01.10.2013 Views

HGS van Genuchten - FEFlow

HGS van Genuchten - FEFlow

HGS van Genuchten - FEFlow

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

2 nd<br />

Well Vulnerability:<br />

A Model Comparison<br />

–<br />

Issues and Pitfalls<br />

E.O. Frind, M. Sousa, J.P.<br />

Jones, and D.L. Rudolph,<br />

University of Waterloo<br />

International FEFLOW User Conference<br />

September 14-18, 2009<br />

Potsdam/Berlin, Germany


Wellhead<br />

protection,<br />

standard<br />

approach<br />

Apply particle<br />

tracking,<br />

delineate<br />

capture zone<br />

Possible<br />

sources?


The Real<br />

World:<br />

Cape Cod<br />

example<br />

(USGS Circular<br />

1174, 1998)<br />

A: K=50 ft/d<br />

B: K=150 ft/d<br />

Small change<br />

in head can<br />

cause large<br />

change in<br />

particle tracks


•<br />

•<br />

•<br />

•<br />

•<br />

Key Issues<br />

Sensitivity: Small changes in heads can cause large<br />

changes in capture zone<br />

Uncertainty:<br />

–<br />

–<br />

Effect of heterogeneities<br />

Stochastic approaches (mostly 2D)<br />

Physical approach: represent small-scale heterogeneities by<br />

macrodispersion (Gelhar and Axness 1983); large-scale by scenario<br />

analysis (always 3D)<br />

Determinism: Meaning of a source just inside vs. a source<br />

just outside of capture zone boundary<br />

Impact:<br />

–<br />

How critical will a contaminant source be?<br />

Well Vulnerability Concept gives actual impact on well<br />

Model performance:<br />

against each other?<br />

How do commercial models stack up


Illustrative Example: Mannheim Well Field<br />

Grand River<br />

Watershed<br />

Chicago<br />

Ontario<br />

Detroit<br />

Cleveland<br />

Toronto<br />

Grand River<br />

Conservation Authority<br />

Dundalk<br />

Waterloo<br />

Kitchener<br />

Kitchener<br />

Waterloo<br />

50 km<br />

Guelph<br />

Cambridge<br />

Brantford<br />

• 6800 km2 drainage area<br />

• 10% of Lake Erie<br />

Local Drainage<br />

• Population = 800,000<br />

• 38 local/regional<br />

Grand River<br />

Watershed<br />

governments<br />

• Major reservoirs<br />

regulate 27% of<br />

watershed area<br />

•82% of watershed residents<br />

reliant on groundwater for<br />

their drinking water supply.<br />

Waterloo Moraine model<br />

Mannheim<br />

well field<br />

Dunnville


Waterloo Moraine Model<br />

N<br />

Paul Martin,<br />

1995<br />

Mannheim<br />

well field<br />

Alder<br />

Creek<br />

watershed


Uncertainty due to Local Heterogeneities<br />

N<br />

Total Particles 1760<br />

Reaching Limit 17.95 %<br />

Reaching Top 82.05 %<br />

Reaching Side 0.00 %<br />

Well<br />

K22<br />

Mannheim North<br />

K23<br />

K93<br />

K24<br />

K94<br />

K91<br />

K92<br />

K26<br />

K21<br />

K25<br />

K29<br />

Mannheim South<br />

2km<br />

Mannheim Well Field,<br />

100-yr Capture Zone<br />

by backward particle<br />

tracking (older model)<br />

Can draw envelope<br />

curve around particle<br />

end positions<br />

But first consider<br />

uncertainty due to<br />

local heterogeneities


Uncertainty due to Local Heterogeneities<br />

N<br />

0.1<br />

0.01<br />

Well<br />

0.3<br />

0.5<br />

0.9<br />

0.7<br />

K22<br />

Capture probability<br />

0.01 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9<br />

K93<br />

K23<br />

K24<br />

K94<br />

K92<br />

K91<br />

K26<br />

K21<br />

