basal — primary-source ecological data for loan-review committees

How it runs

SD card in. Audit-traceable
exposure record out.

Every transformation is a cited method. Every intermediate value persists, so an external auditor can inspect any step — detection rate, bias correction, density estimate, tier call — independently.

pipeline inputs

detections · camera-days · placement_context · boundary

methodology outputs

rate · adj_rate · density · CI · tier · score

supplementary

modeled $ · crop modifier · ESA flags

Worked example · Farm Credit of Central Texas

Riverbend Farm, Brazos Co. — 650 acres of corn.

Severe TX-BRA-2026-00012

83.7 / 100 Feral hog exposure score

13.47 animals/km² 95% CI 3.49 – 36.56

1.526 raw rate/cam-day pre-bias correction

1.042 IPW-adjusted rate REM consumed

Trajectory

Fall 2025 Elevated 5.66 /km² ⟶ Spring 2026 Severe 13.47 /km²

+138% density since the prior survey. The trajectory a $40K point-in-time field survey would have missed.

Open full report PDF JSON API

CAM-RB-RAND-01 · IR

random anchor

REC

event 94 of 354

2026-02-28 · 23:14:51

Modeled projection

$22,955/year

not a pipeline output

Scaled from Anderson 2016 per-hog damage × 650 ac × corn modifier 1.60. Third-party loss data; a committee with an internal damage model should consume the pipeline outputs directly.

Methodology

Published math.
Inspectable every step.

An external auditor can recover every value we report. The random-placement reference camera on each parcel anchors the IPW correction; the bootstrap propagates the weighting uncertainty into the CI; the tier cutoff is a published binning.

01 / index

Detection frequency — independent events per camera-day. Pre-REM relative abundance. Zero movement assumption. Directly comparable across identical deployments.

02 / debias

IPW correction — per-camera rate deflated by a literature-prior inflation factor (Kolowski & Forrester 2017) keyed to placement_context. Hájek IPW reported alongside as a diagnostic. Random-placement cameras anchor the correction against an unbiased reference.

03 / density

Random Encounter Model — D = (y/t) · π / (v · r · (2 + θ)). Per-species v: hog 6.0 km/day (Kay 2017), deer 1.5 (Webb 2010), coyote 10.0 (Andelt 1985). 1000-iteration bootstrap over cameras with truncated v-perturbation.

04 / tier

Mayer-Brisbin 2009 cutoffs. <2/km² Low · 2-5 Moderate · 5-10 Elevated · ≥10 Severe. Feral hog only at v1; other species surfaced as Informational with density where v is published.

05 / flag

Recommendation — insufficient_data below 100 cam-days or 20 events; recommend_supplementary_survey when CI upper/lower > 1.5; else sufficient_for_decision.

Pricing

Orders of magnitude cheaper than a field survey.
Continuous instead of point-in-time.

A loan-review committee's alternative is an independent biologist survey that runs into the tens of thousands and is stale in six months. We refresh on any new upload.

Per verification

Pilotengagement

One report per survey window. Use it for a single collateral review. Contact for pilot terms.

Full Nature Exposure Report (HTML + JSON)
Methodology appendix + 12 citations
JSON API for internal portfolio systems
Camera-day-level audit trail

Portfolio

Pilotengagement

Unlimited parcels, continuous refresh on every upload. Contact for pilot terms.

Every parcel in your portfolio
Survey-trend delta across all periods
Cross-portfolio tier distribution
Ag-lender dashboard + exports
Regulatory-affairs methodology briefing

SD card in. Audit-traceable exposure record out.

Riverbend Farm, Brazos Co. — 650 acres of corn.

Published math. Inspectable every step.

Orders of magnitude cheaper than a field survey. Continuous instead of point-in-time.

SD card in. Audit-traceable
exposure record out.

Published math.
Inspectable every step.

Orders of magnitude cheaper than a field survey.
Continuous instead of point-in-time.