Surface Water as Boundary Conditions for Groundwater

14.1 Why Surface Water Dominates Groundwater Models

The treatment of surface water deserves its own chapter because SW is not just one feature among many — for most regional groundwater models, it is the dominant boundary condition. Getting SW right is not optional detail; it is central to whether the model is defensible. Getting SW wrong produces models where the head contours look convincing but the underlying mass balance is physically absurd.

14.1.1 The mass-balance argument

Consider a regional groundwater system in a humid or temperate climate. Water enters the system primarily through recharge from precipitation — a process distributed over the entire land surface at rates typically in the range of inches per year to a few feet per year. Water must leave the system at an equal rate over the long term (the system is approximately steady over years to decades). Where does it leave?

The options are limited. Some groundwater is extracted by wells (but pumped volumes are usually small compared to natural recharge over a regional area). Some is consumed by evapotranspiration from shallow water tables (important in wetlands and riparian zones, but typically a smaller fraction of the total). Some discharges laterally across the domain boundary (which for a well-designed regional model should be a minor component). The dominant exit — often 70–95% of the total groundwater discharge in humid regions — is to surface water features: rivers and streams, lakes and reservoirs, wetlands and springs.

This is not an empirical observation about specific regions; it is a consequence of mass balance applied to any regional aquifer receiving meaningful recharge. Whatever recharge comes down has to leave, and surface water is the dominant exit route. Which means:

14.1.2 What SW controls in the groundwater flow pattern

If SW is the dominant boundary condition, what specifically does it control?

• The head field's overall shape. Regional head contours are pulled down toward rivers, lakes, and valleys; the large-scale pattern of "water flows from uplands toward valleys" is a direct consequence of where SW discharge occurs. Without SW features, the head field would be a nearly-flat raised surface sitting atop bedrock; with SW, it has the familiar "follows topography, smoother" shape.
• The flow directions in the near-surface aquifer. Groundwater flow converges toward SW discharge features. In any small area, which direction water flows is primarily set by which SW feature is nearest and how big it is.
• Baseflow to streams. The rate of GW discharge to streams is the stream's baseflow — the portion of streamflow that isn't from surface runoff. This is a measurable quantity; if the GW model's simulated baseflow to streams doesn't match observed stream gauge data, the model is demonstrably wrong regardless of how good the head contours look.
• Local gradients near SW features. Pumping from a well near a river can induce water from the river (stream depletion); the resulting local gradients are entirely governed by the SW-aquifer exchange physics. Well-capture calculations, wellhead protection areas, and contaminant fate near rivers all depend on getting SW right.
• Seasonal and transient behavior. As stream stages rise in spring and fall in summer, groundwater heads nearby respond. The amplitude and phase of this response is set by aquifer K and the SW-aquifer exchange physics.

14.1.3 Why getting SW wrong is so insidious

A groundwater model with badly-configured SW boundary conditions can still produce plausible-looking head contours. This is the insidious part: SW boundary conditions are so strong that they pull the head solution toward whatever stage you told them to pull toward, even when the aggregate fluxes implied are physically nonsense. The model dutifully solves the equations it was given. The plot of hydraulic head contours looks smooth, continuous, and physically reasonable. The water-balance error is invisible in the head plot.

The only way to see the error is to check the water balance directly: does the total SW flux into and out of the aquifer balance against rainfall recharge and well extraction in a physically sensible way? If not, no amount of calibration of other parameters can fix the fundamental problem — the SW configuration is wrong, and until you fix it, the model is producing fiction that happens to look plausible.

14.2 Three Levels of Surface-Water Representation

IGW-NET supports three qualitatively different approaches to representing surface water, each appropriate in different modeling contexts. Understanding the three levels — and knowing which one your current model needs — is the foundation for everything that follows in this chapter.

14.2.1 Level 1 — DEM as drain (the platform default)

The simplest and most robust representation is IGW-NET's default: the land surface (from the DEM) acts as a one-way drain. The mechanism: whenever the groundwater head at a cell exceeds the DEM elevation at that cell, water discharges from the aquifer at a rate controlled by the Surface Drain Leakancy parameter (Ch. 5 §5.8). If head is below DEM, nothing happens — no flux.

Level 1 is beautiful because of what it captures for free:

• Every gaining SW feature, everywhere in the domain — rivers, lakes, wetlands, springs, seeps — is represented implicitly because they all sit at DEM lows. The DEM already encodes their locations.
• Spatially-distributed drainage — water can discharge wherever head rises above DEM, not just at explicit drawn features. This catches small streams and wetland margins that would be tedious to draw individually.
• Robust numerics — one-way drains are well-behaved; there is no way for them to inject spurious inflow. See §14.7 for why this matters.
• Zero setup — the DEM is already loaded (from the Global Base Model), and the drain behavior is on by default.

What Level 1 can't do: represent losing streams (where SW provides water to GW), specify stage different from DEM, handle lakes with meaningful two-way exchange (lake recharging aquifer near a pumping well), or any transient stage effect. If those matter, you escalate to Level 2.

