Calibration
Calibration is adjusting model parameters until predictions match measurements.
Before you start
Section titled “Before you start”Import the observed series first (see Importing monitoring data), and check the obvious things by eye. Automated calibration will happily tune a model whose rain gage is wrong, and give you a well-fitted wrong answer.
What gets tuned
Section titled “What gets tuned”The wizard adjusts the parameters that genuinely carry uncertainty:
- Conduit roughness - affects travel time and peak attenuation.
- Percent imperviousness - affects runoff volume more than anything else.
- Dry-weather base flow - the sanitary baseline.
It uses coordinate descent: change one parameter, keep what improved, move on, repeat.
Judging the fit
Section titled “Judging the fit”Two measures, reported together:
NSE (Nash-Sutcliffe efficiency) compares the model against simply using the mean of the
observations. 1.0 is perfect, 0 means the model is no better than the mean, and negative
means worse than the mean.
PBIAS (percent bias) is systematic over- or under-prediction. Near zero is good; positive means the model consistently under-predicts.
The results are rated against published performance bands, so “is this good enough” is a stated standard rather than an opinion.
Calibrating a foul sewer
Section titled “Calibrating a foul sewer”Separate the two regimes:
- Dry days first. Fix the sanitary baseline and its diurnal shape. Nothing about wet weather can be judged until the dry baseline is right.
- Then storms. Whatever arrives above the dry-weather baseline is rainfall-derived, and that is what the RDII parameters have to reproduce - volume first, then timing, then recession.