Hydrologic Simulation Models, Parameter Calibration and Uncertainty

Case study: Hydrological model review board

A model calibrated on one intense storm appears to reproduce the observed peak, but the discharge integral is biased upward and low-flow periods are systematically overpredicted. A high NSE on a few peak-dominated values does not ensure the model is suitable for monthly drought allocation. Different stakeholders will use different parts of the hydrograph.

Investigation task. Create a validation dashboard with volume ratio, peak magnitude, peak time, NSE and uncertainty, and select a safe first decision.

The unit as part of a watershed investigation

This unit asks an engineer to defend validation and transparency 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 hydrological model review board, not merely against an arbitrary calculator threshold.

Field evidence, scale and uncertainty

The most important practical complication in hydrologic modeling and field practice 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 hydrological model review board, 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 8.1 — Model selection, calibration and parameter meaning

Hydrological mechanism

A lumped event model, distributed physics model and empirical statistical model answer different questions. Complexity is not synonymous with accuracy; flexible parameters can compensate for poor rainfall inputs or unrealistic catchment connectivity. Calibration should be constrained by physically plausible values and assessed using metrics for volume, peak, timing and low flow. Transfer to ungauged sites requires honest limitations and independent regional evidence.

Governing expression and dimensions

Teaching relation: NSE=1-sum((Qobs-Qsim)^2)/sum((Qobs-Qmean)^2). 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 model selection, calibration and parameter meaning, 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: Review an overfitted catchment model after urban expansion

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 parameter identifiability cannot be established from a single peak alone. 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 model selection, calibration and parameter meaning, 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 time-step and spatial discretization because a model transfers its assumptions through the wider analysis. It also affects reproducible analysis, ethics and final synthesis, 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 review an overfitted catchment model after urban expansion; mark every measured and assumed boundary.

1. Calculate the result from NSE=1-sum((Qobs-Qsim)^2)/sum((Qobs-Qmean)^2) 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. Hold out at least one distinct event for validation when data permit. 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 review an overfitted catchment model after urban expansion. 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 review an overfitted catchment model after urban expansion, 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 NSE=1-sum((Qobs-Qsim)^2)/sum((Qobs-Qmean)^2) defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with parameter identifiability cannot be established from a single peak alone.

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 model selection, calibration and parameter meaning 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 8.2 — Time-step and spatial discretization

Hydrological mechanism

Hydrology models operate at a chosen temporal and spatial resolution. Storms shorter than the computational step may be smoothed away; very small steps increase processing cost and amplify data requirements. Aggregating heterogeneous terrain into one parameter set can erase crucial connectivity. Convergence checks ask whether predictions stabilize as resolution changes, but an incorrect conceptual model may converge consistently to the wrong answer.

Governing expression and dimensions

Teaching relation: For a time series, event volume approximates sum(Q Delta t). 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 time-step and spatial discretization, 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: Test an hourly versus 5-minute runoff representation

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 sampling frequency, measurement response and model time step are separate concepts. 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 time-step and spatial discretization, 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 model checking, skill and uncertainty because a model transfers its assumptions through the wider analysis. It also affects model selection, calibration and parameter meaning, 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 test an hourly versus 5-minute runoff representation; mark every measured and assumed boundary.

1. Calculate the result from For a time series, event volume approximates sum(Q Delta t) 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. Quantify peak dilution when a short burst is aggregated into a daily total. 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 test an hourly versus 5-minute runoff representation. 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 test an hourly versus 5-minute runoff representation, 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 For a time series, event volume approximates sum(Q Delta t) defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with sampling frequency, measurement response and model time step are separate concepts.

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 time-step and spatial discretization 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 8.3 — Model checking, skill and uncertainty

Hydrological mechanism

Model evaluation should include water-balance closure, peak timing, bias, graphical residuals and skill against a defensible baseline. Nash–Sutcliffe efficiency can become negative; a high value does not guarantee correct extremes. A four-point illustrative exercise reveals how bias changes a score but cannot establish real forecasting skill. Examine performance on conditions not used in calibration and document uncertainty sources.

Governing expression and dimensions

Teaching relation: NSE=1-SSE/SST with nonconstant observations. 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 model checking, skill and uncertainty, 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 biased and unbiased simulations of a small observed hydrograph

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 a single metric is insensitive to some important error structures. 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 model checking, skill and uncertainty, 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 reproducible analysis, ethics and final synthesis because a model transfers its assumptions through the wider analysis. It also affects time-step and spatial discretization, 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 biased and unbiased simulations of a small observed hydrograph; mark every measured and assumed boundary.

1. Calculate the result from NSE=1-SSE/SST with nonconstant observations 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. Report peak error, timing error, volume bias and a simple uncertainty statement. 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 biased and unbiased simulations of a small observed hydrograph. 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 biased and unbiased simulations of a small observed hydrograph, 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 NSE=1-SSE/SST with nonconstant observations defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with a single metric is insensitive to some important error structures.

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 model checking, skill and uncertainty 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 8.4 — Reproducible analysis, ethics and final synthesis

Hydrological mechanism

Reproducible hydrology means preserving the raw data, units, site metadata, software and formulas, calibration settings and every data-cleaning decision. Ethical application communicates what the analysis cannot resolve, especially where flood decisions affect vulnerable communities. A transparent notebook with independent spot checks is more valuable than a persuasive graphic with hidden assumptions. Distinguish public data, locally measured data, synthetic examples and unverified inputs.

Governing expression and dimensions

Teaching relation: Trace input -> transformation -> model -> decision -> independent check. 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 reproducible analysis, ethics and final synthesis, 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: Prepare a watershed report for a community flood committee

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 data licensing and privacy matter when household-level flood exposure is included. 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 reproducible analysis, ethics and final synthesis, 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 model selection, calibration and parameter meaning because a model transfers its assumptions through the wider analysis. It also affects model checking, skill and uncertainty, 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 prepare a watershed report for a community flood committee; mark every measured and assumed boundary.

1. Calculate the result from Trace input -> transformation -> model -> decision -> independent check 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. Publish a model limitations register alongside the simulated results. 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 prepare a watershed report for a community flood committee. 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 prepare a watershed report for a community flood committee, 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 Trace input -> transformation -> model -> decision -> independent check defensible? Answer guidance: specify the governing conservation or calibration assumptions, a physical range of inputs and at least one failure condition associated with data licensing and privacy matter when household-level flood exposure is included.

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 reproducible analysis, ethics and final synthesis 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 8 — Hydrological model review board

A model calibrated on one intense storm appears to reproduce the observed peak, but the discharge integral is biased upward and low-flow periods are systematically overpredicted. A high NSE on a few peak-dominated values does not ensure the model is suitable for monthly drought allocation. Different stakeholders will use different parts of the hydrograph.

Project deliverable. Create a validation dashboard with volume ratio, peak magnitude, peak time, NSE and uncertainty, and select a safe first decision. 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.