Flood Frequency Analysis, Extreme Values and Hydrologic Risk

Case study: Regulated river flood frequency

A 38-year annual-maximum record includes two periods with different upstream reservoir operations. Pooling them yields a mean and standard deviation, but the stationarity assumption is not established. The 100-year quantile may be convenient to plot yet scientifically ambiguous without a treatment of regulation and uncertainty.

Investigation task. Fit a classroom Gumbel estimate, then write two conflicting explanations for the apparent tail change and identify evidence distinguishing them.

The unit as part of a watershed investigation

This unit asks an engineer to defend tail inference as a quantitative argument rather than a number inserted into a generic formula. The natural control volume is the selected hydrologic subsystem and its actual inflow and outflow boundaries. Write the unit of every parameter and specify the temporal support of the observations; an hourly intensity, a daily depth and an annual volume cannot be substituted for each other without integration. A concept sketch should show the site boundaries, measurement stations, surface and subsurface pathways, relevant human infrastructure and the time interval over which the process operates. Interpret the final result relative to the real decision at regulated river flood frequency, not merely against an arbitrary calculator threshold.

Field evidence, scale and uncertainty

The most important practical complication in flood frequency and risk is that model inputs are estimates of spatially and temporally varying states. Two instruments reporting the same variable may not have comparable averaging periods or calibration. The impact of an uncertain coefficient depends on the derivative of the output with respect to that coefficient and on where the operating point falls relative to a threshold. If a predicted response appears robust, deliberately test a neighboring process that the simple model omits: channel blockage, antecedent saturation, high tailwater, sensor drift, groundwater exchange or a different storm shape. Record failures of the model openly, rather than editing the inputs until a visually pleasing curve appears.

Professional inference standard

An educational simulation is useful when it generates a falsifiable prediction. Before using any of the four lessons to make a design claim, identify an independently measurable output, a second approximation with different assumptions and a failure consequence. For regulated river flood frequency, a field campaign might require time-synchronized precipitation and flow observations, a georeferenced inventory of hydraulic structures, separate dry- and wet-weather site visits and a provenance ledger for every parameter. The exercises below are intentionally synthetic and no parameter set is asserted to describe a real Bangladesh, US or other watershed. Local codes and agency-approved hydrologic methods supersede these teaching screens.

Lesson 6.1 — Peak-flow records and annual maxima

Hydrological mechanism

Flood frequency inference begins with a consistent definition of independent events and a trustworthy streamflow series. Annual maxima and peaks over threshold support different statistical models; treating multiple correlated peaks as independent biases inference. Changes in river regulation, land cover and gauge rating may make a long record nonhomogeneous. The engineering objective is not merely fitting a smooth curve but establishing what the fitted tail is allowed to represent.

Governing expression and dimensions

Teaching relation: AEP=p and return period T=1/p for a stationary event definition. Every term must be traced to its physical definition before numerical evaluation. A hydrologic depth expressed in millimeters corresponds to one thousand cubic meters for every square kilometer of area. A discharge expressed in cubic meters per second becomes volume only after integrating over seconds. A rate reported over a nominal day or hour is not a spatial average by itself. State whether the equation is a conservation balance, idealization, calibrated empirical law or fitted statistical relation; that classification determines what evidence is needed before transfer to a new catchment.

Units and source audit. For peak-flow records and annual maxima, list the observed quantities, estimated coefficients and chosen reference conditions separately. Include the dates and methods used to obtain each. Calculate one reference state using all quantities in compatible units and round only after a dimensional check. Then reverse a sign convention or convert to a different set of units and verify that the physical meaning survives. A unit-consistent equation may still be wrong because its assumptions do not match the setting.

Applied decision: Evaluate the sensitivity of a flood estimate to a newly discovered rating change

The central decision is not whether a formula can be evaluated; it is whether the value carries information sufficient for an engineering action. In this situation, first identify the outlet, observation period and decision variable. Establish a low, central and high forcing scenario using values that can be defended by measurements or explicitly labeled hypotheses. Predict how the response changes before running the model. A prediction that contradicts conservation of water, plausible timing or an independently measured constraint requires investigation even when the graph looks smooth.

