NAWQA Cycle 3 Groundwater Studies in the Glacial Aquifer Paul Juckem, USGS-WiWSC Contributions from: Sandy Eberts, Ken Belitz, Daniel Feinstein, Mike Fienen, Leon Kauffman, Jim Tesoriero, Chris Shope, Jeff Starn, Dick Yager
NAWQA-Cycle 3: Groundwater Goals In Cycle 3, NAWQA groundwater assessment activities will focus on the quality of groundwater resources available for public and domestic drinking supply, and on the contribution of contaminants from groundwater to streams. In other words: 1. Groundwater supply to humans Cycle 2 = well vulnerability Cycle 3 = aquifer vulnerability Vulnerability = f (contaminant input, contaminant mobility & persistence, intrinsic susceptibility) Intrinsic Susceptibility = ease of transport through UZ and aquifer (R, T, n, I, Q) 2. Groundwater supply to ecosystems Cycle 2 = local nutrient transport and geochemical processes Cycle 3 = integration of groundwater processes into regional forecasting tools (SPARROW)
Cycle 2: Measures of Vulnerability at the Well Scale Useful Measures for the Individual Well Scale: Sources of Recharge A measure of contaminant input Geochemical Conditions (redox, ph, alkalinity) A measure of contaminant mobility and persistence Groundwater-Age Mixture A measure of intrinsic susceptibility Individual Well Scale: Age of water in sample from well National Scale: Soil permeability
The Glacial Principal Aquifer How can Intrinsic Susceptibility (e.g., age distributions) be mapped at this scale? Glacial aquifers
Up-scale & Extrapolate Goal: Up-scale & extrapolate intrinsic susceptibility (GW age) at depths of domestic and supply wells in the glacial aquifer 1. Evaluate improved predictive power with increased model complexity & scale a. How complex do our flow and transport (CBPT) models need to be? b. What range of variability is needed for extrapolating understanding to other areas? 2. Develop tools for estimating groundwater age across scales w/o flow models a. Idealized equations (GWstrat) b. Metamodels (hybrid process & statistical models; e.g., Bayesian Networks)
Model Complexity and Scale Soller, Packard, & Garrity, 2012
Up-scale & Extrapolate Goal: Up-scale & extrapolate intrinsic susceptibility (GW age) at depths of domestic and supply wells in the glacial aquifer 1. Evaluate improved predictive power with increased model complexity & scale a. How complex do our flow and transport (CBPT) models need to be? b. What range of variability is needed for extrapolating understanding to other areas? 2. Develop tools for estimating groundwater age across scales w/o flow models a. Idealized equations (GWstrat) b. Metamodels (hybrid process & statistical models; e.g., Bayesian Networks)
Idealized Equation Methods Cook, P.G., and Böhlke, J.K., 2000, Determining timescales for groundwater flow and solute transport, chap. 1 in Cook, P.G., and Herczeg, A.L., eds., Environmental tracers in subsurface hydrology: Boston, Mass., Kluwer Academic Publishers, p. 1-30. GWstrat LPM
GW divide Large river GWstrat Conceptualization Groundwater recharge z H Groundwater age increases exponentially with depth Raster cells Mean age t = (H*ε / R) * ln(h / (H z)) Limitations: 1. Dupuit-Forchheimer flow (not valid for 3D flow near pumping centers or streams) 2. Assumes change (H, ε, R) between adjacent raster cells is modest & not systematic 3. Assumes 1-layer aquifer, no weak sinks
Groundwater Stratigraphy Toolbox for ArcGIS GWstrat Groundwater age influences many water quality factors (redox, anthropogenic vs. natural sources, etc.) Can we map groundwater age patterns w/o flow models?
Up-scale & Extrapolate Goal: Up-scale & extrapolate intrinsic susceptibility (GW age) at depths of domestic and supply wells in the glacial aquifer 1. Evaluate improved predictive power with increased model complexity & scale a. How complex do our flow and transport (CBPT) models need to be? b. What range of variability is needed for extrapolating understanding to other areas? 2. Develop tools for estimating groundwater age across scales w/o flow models a. Idealized equations (GWstrat) b. Metamodels (hybrid process & statistical models; e.g., Bayesian Networks)
Young-fraction Bayesian Network 1. Grab slides from DTF and MNF <50 yrs <40 yrs <30 yrs, etc.
