1 · What Is the Earthflow Fire Risk Module?

Earthflow's Physics AI™ Fire Risk Assessment Module is a dual-pathway engine that models surface fire and subsurface peat combustion as independent risk streams, then couples them via multiplicative amplification to capture compound risk. This coupling — surface ignition igniting underlying peat — is the mechanism behind multi-week smoldering fires that burn invisibly underground and cannot be put out with conventional suppression.

Pathway A · Surface Fire

Rothermel spread dynamics (simplified)

LANDFIRE FBFM40 fuel model (30 m) → fuel load class
GridMET ERC & KBDI → fire weather score
Wind (mph), slope (°), ignition (lightning) adjusters
MODIS / VIIRS 5-yr fire history within buffer

Surface_Score (0–100)

+

Pathway B · Peat Fire

Rein smoldering thermodynamics (Rein 2009; Huang & Rein 2017)

SSURGO organic matter, texture, and drainage → USDA Soil Taxonomy (Histosol class)
Peat depth formula (Fibric Peat / Hemic Peat / Sapric Peat)
Sentinel-1 C-band SAR subsurface moisture (10 m) or SMAP rootzone (10 km)
Detection-difficulty factor (burn depth → surface visibility)

Peat_Score (0–100)

Convergence — Combined Score

amplification = min(1.5, 1.0 + Surface_Score / 200) amplified_peat = min(100, Peat_Score × amplification) combined = 0.55 · Surface_Score + 0.45 · amplified_peat compound_bonus = ×1.10 if Surface_Score > 40 AND Peat_Score > 40

References: Rothermel spread dynamics · Keetch-Byram drought physics · Rein smoldering thermodynamics (Rein 2009; Huang & Rein 2017).

Figure 1.1 · Physics AI™ Convergence Flow

The two physics pathways run independently, then couple via multiplicative amplification to capture compound surface→peat risk.

Figure 1.2 · Surface→Peat Amplification Curve

Amplification scales linearly with Surface_Score up to the 1.5× cap. All four Peatville candidates fall in the 1.17–1.22× band.

Figure 1.3 · Surface × Peat Risk Quadrant — Where Each Candidate Sits

The compound-hazard quadrant (shaded) is the decision boundary — Candidate A lives there. C narrowly misses the Surface≥40 threshold; B and D collapse to the Peat_Score=0 axis.

2 · Current Capabilities

Every capability below is live in production today and runs on every site analyzed by the Earthflow platform. Runtime is sub-second per site against pre-collected geospatial data.

19+ Geospatial Data Sources Live

SSURGO soils (30 m) · LANDFIRE FBFM40 (30 m) · GridMET ERC/KBDI (daily) · Sentinel-1 C-band SAR (10 m, 6–12 d revisit) · SMAP (10 km, 2–3 d) · MODIS/VIIRS fire history · USGS RCMAP · terrain · wind · vegetation indices.

49+ Output Fields in <1 s Live

Peat classification (Fibric Peat / Hemic Peat / Sapric Peat / Organic-rich Loam / Mineral Soil), smoldering spread rate (0.5–5.0 cm/hr, superlinear), burn depth, days-to-extinguish, detection difficulty, fire-season windows, peak months.

Dual-Pathway Physics Coupling Live

1.5× surface→peat amplification with a +10% compound-hazard bonus when both pathways exceed threshold. One of the only solar-industry platforms to explicitly model the surface- ignition-to-peat pathway.

Hierarchical Fallback + Provenance Live

Hierarchical fallback chains on every input (KBDI: GridMET → ERC×5.5 → simplified K&B). Every output carries a data-source tag and contributes to a 0.30–0.95 confidence score. Graceful degradation, not errors.

Peat-Specific Physics Calibration Live

USDA Soil Taxonomy (Histosol class) from SSURGO organic matter + texture + drainage. Continuous burn-depth function (0.05 → 2.0 m). Superlinear smoldering spread: 0.5 + 4.5 × (deficit)^1.5 cm/hr. Days-to-extinguish up to 30 d.

