How to Calibrate an EPA SWMM Model Against Flow Data

What does calibrating a SWMM model mean?

It means adjusting a model's uncertain parameters until simulated flows and depths match what field monitors recorded, then checking the result against data you did not tune to. The UK CIWEM Urban Drainage Group's Code of Practice for the Hydraulic Modelling of Urban Drainage Systems (2017) separates three things. Verification checks a model against independent data. Calibration adjusts parameters to fit measured conditions and should be followed by verification on a different data set. Force-fitting makes arbitrary changes to match the data, and the Code says it should not be done.

The EPA SWMM Reference Manual, Volume I: Hydrology (revised 2016) notes that most of SWMM's 150 or more constants can be measured, tested or looked up, and only a smaller number may need calibration.

What data do you need before you start?

You need rainfall from gauges in or next to the catchment, which the hydrology manual calls essential because convective storms can be very local, and flow or depth records at sites with a clear response. The CIWEM Code asks a short-term flow survey to aim for three acceptable storm events and a run of dry days covering weekdays and weekends.

Check the records first: compare monitors on the same branch for continuity of flow, and check each scattergraph against the Colebrook-White equation, as the Code suggests. A poor fit can mean a wrong invert, gradient or pipe size, or a control downstream.

Should you calibrate dry weather or wet weather flow first?

Dry weather first. The CIWEM Code says sewer models should generally be verified for dry weather flow before storm verification, then against long-term data and historical events. In a foul or combined sewer the storm hydrograph sits on the dry weather flow, so a base flow error carries into every storm statistic.

In SWMM, a node's dry weather inflow is an average rate multiplied by up to four time patterns: monthly, daily, hourly and weekend. The Code's method is to overlay the recorded dry days, draw minimum and maximum envelopes for weekdays and weekends, and adjust until the simulated profile, including the timing of its peaks and troughs, falls inside them. For groundwater infiltration it suggests starting from the treatment works' daily flow records, using the Q80 low flow for each month or season.

Which SWMM parameters should you adjust, and in what order?

Fix runoff volume first, then hydrograph shape and timing, and touch pipe roughness last. Table 3-6 of the hydrology manual gives the reasoning:

Parameter	Effect of an increase (EPA Reference Manual Vol. I, Table 3-6)
Imperviousness	Significant: more runoff volume and a higher peak
Width	Affects shape: higher and earlier peaks for rain of varying intensity
Slope	As for width, but less sensitive, since flow goes with the square root of slope
Overland Manning n	The inverse of width
Depression storage	Moderate: less volume and a lower peak, significant only for small storms

A workable order follows from that table:

1. Imperviousness. The manual says runoff volume and flow rates are strongly sensitive to it, so check it against aerial imagery before treating it as free.
2. Depression storage and infiltration. The manual treats depression storage as a calibration parameter "particularly to adjust runoff volumes". Infiltration parameters set the pervious response, which the CIWEM Code expects to need some calibration.
3. RDII in separate sewers. SWMM's unit hydrographs take R, the fraction of rainfall volume that enters the sewer, then T, the time to peak, and K, the ratio of recession time to T. R sets volume, T and K set shape.
4. Width. "For SWMM, it is common to fix estimates of slope and Manning's n and calibrate with the subcatchment width," says section 3.8.6 of the manual. Width moves peaks and their timing with little effect on volume.
5. Conduit roughness. Change it only where the pipe's condition justifies it.

Simulate the full survey period, not isolated storms, so each event starts from the right antecedent conditions, as the CIWEM Code asks.

Which goodness-of-fit statistics should you report?

Report a shape statistic and the peak, volume and timing errors for every monitor and event. With observed O and simulated S:

• Nash-Sutcliffe efficiency (NSE), from Nash and Sutcliffe (1970), Journal of Hydrology 10(3): 282 to 290: 1 minus the sum of (O minus S) squared over the sum of (O minus the mean of O) squared. A value of 1 is a perfect match; 0 means the model does no better than the observed mean.
• RMSE: the root mean squared difference, in the record's units.
• Peak error: simulated minus observed peak, as a percentage, or in metres for depth.
• Volume error: simulated minus observed volume, as a percentage. Moriasi's PBIAS has the opposite sign; there, a positive value means the model underestimates.
• Timing error: the time of the simulated peak minus the time of the observed peak.

NSE squares the differences, so a slightly late peak can pull it down even when peak and volume are right. Read it beside the hydrographs.

What NSE value counts as a good SWMM calibration?

For a sewer model, the most specific published targets are in Table 5-1 of the CIWEM Code, which asks a model to meet them for at least two of the three verification storms:

Measure	General target	Critical locations
Shape	Good match (NSE above 0.5 if used)	Good match (NSE above 0.5 if used)
Timing of peaks and troughs	±0.5 hour	±0.5 hour
Peak depth, unsurcharged	±0.1 m or ±10%, whichever is greater	±0.1 m
Peak depth, surcharged	+0.5 m to -0.1 m	±0.1 m
Peak flow	+25% to -15%	±10%
Flow volume	+20% to -10%	±10%

Critical locations, agreed with the client, typically include flooding sites, overflows and treatment works. The Code calls these targets a general guide.

The NSE bands often quoted come from catchment hydrology. Moriasi et al. (2007), Transactions of the ASABE 50(3): 885 to 900, rate NSE above 0.75 very good, 0.65 to 0.75 good and 0.50 to 0.65 satisfactory, for a monthly time step. Moriasi et al. (2015), Transactions of the ASABE 58(6): 1763 to 1785, rate watershed-scale flow at daily, monthly or annual steps as very good above 0.80, good from 0.70 to 0.80 and satisfactory from 0.50 to 0.70, with PBIAS within ±15% for satisfactory. The 2015 bands come from SWAT, HSPF and WARMF studies, and neither paper was written for 5 minute sewer hydrographs. The EPA SWMM manuals set no numeric targets.

How do you run a sensitivity analysis?

Move one parameter at a time by a fixed percentage, rerun the model, and record how far peak and volume move. Relative sensitivity is the fractional change in the result divided by the fractional change in the parameter: a value of 1 means a 10% change in the parameter moves the result by 10%. Parameters that barely move the result are not worth calibrating, and two that move it the same way cannot both be fixed from one record.

How do you calibrate a SWMM model in DrainPilot?

DrainPilot runs EPA SWMM 5.2.4 in the browser. Desktop EPA SWMM plots calibration files beside simulated series; DrainPilot adds the statistics and the search.

1. Open Tools > Calibration... and load flow monitor, level logger or rain gauge records from CSV or Excel.
2. Run the model. The Statistics tab gives NSE, RMSE, R2, peak, volume and timing errors per record, with checks against peak flow +25% to -15%, volume +20% to -10% and peak depth within 0.1 m.
3. Add parameters: imperviousness, width, slope, overland and conduit Manning n, depression storage, the infiltration parameters in use, and RDII R, T and K. Each varies by a multiplier around your own values.
4. Run the sensitivity sweep, then the automatic calibration: a seeded genetic search that minimises weighted (1 minus NSE), volume error and peak error over the flow and depth records.
5. Apply the multipliers, then verify on a different event without further changes.

DrainPilot does not calibrate dry weather flow: average inflows and time patterns are not in the parameter list, and it draws no dry weather envelopes. It does not fit rainfall records or split your events into calibration and verification sets, and it applies the same 0.1 m depth check whether or not the pipe is surcharged.

DrainPilot is free to use at geopilotai.com/drainpilot. The calibration and scenarios docs page explains every field in the Calibration window.