Inverse Module API

The retrieve() function estimates ice physical properties from observed albedo.

from biosnicar.inverse import retrieve

retrieve()

retrieve(observed, parameters, emulator=None, forward_fn=None,
         platform=None, observed_band_names=None, obs_uncertainty=None,
         bounds=None, x0=None, regularization=None, wavelength_mask=None,
         method="L-BFGS-B", mcmc_walkers=32, mcmc_steps=2000, mcmc_burn=500,
         fixed_params=None, ssa_rho=None) -> RetrievalResult

Parameter	Type	Default	Description
`observed`	ndarray	required	480-band spectral albedo or N-band satellite array
`parameters`	list[str]	required	Names of parameters to retrieve
`emulator`	Emulator	None	Trained emulator (fast mode)
`forward_fn`	callable	None	Direct forward function (slow mode)
`platform`	str	None	Satellite platform key. Activates band mode.
`observed_band_names`	list[str]	None	Band names matching `observed`. Required with `platform`.
`obs_uncertainty`	ndarray	None	Per-element 1-sigma measurement uncertainty
`bounds`	dict	None	`{name: (lo, hi)}` overriding defaults
`x0`	dict	None	`{name: value}` initial guesses
`regularization`	dict	None	`{name: (prior_mean, prior_sigma)}` Gaussian priors
`wavelength_mask`	ndarray[bool]	None	True = include wavelength (spectral mode only)
`method`	str	`"L-BFGS-B"`	Optimisation method
`mcmc_walkers`	int	32	MCMC walkers
`mcmc_steps`	int	2000	MCMC steps
`mcmc_burn`	int	500	MCMC burn-in steps
`fixed_params`	dict	None	Known parameters not to be optimised
`ssa_rho`	float	None	Reference density for SSA decomposition

Either emulator or forward_fn must be provided.

RetrievalResult

Attribute	Type	Description
`best_fit`	dict	`{param_name: optimal_value}`
`cost`	float	Final chi-squared value
`uncertainty`	dict	`{param_name: 1_sigma}`
`predicted_albedo`	ndarray (480,)	Spectrum at best-fit point
`observed`	ndarray	Input observations
`converged`	bool	Optimiser convergence flag
`method`	str	Method used
`n_function_evals`	int	Number of forward evaluations
`derived`	dict	Internal decomposition (SSA mode: `rds_internal`, `rho_ref`)
`chains`	ndarray or None	MCMC chains (steps, walkers, params)

result.summary() returns a human-readable string.

Default bounds

Parameter	Bounds	Default x0	Units
`ssa`	(0.01, 300)	2.0	m²/kg
`rds`	(100, 5000)	1000	µm
`rho`	(100, 917)	500	kg/m³
`solzen`	(20, 80)	50	degrees
`black_carbon`	(0, 100000)	100	ppb
`snow_algae`	(0, 500000)	10000	cells/mL
`glacier_algae`	(0, 100000)	100	cells/mL
`dust`	(0, 500000)	100	ppb

Constraints

direct cannot be in parameters — it’s binary. Pass via fixed_params.
ssa cannot appear alongside rds or rho — SSA replaces both.
Impurity concentrations and SSA are optimised in log₁₀(x+1) space internally.

Emulator Band Convolution