Appearance
When reflectance spectroscopy goes wrong, the cause is nearly always referencing, geometry or light source — not the sample. The fastest diagnostic is to re-take your white and dark references under the exact conditions of the measurement and re-process: roughly two-thirds of "weird spectra" resolve at that step. This guide walks the common failures in the order you should check them, from above-100% reflectance to mismatched pigment identifications.
Why does my spectrum read above 100% reflectance?
Reflectance is a ratio: sample signal divided by white-reference signal, after subtracting the dark. If the spectrum climbs above 1.0, the denominator is too small. The usual culprit is a white reference taken while the lamp was still warming, then a measurement minutes later once output had risen.
python
reflectance = (sample - dark) / (white - dark)
# values > 1.0 after this => white reference is stale or mispositionedFix: warm the lamp 15-20 minutes, re-take white and dark, keep the Spectralon standard at the same stand-off as the sample.
What causes noisy or spiky near-infrared spectra?
Detector noise dominates where the light source is weak. Many portable rigs use white LEDs that emit almost nothing past 700 nm, so the 700-1000 nm region is pure noise dressed up as data. Diagnose by capturing a white reference alone and looking at raw counts — if they collapse in the NIR, no amount of averaging recovers real signal there.
| Symptom | Likely cause | Fix |
|---|---|---|
| Spikes only in NIR | Weak source past 700 nm | Halogen/tungsten source |
| Uniform high noise | Integration time too short | Longer integration, more averages |
| Sharp single spikes | Cosmic ray / detector glitch | Median of N captures |
Why do repeat scans of one spot disagree?
This is the failure unique to point-based FORS. The probe footprint is small (often 1-3 mm) and parchment is heterogeneous at that scale. Move 1 mm and you sample different fibres, sizing or pigment grains.
Mitigations: a rigid stand-off cone so geometry is constant, a pencil-light registration mark, and averaging 5-10 captures with the probe lifted and replaced between each to estimate real variance.
How do I fix a sloping or tilted baseline?
A baseline that rises or falls monotonically across all wavelengths signals an additive contamination — stray room light, specular glare, or a contaminated fibre. Switch to a 45/0 geometry (illuminate at 45 degrees, collect at 0) to reject the mirror-like surface reflection that carries no pigment information. Shroud the setup with black cloth and re-reference.
If the slope survives all that, clean the fibre tip; a fingerprint adds a broad scattering offset.
Why don't my spectra match reference pigment libraries?
FORS identification compares shape, not absolute level. Published references (e.g. the IFAC-CNR or Kremer libraries) are true, calibrated reflectance. If you compare raw intensity, or skipped referencing, the inflection points still exist but the curve sits at the wrong level. Normalise and take the first derivative:
python
import numpy as np
d1 = np.gradient(reflectance, wavelengths) # first derivative
# match inflection positions, not absolute reflectanceIndigo's characteristic absorption near 660 nm and the ~735 nm "red edge" of some greens survive normalisation; absolute height does not.
What is a fast triage order for any bad FORS spectrum?
- Re-take white and dark references — fixes most cases.
- Confirm the source still emits across your full range.
- Check geometry and stand-off are unchanged.
- Shroud against ambient and specular light.
- Average several repositioned captures.
- Compare normalised derivatives, never raw intensity.
Key Takeaways
- Above-100% reflectance means a stale or mispositioned white reference — re-reference first.
- NIR noise usually means a weak source past 700 nm, not a software bug; use a halogen lamp.
- Point-based FORS adds repositioning error; fix the geometry and average several captures.
- A tilted baseline points to stray or specular light; use 45/0 geometry and shroud the rig.
- Match pigment libraries on normalised first-derivative shape, not absolute reflectance.
- Warm the lamp 15-20 minutes before referencing to avoid drift-induced artefacts.
- Triage in order: references, source, geometry, stray light, averaging, normalisation.
Frequently Asked Questions
Why does my reflectance spectrum read above 100%?
Almost always a stale or mismatched white reference. The sample is reflecting more than the white tile recorded earlier, usually because the lamp warmed up or the geometry shifted; re-take the white reference under identical conditions and the values drop back into range.
What causes a noisy, spiky spectrum in the near-infrared?
Low signal at the detector's band edges combined with short integration time. Increase integration time, average more scans, and check that your light source still emits usefully out to 1000 nm — many white LEDs collapse past 700 nm.
Why do two scans of the same spot disagree?
Repositioning error or a moving sample. Even a 1 mm shift on a textured page changes which fibres the probe sees. Use a fixed stand-off, mark the spot, and average several captures rather than trusting one.
How do I fix a sloping baseline across the whole spectrum?
A tilted baseline is usually stray ambient light or specular glare entering the probe. Shroud the measurement, use a 45/0 geometry to reject the specular component, and re-reference; if it persists, check for a fingerprint on the fibre tip.
My FORS spectra do not match published pigment references — why?
Reference libraries assume a calibrated, illuminant-independent reflectance. If you skipped white/dark referencing or used a different geometry, shapes shift. Re-process as true reflectance and compare normalised first-derivative spectra, not raw intensity.
Is reflectance spectroscopy point-based or imaging?
Both exist. Fibre-optic reflectance spectroscopy (FORS) is point-based — one spectrum per spot. Hyperspectral imaging gives a full spectrum per pixel. Many problems above are shared, but FORS adds repositioning error that imaging avoids.