Insurance underwriting - Mireye Earth

A wildfire-and-flood underwriter at a small-business carrier needs to score thousands of rural properties without sending an adjuster on-site. The traditional path — proprietary CAT models, ESRI add-ons, a half-dozen specialty data brokers — produces a number with no audit trail. The regulator asks “where did the 18% loss probability come from?” and the underwriter has to email three vendors and hope for the best. Mireye Earth gives the same underwriter a per-field, per-coordinate answer with the federal source cited inline.

The query

Address geocodes to lat=46.6, lng=-93.7 (Aitkin County, Minnesota — rural, open herbaceous cover, heavy wetlands). Run the wildfire_underwrite preset:

curl -s https://api.mireye.com/v1/fetch \
  -H "Authorization: Bearer $MIREYE_API_TOKEN" \
  -H 'content-type: application/json' \
  -d '{
    "lat": 46.6,
    "lng": -93.7,
    "preset": "wildfire_underwrite"
  }' | jq

The response

{
  "lat": 46.6,
  "lng": -93.7,
  "fetched_at": "2026-06-12T07:28:11.332523+00:00",
  "fields": {
    "lcms_class": {
      "value": "Grass/Forb/Herb",
      "unit": null,
      "source": "USFS_LCMS",
      "source_url": "https://developers.google.com/earth-engine/datasets/catalog/USFS_GTAC_LCMS_v2024-10",
      "confidence": "high",
      "fetched_at": "2026-06-12T07:28:08.687077+00:00",
      "dataset_vintage": "2025-11",
      "ttl_seconds": 2592000,
      "notes": null
    },
    "tree_canopy_pct": {
      "value": 0.0,
      "unit": "percent",
      "source": "USGS_NLCD_TCC_V2023_5",
      "source_url": "https://developers.google.com/earth-engine/datasets/catalog/USGS_NLCD_RELEASES_2023_REL_TCC_v2023-5",
      "confidence": "high",
      "fetched_at": "2026-06-12T07:28:11.331689+00:00",
      "dataset_vintage": "NLCD TCC 2023",
      "ttl_seconds": 2592000,
      "notes": null
    },
    "ndvi_current": {
      "value": 0.18159930408000946,
      "unit": null,
      "source": "COPERNICUS_S2_SR_HARMONIZED",
      "source_url": "https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S2_SR_HARMONIZED",
      "confidence": "high",
      "fetched_at": "2026-06-12T07:28:11.331727+00:00",
      "dataset_vintage": "S2 SR 60d window @ 2026-06-12",
      "ttl_seconds": 604800,
      "notes": null
    },
    "ndvi_change_5y": {
      "value": -0.492433562874794,
      "unit": null,
      "source": "COPERNICUS_S2_SR_HARMONIZED",
      "source_url": "https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S2_SR_HARMONIZED",
      "confidence": "medium",
      "fetched_at": "2026-06-12T07:28:11.331776+00:00",
      "dataset_vintage": null,
      "ttl_seconds": 604800,
      "notes": null
    },
    "slope_degrees": {
      "value": 0.053343165665864944,
      "unit": "degrees",
      "source": "USGS_3DEP_COG",
      "source_url": "https://www.usgs.gov/3d-elevation-program",
      "confidence": "medium",
      "fetched_at": "2026-06-12T07:28:07.980417+00:00",
      "dataset_vintage": null,
      "ttl_seconds": 31536000,
      "notes": null
    },
    "elevation": {
      "value": 366.107299805,
      "unit": "meters",
      "source": "USGS_EPQS",
      "source_url": "https://epqs.nationalmap.gov/v1/json?x=-93.7&y=46.6&wkid=4326&units=Meters",
      "confidence": "high",
      "fetched_at": "2026-06-12T07:28:09.139114+00:00",
      "dataset_vintage": "3DEP dynamic service",
      "ttl_seconds": 31536000,
      "notes": null
    }
  },
  "partial_failures": []
}