14.2.2 Level 2 — Explicit head-dependent features

For specific SW features where two-way exchange matters hydrologically — a large river that can recharge the aquifer during high flow, a lake with significant storage affecting nearby wells, a controlled reservoir with stages different from the DEM — you draw explicit features with their own attributes. IGW-NET supports two geometries:

• Polyline (line) features — rivers, streams, drains, diversions. Ch. 6 §6.3 introduced the nine polyline types; for SW, the main ones are Two-way Stream (Type 5, head-dependent bidirectional), One-way Drain (Type 6, discharge-only), and variants of Prescribed Head (Types 1–3). See Fig. 14.6 below.
• Zone (polygon) features — lakes, ponds, reservoirs, wetland complexes. Configured via the Zone Attributes dialog with Sources and Sinks → Head Dependent → Two-way Head Dependent. See Fig. 14.4 below.

Each feature specifies its own:

• Stage — the water-surface elevation (constant in space and time, or variable along the feature, or from DEM via the "TopE" option)
• Leakance — how easily water exchanges across the bed (units of m/day for lines, 1/day for zones; see Ch. 6 §6.3.4)
• Bottom elevation — the streambed or lake-bed elevation below which gravity-limited disconnect kicks in
• Geometry — width for lines, area for zones

Level 2 is appropriate when specific features need to be represented explicitly — typically larger rivers, meaningful lakes, or where Level 1's drain-only default misses physically-important recharge from SW to GW.

14.2.3 Level 3 — Coupled modeling (Chapter 15)

At Level 3, stage stops being an input and becomes a solution variable. The SW-aquifer system is solved together as a coupled hydrologic model: lake stages respond to aquifer fluxes and surface inflows/outflows; wetland stages are solved consistently with the GW head field; stream-aquifer exchange includes the stream's own water balance equation. This is the approach needed when the coupling itself is the question — lake levels responding to pumping scenarios, wetland stages under climate change, reservoir operations integrated with aquifer management.

Level 3 is the subject of Chapter 15. This chapter (Ch. 14) is concerned with Levels 1 and 2, which are the standard approaches for the great majority of groundwater modeling work.

14.2.4 Choosing between Levels 1 and 2

14.3 Head-Dependent Flux — The Physics

Both Level 1 (DEM-drain) and Level 2 (explicit features) are head-dependent-flux mechanisms; they differ in what the "stage" is (DEM vs user-specified) and whether flux is one-way or two-way. Understanding the underlying physics is essential for configuring either correctly.

14.3.1 The governing equation

This is the linear "head-dependent" approximation — flux scales linearly with the head difference. It's the standard approach across groundwater modeling (MODFLOW's RIV, DRN, GHB packages all implement variants). IGW-NET uses it throughout both Level-1 DEM-drain and Level-2 explicit features.

14.3.2 Two regimes — connected and disconnected

The linear relationship above holds when the aquifer is hydraulically connected to the surface-water feature — that is, when the aquifer head is above the bottom elevation of the feature's bed. When the aquifer head drops below the bed bottom, the regime changes:

The disconnect regime occurs mostly in arid settings or near heavily-pumped wells where the water table has been drawn down below the streambed. In that case, water falls from the stream through an unsaturated zone before reaching the water table; the flux is limited by gravity and bed permeability, not by the aquifer head. The stream becomes effectively a fixed-rate recharge source to the aquifer.

IGW-NET handles the regime transition automatically — you don't need to specify which regime applies; the solver checks h_aq against the bed bottom at every simulation step and switches behavior as needed. What you do need to specify, for Level-2 features, is the bed bottom elevation (for rivers, typically the streambed elevation; for lakes, the lake-bed elevation).

14.3.3 Level 1 (DEM-drain) in head-dependent terms

The DEM-drain default is a specific case of head-dependent flux where:

• h_sw = DEM elevation at each cell — the land surface is treated as the surface-water stage
• Flux is one-way only — Q is computed and applied only when h_aq > h_sw (aquifer above land surface); otherwise Q = 0
• Leakance = Surface Drain Leakancy × cell area — distributed over every cell in the domain
• Bed bottom = h_sw — the drain activates at the land surface; there is no disconnect regime (because there's no stream channel below the surface to disconnect from)

This cellwise one-way drain, operating everywhere in the domain, collectively represents all the gaining SW features implicitly. The DEM's topographic lows — rivers, lakes, wetland areas — are where h_aq most commonly rises above h_sw and drainage occurs, which is exactly where real surface water is located.

14.3.4 Level 2 (explicit features) in head-dependent terms

Level-2 features replace the DEM-based one-way drain at specific locations with explicit head-dependent flux:

• h_sw is user-specified (constant, variable along the feature, or from DEM via TopE; see §14.4)
• Flux can be two-way (Type 5 Two-way Stream, or Two-way Head Dependent for lake zones) or one-way (Type 6 One-way Drain)
• Leakance is per-feature, reflecting the specific bed sediment and geometry (see Ch. 6 §6.3.4)
• Bed bottom is explicitly specified per feature, enabling the disconnect regime where appropriate

14.4 Where Does Stage Come From?

Head-dependent flux needs a stage value — h_sw in the governing equation. For every river cell, every lake vertex, every drain location. Where does this stage come from, and when is its source reliable enough for the work at hand?

