Calibration examines agreement with observed events; validation tests representation of other observations. Numerical verification checks computation behaviour. These complement one another but are distinct. Owners should know the supporting observations and limits where none exist. Realistic-looking maps or agreement at one point do not establish domain-wide reliability.
Assess observation quality first
Record date, location, units and references for measured flow, stage, flood marks and extent. Investigate waves, splash or post-event deposition behind marks, and include flow uncertainty. Note photograph times and control points. Do not combine marks from different events as simultaneous levels. State reliability; precise notation cannot remove weak measurement evidence.
Check inputs before adjusting parameters
Investigate terrain, sections, openings, flow and boundaries before changing roughness. Datum or timing errors can be concealed by tuning without being corrected. Check domain coverage, software, equations and settings. Then select parameters and plausible ranges, recording each change so agreement is traceable to evidence and assumptions.
Use multiple criteria rather than one point
Matching one stage does not prove correct velocities or timing. Assess suitable levels, hydrograph shape, peak time, extent and volume at multiple locations. Two models can match a peak but differ in arrival time. Select errors and plots suited to observations, with justified acceptance criteria reflecting data accuracy and purpose. No universal fit percentage replaces physical interpretation.
Independent events and overfitting
Where possible, validate against an unused event or dataset. This exposes overfitting and behaviour under other flows, vegetation or downstream conditions. Re-tuning for every validation observation weakens independence. Investigate poor fit and redefine applicability. State limited evidence; sensitivity tests or benchmark checks without observations are not observational calibration.
Delivering the checking record
Deliver observation inventory, initial model, parameter changes, tested ranges and comparisons with shared time and elevation references. Include balance, warnings and mesh/time sensitivity alongside fit. Identify unchecked areas and flow ranges. Versioned records show which checks need renewal after revisions and preserve the evidence and limits for future engineering use.
Delivering a reproducible model
A hydraulic-model delivery package should include the software version, terrain, geometry, flow files, computation plan and output settings. Screenshots alone do not allow the model to be rerun or later project revisions to be assessed. Define the coordinate system, elevation reference and units clearly in the report. Each scenario's name, data source and changed parameters should be traceable. This allows the design team to identify the calculation conditions associated with every map.
Review volume balance, computation warnings, water-level and flow time series at important locations, and the physical meaning of the results together. A run that finishes without errors is not, by itself, engineering verification. Explain areas with insufficient data and processes the model cannot represent. For the owner, the delivery scope can be defined through the required maps, technical report, model files and any alternative-design comparisons. Sharing the study boundary and available surveys with ArchSu helps describe the modelling requirements realistically.
Sources and related articles
Let's discuss your project's data requirements and reporting scope.
Request a project quote