Question 7 of 7: Channel Routing and Flood Wave Behaviour

Problem 7: Channel Routing and Flood Wave Behaviour (20 marks)

Question text not reproduced: the examination questions are © Engineers and Geoscientists BC. Open the official past paper (linked at the top of this page) to read the question, then follow the worked solution below.

Part (i) — Key steps in channel routing, with two advantages and two issues (7 marks)

Channel routing predicts how a flood hydrograph is modified — translated in time and attenuated in peak — as it travels down a reach. The Muskingum method is the common hydrologic form, in which reach storage is split into a prism proportional to outflow and a wedge proportional to the difference between inflow and outflow:

where K is the travel time through the reach and X a dimensionless weighting factor between 0 and 0.5.

Key steps.

1. Define the reach and assemble its data. Choose reach limits between which the channel is reasonably uniform and no major tributary enters; survey the cross-sections, slope and roughness. Break the river at every significant confluence, so that lateral inflows are added at reach boundaries rather than smeared along a reach.
2. Obtain the inflow hydrograph. Either gauged at the upstream section or computed by rainfall–runoff modelling of the upstream catchment.
3. Determine the routing parameters. From a pair of observed inflow–outflow hydrographs, K is estimated as the lag between their centroids and X by trial until a plot of $S$ against $[XI+(1-X)O]$ collapses to a single line rather than a loop. Where no gauged pair exists, Muskingum–Cunge derives both from channel geometry and wave celerity, which is why it is the preferred variant for ungauged reaches.
4. Compute the routing coefficients and check their closure. With $K = 12$ h, $X = 0.20$ and $\Delta t = 6$ h, the denominator is $D = 2K(1-X)+\Delta t = 25.2$, giving $$C_0=\frac{\Delta t-2KX}{D}=0.048,\qquad C_1=\frac{\Delta t+2KX}{D}=0.429,\qquad C_2=\frac{2K(1-X)-\Delta t}{D}=0.524$$ The three must sum to unity, and here $0.048+0.429+0.524 = 1.000$ — the standard arithmetic check, since a routed steady inflow must produce an equal steady outflow. The time step must also satisfy $2KX \le \Delta t \le 2K(1-X)$, that is $4.8 \le 6 \le 19.2$ h, which it does; violating this window produces negative coefficients and a spurious dip at the start of the outflow hydrograph.
5. Route the hydrograph step by step. Apply $O_2 = C_0 I_2 + C_1 I_1 + C_2 O_1$ successively down the reach, add any lateral inflow, and pass the result to the next reach.
6. Check. Confirm that the outflow volume equals the inflow volume plus lateral inflow, that the peak is attenuated and lagged, and that the computed hydrograph matches an observed event at the downstream gauge.

Two advantages. First, the method is computationally cheap and needs only two parameters, so it can be applied to a whole river network of dozens of reaches, run continuously for decades, and embedded in a real-time forecasting system that must produce an answer in seconds. Second, because K and X can be fitted directly to observed hydrographs, a well-calibrated Muskingum reach reproduces measured floods accurately without any survey of cross-sections at all — a decisive practical advantage on rivers where survey data are thin.

Two issues. First, the method omits momentum entirely, so it cannot represent backwater effects, flow reversal, or the influence of a downstream control such as a tidal boundary, a reservoir, or a confluence with a larger river in flood. Where these govern, Muskingum will be confidently wrong and a full dynamic-wave model is required. Second, the parameters are calibrated over the range of the observed events, and X in particular changes when the flood goes overbank and the floodplain begins to store water; extrapolating a set of parameters fitted to in-bank floods to a design event that spills onto the floodplain will under-estimate attenuation. Related to both is the constraint that K and $\Delta t$ are locked together by the stability window above, so refining the time step is not free — the reach may need subdividing.

Part (ii) — Methodology for predicting a flood event, and two calibration and verification methods (8 marks)

A defensible flood prediction proceeds through six stages:

1. Define the design event and the objective. Fix the return period or the design flood standard from the consequence of failure — for a dam, through the Canadian Dam Association consequence classification, which can require the probable maximum flood for a very high consequence structure. Decide whether a peak flow, a full hydrograph, or an inundation extent is needed, because that determines everything downstream.
2. Assemble and screen the data. Precipitation and temperature records, the annual maximum series from HYDAT, DEM, soils and land cover, channel and floodplain geometry, and any reservoir operating rules. Screen the flow record for homogeneity, stationarity and rating-curve shifts before using it.
3. Generate the flood-producing input. Either statistically, by frequency analysis of the annual maximum series to get the design peak directly, or by design storm: build the storm from IDF curves with an appropriate temporal pattern and areal reduction factor, and for a spring event include snowmelt using a degree-day or energy-balance melt model.
4. Transform rainfall to runoff. Remove losses (SCS curve number, or Green–Ampt), convert the excess to a direct-runoff hydrograph by unit-hydrograph convolution or a kinematic-wave routine for each subbasin, and add baseflow.
5. Route the flood downstream. Combine subbasin hydrographs at confluences, route through channel reaches by Muskingum–Cunge, and route through reservoirs and lakes by level-pool storage indication. Where backwater or floodplain storage governs, use a hydraulic model instead.
6. Map and interpret. Convert the routed peak to water levels through a hydraulic (HEC-RAS type) model, delineate the floodplain, and state the uncertainty in the estimate.