K29<br />

K25<br />

2km<br />

Transport parameters:<br />

α L = 20 m, α Th = 5 m, α TV = 0.02 m, D m = 10 -10<br />

(Frind et al., 2002)<br />

100-yr Probability-of<br />

Capture plume by<br />

backward advectivedispersive<br />

transport<br />

modelling<br />

Obtain capture zone<br />

outline by mass balance<br />

(not necessarily 0.5<br />

contour)<br />

m 2 /s


Particle Tracking vs Backward Transport Model<br />

N<br />

100-yr<br />

Total Particles 1760<br />

Reaching Limit 17.95 %<br />

Reaching Top 82.05 %<br />

Reaching Side 0.00 %<br />

Well<br />

K22<br />

Mannheim North<br />

K94 K21<br />

K93<br />

K92<br />

K29<br />

K91<br />

K25<br />

K23<br />

Mannheim South<br />

K24<br />

K26<br />

2km<br />

Capture zone from<br />

transport model is<br />

larger due to<br />

dispersion<br />

- accounts for<br />

uncertainty due to<br />

local scale<br />

heterogeneity<br />

- more conservative<br />

than particle tracking


N<br />

100-yr<br />

Source<br />

Well<br />

The Well Vulnerability Concept<br />

K22<br />

K23<br />

K94 K21<br />

K93 K92<br />

K91<br />

K29<br />

K25<br />

K24<br />

K26<br />

2km<br />

(Frind et al., 2006)<br />

Known source:<br />

- Apply pulse at source<br />

- Record breakthrough at<br />

well (forward method)<br />

Unknown source:<br />

- Apply pulse at well<br />

- Record breakthrough at<br />

any point within capture<br />

zone (backward method)<br />

Forward and backward<br />

breakthrough curves will be<br />

the same (Adjoint theory)


Concentration (gm -3 )<br />

Well Vulnerability: Forward Mode<br />

5.0E-04<br />

4.0E-04<br />

3.0E-04<br />

2.0E-04<br />

1.0E-04<br />

C<br />

dws<br />

C<br />

peak<br />

0.0E+00<br />

0<br />

Texceed<br />

50<br />

Tpeak<br />

100 150<br />

Time (days)<br />

200<br />

T<br />

expo<br />

Breakthrough curve<br />

at well due to pulse<br />

source within<br />

capture zone:<br />

• Max concentration<br />

expected at well<br />

• Time to reach max<br />

concentration<br />

• Time to breach<br />

drinking water limit<br />

• Time of exposure


•<br />

•<br />

FEFLOW<br />

–<br />

–<br />

Model Comparison:<br />

Apply to Prediction of Well<br />

Galerkin<br />

Vulnerability<br />

Finite Elements<br />

requires specified recharge function<br />

HydroGeoSphere (<strong>HGS</strong>) (Therrien<br />

–<br />

–<br />

–<br />

Control Volume Finite Elements<br />

and Sudicky, 2003)<br />

integrated groundwater / surface water model<br />

distributes recharge by means of diffusion-wave<br />

equation


•<br />

•<br />

•<br />

Model Functions Tested<br />

Unsaturated flow/transport<br />

–<br />

1D column<br />

3D homogeneous system<br />

–<br />

box model, average Mannheim aquifer properties<br />

3D heterogeneous system<br />

–<br />

actual Mannheim aquifer, Alder Creek model


Recharge 200 mm/yr<br />

Flux concentration 1 mg/L for 20 years<br />

1 m 2<br />

Water<br />

Table<br />

Unsaturated Flow/Transport<br />

40m<br />

Silty Sand:<br />

K = 5 x 10-4 m/s<br />

Porosity = 0.37<br />

Residual Saturation = 0.049<br />

Van <strong>Genuchten</strong> parameters:<br />

A = 3.475, n = 1.746<br />

Molecular diffusion = 1.2 x 10-9 m2 /s<br />

Longitudinal dispersivity = 10 m<br />

Transverse dispersivity = 1 m<br />

Vertical discretization 20m


Van <strong>Genuchten</strong> vs Modified <strong>van</strong> <strong>Genuchten</strong> Model<br />

