16-Civ-B4 Engineering Hydrology · May 2015
Nivaar worked solution (AI-drafted; not reviewed by a licensed engineer)
Paper format. National Examination, May 2015, 98-Civ-B4 Engineering Hydrology, three hours’ duration, closed book with one two-sided candidate-prepared aid sheet (8½″ × 11″) and an approved Casio or Sharp calculator whose model must be declared in the work book. Seven problems are printed; page-1 Note 4 states that any five (5) questions constitute a complete paper and that only the first five answers appearing in the work book are marked. Each problem is weighted at twenty (20) points, so the examinable total is 5 × 20 = 100 points. The page-6 marking scheme gives the sub-part split for every problem. All seven problems are solved here, because this set is a study resource rather than a timed sitting; the sub-part mark values shown below are the printed ones.
Reference texts. V. T. Chow, D. R. Maidment and L. W. Mays, Applied Hydrology (hydrologic cycle, unit hydrographs, routing, frequency analysis, infiltration); L. W. Mays, Water Resources Engineering, 3rd ed. (design application, rainfall–runoff, reservoir operation); W. Viessman and G. L. Lewis, Introduction to Hydrology, 5th ed. (measurement, areal precipitation, energy budget); V. T. Chow, Open-Channel Hydraulics (1959) (flood-wave propagation, dam-break and gradually varied unsteady flow); C. W. Fetter, Applied Hydrogeology, 4th ed. (Darcy’s law, hydraulic conductivity, advective transport). Canadian practice references: Environment and Climate Change Canada / Water Survey of Canada HYDAT archive and the ECCC Engineering Climate Datasets (IDF curves and the IDF_CC climate-adjustment tool); ISO 1100 / WMO Manual on Stream Gauging (stage–discharge rating practice); and the Canadian Dam Association Dam Safety Guidelines (inflow design flood, dam-break inundation mapping).
Check — Problem 7(i) source data are internally inconsistent. As printed, the 10 000 km2 basin receives 50 mm of rain in a year while its river carries 200 m3/s, which is a runoff depth of 630.72 mm — about 12.6 times the stated rainfall. No basin can discharge more water than it receives, so the printed precipitation is in error (almost certainly a lost order of magnitude), not the discharge. Problem 7(i) below boxes the runoff depth first, demonstrates that the balance cannot close, and then adopts a declared corrected annual precipitation of 1000 mm/a to complete the estimate. Every number is flagged where the correction is used.
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.
Note on the printed marking scheme: page 6 lists Problem 6 as “(i) 5, (ii) 5 (iii) 5, (iv) 5 marks, 6 marks total”. The stated total is a typographical error — the four sub-parts sum to 20, consistent with every other problem on the paper and with the page-1 statement that each question is weighted at twenty points. The 20-mark total is used here.
An intensity–duration–frequency curve summarises, for one location, the rainfall intensity expected for each combination of storm duration and return period. It is derived in five steps from a long record of continuous rainfall (in Canada, from an ECCC tipping-bucket or weighing-gauge station with at least 20–25 years of record).
First, for each standard duration — 5, 10, 15, 30 min, 1, 2, 6, 12, 24 h — scan the record with a moving window and extract the annual maximum depth for that duration in every year. Second, convert each depth to an intensity, $i = P/t_d$. Third, fit a probability distribution to each duration’s annual-maximum series independently; the Gumbel (EV1) distribution is standard Canadian practice for this purpose. Fourth, evaluate the fitted distribution at the required return periods (2, 5, 10, 25, 50, 100 years) to obtain the intensity for each duration–frequency pair. Fifth, plot intensity against duration on log–log axes with one curve per return period, and fit a smooth analytical form for design use, commonly the Sherman-type equation
$$i = \frac{a}{(t_d + b)^{c}}$$
with $a$, $b$, $c$ fitted per return period. For example, with $a = 1200$, $b = 8$ min and $c = 0.75$, a 30-minute storm gives
$$i = \frac{1200}{(30 + 8)^{0.75}} = \boxed{78.4\ \text{mm/h}} \qquad\Rightarrow\qquad P = 78.4 \times \tfrac{30}{60} = 39.2\ \text{mm}$$
Limitation / assumption. The controlling assumption is stationarity — that the statistical properties of the rainfall record do not change with time, so the past is a valid sample of the future. Under a warming climate that assumption is increasingly untenable: short-duration extreme intensities are expected to increase with atmospheric moisture capacity, so an IDF curve fitted to 1960–2000 data may understate the design intensity for an asset built now and operated to 2075. This is precisely why ECCC publishes the IDF_CC climate-adjustment tool alongside the standard IDF datasets. Two further assumptions deserve mention: the curve is a point rainfall statistic and must be reduced by an areal-reduction factor before being applied to a catchment of any size; and the several durations are fitted independently, so an IDF curve does not describe a single real storm — a design storm profile (Chicago, SCS, or an observed pattern) must be constructed from it before it can be used as model input.
