Configuration arguments

Calibration settings are passed through CalibrationConfig. The API reference contains the exact constructor signature.

Argument	Default	Meaning
`method`	`"de"`	Differential evolution; `"lhs"` performs Latin-hypercube candidate search.
`objective`	`"nse"`	NSE, KGE, RMSE, log NSE, or supported weighted objective mapping.
`warmup`	`365`	Initial model steps excluded from scoring, not always days.
`calibration_end`	`None`	Explicit training end date; do not combine with `calibration_fraction`.
`calibration_fraction`	`None`	Optional chronological training fraction strictly between zero and one.
`validation_start`	`None`	Optional later validation start, after training.
`seed`	`42`	Optimizer random seed.
`maxiter`	`100`	Differential-evolution iteration budget.
`popsize`	`10`	Differential-evolution population multiplier.
`samples`	`1000`	Latin-hypercube candidate count.
`polish`	`False`	Optional local polishing after differential evolution.
`bounds`	`None`	Override supported calibration bounds by parameter name.
`fixed`	`{}`	Explicit parameter values held fixed during fitting.

Study choices

The default settings are software defaults, not basin-specific recommendations. Use model steps for warmup: a monthly model needs a different choice than a daily one. Without an explicit split, do not interpret training metrics as held-out validation.

Parameter names, ranges, fixed initial stores and temperature requirements are listed on each model page. The model must match the forcing time step.

Optional engine settings

configure_marrmot(octave=..., package_list=..., timeout=300) configures the Octave backend. See Installation and the generated API for supported arguments. This backend is optional and is not a Python-only execution engine.