14.4.1 The DEM-as-stage convention

The foundational convention in regional-scale groundwater modeling: use the DEM as the stage source. The DEM at a river cell was, at the moment the DEM was captured, exactly the water-surface elevation at that location — the DEM is a snapshot of the landscape including water surfaces. LiDAR DEMs capture the water-surface return for water bodies; photogrammetric DEMs similarly capture the visible water surface. Same for lakes: the DEM over a lake is the lake surface elevation at capture time.

This convention has powerful consequences for groundwater modeling:

• Stage is available everywhere for free. No gauge data, no surveyed profiles, no stream-routing required — just the DEM, which you have anyway.
• Stage has the right spatial structure. The DEM's longitudinal profile along a river decreases downstream (correctly); lakes are flat (correctly); tributary junctions are at the right elevations (approximately).
• Stage is consistent with topography. There's no risk of stages being higher than the adjacent land surface — they can't be, because the DEM is the land surface, and water bodies are topographic lows within it.
• The Level-1 DEM-drain default works because of this. Without the DEM-as-stage convention, there would be no way to represent all SW features implicitly via the DEM; the Level-1 default is entirely built on this convention.

14.4.2 How IGW-NET implements it — the TopE option

In every Level-2 feature attribute dialog — Zone Attributes for lakes, Polyline Attributes for rivers — you have a stage input with a radio-button choice between:

• Const — user-specified constant stage value
• TopE — stage equals Top Elevation (from DEM), optionally minus a user-specified offset
• Variable — stage varies along the feature (for polylines with distance-varying stage)

The TopE option is IGW-NET's implementation of the DEM-as-stage convention: click TopE and the stage at every vertex is automatically the DEM elevation at that vertex — a spatially-complete stage field pulled from the data you already have. For most regional modeling of reasonably large rivers and lakes, this is the right choice.

14.4.3 When DEM-as-stage is good enough

DEM-as-stage works well in these conditions:

• Regional-scale modeling. Basin-wide or watershed-scale flow models, where you care about the big-picture head pattern. DEM-stage errors of a few meters at any single river cell average out; the overall pattern is correct.
• Reasonably large rivers. 3rd-order or higher streams (§14.5) typically have channels wide enough that LiDAR or photogrammetric DEMs capture the water surface with sub-meter precision. The stage value is physically meaningful.
• Lakes large enough to be well-represented in the DEM. Anything with area above ~10 hectares (see Ch. 6) has a stable, flat, well-captured surface in typical DEMs.
• Long-term / average conditions. The DEM captures stage at one moment; if you're modeling long-term average hydrologic conditions (typical baseflow, approximate steady state), the DEM-moment stage is close to typical stage.

14.4.4 When DEM-as-stage breaks down

DEM-as-stage becomes unreliable or wrong in these cases:

• Small streams. 1st- and 2nd-order streams are often narrower than the DEM cell size (30 m, 10 m, or even finer). The DEM cell over a 2 m-wide stream likely gives you the floodplain elevation, not the water surface. DEM-stage error can be meters. This is the source of the 10×-rainfall pitfall in §14.7 — and why small streams should be one-way drains (§14.5), not two-way head-dependent.
• Ephemeral or seasonal systems. The DEM captured a specific moment; if you're modeling dry-season conditions and the DEM was flown during wet season (or vice versa), the stage you get is for the wrong condition.
• Engineered systems. Dams, weirs, locks, and controlled reservoirs have stages that change with operations. The DEM captured one operating level. Modeling any other operating condition requires explicit stage input, not DEM-based.
• Site-scale precision matters. If you need stage accurate to tens of centimeters (remediation work, precise well capture, detailed river-aquifer exchange for a specific reach), DEM stage is too coarse. Use surveyed profiles or gauge data.
• Transient stage modeling. The DEM is a snapshot. Modeling how aquifer response changes as stream stages rise and fall through time requires time-varying stage input (from transient file, or Level-3 coupled modeling).

14.4.5 Alternatives to DEM-as-stage

When DEM-as-stage isn't appropriate, IGW-NET supports other stage sources:

• Constant user-specified stage (Const radio button) — for lakes or reservoir operations where you know the operating level, or short reaches where you have survey data
• Variable stage along feature (Variable, with distance-elevation table) — for longer reaches with known longitudinal profile from gauge data or routing
• DEM minus constant offset (TopE with non-zero offset) — the river surface sits some meters below the bank; tells IGW-NET to use DEM minus that offset as stage
• Transient stage file (in transient simulations) — time series from gauge data, loaded as a CSV
• Coupled modeling (Level 3, Ch. 15) — stage is a solution variable, not an input

For most regional modeling work in humid climates using the standard three-level framework, DEM-as-stage via TopE is the default and is the right choice. The alternatives are for specific situations where DEM-stage's limitations matter.

14.5 Stream-Order Hierarchy — Which Features Should Be One-Way vs Two-Way

When you have a hydrography network with many stream reaches of different sizes, you need a rule for which ones to configure as what type. The stream-order hierarchy — a concept from classical hydrology applied directly to modeling decisions — is how IGW-NET handles this.