δ = 5.4 δ = 4.0 δ = 1.1<br />

VG<br />

Improving<br />

convergence<br />

MVG<br />

δ


Effect of <strong>van</strong> <strong>Genuchten</strong> δ<br />

Changing δ from 5.4 to 1.1 will change the unsaturated<br />

hydraulic conductivity by up to 7 orders of magnitude<br />

This should have a big impact on travel time?<br />

Not necessarily!<br />

With a specified recharge boundary condition, the same flux<br />

will enter the column, regardless of the K value<br />

–<br />

–<br />

changing K will simply change the gradient<br />

travel time and breakthrough curve will be the same


10m<br />

20m<br />

Breakthrough curve at water table<br />

1m<br />

FEFLOW<br />

5m Modified <strong>van</strong> <strong>Genuchten</strong><br />

Effect of vertical<br />

discretization<br />

Solution converges for<br />

Δz = 1m


Breakthrough curve at water table<br />

1m<br />

<strong>HGS</strong><br />

5m Modified <strong>van</strong> <strong>Genuchten</strong><br />

10m<br />

20m<br />

Effect of vertical<br />

discretization<br />

Solution converges for<br />

Δz = 1m


10m<br />

20m<br />

1m<br />

5m<br />

FEFLOW vs. <strong>HGS</strong><br />

Modified <strong>van</strong><br />

<strong>Genuchten</strong><br />

FEFLOW<br />

<strong>HGS</strong>


1m<br />

5m<br />

10m<br />

20m<br />

<strong>HGS</strong><br />

<strong>van</strong> <strong>Genuchten</strong><br />

<strong>HGS</strong> Solution<br />

converges for<br />

Δz = 1m<br />

Feflow requires<br />

Δz = 0.005m for<br />

convergence<br />

(not practical)