Site-Actionable Economics Live

Risk tier → mitigation $/acre (+$500–$4,000 peat adder), insurance surcharge (0–40%), construction delay days, tiered monitoring protocols (thermal cadence, sensor spacing, KBDI halt thresholds).

Figure 2.2 · Data Stack · Spatial Resolution × Revisit Cadence

Earthflow stacks 8 production data sources today (solid) with a 3-source roadmap tier (dashed) — FireSat, GOES-16 ABI/FDCA, and GOES-16 GLM — that extends active monitoring from daily into the sub-hourly regime. Geostationary thermal (5-min) and lightning detection (continuous) close the gap between ignition and detection.

Figure 2.1 · Superlinear Smoldering Spread Rate

Dry peat doesn't burn linearly — it accelerates. A 5× drier deficit produces an ~8× faster smoldering front.

3 · Future Detection Capabilities

Today Earthflow answers "how likely is ignition?" The next wave of capability answers "is it burning right now, and where?" This is the response to the third-party review that flagged the gap between pre-ignition risk scoring and post-ignition confirmation — especially for smoldering peat fires that can burn for weeks underground. All items below are on the active roadmap.

Figure 3.1 · From Risk Scoring to Active Detection

Today's risk-scoring engine provides the foundation. The detection roadmap closes the loop with active observation of in-progress events.

Active Thermal Anomaly Detection Roadmap

VIIRS I-band (375 m), Landsat 8/9 TIRS (100 m), ECOSTRESS (70 m) for near-real-time detection of in-progress smoldering fronts that surface fire-watch crews cannot see.

Sentinel-1 InSAR Subsidence Mapping Roadmap

Peat combustion consumes organic soil volume, producing 10–50 cm ground subsidence detectable with Sentinel-1 InSAR — physical confirmation of subsurface burn. Genuinely novel signal for solar O&M.

SAR Moisture "Dry Halo" Anomaly Roadmap

Smoldering fronts desiccate surrounding peat, producing a characteristic moisture signature. Automated per-pixel baseline deviation against historical Sentinel-1 SAR.

dNBR Burn-Severity Validation Roadmap

Differenced Normalized Burn Ratio (Landsat/Sentinel-2 SWIR) to back-check predicted burn depth against observed organic consumption after confirmed events.

MTBS / NIFC Historical Back-Testing Roadmap

Run Earthflow against USGS MTBS perimeters + NIFC/IRWIN records. Publish ROC / AUC and calibration curves. Pair with DTE fire-dept records for Peatville back-check.

ML Calibration of SAR → Peat Moisture Roadmap

Replace the linear Sentinel-1 dielectric transform with an ML layer trained on co-located SMAP + in-situ peat observations. Sharpens ignition-threshold crossings.

4 · Peatville Analysis — Key Results

DTE Peatville Solar Park, Minden Township, Sanilac County, MI. Four candidate coordinates run through the Fire Risk Module (Cloud Run, live endpoint). Sites span ~9 km inside and adjacent to the historical Minden Bog (15,000 ac peatland, >50% drained to agriculture mid-20th century).

300×

Peat Fire Score Spread

~9 km

Geographic Footprint

4

Candidates Analyzed

0.85

Confidence (All Sites)

Figure 4.1 · Candidate Coordinates · Minden Township, Sanilac Co MI

Mapbox satellite view of Minden Township, MI with 4 candidate coordinates labeled A–D

Peat soil (A, C) — Fibric Peat adjacent to bog

Mineral soil (B, D) — Drained cropland

A 43.620, -82.830 · Fibric Peat · 75.6 HIGH

B 43.595, -82.820 · Mineral · 2.0

C 43.640, -82.845 · Fibric Peat · 57.0 MOD

D 43.672, -82.776 · Mineral · 0.5

Mapbox satellite basemap · 4 candidates span ~9 km inside Minden Township. A and C (green pins) sit on intact Fibric Peat adjacent to the Minden City State Game Area bog remnant; B and D (amber pins) are on historically drained farmland.