14.5.1 The Strahler stream-order concept

The Strahler stream order is a classical measure of a stream's position in the drainage network. A 1st-order stream has no tributaries — it's a headwater. When two 1st-order streams meet, they form a 2nd-order stream. When two 2nd-order streams meet, they form a 3rd-order; and so on. (When two streams of different orders meet, the resulting stream keeps the higher of the two orders.) Stream order thus captures how "upstream" or "downstream" a reach is in the overall network hierarchy.

Stream order correlates with physical properties relevant for groundwater modeling:

• Channel width. 1st-order streams are typically 1–3 meters wide; 3rd-order perhaps 10–30 m; 5th-order and higher tens to hundreds of meters wide.
• DEM resolvability. Lower-order streams are narrower than DEM cells and thus poorly captured; higher-order streams have surfaces visible and well-captured in typical LiDAR DEMs.
• Flow magnitude. 1st-order streams have small, often intermittent baseflow; higher-order streams have year-round substantial flow.
• Hydraulic stability. Higher-order streams have stage that changes slowly and predictably; lower-order streams may be dry, flashy, or have rapidly-changing stages.

Hydrography datasets (NHD, HydroRIVERS; see §14.6) carry stream-order information as an attribute on each reach. IGW-NET's DataNET aggregation preserves this information — when you import a hydrography dataset, you know which reaches are 1st-order, 2nd, 3rd, and so on.

14.5.2 The decision rule

14.5.3 What IGW-NET does by default

When you import hydrography features from the Data Center (the "auto-populated surface water features" mechanism), IGW-NET applies the stream-order decision rule automatically based on the imported attributes:

• Small streams come in as one-way drains with stage from DEM and default drain leakance
• Larger rivers come in as two-way head-dependent streams with stage from DEM and default stream leakance (Ch. 6 §6.3.4)
• Small water bodies come in as one-way drains over their polygon extent
• Large lakes come in as two-way head-dependent zones with stage from DEM and default lake leakance

This gives you a sensible Level-1 + Level-2 mix as a starting point, without requiring you to make the stream-order decisions manually. You can override individual features to adjust the treatment, but the default is designed to avoid the 10×-rainfall pitfall out of the box.

14.5.4 Edge cases and exceptions

The stream-order rule is a rule of thumb; specific situations may call for adjustment:

• A small stream where you specifically need two-way exchange (e.g., a small stream near a pumping well where induced infiltration matters). In this case, override to two-way, but provide surveyed or gauge-based stage, not DEM. The DEM-stage unreliability is the issue; if you can get better stage data, two-way becomes viable even on smaller streams.
• A large river with known disconnect (perched in an arid region). Configure as two-way (it's a large feature with good DEM stage), but the disconnect regime (§14.3.2) will kick in automatically when aquifer head drops below streambed bottom; the flux becomes gravity-limited downward seepage.
• High-elevation headwater streams or upland lakes known to be losing features. Even in humid climates, the losing 5-10% of the network is concentrated in uplands (§14.8.1). A kettle lake on a moraine that recharges a valley aquifer, a perched headwater stream above a municipal wellfield, or a seep that leaks downward rather than discharging outward — these are losing features whose recharge contribution matters hydrologically. Escalate to Level 2 two-way; use surveyed or observed stage (DEM-stage on small upland streams is unreliable); leakance captures the perched-seepage rate. This is a physically-motivated escalation distinct from the pumping-near-SW escalation.
• Engineered controlled reservoirs. Always specify stage explicitly rather than using DEM — the operating level may differ from DEM moment, and you may want to model alternative operating scenarios.
• Canals. These are often small but highly-controlled features with well-known stages; explicit stage (Const) and two-way is appropriate because the stage is a known quantity independent of DEM.

14.6 Hydrography Datasets — NHD, HydroSHEDS, and DataNET

The raw material for SW representation — where streams are, where lakes are, what order they are, how they connect — comes from hydrography datasets. IGW-NET integrates major US and global datasets through its DataNET layer.

14.6.1 The main datasets

14.6.2 How DataNET handles them

IGW-NET's DataNET layer provides access to hydrography through a specific architecture:

• Out-of-memory, chunk-by-chunk processing. Large hydrography datasets are not loaded wholesale — DataNET processes chunks on demand, aggregates to the user's current model grid resolution, and returns grid-ready features. A continental-scale stream network can be queried for a small watershed without memory issues.
• Pre-aggregation to grid resolution. When NHD is imported into a model at NX = 40 over a 100 km × 100 km domain (2.5 km cells), DataNET aggregates the fine-resolution stream lines to the cell scale — small streams inside a cell are merged or represented collectively. The user sees cell-appropriate features, not raw fine-scale data.
• Automatic classification. Stream order, reach type, feature size, and other attributes are preserved through the aggregation, enabling IGW-NET's default one-way-vs-two-way decisions (§14.5.3).
• Global base model integration. Hydrography is part of the MAGNET4WATER Global Base Model — available everywhere on Earth as a pre-assembled layer, not something the user has to hunt down and import per-session.

