A cross-scale bridge from lab chemistry to forage decisions.
Benchtop NIR spectroscopy, hyperspectral and lidar UAV surveys, satellite data, and interpretable ML models, combined into a transferable, field-validated pipeline for forage quality and invasive-grass detection.
The sensing stack
Multiple platforms, calibrated to work together.
Benchtop NIR Spectroscopy
- Full-range (450–2500 nm) reflectance on dried, ground samples
- Calibrations for silica, nitrogen (crude protein), and fiber (NDF)
- Chemometrics identify the wavelengths that carry stable signal
Hyperspectral & Lidar UAV
- Full-spectrum hyperspectral plus lidar structure from the air
- Centimeter-to-meter resolution over large coverage areas
- Species discrimination, forage chemistry, and biomass at canopy scale
Satellite Data
- Sentinel-2, Landsat, and Planet for landscape-scale coverage
- Time-series analysis for change detection and phenology
- Coarser resolution anchored by UAV and field ground truth
ML model pipeline
From raw spectra to calibrated, validated outputs.
Spectral preprocessing and feature engineering
- Baseline correction, derivative transforms, dimensionality reduction
- Controlled for geometry, moisture, and sensor drift
- Preprocessing tailored to sensor type and target medium
Model types and validation
- PLS, PLS-DA, SVM, and lightweight neural networks
- Performance targets set per engagement with documented failure modes
- Validated against held-out samples and real field conditions
Data delivery
Outputs designed for the programs and workflows clients actually use.
GIS-compatible mapping outputs
Shapefiles, GeoTIFFs, and KMZ files ready for program submissions, management planning, and agency review.
Program-formatted reports
Documentation designed to support EQIP applications, Defend the Core submissions, and internal management records.
Decision outputs with uncertainty
Per-pasture grazing-days, AUMs, nutritional-risk classes, and treatment-priority scores, with calibrated per-pixel uncertainty so low-confidence areas defer rather than mislead.
The science behind it
NIR spectroscopy and machine learning make fast, field-ready plant identification and chemistry analysis possible.
Reflectance encodes chemistry & structure
Across 400 to 2500 nm, nitrogen and fiber express diagnostic SWIR absorption features, while silica, deposited in grasses as amorphous opal (phytoliths), leaves fingerprints in the VNIR and SWIR. Where those wavelengths sit decides sensor cost and feasibility.
Transferability is the hard part
Models calibrated in one community or growth stage rarely hold in the next. We use stratified sampling, feature selection, and spectral unmixing (separating green vegetation from litter, soil, and shadow) to find signal that actually transfers across species, season, and site.
The asset is the calibration, not the hardware
Zubr's durable advantage is a proprietary, community-stratified calibration dataset and the workflow that builds it, carrying forage-chemistry signals from bench to UAV to satellite. No open dataset fills that gap.
How a typical engagement works
From scoping call to delivered data products.
Scope the data need
Define species targets, geography, program requirements, delivery formats, and timeline.
Collect and calibrate
UAV flights, satellite data pulls, and stratified field sampling with lab chemistry, per the agreed plan and sampling design.
Process and model
Spectral preprocessing, model training or application, and output QA against documented performance targets.
Deliver and document
GIS files, analysis reports, and program-ready documentation, with support for any submission or review process.
Need a custom technology approach?
Tell us your target species, geography, and program requirements.
Request a consultation