere in between. For the cases that we investigated, apparent ages inferred from SF6 measurements while assuming piston-flow transport assumptions for the SF6 were biased young, with biases especially pronounced with layered hydrogeology in which discharge consists of shallow surficial flow mixed with a contrasting regime of much older water. For one of the subcatchments in our Maryland study site we found that seasonal changes in recharge may only result in changes in base-flow age of 3 to 4 years, but that SF6 apparent ages based on piston-flow assumptions of tracer transport may underestimate the mean base-flow age by 60% and more closely resemble the median system age.
In the fourth and final study, we consider the impact of assumptions about age transport on the automated calibration of groundwater flow and transport models. As illustrated in the calibration of the flow and transport model used throughout this dissertation, parameters such as porosity and dispersivity must often be estimated through model calibration against data describing groundwater age. In such cases, the groundwater age observed at a point in the subsurface is often assumed to be a function of purely hydraulic processes. In this study we use the automated calibration of several synthetic aquifers to investigate the impact of that assumption on the resulting calibrated model under a variety of heterogeneity and dispersivity scenarios. We also consider the impact of applying advective-dispersive methods to the same range of scenarios. We show that as true system dispersivity increases, the capacity of kinematic simulations of age to translate the available system information into accurate parameter estimates decreases.