14.6.3 Using hydrography datasets in IGW-NET

Three pathways for getting hydrography into your model:

• Auto-population from the Data Center (the default). When you draw a new model domain, the Data Center automatically populates SW features from the best available hydrography for that location — NHD in the US, HydroSHEDS/HydroRIVERS/HydroLAKES globally. You get a sensible starting mix of one-way drains and two-way features based on stream order and lake size.
• Import Shapefile. Load a user-supplied shapefile with stream lines or lake polygons. IGW-NET reads the geometry and classifies features using default rules or user-provided attribute mapping.
• Manual drawing (Ch. 6). Draw specific features interactively — useful for refining auto-populated features, adding features missing from the datasets, or specifying explicit geometry for specialized studies.

14.6.4 The Server River Options and Server Lake Options dialogs

The abstract teaching above (§14.5 stream-order rule, §14.6.1 datasets, §14.6.3 auto-population) becomes concrete in two dialog boxes. Understanding these dialogs is essential because they are where you actually control how hydrography is imported and classified.

The top-level import toggle

Server River Options — the stream-order rule as UI

Walking through the dialog from top to bottom:

• Source selector — USA NHDplus, Global HydroSHEDS, or User Data (for a user-supplied shapefile, see §14.6.3). Pick whichever dataset is authoritative for your region.
• Stream Orders checkboxes — 1st-and-0th, 2nd, 3rd, 4th, 5th, 6th, 7th, 8th, 9th, 10th-and-up. You choose which orders to import. The note in the dialog is important: "selecting smaller stream orders in a big area is not necessary and may cause a long waiting time or memory issues during extracting data from servers." For regional models, you typically uncheck the lowest orders (1st-and-0th, often 2nd too) — they add massive numbers of small features that will go straight to one-way drain or Level-1 DEM-drain anyway. For detailed site-scale work over small areas, you can include them.
• Drawing color and thickness per order — visualization; color and line thickness scale with order so you can see the network hierarchy at a glance.
• Order Based table — the heart of the dialog. Six rows covering six order groups (≤1, 2, 3, 4, 5, ≥6). For each group:
  • Leakance — the stream-aquifer exchange parameter. Default values scale with order: 1 per m for order ≤1, up to 50 per m for ≥6. Larger/deeper channels have higher leakance reflecting better aquifer connection.
  • Two-way / One-way radio buttons — the operational expression of the stream-order rule. Low-order streams default to one-way drain (mass-balance protective; §14.7); high-order streams default to two-way head-dependent (real two-way exchange, reliable stage). You can override per row, but the defaults embody the §14.5 decision rule directly.
  • Depth — used to set streambed elevation = stage − depth. Scales from 0.3 m (order ≤1) to 2 m (order ≥6).
  • Width — channel width, used in the leakance × width computation for the line-feature flux (Ch. 6 §6.3.4). Scales similarly.
  • Manning coefficient — surface roughness for the Watershed Solver coupling (if Watershed Solver is active; see Ch. 16). Default 0.045 for small channels, 0.025 for major rivers.
  • Order Name / Alias Name — labels used in the water-balance output and map display, so you can distinguish e.g. "Kalamazoo River" from "small tributaries" in the results.
• Constant option — alternative to the Order Based table; use a single leakance and a single one-way/two-way classification for all orders. Useful for simple demonstrations or when you don't have stream-order attributes in a user-supplied dataset.
• TopE as Stage/Drain Bottom — the DEM-as-stage convention (§14.4) expressed as a checkbox. On by default. Stage at every stream vertex is the DEM at that vertex.
• Allow to add recharge to RIV cell — protective flag, off by default. When off, RIV cells (two-way stream cells) cannot receive distributed rainfall recharge in addition to their head-dependent exchange. Turning this on can double-count water. Leave it off unless you specifically need both mechanisms.
• RIV cell overrides DRN cell — on by default. If a cell is both a two-way river (RIV) and a one-way drain (DRN), the RIV treatment wins. Prevents conflicts when the same geographic location is classified both ways in different datasets.

Server Lake Options — the lake-size rule as UI

The lake dialog parallels the river dialog with lake-appropriate adjustments:

