EPAnet Fire Flow Models Need Accurate Network Data to Produce Defensible Results
EPAnet fire flow modeling delivers actionable results only when your network geometry matches reality and your demand allocation reflects actual consumption patterns. I’ve run hundreds of EPAnet simulations for California water districts, and the models that fail do so at the data layer—not the solver. This article walks through node setup, demand allocation, C-factor validation, and the calibration steps that separate credible models from screen-generated fiction.
Why Node Spacing Matters for Fire Flow Accuracy
Node density directly controls pressure gradient fidelity. Space nodes too far apart and you’ll overestimate available pressure at fire test locations; space them too close and you’ll spend weeks on a model that runs slower without gaining precision.
For distribution mains under 6 inches, place nodes every 200–300 feet. For 8–12 inch mains, 300–500 feet is standard. For transmission and trunk lines larger than 12 inches, 500–1,000 feet works if you’re not trying to capture localized pressure swings. I ran a model for a Placer County system in 2022 where the engineer had spaced 12-inch nodes 2,000 feet apart; when we remeshed at 600 feet, the fire flow at a critical hydrant dropped from 1,850 gpm to 1,420 gpm—a 23% correction that changed whether a high-rise met CBC Section 902 supply requirements.
Elevation accuracy is non-negotiable. Pull elevations from USGS quad data or surveyed datum points, not Google Earth. A 50-foot elevation error on a node creates 22 psi of phantom pressure. California’s CGP (Caltrans Stormwater Program) and local water quality rules often require hydrostatic testing; if your model’s elevation is wrong, your predicted fire flows won’t match field tests, and you’ll face pushback during plan check.
Pipe Length and Diameter: The Foundation Layer
Pipe data errors compound fast. I’ve inherited models where GIS shapefiles showed pipe segments that didn’t exist on the ground, or where 50-year-old cardboard records listed diameters that field inspections contradicted. Before you model, walk your critical mains or pull CCTV footage.
Capture actual pipe length, not straight-line distance. A 400-foot main that curves through a neighborhood is 480 feet of real pipe. Use calibrated odometer data from water company crews, or pay for GPS survey work if the system is new or recently rebuilt. Underestimating length reduces friction loss and inflates fire flow predictions.
Diameter changes must be explicit in your network. If a 12-inch main reduces to 10 inches at a tee, model it as two separate links with a node at the transition. EPAnet will apply the Hazen-Williams equation separately to each segment; if you collapse them into one 400-foot link, you’re averaging the diameter and losing the real velocity spike at the contraction.
Demand Allocation: Peak Hour, Average Day, and Fire Demand
Baseline demand calibration is the difference between a model that validates against field tests and one that’s fiction. I use three sources: meter records from the district’s SCADA system, NRCS (Natural Resources Conservation Service) irrigation demand data for agricultural zones, and CBC Table C101.1 fixture count estimates for residential areas.
Peak hour demand is typically 1.5–2.5 times average daily demand for residential systems; commercial and light industrial systems can spike 2.0–3.0x depending on process loads. Don’t assume. Pull your district’s 12-month consumption record from the billing system and divide total volume by 365 to get average daily. Then identify the single highest daily consumption in that year—that’s your basis for peak demand.
For fire flow modeling, apply baseline demand at all demand nodes, then add fire demand (usually 750–2,500 gpm depending on occupancy class per CBC Table C102.1) at the test node. If you’re modeling a system with multiple simultaneous fire scenarios, stack fire demands at each location. A district in Tuolumne County asked us to model a 1,500 gpm fire at Node A and a concurrent 1,500 gpm demand at Node B; without stacking, their plan check engineer rejected the model as incomplete.
C-Factor Calibration: Honest Friction Loss Data
The Hazen-Williams C-factor is your friction loss proxy. New clean pipe runs C = 130–140. Corroded or incrusted pipe runs C = 70–100. Don’t guess. I’ve seen consultants use C = 120 across an entire 40-year-old system and predict fire flows 30% higher than what field testing showed.
Calibrate C-factors using field fire tests. Run an actual hydrant test at known flow rates (your district’s hydrant testing program will have this data) and compare measured pressure to your EPAnet prediction at that same node. If the model overestimates pressure at 1,000 gpm, your C-factor is too high—lower it and re-run. Iterate until your model pressure matches field pressure within ±5 psi across a range of flows.
For very old pipe, conduct internal inspections. CCTV footage will show tuberculation and deposit buildup; a main with heavy scaling visible at 50% of its internal diameter isn’t C = 100—it’s C = 65. Document this in your model notes. When a Fresno County district pushed back on our low C-factor assignment, CCTV stills won the argument instantly.
Validation: Field Testing Against Model Predictions
A validated model predicts fire flow ±10% of actual field test results under identical demand conditions. Schedule field tests during the same time-of-day and season as your model baseline. If you calibrate your model in July at 2 PM (peak demand) but test in March at 6 AM (minimum demand), you won’t validate—the demand profiles are different.
For each fire flow test, record: hydrant location, static pressure, residual pressure at measured flow (measure with calibrated pitot tube and flow meter), ambient temperature, and system demand at that moment (pull from SCADA or call the district). Run your EPAnet model with identical demand assumptions and compare predicted residual pressure to measured residual pressure at that flow rate.
If your model predicts 45 psi residual and the field test shows 38 psi, that’s 15% error—too high. Adjust C-factors downward on the mains supplying that hydrant, re-calibrate, and test again. We validate models across 6–8 field test locations on typical systems; larger or more complex systems need 10–12 test points.
Validation Tools and Real-World Constraints
EPAnet 2.2 (USEPA’s free hydraulic solver) runs on Windows and Linux, though we’ve moved to water distribution modeling platforms like OpenFlows WaterGEMS for systems over 500 nodes because the batch processing and sensitivity analysis tools save weeks on large fire flow studies. For simple systems under 200 nodes, EPAnet 2.2 is sufficient and transparent—every equation is documented.
Be honest about your model’s limits. If your system has pressure-reducing valves (PRVs), check that your model includes them with correct setpoints from the valve nameplate or district maintenance records. If you have variable-speed pumps, model their performance curves from pump test data—don’t assume linear pump behavior. A 500-node model for a Nevada County system we validated in 2023 had pump curves coded as linear; actual pump curves were flat, which meant we were overestimating head at design flow by 12 feet—enough to fail fire flow in two zones.
Document all assumptions. Your model report should include: node spacing rationale, elevation source and accuracy, pipe length verification method, baseline demand calculation, C-factor sources and calibration results, field test data and validation errors, and sensitivity analysis (how much does fire flow change if you drop system pressure 2 psi, or lose a major supply line). This isn’t busywork—it’s your defense when a plan check engineer or fire marshal questions your results.
Start Your Fire Flow Validation Today
We validate EPAnet models against field data and help districts defend fire flow predictions in plan check. Contact us to schedule a model review or field calibration program.