Flood-frequency analysis converts a record of past floods into a probability statement about future ones. The procedure is: extract the annual maximum instantaneous peak discharge from each year of the gauged record (the annual-maximum series; a partial-duration or peaks-over-threshold series is used when the record is short); screen for independence, homogeneity, trend and outliers, and for regulation or diversion that would make the series non-homogeneous; rank the series and assign plotting positions, commonly Weibull,
$$P(X \ge x_m) = \frac{m}{n+1}, \qquad T = \frac{1}{P}$$
so that, in a 40-year record, the third-largest flood plots at $p = 3/41 = 0.073$, or $T = 13.7$ years; fit a probability distribution (Gumbel, log-Pearson Type III, GEV) by moments, L-moments or maximum likelihood; and evaluate the fitted distribution at the required return periods using the frequency-factor form
$$x_T = \bar{x} + K_T\,s$$
where $K_T$ depends on the distribution and the return period. For a Gumbel fit,
$$K_T = -\frac{\sqrt{6}}{\pi}\left[0.5772 + \ln\ln\left(\frac{T}{T-1}\right)\right]$$
With $\bar{x} = 420\ \text{m}^3/\text{s}$ and $s = 95\ \text{m}^3/\text{s}$, the 100-year flood is $K_{100} = 3.137$ and
$$Q_{100} = 420 + 3.137(95) = \boxed{718\ \text{m}^3/\text{s}}$$
What it predicts, and what it does not. The return period is not a schedule — it is the reciprocal of the annual exceedance probability, so a 100-year flood has a 1 % chance in any year regardless of when the last one occurred, and two can occur in consecutive years. What the analysis genuinely delivers is the design flood magnitude for a stated risk level, the floodplain extent when the flood is routed hydraulically, and the risk over a project life:
$$R = 1 - \left(1 - \frac{1}{T}\right)^{n}$$
For the 100-year event over a 50-year design life, $R = 1 - 0.99^{50} = 0.395$ — a 40 % chance of at least one exceedance, which is the number that actually informs a decision about protection level. Confidence limits widen sharply beyond about twice the record length, so estimating a 200-year flood from 30 years of data carries large uncertainty; regional analysis, which pools records from hydrologically similar gauged basins, is the standard remedy and is also how estimates are made at ungauged sites.
Two variables carry most design work: annual maximum flood peak and annual maximum rainfall depth (or intensity) for a fixed duration. Both are maxima of many independent events within a year, so extreme-value theory applies to both, and both are bounded below by zero and positively skewed.
Variable 1 — annual maximum flood peak $Q$, characterised by the Gumbel (EV1) or log-Pearson III distribution. The Gumbel cdf and its inverse are
$$F(x) = \exp\!\left[-\exp\!\left(-\frac{x-u}{\alpha}\right)\right], \qquad x_T = u - \alpha\ln\ln\!\left(\frac{T}{T-1}\right)$$
with $\alpha = \sqrt{6}\,s/\pi$ and $u = \bar{x} - 0.5772\alpha$. Gumbel is the asymptotic distribution of the maximum of many independent samples, which is exactly what an annual flood peak is, and it plots as a straight line on Gumbel probability paper — the visual fit check used in practice.
Variable 2 — annual maximum rainfall depth/intensity $P$ for a fixed duration, characterised by Gumbel or the generalised extreme value (GEV) distribution. The same reasoning applies: it is an annual maximum, so an extreme-value family is the right choice, and the fitted quantiles are what populate the IDF curves of part (i). The GEV generalises Gumbel with a shape parameter $k$,
$$F(x) = \exp\!\left\{-\left[1 - \frac{k(x - u)}{\alpha}\right]^{1/k}\right\}$$
reducing to Gumbel when $k \to 0$, and it is now the preferred fit in Canadian IDF updating because it allows a bounded or heavier upper tail as the data indicate.
In both cases the design step is identical, and is what the two diagrams show: choose the annual exceedance probability $1/T$ that matches the acceptable risk, read the corresponding quantile $x_T$ from the fitted cdf, and use it as the design value. The log-normal distribution is a common alternative for variables that are products of many factors (rainfall depths, low flows), and the exponential or Poisson family is used for the arrival of events rather than their magnitude in partial-duration analysis.
The Pearson Type III (gamma with a location shift) distribution is a three-parameter family described by its mean, standard deviation and skewness. Its importance is exactly that third parameter: real flood series are positively skewed, and a two-parameter distribution such as the normal or Gumbel imposes a fixed skew that the data may not have. Pearson III lets the skew be estimated from the data, so a single family can fit series ranging from nearly symmetric snowmelt-dominated records to strongly skewed convective-flood records.
In practice the log-Pearson Type III (LP3) form is used: the logarithms of the annual peaks are fitted with a Pearson III distribution. The frequency-factor equation is applied in log space,
$$\log Q_T = \overline{\log Q} + K_T\,s_{\log Q}, \qquad Q_T = 10^{\,\log Q_T}$$
where $K_T$ is read from the standard Pearson III table as a function of return period and the coefficient of skewness $C_s$ of the logarithms. When $C_s = 0$, $K_T$ reduces to the standard normal deviate and LP3 collapses to the log-normal distribution — a useful internal check.
Worked illustration. For a station whose log-transformed annual peaks have $\overline{\log Q} = 2.65$, $s = 0.18$ and $C_s = -0.20$, the tabulated $K_{100} = 2.159$, so
$$\log Q_{100} = 2.65 + 2.159(0.18) = 3.0386 \qquad\Rightarrow\qquad Q_{100} = 10^{3.0386} = \boxed{1093\ \text{m}^3/\text{s}}$$
Had the skew been taken as zero (log-normal, $K_{100} = 2.326$), the same data would give 1171 m3/s — about 7 % higher. That sensitivity is the point: the skew coefficient materially changes the design flood, and it is the least reliably estimated of the three parameters from a short record, since sample skew has a large sampling variance. Standard practice (US Bulletin 17B/17C, and the equivalent Canadian regional procedures) therefore weights the station skew with a regional or generalised skew value, weighting by the relative reliability of each. LP3 is the recommended distribution for flood-frequency analysis in a great deal of North American practice for these reasons: it accommodates skew, it constrains the variable to positive values through the log transform, and it fits the upper tail of most annual-maximum flood series well. Its limitations are the usual ones — sensitivity to low and high outliers, the need for a reasonably long record (25 years or more), and the assumptions of independence, homogeneity and stationarity that underlie all frequency analysis.