For this particular case the dominant caution is that censored historical observations and mixed populations can distort the sample. Assess which boundary condition is most likely to be violated. For an event analysis, compare the timing of inputs and outputs; for a seasonal balance, separate fluxes from storage; for statistical estimates, distinguish frequency assumptions from observed counts. Do not claim a coefficient calibrated in one environment is universal. State one missing measurement and the practical difficulty of obtaining it.

Mechanism checks and falsification

Counterfactual A — scale. Suppose the area or response time is changed while keeping some input depth or intensity fixed. Decide whether the output should scale linearly, sublinearly or nonlinearly, and explain which physical assumption controls that conclusion. For peak-flow records and annual maxima, the relevant limiting cases include a nearly dry condition, a very large forcing and a coefficient approaching the edge of its documented range. A model that produces negative water depth, impossible storage or negative coefficients without a clear warning is not fit for student use.

Counterfactual B — observation. Ask how a stage recorder, pluviometer, soil-moisture probe or properly maintained flow measurement could disprove the chosen interpretation. Compare two alternatives: an error in the input and an error in the process structure. Error in a timestamp may alter the inferred lag without changing storm totals; an unmodeled flow diversion may alter volume without a rainfall error. Use these fingerprints to choose the next data collection action.

Cross-topic connection. This lesson connects to extreme-value screening and gumbel quantiles because a model transfers its assumptions through the wider analysis. It also affects uncertainty, nonstationarity and decision thresholds, which may rely on the same boundary records but at a different temporal or spatial resolution. A strong report shows these dependencies explicitly, identifies the quantities shared by two calculations and avoids treating correlated errors as independent measurements.

Field, laboratory and spreadsheet workflow

1. Draw a control-volume diagram for evaluate the sensitivity of a flood estimate to a newly discovered rating change; mark every measured and assumed boundary.

1. Calculate the result from AEP=p and return period T=1/p for a stationary event definition where its inputs are available. Show the time-base and area conversions with intermediate numbers rather than only the software answer.

1. Perform a one-at-a-time sensitivity screen for the most uncertain input. State which physical relationship predicts the direction of the response; explain any threshold or plateau.

1. Design an independent field check. Plot a record chronologically before fitting its frequency distribution. Explain exactly what would count as disagreement and how you would respond before changing model parameters.

1. Write a limitation paragraph and a separate decision paragraph. The limitation describes what has not been established; the decision identifies the next action that is justified by the evidence available.

Worked investigative thought experiment

Consider two plausible observations from evaluate the sensitivity of a flood estimate to a newly discovered rating change. In the first, the measured quantity changes noticeably after a storm, whereas in the second, a neighboring instrument records a muted response. One explanation is spatial variability in forcing; another is storage and travel-time filtering. A third is mismatched sampling intervals. Quantify the first explanation with an appropriately scaled synthetic input, then test the second through a changed process coefficient. Keep the third as a data-quality hypothesis until the metadata are checked. This separation protects against false certainty: a model that can fit both instruments with different parameters has not identified the underlying cause.

In-class questions and solutions guidance

Question 1. In the context of evaluate the sensitivity of a flood estimate to a newly discovered rating change, identify all inputs and outputs of the selected hydrologic control volume. Answer guidance: give measured depths, rates, storage changes and known management transfers; justify the time interval and boundary instead of simply listing variables.

Question 2. Under what conditions is AEP=p and return period T=1/p for a stationary event definition defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with censored historical observations and mixed populations can distort the sample.

Question 3. How could you distinguish genuine process sensitivity from measurement bias? Answer guidance: formulate competing mechanisms, specify an independent sensor or additional event, and compare timing, volume and directional response.

Question 4. What changes if the same equation is transferred from a small engineered catchment to a large heterogeneous river basin? Answer guidance: evaluate spatial averaging, storage, travel time, changing active area, scale dependence and available calibration evidence.