• Source selector — same three options (USA NHDplus, Global HydroSHEDS, User Data).
• Lake Orders categories — Small, Middle, Large, Very Large, Great Lakes, River Polygon. Not Strahler-like; these are size bins. "River Polygon" is for features that appear as polygons in the dataset even though they're really large river reaches (e.g., braided rivers, wide lower Mississippi).
• Size Category Based on — choice of Percentile or Area. Percentile partitions the lakes in the current domain by their size-rank (10% smallest get one treatment, 10-20% next, etc.). Area uses absolute area thresholds. Percentile is more adaptive — the same settings work across different regions.
• 10-row table — for the Percentile option, 10 rows for the 10%, 20%, 30%, ..., 100% percentile bins. For each:
  • Area — the area threshold for that percentile in this domain (computed automatically).
  • Leakance — scales from 0.005 per day for the smallest lakes (10%) to 100 per day for the largest (100%). Note the much wider range than streams; lake-aquifer exchange is highly size-dependent.
  • Two-way / One-way radio buttons — the lake equivalent of the stream-order decision. Smallest lakes default to one-way drain; largest default to two-way head-dependent. The crossover is similar in spirit to the stream-order threshold: features big enough to have reliable DEM stage and meaningful bidirectional exchange get two-way; smaller features get protective one-way.
  • Depth — scales from 0.3 m for smallest to 6 m for largest, used for lake-bed elevation.
  • Lake Name — label for water-balance output.
• Constant option — as in the river dialog, use a single setting for all sizes if you don't have size attributes.
• TopE as Stage/Drain Bottom — DEM-as-stage convention. On by default.
• Keep lakes in a cell when total lake areas > [0.25] cell area — the cell-fraction threshold. Small lakes whose total area within a grid cell is less than this fraction (default 25%) are dropped from the import (they're not resolvable at this grid resolution and would add clutter). Raising this threshold drops more small lakes; lowering it keeps more.

The flow-through lake — why lakes must usually be two-way

One scenario deserves explicit attention because it's easy to miss: flow-through lakes. A lake sitting in a regional topographic gradient can be simultaneously gaining on one side and losing on the other — the upstream portion receives GW discharge (h_aq > h_lake, gaining); the downstream portion provides GW recharge (h_aq < h_lake, losing). Net flow passes through the lake from the upstream aquifer to the downstream aquifer.

Flow-through behavior is common in landscapes with meaningful topographic gradient — kettle lakes in outwash plains, glacial lakes in moraine terrain, reservoir-like lakes in sloping valleys. Even in humid regions like Michigan, many lakes exhibit flow-through rather than pure gaining behavior.

The important modeling consequence: a one-way drain cannot represent a flow-through lake at all. One-way drain can only discharge, so it captures the gaining upstream side but misses the losing downstream side entirely. The lake's role as a throughflow feature — contributing to both aquifer drawdown upstream and aquifer recharge downstream — is lost. Water balance across the lake is broken.

A two-way head-dependent lake handles flow-through correctly because the flux is computed per cell within the lake polygon, using the cell-specific aquifer head. With a single lake stage and spatially-varying aquifer head beneath the lake:

14.7 The Central Pitfall — Why Small Streams Must Be One-Way Drains

This section is the flagship teaching of the chapter. It names the specific water-balance failure mode that comes from combining dense hydrography networks, DEM-based stages, and two-way head-dependent flux — and explains why the one-way-drain convention is an active protection mechanism rather than just an approximation.

14.7.1 The failure mode — mechanism

The failure mode is precise and reproducible. It arises whenever all three of these conditions apply:

1. You have a dense hydrography network with many small streams (1st- and 2nd-order)
2. Those small streams are configured as two-way head-dependent flux
3. Their stages come from the DEM (the TopE option or equivalent)

Here's what happens numerically:

14.7.2 Why one-way drains don't have this failure mode

A one-way drain behaves fundamentally differently. The flux is:

Consider the same small-stream-with-DEM-error scenario, but with the stream configured as one-way drain:

1. If the DEM gives a spuriously-high drain elevation (DEM sits above true water table): haq < hdrain, so Q = 0. The drain does nothing. No spurious inflow. Water balance unaffected.
2. If the DEM gives a spuriously-low drain elevation (DEM sits below true water table): the drain does discharge, perhaps at an elevated rate compared to reality. But discharge is physically possible (groundwater above land surface drains to streams) — the worst case is overstated baseflow, which is a much softer failure than nonphysical inflow.
3. Aggregate effect: thousands of small-stream drain cells mostly do nothing most of the time; they activate only where real discharge is occurring. Water balance stays physically sensible.

14.7.3 The protective framing

14.7.4 Detecting the failure

Because the failure is invisible in head plots, you detect it through water-balance checks (§14.9). The diagnostic question: is the head-dependent flux from SW INTO the aquifer a plausible fraction of total rainfall recharge?

• Plausible: 0 (pure drain behavior, no two-way features). Or a small fraction (5–20%) representing real SW-to-GW recharge in specific locations.
• Plausibly in arid regions: up to ~100% or more, where SW-source behavior dominates (§14.8).
• Not plausible: 5× rainfall. 10× rainfall. 100× rainfall. These values are immediate red flags — your SW configuration is broken.

14.7.5 The fix

When you detect the 10×-rainfall failure, the fix is usually straightforward:

1. Identify the small-stream features configured as two-way head-dependent. Look at the polyline list in the Feature Manager; any small-stream reaches marked as Type 5 Two-way Stream are candidates.
2. Convert them to one-way drains (Type 6) or delete them entirely (letting Level-1 DEM-drain handle them).
3. Re-simulate and re-check the water balance. The SW-to-GW flux should drop dramatically, back into plausible range.
4. Keep only 3rd-order and larger streams as two-way, and reconfirm those have correct stages (from DEM or survey/gauge data) and appropriate leakance.