<strong>HGS</strong><br />

Van <strong>Genuchten</strong> vs<br />

Modified <strong>van</strong> <strong>Genuchten</strong><br />

Mod. <strong>van</strong> <strong>Genuchten</strong><br />

Van <strong>Genuchten</strong>


<strong>HGS</strong><br />

Van <strong>Genuchten</strong> vs<br />

Modified <strong>van</strong> <strong>Genuchten</strong><br />

Effect of unsat zone thickness<br />

Van <strong>Genuchten</strong><br />

40m<br />

5m<br />

Mod. Van <strong>Genuchten</strong><br />

40m<br />

5m


Findings: Unsaturated Flow/Transport<br />

•<br />

•<br />

•<br />

•<br />

•<br />

Feflow works best with MVG<br />

<strong>HGS</strong> works with both options<br />

Difference between VG and MVG declines with<br />

unsat zone thickness<br />

δ<br />

used with MVG does not affect results if flux<br />

boundary is chosen<br />

Overall, both models give equivalent results


3D Homogeneous System:<br />

Box Model representing average<br />

conditions for Mannheim aquifer<br />

4km<br />

Pumping Well<br />

500 m 3 /day<br />

1km<br />

Top BC<br />

Recharge 200 mm/yr<br />

400m<br />

25m<br />

Default BC No-Flow<br />

Side:<br />

Constant Head 20m


Finite Element Model<br />

Horizontal grid spacing: 15m


Source Conditions<br />

Source centered within capture zone<br />

Apply 1yr-pulse of specified recharge concentration of 1 g/m 3<br />

Source outside capture zone (OCZ)<br />

Apply 1yr-pulse of specified recharge concentration of 1 g/m 3<br />

Well<br />

Well


Inside Source<br />

FEFLOW vs <strong>HGS</strong>,<br />

Upwind vs No-Upwind<br />

Upstream weighting<br />

Feflow no-upwind option<br />

provides best accuracy<br />

Feflow full-upwind and<br />

<strong>HGS</strong> upstream weighting<br />

are equivalent


Outside Source FEFLOW vs <strong>HGS</strong>,<br />

Upwind vs No-Upwind<br />

Upstream weighting<br />

Mass spread to outside<br />

source by transverse<br />

dispersion: upwind<br />

options spread more<br />

mass


FEFLOW, Plume at 8 years<br />

No upwind<br />

Some oscillations and<br />

negative concentrations


FEFLOW, Plume at 8 years<br />

Full upwind<br />

Negligible oscillations but<br />

significant smearing


•<br />

•<br />

•<br />

•<br />

•<br />

Findings: Homogeneous System<br />

Feflow no-upwind option gives the best accuracy<br />

Feflow full-upwind and <strong>HGS</strong> upstream weighting<br />

give essentially the same results<br />

Upwinding reduces oscillations, but increases<br />

smearing<br />

Transverse dispersion transfers mass to outside<br />

source, upwinding increases the transfer<br />

Overall, model results are comparable


3D Heterogeneous System<br />

N<br />

Waterloo<br />

Moraine<br />

model<br />

Mannheim<br />

well field<br />

Alder<br />

Creek<br />

watershed


•<br />

•<br />

•<br />

Setup:<br />

–<br />

–<br />

–<br />

3D Heterogeneous System:<br />

Alder Creek Model<br />

recharge distribution generated by <strong>HGS</strong>, transferred<br />

to Feflow<br />

MVG for Feflow and <strong>HGS</strong><br />

dispersive form for Feflow (mass flux specified)<br />

Upwinding:<br />

–<br />

–<br />

Feflow: full upwinding (no-upwind required very small<br />

time step (0.1 day)<br />

<strong>HGS</strong>: upstream weighting<br />

Max time step:<br />

–<br />

–<br />

10 days: 3 weeks execution time<br />

100 days: 6 days execution time, same results


Alder Creek Watershed: Ground Surface Elevations


Alder Creek Watershed: Municipal Wells


Alder Creek Watershed: Recharge Distribution<br />

Recharge distribution generated by <strong>HGS</strong> from DEM


87 layers (incl unsat zone)<br />

7216 elements/layer<br />

1204428 elements total<br />

3D Finite<br />

Element Grid<br />

J.P. Jones, 2004<br />

Study<br />

Area<br />

Elevation masl<br />

Area 78 km 2


Alder Creek Watershed: Finite Element Grid<br />

Study<br />

Area


Sources for Vulnerability Study<br />

2 nd<br />

source<br />

1 st<br />

source<br />

K22A<br />

K23<br />

K24<br />

2 nd<br />

source:<br />

14m unsat<br />

zone<br />

1st source:<br />

12m unsat<br />

zone


2 nd<br />

Recharge Distribution<br />

source<br />

1 st<br />

source<br />

K22A<br />

K23<br />

K24<br />

Recharge distribution transferred from <strong>HGS</strong> to Feflow


Mannheim Well Field: Particle Tracks<br />

N<br />

Total Particles 1760<br />

Reaching Limit 27.44 %<br />

Reaching Top 65.91 %<br />

Reaching Side 6.65 %<br />

Well<br />

K22<br />

Mannheim North<br />

K94 K21<br />

K93<br />

K92<br />

K29<br />

K91<br />

K25<br />

K23<br />

K24<br />

Mannheim South<br />

K26<br />

2km<br />

From older<br />

model,<br />

uniform<br />

recharge<br />

100-year<br />

particle tracks<br />

2 nd<br />

1 st<br />

source<br />

source


30 y<br />

K24<br />

K24<br />

Other<br />

wells<br />

FEFLOW<br />

<strong>HGS</strong><br />

1 st<br />

Source<br />

Mass at<br />

300 yrs Feflow <strong>HGS</strong><br />

input (g) 9788 100% 9828 100%<br />

K22A 4 0% 10 0%<br />

K23 70 1% 382 4%<br />

K24 1142 12% 1552 16%<br />

K26 131 1% 281 3%<br />

First arrival of critical conc ?<br />

Arrival of max conc: ~30 years<br />

Max concentration ?<br />

Exposure time ?