Extended engineering reflection

A credible result for peak-flow records and annual maxima needs more than an apparently precise scalar. Provide a unit table, parameter source table, version of the governing equation, uncertainty register and plot of response against at least one input. Compare the output to a basic conservation or range check; no negative runoff depth can be created from positive water inputs without explaining a signed flux convention. Consider whether the decision is reversible, such as adjusting a monitoring schedule, or hard to reverse, such as constructing a hydraulic structure. For high-consequence decisions, missing data should lead to targeted measurement and professional review rather than a higher number of decimal places.

Lesson 6.2 — Extreme-value screening and Gumbel quantiles

Hydrological mechanism

Gumbel quantiles provide a transparent introduction to an extreme-value distribution. A method-of-moments fit translates a sample mean and standard deviation into location and scale, then computes a quantile for an assumed annual exceedance probability. This is only a classroom comparison, not a universal choice for regulated flood estimates. Tail uncertainty, skewness, outliers and the adopted regional method can be decisive.

Governing expression and dimensions

Teaching relation: Q_T=mean+(sqrt(6)/pi) sd[-ln(-ln(1-1/T))-gamma]. Every term must be traced to its physical definition before numerical evaluation. A hydrologic depth expressed in millimeters corresponds to one thousand cubic meters for every square kilometer of area. A discharge expressed in cubic meters per second becomes volume only after integrating over seconds. A rate reported over a nominal day or hour is not a spatial average by itself. State whether the equation is a conservation balance, idealization, calibrated empirical law or fitted statistical relation; that classification determines what evidence is needed before transfer to a new catchment.

Units and source audit. For extreme-value screening and gumbel quantiles, list the observed quantities, estimated coefficients and chosen reference conditions separately. Include the dates and methods used to obtain each. Calculate one reference state using all quantities in compatible units and round only after a dimensional check. Then reverse a sign convention or convert to a different set of units and verify that the physical meaning survives. A unit-consistent equation may still be wrong because its assumptions do not match the setting.

Applied decision: Screen the spread between 10-year and 100-year illustrative floods

The central decision is not whether a formula can be evaluated; it is whether the value carries information sufficient for an engineering action. In this situation, first identify the outlet, observation period and decision variable. Establish a low, central and high forcing scenario using values that can be defended by measurements or explicitly labeled hypotheses. Predict how the response changes before running the model. A prediction that contradicts conservation of water, plausible timing or an independently measured constraint requires investigation even when the graph looks smooth.

For this particular case the dominant caution is that gumbel shape is restrictive relative to other flood-frequency families. Assess which boundary condition is most likely to be violated. For an event analysis, compare the timing of inputs and outputs; for a seasonal balance, separate fluxes from storage; for statistical estimates, distinguish frequency assumptions from observed counts. Do not claim a coefficient calibrated in one environment is universal. State one missing measurement and the practical difficulty of obtaining it.

Mechanism checks and falsification

Counterfactual A — scale. Suppose the area or response time is changed while keeping some input depth or intensity fixed. Decide whether the output should scale linearly, sublinearly or nonlinearly, and explain which physical assumption controls that conclusion. For extreme-value screening and gumbel quantiles, the relevant limiting cases include a nearly dry condition, a very large forcing and a coefficient approaching the edge of its documented range. A model that produces negative water depth, impossible storage or negative coefficients without a clear warning is not fit for student use.

Counterfactual B — observation. Ask how a stage recorder, pluviometer, soil-moisture probe or properly maintained flow measurement could disprove the chosen interpretation. Compare two alternatives: an error in the input and an error in the process structure. Error in a timestamp may alter the inferred lag without changing storm totals; an unmodeled flow diversion may alter volume without a rainfall error. Use these fingerprints to choose the next data collection action.

Cross-topic connection. This lesson connects to flood hazards, exposure and consequences because a model transfers its assumptions through the wider analysis. It also affects peak-flow records and annual maxima, which may rely on the same boundary records but at a different temporal or spatial resolution. A strong report shows these dependencies explicitly, identifies the quantities shared by two calculations and avoids treating correlated errors as independent measurements.

Field, laboratory and spreadsheet workflow

1. Draw a control-volume diagram for screen the spread between 10-year and 100-year illustrative floods; mark every measured and assumed boundary.