In auto-populated IGW-NET models, this failure should not arise in the first place — the stream-order rule is applied during auto-population. But it can arise from:

• Importing a user-supplied shapefile without stream-order classification, and defaulting everything to two-way
• Manually drawing many small-stream polylines and choosing Two-way as the default type
• Overriding auto-populated one-way drains to two-way without providing non-DEM stage data
• Using older model configurations that predated stream-order-aware defaults

14.8 The Climate Gradient — From Humid (Drain-Dominated) to Arid (SW as Source)

Everything above assumed a humid-climate context. The defaults (DEM-drain as Level 1, stream-order rule for Level 2 escalation) are optimized for regions where precipitation exceeds evapotranspiration and surface water functions as the drain for the groundwater system. In drier climates, this assumption progressively fails and the right defaults change.

14.8.1 The humid end — Michigan as archetype

In humid temperate climates — Michigan, the upper Midwest, New England, the Pacific Northwest, most of the eastern half of North America — the SW-as-drain framework is empirically dominant:

• Annual precipitation typically exceeds evapotranspiration; the water balance has positive net groundwater recharge
• Surface water is dense and well-distributed; dense hydrography networks cover the landscape
• Approximately 90–95% of stream reaches at any moment are gaining (receiving water from groundwater); approximately 5–10% are losing (providing water to groundwater), usually seasonally and locally
• Lakes and wetlands generally function as discharge points for the aquifer
• The Level-1 DEM-drain default captures this correctly with minimal configuration

In this regime, the main modeling decisions are: when to escalate specific features to Level 2 for two-way exchange (large rivers, meaningful lakes, near pumping wells), and how to avoid the stream-order pitfall (§14.7). Almost every region from which groundwater modeling conventions developed — the US Midwest, the UK, Germany, the Netherlands — is in the humid end of this spectrum, which is why the one-way-drain defaults are so entrenched.

The topographic organization — where the losing 5-10% actually lives

The "5–10% losing" figure is a network-wide average, and it masks an important spatial pattern. Losing reaches are not scattered randomly across the network; they are concentrated at the top of the landscape. The pattern, from uplands to valleys:

• High-elevation headwater reaches and small upland lakes — often losing. These features sit on upland terrain, perched above or near the regional water table. In humid regions the water table is shallow enough that many headwater streams are still gaining, but a meaningful fraction are perched-losing, leaking water downward through permeable upland soils toward the deeper water table below. The same is true of many small upland lakes — even in Michigan, some sit above the regional water table and function as recharge features, not discharge features.
• Mid-elevation streams — transitional. Gaining in wet seasons, losing in dry; gaining in humid sub-basins, losing where they cross permeable outwash deposits; generally gaining but with meaningful variability.
• Low-elevation mainstem streams and valley rivers — strongly gaining. These sit at or near the regional water table, receiving the accumulated discharge of the aquifer system. In mainstem reaches of a typical Michigan watershed, gaining behavior approaches 100% of the time and 100% of the reach.
• Valley lakes and wetlands — almost always gaining. Large lakes in valleys, wetland systems at topographic lows, springs and seeps — these are the system's discharge points, receiving GW as a matter of regional mass balance.

This pattern follows directly from the regional flow system: upland recharge → regional groundwater flow toward lowlands → discharge to lowland SW. Surface water features above the regional water table cannot receive GW discharge (GW can't flow uphill), so they must be either isolated from GW or losing. Surface water features below the regional water table are naturally discharge points. Elevation relative to the water table — not absolute elevation, and not stream order alone — is the primary organizing variable.

14.8.2 The arid end — Arizona and Hawaii as archetypes

In arid and semi-arid climates, the balance flips:

• Annual evapotranspiration exceeds or approaches precipitation; direct recharge from precipitation is small or zero over much of the landscape
• Surface water is sparse — perennial streams are rare; ephemeral washes flow only after storms
• Losing streams dominate. Where streams do flow, they typically lose water to the unsaturated zone below, recharging the aquifer — sometimes as the dominant recharge mechanism for the entire basin
• Irrigation canals, irrigated agricultural returns, and urban infrastructure become important SW-to-GW recharge sources
• The water table may be hundreds of meters below the DEM in interstream areas; the connection between DEM topography and groundwater is weak

Examples where this regime applies:

• Arizona — Sonoran Desert basin-fill aquifers recharged primarily by mountain-front recharge and losing streams; perennial rivers in deep valleys are frequently losing
• Hawaii — Mountain streams on leeward volcanic slopes often lose most of their flow to the subsurface before reaching the coast; the basal aquifer is recharged mainly by these losing stream and by direct mountain-recharge
• Central Valley, California — Irrigation and intermittent surface-water flows dominate the recharge regime for the agricultural basin
• Interior West generally — Utah, Nevada, New Mexico, parts of Colorado, Wyoming — all have regions where losing streams are significant

14.8.3 What changes in the modeling approach

14.8.4 The transition zones

The humid/arid dichotomy is useful but real climates are a continuum. Semi-arid regions (Texas, Oklahoma panhandles, eastern Colorado, much of the interior Southwest) have characteristics of both — humid-seasonal (spring/early summer) and arid-seasonal behavior. Mediterranean climates (coastal California, Oregon) have winter-wet humid-like behavior and summer-dry arid-like behavior.

