NEON flightlines · drone imagery · bulk analysis
SpectralBridge
Connect reflectance processing, spectral libraries and sensor comparisons.
Airborne, drone and satellite observations differ in wavelength coverage, ground footprint, geometry and processing history. SpectralBridge is an open-source Python package for processing, harmonizing and comparing these observations, with explicit assumptions, provenance and QA.
Install from a source checkout with python -m pip install -e .;
follow the setup instructions.
API Reference · How to cite · AI transparency
Common framework. Sensor-specific treatment.
Why it exists
One landscape, different measurements.
NEON airborne instruments record reflected light across many narrow wavelength ranges. Drones such as those carrying MicaSense cameras provide fine local observations. Landsat satellites observe wider areas with larger ground footprints. SpectralBridge gives these systems a common framework while treating each according to how it was collected. Configured Landsat-like bands do not prove agreement with Landsat observations; aligning or aggregating ground footprints remains a separate scientific decision.
SpectralBridge applies configured corrections, constructs sensor-like wavelength bands from NEON observations, and compares completed products. Drone observations follow their own preparation path. The software preserves the meaning of each measurement and records the choices needed to review it.
Choose your way in
Choose the pipeline for your inputs.
NEON flightlines
Prepare detailed airborne spectra, apply configured corrections, construct sensor-like bands, and inspect outputs.
Run the NEON vignette → 02Drone imagery
Prepare supported local imagery while retaining the drone observation’s identity and its own correction path.
Run the drone vignette → 03Bulk analysis
Compare completed products across flightlines and sites; inspect how relationships and uncertainties vary.
Run the bulk vignette →Scientific products
From measurements to interpretable spectra.
The three workflows produce corrected observations, configured sensor-like representations and compact comparison evidence. Polygon extraction and spectral-library analysis are downstream products and modules, not additional processing pipelines.
Corrected observations
Inspect persisted raw, corrected and derived sensor products together with their QA and provenance.
Read the output contracts →Spectral libraries
Carry pixel and polygon identities into labeled libraries; inspect species or population variation with explicit sampling units.
Inspect a spectral library →Evidence and limitations
Separate successful execution and internal QA from agreement with independently collected observations.
Read the evidence hierarchy →
Follow optional downstream analyses without changing the core NEON, drone and bulk processing routes. Sensor information retention and one real-Landsat pilot supply different evidence; operational abundance validation remains incomplete.
Explore the technical figures
The system at a glance
Three technical views. Read them one at a time.
The original scientific figure is separated into enlarged panels here. The plots, wavelength ranges, processing stages, and translation relationships remain unchanged; the surrounding text provides a readable path through them.
Measure at three scales.
NEON airborne imaging spectroscopy, MicaSense UAS observations and Landsat each sample reflectance differently. The supplied conceptual artwork uses NBAR terminology; the independent pilot documented here uses Collection 2 Level-2 surface reflectance and does not establish NBAR agreement. The first panel keeps their spectral ranges and sampling patterns visible together.
Why cross-sensor calibration matters →
Correct before you compare.
The supplied processing diagram is a conceptual view. In the current package, NEON hyperspectral products use spectral convolution, while drone products use optional post-correction affine translation. Bulk analysis estimates and reviews the relationships from completed products.
Walk through the full pipeline →Use NEON as the hyperspectral anchor.
Matched MicaSense and Landsat products generated from the same corrected NEON source support diagnostic regressions. Actual Landsat comparison is a separate, optional validation step; a strong synthetic fit alone does not prove sensor interchangeability.
Inspect sensor harmonization →- DeterministicFixed inputs and configuration
- RestartableReuse validated products
- VersionedRecord software and parameters
- Provenance trackedInputs, models, and outputs
- Open evidenceInspect sources and gaps
The NEON processing path
Source observations in. Traceable products out.
The five steps below describe the individual-flightline NEON workflow. Drone and bulk follow their own linked vignettes above.
Evidence, not mystery
Every transformation should leave a trail.
Validation campaigns record input variation, expected behavior, observed diagnostics, runtime, provenance, and failures. QA panels turn those records into something scientists can inspect.
Distinguish tests, QA and independent validation →Already halfway there?
Carry on from the files you have.
Individual-flightline processing validates existing artifacts and resumes at the first missing or invalid stage.
Carry On My Wayward Son →Ready when you are
Translate reflectance across sensors and scales.
What have we learned?¶
Difference → identity → quantity. All 10 researcher-category pairs differ in the observed matched cohorts. Full-NEON recognition falls from 85.2% for unseen polygons to 73.8% for held sites. The original synthetic Ridge120 advantage disappears under the matched15 NEON library control: retained diversity matters. Better prediction does not imply unique composition. Read the scientific synthesis for the evidence, reusable methods, mathematical explanation and remaining field-validation gap.