01 K Variability
Why does a contaminant plume spread far more than simple theory predicts?
What you’re watching Four runs of the same plume stacked top to bottom, each in an aquifer with greater variability in hydraulic conductivity (ln K variance 0, 0.5, 1.0, 3.0). The top aquifer is uniform and the magenta plume stays compact; as variability grows downward, the plume stretches, fingers, and channels through the colored conductivity field.
The mechanism Variability in conductivity makes water race through high-K zones and stall in low-K ones. That spread of velocities pulls the plume apart into fingers — an apparent spreading (‘macrodispersion’) that grows with the variance of ln K and dwarfs pore-scale mixing.
Why it matters This is why real plumes arrive sooner, reach farther, and resist containment more than homogeneous models predict — a core reason groundwater cleanups overrun their schedules and budgets.
IGW-NET You set a variance; IGW-NET generates the conductivity field and runs the transport, and the plume renders as the simulation advances. The act of simulating is the act of seeing — an abstract statistic becomes a plume you watch finger across the screen.
Dig deeper — the rigorous version
For a statistically stationary, lognormal K field, stochastic theory predicts that field-scale (macro)dispersion grows with the log-conductivity variance σ²lnK and the correlation scale λ. Gelhar’s classic result gives an asymptotic longitudinal macrodispersivity on the order of A ≈ σ²lnK·λ, reached only after the plume has sampled many correlation scales.
At high variance (σ²lnK ≳ 1) the Gaussian, Fickian picture breaks down: transport becomes strongly non-Fickian, with pronounced tailing and channeling, and a single effective macrodispersion coefficient poorly represents the real spreading. This is the regime the bottom panel illustrates.
Deeper-dive framing after Li & Liu (2004) and Gelhar, Stochastic Subsurface Hydrology. Replace with the original topic write-up when available.02 K Variability w/ & w/o Pore Scale Dispersion
Once heterogeneity takes over, does local mixing still matter?
What you’re watching The same heterogeneous plume run two ways — with and without pore-scale (local) dispersion added on top of the conductivity-driven spreading.
The mechanism Pore-scale dispersion mixes solute across streamlines locally. Under strong heterogeneity the large-scale velocity contrasts set how far the plume spreads; local dispersion mainly smooths the fingers and controls dilution within them.
Why it matters Spreading and dilution are not the same: a plume can spread enormously yet stay concentrated within fingers, which is why dangerous peak concentrations can persist far downstream.
IGW-NET Toggle local dispersion on and off and re-run; because IGW-NET simulates and visualizes in one step, the separate roles of spreading and dilution appear side by side.
Dig deeper — the rigorous version
Spreading is measured by the growth of the plume’s spatial moments (second central moment); dilution is measured by the decay of concentration variance, or equivalently the rise of a dilution index. Heterogeneity-driven advection increases spreading dramatically while doing little to dilute; local (pore-scale) dispersion is what actually destroys concentration variance.
The practical consequence: peak concentration and spatial footprint must be tracked separately. Risk often hinges on the peak, which local dispersion controls, not on the footprint, which heterogeneity controls.
Framing after Li & Liu (2004). Original topic text to be inserted here as the rigorous reference.03 K Variability w/ & w/o Pore Scale Dispersion
What does pore-scale mixing change about the plume’s internal structure?
What you’re watching A continued with- and without-local-dispersion comparison, focusing on the internal structure of the plume’s fingers.
The mechanism Without local dispersion the fingers stay sharp and isolated; with it, adjacent fingers blur and merge, raising internal mixing while the outer envelope changes little.
Why it matters Whether fingers mix or stay separate controls reaction rates — for natural attenuation or bioremediation — and the concentrations a monitoring well actually records.
IGW-NET Watching fingers merge or stay crisp as local dispersion is added reveals a subtle process that numbers alone hide — simulation as visualization.
Dig deeper — the rigorous version
Many subsurface reactions are mixing-limited: they proceed only where solutes are brought into contact. In a heterogeneous plume that contact happens mainly by transverse local dispersion across finger boundaries, so the apparent reaction rate is controlled by mixing, not by the reaction kinetics themselves.
This distinction — transport-limited versus reaction-limited behavior — governs whether natural attenuation can be relied upon, and is invisible if one only tracks bulk plume extent.
Deeper-dive framing; original topic content to replace.04 K Correlation Scale
Does it matter whether an aquifer’s patchiness is fine-grained or coarse?
What you’re watching Two heterogeneous aquifers with the same variance but different correlation scale. The fine-textured field (top) shreds the plume into many thin fingers; the coarse-textured field (bottom) channels it into a fatter, more coherent body.
The mechanism Correlation scale is the typical size of high- and low-K patches. Small patches scatter the plume into short fingers that re-mix; large patches build long, connected fast paths that carry the plume in coherent channels far downstream.
Why it matters The same average aquifer can deliver a contaminant to a well in a year or a decade depending on patch size — correlation scale matters as much as variance, and is measured far less often.
IGW-NET Change one parameter — the correlation scale — and re-run; IGW-NET redraws both the field and the plume, so you see directly how patch size reroutes transport.
Dig deeper — the rigorous version
The integral (correlation) scale λ sets both the asymptotic macrodispersivity (∝ λ) and the travel distance required to reach it — a plume must cross many λ before ‘effective’ behavior emerges. Beyond variance, the connectivity of high-K material matters: percolation-like connected fast paths produce early arrivals that variance alone does not capture.
Framing after Li & Liu (2004); Dagan; Gelhar. Original write-up to be inserted.05 K Correlation Scale
How far must a plume travel before ‘average’ behavior appears?
What you’re watching A continued look at correlation scale, tracking how the plume organizes as patch size changes.
