CPU lab · synthetic measurement operators

What does your new sensor actually measure?

Compare known target-band truth with two estimates from source band averages. This runs numerical response integration on your device. It does not run DOFA, SpectralEarth or a trained adapter.

Choose a counterexample

Source FWHM 8nm · width of each source band Source centre spacing Target FWHM 80nm · same Gaussian assumption for both estimators Absorption feature σ 8nm · feature width, not sensor FWHM True target-centre shift 0nm · estimators still use nominal centres Source channels retained

Band measurements and resampled estimates

● Black: reference target□ Blue: response resampling× Orange: centre interpolation○ Grey: source bandsGrey line: synthetic truth

All signals are original analytic curves. Gaussian SRFs are assumed, truncated at ±4σ. Targets need ≥99.9% modeled response coverage; that threshold is a numerical choice, not confidence. No extrapolation or interpolation across missing source neighbours. Errors are absolute reflectance fractions, not classification scores.

Inspect all 13 target bands

Reference uses true target centres. Estimates use nominal centres. Blank source support is never filled with zero.
Nominal nm	True nm	Reference	Centre estimate	Response estimate	Coverage

Download runnable CPU source · Download Python notebook

Static default results shown. Enable JavaScript for controls; the Python download works independently.