Session 7 · Inside one day

Groundwater: drainage, spill, baseflow

Session 1 noticed that the Leaf River never stops flowing, even after weeks without rain. That water comes from groundwater. This session follows water from the soil down into dCREST's aquifer store and back out as baseflow, and shows why the shape of the baseflow rule controls how a river recedes in a drought.

Time about 25 minutesNeeds Sessions 5 and 6
Watch this session first: a 5-minute narrated video with English subtitles. The page below goes further and lets you try each step yourself.

You will be able to

Where we are: the aquifer receives drainage from the soil and releases baseflow (and spill) to the river; on dry days it also supplies some evaporation (Session 6).

Drainage: from soil to aquifer

After the soil step, a small fraction of the soil water seeps downward every day, wet or dry, to the groundwater below:

\[ D_s = \min(\mathit{ksoil}\cdot \mathit{SM}',\; \mathit{SM}'), \qquad \mathit{SM} \leftarrow \mathit{SM}' - D_s, \qquad G \leftarrow G + D_s \]

ksoil is a rate per day: the fraction of the soil water that drains each day. The minimum only guards against draining more than is there, which cannot happen while ksoil is below 1. For the Leaf River ksoil = 0.00423 per day: about 0.42% of the soil water drains each day, so on its own the soil would lose half its water in about 163 days. On 15 August 1993, with 159.0 mm in the soil after evaporation, 0.67 mm drained to the aquifer.

The aquifer fills and spills

The aquifer, holding G mm, has a capacity hmaxaq. Water drained into a full aquifer cannot be stored and leaves at once as spill:

\[ Q_{sp} = \max(G - \mathit{hmaxaq},\; 0), \qquad G \leftarrow \min(G,\; \mathit{hmaxaq}) \]

Spill is treated as part of baseflow. In very wet periods it lets groundwater respond quickly once the aquifer is full. For the Leaf River hmaxaq = 106.9 mm; spill provides 24% of the groundwater contribution over 1989–1999, most of it (84%) from February to May, when the aquifer is full.

Baseflow

The aquifer then releases baseflow. dCREST does not use a linear reservoir here; it uses an exponential rule in the aquifer's relative fullness \(G/\mathit{hmaxaq}\):

\[ Q_b = \min\!\Big( \mathit{gwc}\,\Big[\exp\!\Big(\mathit{gwe}\,\frac{G}{\mathit{hmaxaq}}\Big) - 1\Big],\;\; G \Big), \qquad G \leftarrow G - Q_b \]

Three properties make this a good choice:

Try itThe baseflow rule and the recession it makes
Baseflow against aquifer fullness
A drought: aquifer starts full, no input for 365 days (log scale)

Grey: a linear reservoir that releases the same amount when the aquifer is full. Under the dCREST rule the flow falls steeply at first, because outflow shrinks faster than the store, and the remaining water then drains very slowly. With the Leaf River values, after a year without input the dCREST aquifer has released 61 mm and still holds 46 mm of its 107; the linear one has released 90 mm and holds 17 mm. Raise gwe to make that contrast stronger.

The aquifer at the Leaf River

DataTen years of groundwater, from the model

Top: aquifer storage G with its capacity dashed. Bottom: baseflow, spill and the observed river flow on a log scale. In late summer the observed flow falls to a few tenths of a millimetre per day; in the model that flow is almost all baseflow.

The aquifer fills in winter, when evaporation is low and drainage outpaces baseflow, and empties slowly through summer, when baseflow and aquifer evaporation (Session 6) exceed the small drainage from a drying soil. Over 1989–1999 groundwater supplies 25% of the model's runoff. That is a small share of the total, but on most late-summer days (90% of August and September days) it is almost all of the river.

In the code: src/dcrest/kernels/soil.py (end, shortened)
# soil drainage to the aquifer
soil2gw = torch.minimum(KSOIL * kd * drainable, drainable)
Wo = Wo - soil2gw
gw = gw + soil2gw
# fill-and-spill aquifer, exponential baseflow
spill = torch.clamp(gw - HMAX, min=0.0)
gw = torch.minimum(gw, HMAX)
base = torch.minimum(GWC * kd * torch.expm1(GWE * torch.clamp(gw, min=0.0) / HMAX), gw)
gw = gw - base
return over, inter, base + spill, Wo, gw

torch.expm1(x) computes \(e^x - 1\) accurately even when \(x\) is tiny, which matters when the aquifer is nearly empty. Spill is returned together with baseflow. The aquifer step follows CRESTPhys, an extension of CREST that exists only in a fork of EF5 (github.com/mchen15ouedu/EF5), not in the official EF5; it added a groundwater store to CREST's single soil store. (In CRESTPhys the excess water within the interflow cap also recharges that store; dCREST v1.0, like the official CREST, makes it interflow.) The code above is shortened: drainable is the soil water after the day's infiltration or evaporation, \(\mathit{SM}'\), with the default options.

In Python

Check your understanding

The soil holds 200 mm after the soil step and ksoil = 0.01. How much drains to the aquifer today?

\(0.01 \times 200 = 2\) mm.

The aquifer holds 100 mm with hmaxaq = 100 mm, and 3 mm drain into it. What happens?

G becomes 103 mm, so 3 mm spill and G is set back to 100 mm. Then baseflow is computed from the full aquifer.

Why does a river with exponential baseflow recede quickly at first and then very slowly?

Because baseflow grows exponentially with storage: at high storage it is large and drains the aquifer fast; as storage falls, baseflow drops sharply and the remaining water drains slowly.

Why use \(e^{x} - 1\) instead of \(e^{x}\)?

So that an empty aquifer gives exactly zero baseflow, smoothly, with no threshold.

What to remember