Skip to content

Common workflows

Recipes chaining the pipeline steps into the classic exercises of operational hydrology. Each links to the step pages for the details.

Manual calibration of one model

  1. Stations: pick the calibration station and period; pick the same station as simulation station (any period) so the pipeline can unlock.
  2. Weather: pick a method.
  3. Model: "Single" mode, pick a model, activate CemaNeige.
  4. Calibration: choose the objective and transformation, set "Algorithm" to Manual, open the model's section and iterate on the sliders. Each move re-simulates immediately; judge the fit on the objective chip and on the hydrograph (zoom into floods and recessions).
  5. "Export" the result — the JSON keeps the parameters and your whole trajectory, the CSV holds the simulated series for your figures.

Automatic calibration with a split-sample test

Calibrate on one period, validate on another: performance on data the model never saw is the only honest measure.

  1. Stations: same station for both roles; calibration period = the fitting window, simulation period = a disjoint validation window.
  2. Weather and Model: as above.
  3. Calibration: "Algorithm" SCE, then "Calibrate"; export the fitted parameters.
  4. Simulation: the validation runs automatically — read the KGE chips and the six-metric profile; export the series and metrics.

To study the effect of the calibration period itself, re-run step 3 with a different (for example much shorter) calibration period and compare the validation metrics: the parameters are only as general as the climate they were fitted on.

Comparing weather sources and models

  1. Set up a split-sample calibration as above, with the Model step in "Ensemble" mode over the models to compare — one SCE run calibrates them all, and every chart overlays them with their median.
  2. To change the weather source, revisit Weather and pick another method. Downstream steps turn stale and the calibrated parameters are discarded — recalibrate, then compare validation metrics between runs.
  3. Export the calibration of each configuration to its own file; re-importing a file restores its whole context (import), so you can flip between configurations at will.

Proxy-basin test (spatial transfer)

Can parameters calibrated on one watershed reproduce another?

  1. Stations: calibration station = the donor basin, simulation station = the receiver basin, with their periods.
  2. Calibrate (SCE) on the donor as above.
  3. Simulation evaluates the donor's parameters on the receiver: the metric profile tells you which aspects of the transfer hold (timing, volume) and which degrade.

Reconstructing an unobserved event

A calibrated model plus weather is enough to estimate flow where and when no gauge was recording — for example the July 1996 Saguenay flood at a station whose record starts later.

  1. Calibrate on the closest suitable station and period (manual or automatic).
  2. Stations: set the simulation station to the target and the simulation period around the event — the period may extend outside the observed record.
  3. Simulation: the hydrograph renders from weather alone (metrics stay blank without observations); zoom to the event and export the series.

Saving and presenting your work

  • Every step has an "Export" button; the CSVs (streamflow, weather, simulated series, regime, indicators) are ready to plot in your report figures.
  • The browser keeps only the latest state: export each calibration you care about to its own JSON and re-import to return to it.