Figure 4.2 · Fire Risk Scores by Candidate

Candidate A shows the full dual-pathway signature: moderate surface risk amplifying a HIGH peat score. B and D, on drained mineral soil, collapse on the peat axis.

Table 4.1 · Candidate-Level Fire Risk Comparison

Metric	A · 43.620, -82.830	B · 43.595, -82.820	C · 43.640, -82.845	D · 43.672, -82.776
Peat Classification	Fibric Peat	Mineral Soil	Fibric Peat	Mineral Soil
Peat Depth (m) · Organic Matter %	≥ 2.00 m · 75%	0.00 m · 4%	≥ 2.00 m · 75%	0.00 m · 1%
Subsurface Moisture (Sentinel-1 SAR)	32.1%	43.4%	44.9%	20.5%
Surface Fire Score	43.4 · Moderate	39.1 · Low	39.1 · Low	34.9 · Low
Peat Fire Score	75.6 · High	2.0 · Minimal	57.0 · Moderate	0.5 · Minimal
Combined Score	71.8 · High	35.4 · Low	52.2 · Moderate	31.4 · Low
Smoldering Ignition Risk	Extreme	Very Low	High	Very Low
Burn Depth (m) · Days to Extinguish	0.77 m · 7 d	0.00 m · 0 d	0.30 m · 3 d	0.00 m · 0 d
Detection Difficulty	Very Hard	Easy	Very Hard	Easy
Mitigation Cost ($/ac)	$7,500	$500	$4,000	$500
Insurance Surcharge	+23%	+2%	+13%	+2%

Figure 4.2a · How Candidate A's 71.8 Is Built — Additive Waterfall

True additive waterfall — Candidate A's 71.8 is the sum of three physics-derived contributions: weighted surface (23.9) + weighted amplified peat (41.4) + compound-hazard bonus (6.5). Crosses the HIGH threshold (70) because of the ×1.10 bonus, which only fires when both pathways exceed 40.

Candidate A — Recommended Demo Point

43.620° N, -82.830° W (1 mi NE of Minden City State Game Area). SSURGO confirms Fibric Peat, ≥ 2.0 m depth, 75% organic matter. Peat Fire Score 75.6 (HIGH), Smoldering Ignition EXTREME, 0.77 m predicted burn depth, 7 days to extinguish, Very Hard surface detection. This is exactly the scenario described by the Minden Township fire department — a subsurface smoldering front that persists past surface suppression and cannot be visually spotted during inspection rounds. Mitigation posture: $7,500/acre with a +23% insurance surcharge.

Headline Finding

Within a 3-mile radius inside Minden Township, Earthflow resolves a 300× gradient in Peat Fire Score (0.5 → 75.6) — driven entirely by historical drainage history. Parcels still sitting on intact peat (A, C) sit in the High/Moderate bucket with multi-day smoldering persistence; adjacent drained muck farmland (B, D) is Mineral Soil with Very Low smoldering risk. This is exactly the spatial heterogeneity that unlabeled "township-level fire history" cannot resolve — and exactly where the planned thermal / InSAR detection enhancements close the loop.

Confidence & Reproducibility

All four candidates returned confidence = 0.85 (SSURGO + Sentinel-1 SAR + MODIS/VIIRS + GridMET ERC all available; GridMET native KBDI fell back to ERC-calibrated proxy for all four). Scoring math has been independently reproduced from the peat_fire_risk.py formulas: Combined Score for Candidate A = 71.8 (verified); economic outputs $7,500/ac @ 23% match the hardcoded mitigation + insurance tables exactly.

What Earthflow Delivers Across the Project Lifecycle

Table 4.1's "Very Hard" detection rating is a statement about surface visual inspection — not about what Earthflow provides. The capabilities below map to the phases DTE actually spends money in.

Figure 4.3 · Earthflow Capability Map · Full Project Lifecycle

Orbyfy Earthflow spans siting through incident response — not a one-shot screening tool. 17 capabilities delivered today across siting / underwriting / construction / response; 7 roadmap items close the active-monitoring loop.

