Every signal Orbispect ships is computed with methods grounded in peer-reviewed science, validated against independent ground truth and bounded by quantified uncertainty. We do not ask clients to trust a black box — we show them the principles and the chain from raw signal to figure.
Full methodology and data sources are shared under NDA, in a secure data room.
Four stages, each versioned and reproducible. A regulator or reinsurer can replay any figure we have ever issued.
Raw radar and optical archives read directly at source; atmospheric and geometric corrections applied under our control.
Radar and optical time series aligned per parcel; cross-sensor checks reject artefacts a single instrument would pass.
Physics-based models assimilated with observation. Machine learning corrects residuals — it never replaces the physics.
Every figure ships with uncertainty, lineage and version. APIs and reports expose the same numbers — never two truths.
The core scientific machinery behind each product family, described at the level of capability and principle. Specific algorithms, parameterisations and data sources are shared with clients under NDA, in a secure data room.
Daily soil-water accounting grounded in established agronomic science: reference evapotranspiration, crop water demand per growth stage, and root-zone depletion tracked through every season.
Multi-scale precipitation and water-balance anomalies expressed as properly fitted standardised indices — not naive percentiles. Compound drought combines meteorological, soil and groundwater signals.
A process crop-growth model provides the physical backbone; machine learning corrects only the residual it cannot explain. Quantile outputs with monotonicity constraints — forecasts are ranges, never points.
Soil-loss modelling computed on real terrain: rainfall erosivity from precipitation records, slope factors derived from elevation models, cover factors from satellite — per district, refreshed seasonally.
Satellite gravimetry observes total water storage at coarse scale; physics-constrained downscaling guided by terrain, soils and recharge brings it to district level. Trends separated from climate noise by structural time-series decomposition; uncertainty quantified per pixel.
Parcel structure (boundaries, strips, management zones) is delivered at decision grade via physics-constrained super-resolution with guaranteed consistency. All biophysical values are computed at the sensor's native resolution and never resampled upward. We sharpen geometry, not physics.
Radar tracks canopy structure and moisture through cloud; optical reads pigment and vigor. Agreement between independent sensors is required before any alert fires — disagreement triggers review, not output.
Machine learning corrects models; it does not replace them. A network that learns residuals stays honest when the season looks like nothing in its training data.
A number without an error bar is marketing. Every output carries calibrated uncertainty, and our calibration itself is tested against held-out seasons.
When a source is down we say so. No silent fallbacks, no synthetic filler. An honest gap is worth more than a fabricated value — especially in an audit.
Every released figure is tied to a model version, input snapshot and code revision. Clients under contract can request a replay of any historical output — and receive the same number, byte for byte. This is what "audit-ready" means in practice.