Two methods of calibration and verification.

1. Split-sample calibration and verification against gauged hydrographs. The historical record is divided into two independent periods. On the calibration period, model parameters — loss rates, unit-hydrograph lag, K and X, roughness — are adjusted, either manually or by automatic optimisation, until the computed hydrographs match the observed ones on peak magnitude, time to peak and total volume, judged by an objective criterion such as the Nash–Sutcliffe efficiency. The parameters are then frozen and run on the verification period, which the calibration never saw. Only the performance in that second period is evidence that the model will work on a future event; agreement on the calibration period alone proves only that the model has enough free parameters.

2. Verification against independent physical evidence of past floods. Surveyed high-water marks, debris lines, damage records, dated flood photographs and newspaper accounts provide flood levels for events that predate the gauge record, and can be converted to discharges by a slope-area computation using Manning’s equation over a surveyed reach. Palaeoflood evidence — slack-water sediment deposits and flood-scarred trees — extends the record further still. Because these events are typically much larger than anything in the systematic record, they test the model precisely in the extrapolated range where design decisions are made, which is exactly where split-sample calibration on ordinary events gives least assurance.

A third check is worth naming: comparing the model’s frequency curve with regional flood-frequency relations from hydrologically similar gauged basins, which catches gross errors in area, loss parameters or units.

Part (iii) — Significance and dimensions of the terms in the spillway equation (5 marks)

Given. The standard spillway discharge relation $Q = C L H^{3/2}$, with an illustrative ogee crest of length L = 30 m operating under a head H = 1.50 m with a discharge coefficient C = 2.20 in SI units.

Find. The physical significance and dimensions of each term, and the discharge for the illustrative case.

Q — discharge over the spillway. The volumetric flow rate passing the crest, in m³/s, dimensions L³T⁻¹. This is the design output: it must equal or exceed the routed peak outflow produced by the inflow design flood, since a spillway of insufficient capacity forces the reservoir level above the dam crest.

L — effective crest length. The width of the overflow section measured perpendicular to the flow, in metres, dimension L. “Effective” matters: piers and abutments contract the flow, so the effective length is less than the gross length, conventionally $L_e = L - 2(N K_p + K_a)H$ where N is the number of piers. Discharge is directly proportional to L, so capacity can be bought linearly by widening the crest — the usual design lever when head is constrained by dam height.

H — total head on the crest. The height of the energy line above the crest, in metres, dimension L — that is, the static head plus the approach velocity head $V^2/2g$. The exponent 3/2 arises from integrating the free-discharge velocity $\sqrt{2gh}$ over the depth of the nappe, and it is the most important feature of the equation for design: discharge is far more sensitive to head than to any other term.

C — discharge coefficient. It absorbs the crest geometry, approach conditions and energy losses. It is not dimensionless in this form: with $[Q] = L^3T^{-1}$, $[L] = L$ and $[H^{3/2}] = L^{3/2}$, dimensional homogeneity requires

so C carries units of m^1/2/s and its numerical value depends on the unit system — typically 2.1–2.3 m^1/2/s for a well-designed ogee crest at its design head, against about 1.7 for a broad-crested weir. Quoting a coefficient from a US source without converting is a classic and expensive error, since the same crest has C ≈ 3.97 in ft^1/2/s. Its value also rises modestly above the design head, as the nappe separates slightly and sub-atmospheric pressure develops on the crest.

1. Evaluate the illustrative discharge and test the sensitivity to head. Substituting, $$Q = C L H^{3/2} = 2.20 \times 30 \times (1.50)^{3/2} = 2.20 \times 30 \times 1.837$$ $$\boxed{Q = 121\ \text{m}^{3}\text{/s}}$$ Doubling the head to 3.00 m multiplies the discharge not by 2 but by $2^{3/2} = 2.83$, to 343 m³/s. This non-linearity is the reason a modest increase in freeboard buys a large increase in spillway capacity, and equally the reason that under-estimating the design flood is dangerous: the reservoir level required to pass it rises much more slowly than the flow, so the failure margin is consumed faster than intuition suggests.