1. Calculate the result from Q_T=mean+(sqrt(6)/pi) sd[-ln(-ln(1-1/T))-gamma] where its inputs are available. Show the time-base and area conversions with intermediate numbers rather than only the software answer.

1. Perform a one-at-a-time sensitivity screen for the most uncertain input. State which physical relationship predicts the direction of the response; explain any threshold or plateau.

1. Design an independent field check. Explain why longer record length does not automatically eliminate model uncertainty. Explain exactly what would count as disagreement and how you would respond before changing model parameters.

1. Write a limitation paragraph and a separate decision paragraph. The limitation describes what has not been established; the decision identifies the next action that is justified by the evidence available.

Worked investigative thought experiment

Consider two plausible observations from screen the spread between 10-year and 100-year illustrative floods. In the first, the measured quantity changes noticeably after a storm, whereas in the second, a neighboring instrument records a muted response. One explanation is spatial variability in forcing; another is storage and travel-time filtering. A third is mismatched sampling intervals. Quantify the first explanation with an appropriately scaled synthetic input, then test the second through a changed process coefficient. Keep the third as a data-quality hypothesis until the metadata are checked. This separation protects against false certainty: a model that can fit both instruments with different parameters has not identified the underlying cause.

In-class questions and solutions guidance

Question 1. In the context of screen the spread between 10-year and 100-year illustrative floods, identify all inputs and outputs of the selected hydrologic control volume. Answer guidance: give measured depths, rates, storage changes and known management transfers; justify the time interval and boundary instead of simply listing variables.

Question 2. Under what conditions is Q_T=mean+(sqrt(6)/pi) sd[-ln(-ln(1-1/T))-gamma] defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with gumbel shape is restrictive relative to other flood-frequency families.

Question 3. How could you distinguish genuine process sensitivity from measurement bias? Answer guidance: formulate competing mechanisms, specify an independent sensor or additional event, and compare timing, volume and directional response.

Question 4. What changes if the same equation is transferred from a small engineered catchment to a large heterogeneous river basin? Answer guidance: evaluate spatial averaging, storage, travel time, changing active area, scale dependence and available calibration evidence.

Extended engineering reflection

A credible result for extreme-value screening and gumbel quantiles needs more than an apparently precise scalar. Provide a unit table, parameter source table, version of the governing equation, uncertainty register and plot of response against at least one input. Compare the output to a basic conservation or range check; no negative runoff depth can be created from positive water inputs without explaining a signed flux convention. Consider whether the decision is reversible, such as adjusting a monitoring schedule, or hard to reverse, such as constructing a hydraulic structure. For high-consequence decisions, missing data should lead to targeted measurement and professional review rather than a higher number of decimal places.

Lesson 6.3 — Flood hazards, exposure and consequences

Hydrological mechanism

Flood hazard describes physical conditions such as inundation depth, velocity, duration and sediment loads. Exposure counts people and assets potentially affected; vulnerability concerns how they respond. A high-flow probability by itself is not a social risk estimate. Levee overtopping and floodplain encroachment may create threshold consequences, so mapping probability without failure pathways can be misleading.

Governing expression and dimensions

Teaching relation: Expected annual damage is an integral over event likelihood and consequence, not simply Q times cost. Every term must be traced to its physical definition before numerical evaluation. A hydrologic depth expressed in millimeters corresponds to one thousand cubic meters for every square kilometer of area. A discharge expressed in cubic meters per second becomes volume only after integrating over seconds. A rate reported over a nominal day or hour is not a spatial average by itself. State whether the equation is a conservation balance, idealization, calibrated empirical law or fitted statistical relation; that classification determines what evidence is needed before transfer to a new catchment.

Units and source audit. For flood hazards, exposure and consequences, list the observed quantities, estimated coefficients and chosen reference conditions separately. Include the dates and methods used to obtain each. Calculate one reference state using all quantities in compatible units and round only after a dimensional check. Then reverse a sign convention or convert to a different set of units and verify that the physical meaning survives. A unit-consistent equation may still be wrong because its assumptions do not match the setting.