60 y<br />

K22A<br />

K23<br />

FEFLOW<br />

<strong>HGS</strong><br />

Other<br />

wells<br />

2 nd<br />

Source<br />

Mass at<br />

400 yrs Feflow <strong>HGS</strong><br />

input (g) 8597 100% 8327 100%<br />

K22A 1697 20% 1396 17%<br />

K23 2823 33% 4814 58%<br />

K24 550 6% 289 3%<br />

K26 404 5% 175 2%<br />

First arrival of critical conc ?<br />

Arrival of max conc: ~60 years<br />

Max concentration ?<br />

Exposure time ?


1 st<br />

Source, Plume at 30 years<br />

K22A<br />

K23<br />

K24<br />

Feflow,<br />

full<br />

upwind<br />

Plume<br />

moves<br />

south: -<br />

this<br />

explains<br />

mass<br />

loss


2 nd<br />

Source, Plume at 75 years<br />

K22A<br />

K23<br />

K24<br />

Feflow,<br />

full<br />

upwind<br />

Plume<br />

moves<br />

east and<br />

south


•<br />

•<br />

•<br />

•<br />

Why the smearing in Feflow?<br />

Effect of heterogeneities?<br />

–<br />

smearing increases with Feflow when going to<br />

heterogeneous system, less so with <strong>HGS</strong><br />

Upwinding?<br />

–<br />

both models use upwinding<br />

Numerical approach?<br />

–<br />

–<br />

Feflow uses Galerkin FE (globally mass conservative)<br />

<strong>HGS</strong> uses Control Volume FE (locally and globally<br />

mass conservative)<br />

Increasing smearing with distance:<br />

–<br />

2nd source is twice as far away as 1st , so numerical<br />

dispersion increases


•<br />

•<br />

•<br />

•<br />

•<br />

Findings: 3D Heterogeneous Model<br />

Recharge distribution has controlling influence on flow<br />

field<br />

Placement of sources within capture zone does not<br />

guarantee capture of input mass by wells<br />

Upwinding seems to be a major factor in transverse<br />

spreading of plume<br />

Accuracy of well vulnerability measures is questionable<br />

Max concentrations at well are all much lower than<br />

source concentrations – dispersion plays a major role


•<br />

•<br />

•<br />

•<br />

•<br />

Reflections on Key Issues<br />

Sensitivity: Recharge distribution has a major impact on<br />

flow field and vulnerability functions<br />

Uncertainty: Macrodispersion is useful, but other<br />

sources of uncertainty may be controlling<br />

Determinism: Sources both inside and outside of<br />

capture zone can contribute to well contamination<br />

Impact: Source location inside capture zone does not<br />

mean that contamination will be critical<br />

Model performance:<br />

different results<br />

Different models can give very


•<br />

•<br />

•<br />

•<br />

•<br />

•<br />

•<br />

•<br />

•<br />

What does it all mean?<br />

Be aware of uncertainty and sensitivity in wellhead<br />

protection work<br />

Use the most complete data available<br />

Use realistic recharge distributions and boundary<br />

conditions<br />

Use the best models available<br />

Select model settings carefully<br />

Consider scenario analyses (windows etc?)<br />

Never take model results at face value<br />

Remember that particle tracks only tell half the story<br />

Use sufficient margins of safety in wellhead protection


Thanks, that’s it for now<br />

We have come a long<br />

way in this journey<br />

Acknowledgements:<br />

Hans-Joerg Diersch<br />

Peter Schaetzl

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!