Table 4.2 · Lifecycle Capabilities Checklist

Project Phase	Orbyfy Earthflow Capabilities Delivered
Siting & Screening	✓ Dual-pathway physics score · ✓ SSURGO peat classification · ✓ Parcel-level $/ac mitigation · ✓ Drainage-history resolution (300× gradient in 3 mi)
Underwriting & Permitting	✓ 49+ output fields with provenance · ✓ Confidence score (0–1) · ✓ Reproducible physics formulas · ✓ Insurance surcharge band (0–40%)
Construction Planning	✓ KBDI trajectory for build season · ✓ Huang-Rein ignition thresholds on disturbed peat · ✓ Protected-area setback flags · ⟳ No-dig window alerts
Operations & Monitoring	⟳ Thermal anomaly detection (VIIRS / TIRS / ECOSTRESS) · ⟳ InSAR subsidence alerts (10–50 cm) · ⟳ SAR dry-halo anomaly · ⟳ O&M alert SLAs
Incident Response	✓ Predicted burn depth · ✓ Days-to-extinguish estimate · ✓ Smoldering ignition rating · ✓ Superlinear spread-rate forecast

✓ = available today in v2.0 · ⟳ = active roadmap item (see Section 3)

5 · Proposed Engagement — Next Steps for Peatville

The 4-candidate analysis above is a demonstration of resolution, not a project assessment. To turn this into an operational tool for DTE, we propose a joint engagement that (a) runs the full array footprint, (b) back-tests Earthflow against Minden Township fire-department records, and (c) connects the risk engine to DTE O&M workflows for the roadmap detection capabilities. Each step below is scoped to be deliverable within the existing Earthflow v2.0 stack — no new schema fields, no model retraining, and no changes to the canonical methodology.

Full-Footprint Hazard Map Proposed · Wks 1-2

Swap 4 candidates for DTE's actual array polygon. Run per-cell scoring on a 30 m SSURGO grid across the entire project boundary. Deliverable: parcel-by-parcel Peat Score heatmap + mitigation-cost roll-up at the array level — not point estimates.

Minden Twp Fire-Dept Back-Test Proposed · Wks 2-4

Ingest documented historical peat-fire incidents from the Minden Township fire department plus surrounding bog / muck fires (SGA records, MTBS perimeters ≥ 1,000 ac). Score each ignition location retroactively and publish ROC / recall / calibration against Earthflow thresholds. Closes the "does this actually predict what you've seen?" question.

Construction-Phase Ignition Assessment Proposed · Wks 3-4

Excavation, pile-driving, and vegetation clearing are established ignition mechanisms on disturbed peat. Run Huang-Rein ignition thresholds against projected construction-phase KBDI trajectories for the chosen build season; flag no-dig windows to the EPC schedule.

Dry-Season (KBDI) Analog Lookback Proposed · Wks 2-3

Replay Candidate A / C scoring through 2012 (Seney NWR), 2016 (Sleeper Lakes, UP), and 2023 (Canadian peat megafires) KBDI trajectories to show what these parcels would have scored during known-bad peat years. Quantifies climate-scenario risk for insurer conversations.

Active-Monitoring Tie-In Proposed · Wks 4-8

Pipe the roadmap detection layers (thermal anomaly · SAR dry-halo · InSAR subsidence) into DTE's existing O&M alerting with site-specific SLAs. Uses the Section 4 candidates as the initial false-positive calibration set.

Ground-Truth Validation Visit Proposed · Wk 4

Joint site visit with DTE + Orbyfy to probe peat depths, pull organic-matter lab samples at A / C, and confirm SSURGO classification against in-situ soil. Anchors the model's most uncertain inputs in measured data and feeds the ML-calibration roadmap item.

Engagement Model

All six items use the existing Earthflow v2.0 pipeline — no new schema fields, Peatville numbers are not re-run, and the canonical methodology is unchanged. Estimated total engagement window: ~8 weeks, parallelizable across tracks. The first two deliverables — full-footprint hazard map and fire-dept back-test — are the highest-leverage for DTE's current underwriting and siting decisions.

Earthflow Fire Risk — Capabilities Overview