In transitional regions:

• Seasonal modeling may be essential — the SW-GW behavior changes through the year
• Defaults appropriate for one season may be wrong for another
• Both gaining and losing streams may coexist in the same basin, even on the same reach at different times
• Careful water-balance diagnostics (§14.9) are especially important because the "expected" pattern is itself complex

14.9 Water-Balance Diagnostics — Catching Silent Failures

Head plots cannot reveal SW-configuration failures; water-balance output can. This section covers the specific diagnostic workflow for catching the silent failures that SW misconfiguration produces.

14.9.1 The water balance in IGW-NET

Every IGW-NET simulation produces a water-balance analysis as part of the standard result set (Ch. 8). It itemizes all sources and sinks of water for the model domain:

• Rainfall recharge — distributed over the domain, typically the primary source in humid climates
• Well extraction / injection — negative or positive depending on direction
• Boundary fluxes — flux across the domain boundary (no-flow by default; may be significant for submodels or prescribed-head boundaries)
• Surface-water fluxes — itemized by feature type: Level-1 DEM-drain, Level-2 two-way streams, one-way drains, lakes, etc. Each with its own in/out contribution
• Storage change — for transient simulations only; zero in steady-state
• Residual — the sum of all the above should be zero; any residual is the solver's discretization / convergence error

For SW configuration diagnostics, the critical items are the SW flux breakdown: how much water the model is flowing into the aquifer from SW features, how much it's flowing out to SW features, and the net.

14.9.2 The canary check

14.9.3 Comparing against observed baseflow

Beyond the internal canary, the most powerful SW-configuration diagnostic compares simulated SW fluxes against observed streamflow data:

• Find USGS gauge records (or equivalent) for specific streams within your model domain
• Obtain baseflow estimates for those gauges (baseflow is a well-established hydrologic analysis; separating baseflow from stormflow in gauge records is routine)
• Compare model-simulated flux from aquifer to that stream segment against the observed baseflow

Baseflow should match simulated GW-to-SW discharge for the corresponding reach. Discrepancies indicate:

• Simulated flux >> observed baseflow → model is over-discharging; leakance or stage may be wrong
• Simulated flux << observed baseflow → model is under-discharging; leakance may be too low, or Level-1 drain leakancy may need tuning
• Simulated flux in wrong direction → model has stream as losing but gauge shows gaining (or vice versa); classification is wrong

This is the gold-standard diagnostic. A model whose simulated baseflow matches observed is well-calibrated against the dominant SW-GW interaction; a model that can't match observed baseflow is poorly calibrated regardless of how good heads look.

14.9.4 Other water-balance red flags

14.9.5 The diagnostic workflow

14.10 When Surface-Water-as-BC Is No Longer Enough

Everything in this chapter has treated SW as a boundary condition: stage is an input, SW-aquifer flux is computed, the SW feature's own water balance is not tracked by the model. This framework handles the great majority of groundwater modeling. But there are cases where it is structurally inadequate, and where the move to Level-3 coupled modeling (Chapter 15) becomes necessary.

14.10.1 Symptoms that SW-as-BC is insufficient

Signs that you need Level 3:

• Lake stages need to respond to pumping. You're modeling a scenario where a high-capacity well extracts water, inducing lake drawdown, and you need to know how much the lake drops. With SW-as-BC, lake stage is fixed; no response possible. Level 3 makes stage a solution variable that responds to aquifer fluxes.
• Wetland water tables drive the question. Groundwater-dependent ecosystems (GDEs), wetland habitat assessments, and climate-change wetland-response studies all require wetland stage to be consistently solved with aquifer head — not independently specified.
• Stream baseflow varies seasonally in ways that matter. Fisheries habitat, ecological-flow requirements, and water-allocation studies need the response of stream stage and baseflow to aquifer conditions, not a fixed stage input.
• Reservoir operations interact with aquifer management. Multi-reservoir systems with managed releases, where aquifer storage is part of the integrated operation, need coupled modeling.
• The SW water balance itself is a modeling target. Not just "how does GW move" but "how does the integrated SW-GW system store and transfer water" — this is inherently a coupled question.

14.10.2 What coupled modeling adds

Chapter 15 covers the specifics, but the structural change is:

14.10.3 Reading on to Chapter 15

Chapter 15 (Coupled Lake-Aquifer Modeling) picks up from here:

• How IGW-NET implements coupled lake-aquifer modeling — the computational architecture and the user interface
• When to escalate specific features from Level 2 to Level 3 — decision criteria and physical indicators
• The Barron Lake integrated case study — a real-world coupled model with explicit stage, aquifer exchange, and complete water balance
• Wetland modeling and shallow-groundwater-driven systems
• Transient stage and stress coupling

For most modeling work, Levels 1 and 2 (this chapter) are sufficient, and you will never need Level 3. But when you do need it, the transition is straightforward: the physics is an extension of the same head-dependent flux framework (§14.3), and the operational workflow builds naturally on the Level-2 feature setup. Chapter 15 is the next step.