Applied decision: Prioritize protection options for a school, road and treatment plant

The central decision is not whether a formula can be evaluated; it is whether the value carries information sufficient for an engineering action. In this situation, first identify the outlet, observation period and decision variable. Establish a low, central and high forcing scenario using values that can be defended by measurements or explicitly labeled hypotheses. Predict how the response changes before running the model. A prediction that contradicts conservation of water, plausible timing or an independently measured constraint requires investigation even when the graph looks smooth.

For this particular case the dominant caution is that protecting one location may displace floodwater or risk to another. Assess which boundary condition is most likely to be violated. For an event analysis, compare the timing of inputs and outputs; for a seasonal balance, separate fluxes from storage; for statistical estimates, distinguish frequency assumptions from observed counts. Do not claim a coefficient calibrated in one environment is universal. State one missing measurement and the practical difficulty of obtaining it.

Mechanism checks and falsification

Counterfactual A — scale. Suppose the area or response time is changed while keeping some input depth or intensity fixed. Decide whether the output should scale linearly, sublinearly or nonlinearly, and explain which physical assumption controls that conclusion. For flood hazards, exposure and consequences, the relevant limiting cases include a nearly dry condition, a very large forcing and a coefficient approaching the edge of its documented range. A model that produces negative water depth, impossible storage or negative coefficients without a clear warning is not fit for student use.

Counterfactual B — observation. Ask how a stage recorder, pluviometer, soil-moisture probe or properly maintained flow measurement could disprove the chosen interpretation. Compare two alternatives: an error in the input and an error in the process structure. Error in a timestamp may alter the inferred lag without changing storm totals; an unmodeled flow diversion may alter volume without a rainfall error. Use these fingerprints to choose the next data collection action.

Cross-topic connection. This lesson connects to uncertainty, nonstationarity and decision thresholds because a model transfers its assumptions through the wider analysis. It also affects extreme-value screening and gumbel quantiles, which may rely on the same boundary records but at a different temporal or spatial resolution. A strong report shows these dependencies explicitly, identifies the quantities shared by two calculations and avoids treating correlated errors as independent measurements.

Field, laboratory and spreadsheet workflow

1. Draw a control-volume diagram for prioritize protection options for a school, road and treatment plant; mark every measured and assumed boundary.

1. Calculate the result from Expected annual damage is an integral over event likelihood and consequence, not simply Q times cost where its inputs are available. Show the time-base and area conversions with intermediate numbers rather than only the software answer.

1. Perform a one-at-a-time sensitivity screen for the most uncertain input. State which physical relationship predicts the direction of the response; explain any threshold or plateau.

1. Design an independent field check. Separate hazard, exposure, vulnerability and residual risk in a memo. Explain exactly what would count as disagreement and how you would respond before changing model parameters.

1. Write a limitation paragraph and a separate decision paragraph. The limitation describes what has not been established; the decision identifies the next action that is justified by the evidence available.

Worked investigative thought experiment

Consider two plausible observations from prioritize protection options for a school, road and treatment plant. In the first, the measured quantity changes noticeably after a storm, whereas in the second, a neighboring instrument records a muted response. One explanation is spatial variability in forcing; another is storage and travel-time filtering. A third is mismatched sampling intervals. Quantify the first explanation with an appropriately scaled synthetic input, then test the second through a changed process coefficient. Keep the third as a data-quality hypothesis until the metadata are checked. This separation protects against false certainty: a model that can fit both instruments with different parameters has not identified the underlying cause.

In-class questions and solutions guidance

Question 1. In the context of prioritize protection options for a school, road and treatment plant, identify all inputs and outputs of the selected hydrologic control volume. Answer guidance: give measured depths, rates, storage changes and known management transfers; justify the time interval and boundary instead of simply listing variables.

Question 2. Under what conditions is Expected annual damage is an integral over event likelihood and consequence, not simply Q times cost defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with protecting one location may displace floodwater or risk to another.

Question 3. How could you distinguish genuine process sensitivity from measurement bias? Answer guidance: formulate competing mechanisms, specify an independent sensor or additional event, and compare timing, volume and directional response.

