Close the gap between your model and the field

epanet-js combines multi-method optimization with zonal decomposition to identify the parameters that best match your observed field data. Review the evidence, inspect the reasoning, and accept adjustments with complete control.

Launch epanet-js

No login needed

Book a demo

Calibration assistant

Suggestion 2 of 7

Zone B · 14 pipes

Roughness 110 95

Observed Simulated

Accept Adjust Reject

Calibration turns a hydraulic model into an operational asset

A model only becomes useful for planning and operational decisions once its simulated heads and flows match real-world observations. Yet traditional calibration is notoriously slow, requiring hours of manually adjusting pipe roughness, tweaking localized demands, checking valve statuses, and comparing error plots across repeated runs.

epanet-js automates the repetitive parameter search while keeping final engineering judgment in your hands.

Import observed field data directly into the model

Calibration begins with field observations. Import your time-series records to visualize discrepancies directly on the network map and plot simulated results alongside observed data:

Pressure logger time series: extended-period pressure logs across critical junctions and network extremities.
DMA inflows and boundary meters: continuous flow monitoring from district meters and custody transfer points.
SCADA records: operational tank levels, pump station discharge pressures, and control valve statuses.
Hydrant and fire flow tests: static pressures, residual pressures, and discharge flow rates during flow tests.

Zonal decomposition prevents compensating errors

Attempting to calibrate a large network as a single global system introduces too many unconstrained variables, often leading to compensating errors where incorrect assumptions in one area mask problems in another.

epanet-js divides the network into discrete DMAs and pressure zones. By solving each zone independently against its boundary flows and internal pressures, parameter searches stay strictly bounded by local evidence.

Zone A Zone B Zone C

Fit by zone

Mean error

Zone A 0.9 m

Zone B 1.4 m

Zone C 3.8 m

Multi-method parameter search

For each zone, epanet-js systematically identifies the parameters (pipe roughness or C-factors, localized demand allocations, and valve statuses) that best reconcile the model with observed data using complementary solvers:

Genetic algorithms: search broad parameter combinations to identify viable candidate network states.
Gradient-based methods: rapidly fine-tune continuous parameters to minimize residual head and flow errors.
Probabilistic inference: quantify uncertainty and evaluate the likelihood of each proposed adjustment.

Transparent reasoning and a complete audit trail

Different algorithms often surface competing candidates. epanet-js evaluates these options against the physical evidence, identifies the most probable parameter set, and clearly documents its reasoning.

Every recommendation is backed by a visible chain of logic: the alternative values considered, the error reduction achieved, and the physical justification for the adjustment. You get a transparent record rather than an unexplained black-box number.

Full engineer-in-the-loop control

The system proposes adjustments; you decide what gets applied. Review each recommendation alongside its confidence metrics and supporting plots, then accept, tweak, or reject individual changes before they are committed to the model.

The result is a calibrated model verified against real-world data and supported by a clear, auditable record of every modification.

Calibration assistant

Suggestion 2 of 7

Zone B · 14 pipes

Roughness 110 95

Observed Simulated

Accept Adjust Reject

Why calibration is different in epanet-js

Models you can trust

A model that matches observed pressures and flows, so the decisions you make from it hold up to scrutiny.

Faster than manual tuning

The search work that used to take days of trial and error becomes a focused review of well-reasoned options.

Every decision documented

Each piece of the network keeps a record of the options considered, their likelihood, and why a value was chosen.

You keep the judgment

Accept, adjust, or reject any recommendation. The tools assist and explain; the engineer decides.

Calibrate with confidence

See assisted calibration on your own network, with the engineer in control of every decision.

Launch epanet-js

No login needed

Book a demo

Where calibration fits into the workflow

Calibration bridges raw network data and reliable engineering decisions. Because epanet-js unifies model building, customer demand allocation, and field test planning in one environment, your calibrated network is immediately ready for scenario testing, fire flow assessments, and planning without exporting files or switching tools.

Where to next

A calibrated model unlocks everything downstream.

Frequently asked questions

The questions teams ask most about calibration in epanet-js.

What does the calibration adjust?: Pipe roughness, demands, and asset status and controls. These are the parameters that most often explain the gap between a model and the field, and they're adjusted within ranges you set so the result stays physically sensible.
What field data can it calibrate against?: Pressure logger time series, metered flows and DMA inflows, SCADA, and one-off hydrant or field tests. The more of your network the observations cover, the more the calibration has to work with.
Is it fully automatic?: No, and that's deliberate. The tools do the searching and propose the most likely values, but every change is yours to accept, adjust, or reject. Nothing is applied without your sign-off.
What is the "AI" part, exactly?: For each piece of the network, several search methods — evolutionary, probabilistic, and gradient-based — try many parameter combinations and each returns candidates with a likelihood. A language model weighs those candidates against the evidence, selects the most defensible option or hands the call to you, and records the reasoning. It's an assistant that does the legwork and shows its working, not a black box.
Can I override a recommendation?: Always. You can change any value, pick a different option from the ones considered, or set your own. Because every alternative is recorded, you can see what else was on the table before you decide.
How is this different from calibrating by hand?: Manual calibration means changing a parameter, re-running, and eyeing the result, over and over. Here the combinations are searched and scored for you, broken into manageable pieces, and presented with reasoning, so your time goes into judgment rather than trial and error.
Do I need good demand before I calibrate?: It helps. Calibration is strongest when it starts from a sound demand picture, which is what demand analysis is for, and from a deliberate field campaign, which is what field planning is for.