ORBISPECT
Validation

We publish skill, not promises.

Every Orbispect layer is benchmarked against independent, in-situ national networks before it ships — and the comparisons below are maintained continuously, not produced once for a sales deck. Where the model is weak, this page says so.

01 / ONGOING BENCHMARKS

Three instruments, three independent referees.

STUDY V-01GRAVIMETRY × BOREHOLES

Groundwater vs national well network

Downscaled satellite water-storage trends compared against geological-survey monitoring boreholes across every district we score. Agreement is required before a Water Credit Score is issued; districts failing the check are flagged, not scored.

REFERENCENational geological survey borehole network
CADENCEre-run with every monthly gravimetry release
SAMPLE380 districts · continuous gravimetric record since 2002
FAILURE MODEdistrict flagged "insufficient agreement" — no score shipped
SATELLITE TREND IN-SITU WELLS EACH POINT = ONE DISTRICT · DASHED = PERFECT AGREEMENT · OUTLIERS FLAGGED, NOT HIDDEN
STUDY V-02REANALYSIS × MET STATIONS

Climate inputs vs national met stations

Gridded climate fields that feed every index are benchmarked against the national meteorological station network — bias measured per region and season, documented, and corrected before any downstream model sees the data.

REFERENCENational met service synoptic stations
SCOPEtemperature, precipitation, reference evapotranspiration
SAMPLEdaily series, full national synoptic network, multi-decade
OUTPUTper-region bias tables, applied upstream of all indices
GRIDDED INPUT MET STATIONS TIGHT AGREEMENT AFTER PER-REGION BIAS CORRECTION · RAW INPUT IS NEVER USED BLIND
STUDY V-03FLOW MODEL × GAUGES

Streamflow vs national river gauges

Basin flow forecasts verified against national hydrological gauge readings, river by river. We publish skill per basin — clients see exactly where the model is sharp and where the catchment defeats it. Weak basins are labelled as such in the product.

REFERENCENational hydrological service gauge network
GRANULARITYskill reported per basin, not as one flattering average
SAMPLEnational gauge network · daily flows · verified release by release
HONESTY RULEbasins below skill threshold ship with a "low confidence" label
MODEL FLOW RIVER GAUGES FADED POINTS = BASINS UNDER REVIEW · PUBLISHED, NOT REMOVED
02 / KNOWN LIMITS

What we are not yet good at.

// PUBLISHED LIMITATIONS — UPDATED QUARTERLY
  • Yield anomaly forecasting is in pilot. The AgroTransformer engine ships to design partners only, until out-of-sample skill clears our internal bar. We will publish the bar and the score.
  • Small, fragmented parcels below sensor resolution carry wider uncertainty — flagged per parcel, priced accordingly.
  • Mountain basins with sparse gauging defeat streamflow verification; those basins are labelled low-confidence in the product.
  • Sedimentation monitoring (HydroSed) is validated on a small set of reservoirs so far; coverage grows survey by survey.

Read the full reports.

Complete validation protocols, per-district correlation tables and historical revisions ship as a validation pack with every pilot — and stay updated for the life of the contract.

Request validation pack