Question 4. What changes if the same equation is transferred from a small engineered catchment to a large heterogeneous river basin? Answer guidance: evaluate spatial averaging, storage, travel time, changing active area, scale dependence and available calibration evidence.

Extended engineering reflection

A credible result for flood hazards, exposure and consequences needs more than an apparently precise scalar. Provide a unit table, parameter source table, version of the governing equation, uncertainty register and plot of response against at least one input. Compare the output to a basic conservation or range check; no negative runoff depth can be created from positive water inputs without explaining a signed flux convention. Consider whether the decision is reversible, such as adjusting a monitoring schedule, or hard to reverse, such as constructing a hydraulic structure. For high-consequence decisions, missing data should lead to targeted measurement and professional review rather than a higher number of decimal places.

Lesson 6.4 — Uncertainty, nonstationarity and decision thresholds

Hydrological mechanism

Hydrological uncertainty includes observations, parameter estimation, process structure, spatial transfer and changing future forcing. Sensitivity analysis explores local response; ensembles and scenarios explore a wider range of plausible conditions. A design decision depends on the cost of failure, adaptability and local regulations rather than one universally optimal return period. Governance requires versioned data, documented assumptions and triggers for revisiting a model.

Governing expression and dimensions

Teaching relation: Decision risk depends on hazard, exposure, vulnerability and allowed consequences. Every term must be traced to its physical definition before numerical evaluation. A hydrologic depth expressed in millimeters corresponds to one thousand cubic meters for every square kilometer of area. A discharge expressed in cubic meters per second becomes volume only after integrating over seconds. A rate reported over a nominal day or hour is not a spatial average by itself. State whether the equation is a conservation balance, idealization, calibrated empirical law or fitted statistical relation; that classification determines what evidence is needed before transfer to a new catchment.

Units and source audit. For uncertainty, nonstationarity and decision thresholds, list the observed quantities, estimated coefficients and chosen reference conditions separately. Include the dates and methods used to obtain each. Calculate one reference state using all quantities in compatible units and round only after a dimensional check. Then reverse a sign convention or convert to a different set of units and verify that the physical meaning survives. A unit-consistent equation may still be wrong because its assumptions do not match the setting.

Applied decision: Compare two floodwall options with different maintenance burdens

The central decision is not whether a formula can be evaluated; it is whether the value carries information sufficient for an engineering action. In this situation, first identify the outlet, observation period and decision variable. Establish a low, central and high forcing scenario using values that can be defended by measurements or explicitly labeled hypotheses. Predict how the response changes before running the model. A prediction that contradicts conservation of water, plausible timing or an independently measured constraint requires investigation even when the graph looks smooth.

For this particular case the dominant caution is that tail extrapolation can imply unjustified precision in benefit-cost comparisons. Assess which boundary condition is most likely to be violated. For an event analysis, compare the timing of inputs and outputs; for a seasonal balance, separate fluxes from storage; for statistical estimates, distinguish frequency assumptions from observed counts. Do not claim a coefficient calibrated in one environment is universal. State one missing measurement and the practical difficulty of obtaining it.

Mechanism checks and falsification

Counterfactual A — scale. Suppose the area or response time is changed while keeping some input depth or intensity fixed. Decide whether the output should scale linearly, sublinearly or nonlinearly, and explain which physical assumption controls that conclusion. For uncertainty, nonstationarity and decision thresholds, the relevant limiting cases include a nearly dry condition, a very large forcing and a coefficient approaching the edge of its documented range. A model that produces negative water depth, impossible storage or negative coefficients without a clear warning is not fit for student use.

Counterfactual B — observation. Ask how a stage recorder, pluviometer, soil-moisture probe or properly maintained flow measurement could disprove the chosen interpretation. Compare two alternatives: an error in the input and an error in the process structure. Error in a timestamp may alter the inferred lag without changing storm totals; an unmodeled flow diversion may alter volume without a rainfall error. Use these fingerprints to choose the next data collection action.