The mechanism A plume must traverse many correlation scales before its spreading settles toward an effective (Fickian) rate. Large correlation scales delay or prevent that, keeping transport channelized and non-Fickian across the whole site.
Why it matters Applying a textbook dispersion coefficient to a coarsely heterogeneous site can badly misplace a plume — a classic and expensive modeling error.
IGW-NET Running plumes across fields of different patch size shows how slowly ‘average’ behavior emerges — visible directly rather than argued from theory.
Dig deeper — the rigorous version
This is the pre-asymptotic (non-ergodic) regime: when the plume is not yet large compared with λ, its spreading is uncertain and realization-dependent, and an effective macrodispersivity is not yet meaningful. The MADE-site experiments are the textbook field example, where transport remained strongly non-Fickian and effective-dispersion models failed.
After Li & Liu (2004); Dagan; MADE studies. Original topic text to replace.06 Anisotropic Heterogeneity
Why do plumes so often spread sideways in thin sheets?
What you’re watching Four runs with increasing horizontal correlation length (100, 500, 1000, 5000 m) against a fixed 10 m vertical scale. The conductivity field grows more layered top to bottom, and the plume flattens into thin horizontal lamellae following the bedding.
The mechanism When conductivity is correlated far horizontally but little vertically — statistical anisotropy, the signature of layered sedimentary aquifers — fast layers carry solute long distances laterally while the slow layers between them resist vertical mixing, so the plume stratifies.
Why it matters Most sedimentary aquifers are layered, so contamination commonly spreads in discrete horizontal bands — which determines the depth at which a monitoring or extraction well must be screened.
IGW-NET Stretch the correlation ellipse and re-run; IGW-NET shows the aquifer become layered and the plume stratify in real time — anisotropy made visual instead of a tensor.
Dig deeper — the rigorous version
Statistical anisotropy is captured by the ratio of horizontal to vertical integral scales λx/λz; large ratios produce a strongly anisotropic macrodispersion tensor with longitudinal spreading orders of magnitude above vertical. Because vertical transverse dispersivity is naturally tiny, layering makes vertical mixing the rate-limiting bottleneck for the whole system. This is the same K-contrast physics that bends flow in the Law of Refraction topic.
After Li & Liu (2004); Gelhar. Original topic content to be inserted.07 Anisotropic Heterogeneity
How does layering throttle vertical mixing?
What you’re watching A continued anisotropy comparison, focusing on how little the plume mixes across layers as the horizontal correlation length grows.
The mechanism Vertical transverse dispersion is inherently weak; layering weakens it further, so solute stays trapped in its starting layers and travels far before any vertical spreading occurs.
Why it matters Thin, persistent layered plumes slip between sparse monitoring screens and resist remediation schemes that assume vertical mixing.
IGW-NET Seeing how little the plume thickens vertically, even over long travel, drives home why field plumes are so hard to intercept.
Dig deeper — the rigorous version
Field-measured vertical transverse dispersivities are often orders of magnitude smaller than longitudinal ones, which is why layered plumes remain thin over long distances. The same weak transverse mixing that keeps plumes thin also limits mixing-controlled reactions, tying this panel back to the dilution-versus-spreading distinction. Practical implication: multilevel sampling is usually required to detect such plumes.
Deeper-dive framing; original topic write-up to replace.08 Variability - Different Realizations
If two aquifers have identical statistics, will a plume behave the same in both?
What you’re watching Four aquifers (Realizations 1–4) generated from the same statistics — same variance, same correlation — but different random arrangements. The plume looks dramatically different in each: different fingers, different reach, different position.
The mechanism Statistics constrain an aquifer but don’t determine it; infinitely many arrangements share the same variance and correlation. Each is one plausible version of the real, unknowable aquifer — and each routes the plume differently.
Why it matters Because the true arrangement is never known, a single deterministic prediction is just one draw from many. Honest forecasts are probabilistic — a range of plumes, not one line.
IGW-NET Regenerate the field and re-run, again and again; IGW-NET produces realization after realization in real time, turning ‘uncertainty’ from a word into a stack of visibly different plumes.
Dig deeper — the rigorous version
In the stochastic framework, each realization is one sample of the random K field consistent with the measured statistics. Running many (Monte Carlo) yields an ensemble — a mean plume, a variance, and pointwise probabilities of exceedance. A central finding of this work is that concentration is strongly non-Gaussian and that the concentration standard deviation, the focus of much theory, is an inadequate measure of uncertainty in most practical cases.
After Li & Liu (2004). Original topic content to be inserted as the rigorous reference.09 Variability - Different Realizations
How do we make a trustworthy prediction when every aquifer is different?
What you’re watching A continued set of equally-likely realizations, setting up how an ensemble of plumes becomes a single probabilistic prediction.
The mechanism Running many realizations and compiling them gives an ensemble: the average plume, the spread around it, and the probability of contamination at any point — a Monte Carlo prediction rather than one guess.
Why it matters Probabilistic predictions let managers ask the real question — what is the chance the plume reaches the well above the limit? — and judge which data would most reduce that uncertainty.
IGW-NET Accumulating realizations on the fly and tallying statistics in real time is how IGW-NET makes Monte Carlo and conditional simulation tangible — the bridge to the stochastic topics.
Dig deeper — the rigorous version
Conditioning the realizations on measured data (conditional simulation) shrinks the ensemble spread where data exist, which is the formal basis for valuing new measurements (‘data worth’). Because the concentration distribution is skewed and non-Gaussian, full probability distributions — not mean ± standard deviation — are needed to support defensible decisions.
After Li & Liu (2004). See also the Random Field Representation and Monte Carlo topics. Original write-up to replace.