The audit trail

The underwriter writes their loss memo:

Wildfire risk at 46.6, -93.7 is low. The site is open herbaceous cover (USFS LCMS dominant class “Grass/Forb/Herb”) with 0.0% tree canopy (NLCD TCC) — no forested fuel structure. Current greenness is low (Sentinel-2 NDVI 0.18) and NDVI declined −0.49 over five years, so cured or disturbed fine fuels are plausible and worth a dry-season review, but terrain is flat (0.05°) and there is no woody fuel to carry a high-intensity burn.

The state regulator asks: “where did the 0.0% canopy figure come from?” The underwriter clicks the source_url — the NLCD Tree Canopy Cover dataset page — which documents the product and methodology. The underwriter shares the fetched_at timestamp; the regulator can re-fetch the same value from the same source and confirm the underwriter’s claim to within a 30-day product cadence. Compare with a proprietary CAT model: the underwriter would have to email the vendor, wait for a response, and hope the methodology PDF matches what the model actually does.

Adding flood

Same property, separate call (or batch via fields to combine):

curl -s https://api.mireye.com/v1/fetch \
  -H "Authorization: Bearer $MIREYE_API_TOKEN" \
  -H 'content-type: application/json' \
  -d '{
    "lat": 46.6,
    "lng": -93.7,
    "preset": "flood_risk"
  }' | jq '{fields: (.fields | {elevation, intersects_wetland, wetland_type, wetlands_within_500m_count}), partial_failures}'

{
  "fields": {
    "elevation": {
      "value": 366.107299805,
      "source": "USGS_EPQS",
      "confidence": "high"
    },
    "intersects_wetland": {
      "value": true,
      "source": "USFWS_NWI",
      "confidence": "high"
    },
    "wetland_type": {
      "value": "Palustrine Emergent Nonpersistent Artificially Flooded Farmed",
      "source": "USFWS_NWI",
      "confidence": "high"
    },
    "wetlands_within_500m_count": {
      "value": 25,
      "source": "USFWS_NWI",
      "confidence": "high"
    }
  },
  "partial_failures": [
    {
      "field": "within_floodplain_polygon",
      "source": "FEMA_NFHL",
      "error": "source returned null at this coordinate",
      "retryable": false
    }
  ]
}

The parcel sits on a USFWS-mapped wetland — and the NWI wetland type (“Palustrine Emergent … Farmed”) plus the 25 wetlands within 500 m tell the underwriter this is farmed wetland terrain, not open water. Note the honesty pattern doing its job: FEMA’s NFHL has no digitized panel at this rural coordinate, so within_floodplain_polygon arrives as a partial_failures entry (retryable: false) instead of a silent false. The memo records “SFHA status: undetermined — no NFHL coverage” with the failure record as evidence.

Why this works for insurance

Defensible. Every value cites the originating federal agency. The underwriter never has to defend a proprietary “model says 18%” number.
Reproducible. fetched_at plus source_url means a regulator three years from now can re-fetch the same source and confirm what was used at underwrite time.
Honest. When a source fails (cloudy Sentinel pass, EIA pipeline data lag), the failure surfaces in partial_failures with a retryable flag rather than silently substituting a stale value.
Composable. The carrier’s own scoring model lives on top — Mireye provides the inputs with provenance, the carrier’s actuaries weight them. The audit trail is preserved end-to-end.

What it does NOT do

Not a CAT model. Mireye returns the input data; modeling loss probability is the carrier’s job.
Not a parcel-search service. Geocoding from address to lat/lng is the caller’s responsibility (US Census Geocoder is free and works).
Not real-time event data. No active fire perimeters, no current flood gauge readings. Add those alongside via NIFC / USGS Water APIs if your underwriting requires them.

​The query

​The response

​The audit trail

​Adding flood

​Why this works for insurance

​What it does NOT do