Cross-topic connection. This lesson connects to peak-flow records and annual maxima because a model transfers its assumptions through the wider analysis. It also affects flood hazards, exposure and consequences, which may rely on the same boundary records but at a different temporal or spatial resolution. A strong report shows these dependencies explicitly, identifies the quantities shared by two calculations and avoids treating correlated errors as independent measurements.

Field, laboratory and spreadsheet workflow

1. Draw a control-volume diagram for compare two floodwall options with different maintenance burdens; mark every measured and assumed boundary.

1. Calculate the result from Decision risk depends on hazard, exposure, vulnerability and allowed consequences where its inputs are available. Show the time-base and area conversions with intermediate numbers rather than only the software answer.

1. Perform a one-at-a-time sensitivity screen for the most uncertain input. State which physical relationship predicts the direction of the response; explain any threshold or plateau.

1. Design an independent field check. Create an evidence matrix distinguishing verified facts, scenarios and policy assumptions. Explain exactly what would count as disagreement and how you would respond before changing model parameters.

1. Write a limitation paragraph and a separate decision paragraph. The limitation describes what has not been established; the decision identifies the next action that is justified by the evidence available.

Worked investigative thought experiment

Consider two plausible observations from compare two floodwall options with different maintenance burdens. In the first, the measured quantity changes noticeably after a storm, whereas in the second, a neighboring instrument records a muted response. One explanation is spatial variability in forcing; another is storage and travel-time filtering. A third is mismatched sampling intervals. Quantify the first explanation with an appropriately scaled synthetic input, then test the second through a changed process coefficient. Keep the third as a data-quality hypothesis until the metadata are checked. This separation protects against false certainty: a model that can fit both instruments with different parameters has not identified the underlying cause.

In-class questions and solutions guidance

Question 1. In the context of compare two floodwall options with different maintenance burdens, identify all inputs and outputs of the selected hydrologic control volume. Answer guidance: give measured depths, rates, storage changes and known management transfers; justify the time interval and boundary instead of simply listing variables.

Question 2. Under what conditions is Decision risk depends on hazard, exposure, vulnerability and allowed consequences defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with tail extrapolation can imply unjustified precision in benefit-cost comparisons.

Question 3. How could you distinguish genuine process sensitivity from measurement bias? Answer guidance: formulate competing mechanisms, specify an independent sensor or additional event, and compare timing, volume and directional response.

Question 4. What changes if the same equation is transferred from a small engineered catchment to a large heterogeneous river basin? Answer guidance: evaluate spatial averaging, storage, travel time, changing active area, scale dependence and available calibration evidence.

Extended engineering reflection

A credible result for uncertainty, nonstationarity and decision thresholds needs more than an apparently precise scalar. Provide a unit table, parameter source table, version of the governing equation, uncertainty register and plot of response against at least one input. Compare the output to a basic conservation or range check; no negative runoff depth can be created from positive water inputs without explaining a signed flux convention. Consider whether the decision is reversible, such as adjusting a monitoring schedule, or hard to reverse, such as constructing a hydraulic structure. For high-consequence decisions, missing data should lead to targeted measurement and professional review rather than a higher number of decimal places.

Applied project 6 — Regulated river flood frequency

A 38-year annual-maximum record includes two periods with different upstream reservoir operations. Pooling them yields a mean and standard deviation, but the stationarity assumption is not established. The 100-year quantile may be convenient to plot yet scientifically ambiguous without a treatment of regulation and uncertainty.

Project deliverable. Fit a classroom Gumbel estimate, then write two conflicting explanations for the apparent tail change and identify evidence distinguishing them. Include an explicit data dictionary, local measurement geometry, at least two competing model structures, a water-balance closure or proxy and a sensitivity analysis. Use the two paired laboratories for this unit as exploratory tools only. A peer should be able to reproduce the results using the included synthetic inputs and code without requesting undocumented assumptions.

Unit review and source-of-error audit

Before claiming success, ask whether the result is constrained by water conservation, the measurement period, the sampling interval and the domain of the empirical relation. Identify how a missing observation could change a decision, not merely how it changes the last decimal digit. The strongest student reports explain why a model is unsuitable for a particular site, include a second defensible interpretation and propose a feasible experiment to choose between them.