Fox-Wolf- Peshtigo Model Multiple glacial categories Large range in transmissivity Large range in recharge Much water quality and age tracer data
Fox-Wolf- Peshtigo Model Multiple glacial categories Large range in lithology Large range in recharge Much water quality and age tracer data
Fox-Wolf- Peshtigo Model Multiple glacial categories Large range in lithology Large range in recharge Much water quality and age tracer data Lake Michigan Basin model Feinstein and others,(2010)
Training Model Transfer Function Sample many areas of the semi-structured glacial Lake Michigan Basin model in order to quantify the fraction of young water captured by wells Build transfer function that links model variables to fraction of young water for shallow wells. Mappable candidate variables: Pumping rate (differ by seed well) Local recharge rate & variance (SWB) Local thickness & variance Coarse fraction & variance Glacial category (e.g., Fullerton ) Cross-validate over the training data to estimate predictive power Test transfer function with Fox-Wolf-Peshtigo model (calibrated to age tracers) Test predictive power against estimated ages from well samples
How Will NAWQA s Effort Differ from Prior Efforts? Scale Developed from sampled data Measured concentrations, redox conditions, age distributions Informed by hydrogeologic & geochemical processes (flow & transport, recharge estimates, redox/denitrification) Natural & anthropogenic contaminants Results expected to include probability and uncertainty
Concentration, as percent of input Effect of Age Mixtures on Contaminant Concentrations 100 < 15 yrs 60 1 to 125 yrs 20 100 60 20 Water table Well A lack of young water can cause concentrations to continue to go up long after source reduction 10 to 1,000s of yrs 10s and 1,000s of yrs 100 60 20 100 60 20 A wide range of groundwater ages provides some protection against high levels of nonpoint-source contamination Time, in years 0 50 100
GWRP Glacial Dataset Products Glacial Category Fullerton & Soller and/or RASA products Only surficial maps of broad categories (till, outwash, lake deposits, organics) Recharge from Soil Water Balance code Applied to conterminous USA by FY15 Gridded DAYMET, but soil and landuse pose challenges Calibrate to baseflow from GAGEII? MODPATHu Coarse Fraction & Glacial Thickness Well construction logs available for 17 states done by FY14 Grid resolution not settled; likely 1 km or larger Pumping Rate: Scale of the stress National Water Use compilations by county Not tied to source aquifer LMB-USG refinement limited to layer 1
SPARROW Water-Quality Model: SPAtially Referenced Regression on Watershed Attributes http://water.usgs.gov/nawqa/sparrow Monitoring Data Annual Loads Y variable Steam Network Land Use Sources Base Year Fertilizers Corn Manure Wheat Point Sources Atmospheric Dep. Mass Balance Model with spatially variable deliveries. Hybrid statistical/ mechanistic process structure. Data-driven, nonlinear estimation of parameters Separates land and in-stream processes Predictions of mean-annual flux reflect long-term, net effects of nutrient supply and loss processes in watersheds X variables Once calibrated, the model has physically interpretable coefficients; model supports hypothesis testing and uncertainty estimation
Reconstructing Nitrate Concentrations in Recharge -Use age dating (CFCs, SF 6 ) to determine recharge date. -Measure conc. of nitrate and N 2 from denitrification. -Recharge Nitrate Conc. = [NO 3- ]+[N 2 from denitrif.] 14 NO 3 conc. in recharge 2000 1990 1980 1970 Nitrate (mg/l as N) 12 10 8 6 4 2 0-2 1960 1970 1980 1990 2000
Does the Recharge NO 3 Trend Scale Up? -FSS scaled to LUS Study Area -Evaluate larger study area Recharge Nitrate Concentrations =[NO 3- ]+[Excess N 2 ] Nitrate (mg/l as N) 30 25 20 15 10 5 0 Wisconsin 20% of N application 1950 1960 1960 1970 1970 1980 1980 1990 1990 2000 2000 Landuse Study Area Flow System Study Saad, JEQ, 2008; Tesoriero et al., JCH, 2007
Similar Nitrate Trend in GW Discharging to Streams Can we forecast in-stream NO 3 concentrations at the watershed scale? Nitrate (mg/l as N) 14 12 10 8 6 4 2 Streambed Samples 2009 Streambed Samples 2019? Stream in 2009 Stream in 2019? 0 1960 1970 1980 1990 2000 Nitrate Conc. in Stream in 2009: 3 mg./l Nitrate Conc. in Stream in 2019: 5 mg./l?
Spatially Variable Denitrification Efficiency in Streambed Samples Flow model provides the framework for evaluating trends in NO 3 discharge to streams in a spatial context NEET Streambed samples throughout the TWR watershed with similar age show more denitrification than Tomorrow River NEET data NEET Is there a mappable feature(s) that may explain this? Graph from Guldan Thesis, UW- Stevens Point
How to Compare Tools Across Scales? Compare residuals among tools and scales evaluate spatially TWR-scale MODFLOW GWstrat Ave error 0.4 yr 0.8 yr Standard deviation 5.2 yr 9.1 yr (numbers are fabricated) Fox-Wolf scale MODFLOW GWstrat Ave error 1.4 yr 3.8 yr Standard deviation 8.2 yr 14.1 yr (numbers are fabricated) How to interpret reductions in accuracy/precision with model simplification across scales? How does this information feed into Metamodels or statistical models?