The Hummingbird Structure Inventory (SI) is a unified building-level dataset covering 122.9 million structures across 49 US states. Every field is derived from publicly available government sources — county tax assessor records (CAMA), federal regulatory GIS layers, USGS instrumented measurements, FEMA databases, and US Census data — fused and calibrated into a single record per building.
This document describes every field in the SI, organized by the method by which the data was collected. Confidence is field-specific, not tier-specific. A lidar ground elevation measurement and an NSI tract median year are both technically "government data" — but the first is a direct instrument reading and the second is a statistical proxy. The Confidence column tells you exactly which is which.
Two attributes drive the largest share of wind/hail underwriting variance: year_built (code vintage, post-storm resilience) and construction_type (wood vs masonry vulnerability). The table below shows exactly how much of the 123.9M-building national inventory is covered at each confidence level for each attribute.
| Source | Confidence | Year-Built | Const. Type | Notes |
|---|---|---|---|---|
| CAMA / FDOR (direct) | HIGH | 30.0M (24.2%) | 14.9M (12.0%) | County tax assessor records. Recorded at permit issuance. FL = FDOR NAL (full state). Used as ground truth for all downstream models. |
| AEI / First American | HIGH | 44.9M (36.2%) | — | First American national parcel database (109M parcels). Same source methodology as CAMA — building-level assessor records. AEI vs CAMA MAE = 5.0 yrs (85% within 5 yrs). Applied to buildings not covered by CAMA. |
| HB Model Inference | MED | 13.3M (10.8%) | 74.9M (60.5%) | Year-built: spatial neighbour inference (50m BallTree, CAMA/AEI donors). Construction type: GradientBoosting classifier trained on CAMA ground truth (70% accuracy). Applied only where year_built quality source was available. |
| County/State Median | LOW | 35.7M (28.8%) | — | Fallback for buildings where no CAMA, AEI, or spatial donor was available. County median used where possible; state median as last resort. MAE vs assessor ~8–15 yrs. |
| NSI Census Proxy | LOW | — | 34.1M (27.5%) | Census block-group aggregate. NSI vs AEI MAE = 16.2 yrs (33% within 5 yrs). Construction type: HAZUS default class for tract. Applied only where no quality year_built existed to enable model inference. Do not use for underwriting decisions. |
| Total / Combined | 123.9M (100%) | 89.8M (72.5%) | CT coverage = CAMA (12%) + model inferred (60.5%). Remaining 27.5% is NSI-only. Year-built: HIGH/MED = 88.4M (71.4%); LOW = 35.7M (28.8%). |
Use effective_year_built_source to filter by quality tier. Use construction_type_source + construction_type_inferred to distinguish assessor-recorded from model-predicted construction class.
Highest confidence. These fields were either recorded at the moment of permit issuance (county assessor / CAMA records) or measured directly by physical instruments (lidar elevation surveys, seismograph networks). They reflect what the building actually is — not a statistical estimate of what it might be.
construction_type, year_built, and roof attributes are the most important fields for wind and hail underwriting. Where these come from CAMA or FDOR, confidence is HIGH. The construction_type_source field tells you exactly which source was used for every building — use it to filter your analysis to high-confidence records where it matters.
Florida FDOR NAL note: For Florida's 8.4M structures, the Florida Department of Revenue (FDOR) statewide NAL database is the assessor source — it is the same type of permit-recorded data as county CAMA, but with statewide coverage in a single dataset. Post-Andrew (1992) and post-2004 hurricane season code changes mean that FL FDOR vintage + construction_type enables precise wind vulnerability classification.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| construction_type | Primary structural system. Values: wood_frame, masonry, steel, manufactured. 100% filled: 12.1% from CAMA/FDOR (HIGH confidence — permit-recorded); 87.9% from NSI inference (LOW confidence — 51–60% accuracy vs. assessor). Use construction_type_source to filter to high-confidence records. |
100%
21.9% CAMA/FDOR; 78.1% other
CAMA covered
|
HIGH (CAMA)LOW (NSI)
≥80% CAMA
40–79%
1–39% CAMA
0% (NSI only)
FDOR NAL
County CAMA
NSI fallback
|
The single most important field for wind and hail underwriting. CAMA validation shows NSI achieves only 51–60% agreement vs. assessor ground truth — biased toward wood_frame, with masonry recall as low as 25% in TX and steel recall 14% in FL. Cedant EDMs derived from NSI likely carry the same bias. Map shows CAMA coverage % by state; filter to CAMA records for high-confidence analysis. |
| construction_type_inferred | Model-predicted construction type for buildings where CAMA/FDOR did not provide a value. GradientBoosting classifier trained on CAMA ground truth (~12M buildings). Features: effective_year_built, sqft, census_region, census_division, cbsa_type, num_story. Only applied where effective_year_built_source ∈ {cama, aei, inferred_spatial, inferred_county}. |
—
model inference
|
MED | Replaces NSI-sourced construction_type where AEI/CAMA year_built is now available. Accuracy ~65% vs assessor ground truth (vs 51–60% for NSI baseline). Use construction_type_source to distinguish CAMA-sourced from model-inferred rows. |
| construction_type_source | Identifies which data source provided the construction_type value for each building. Values: fdor_nal, county_assessor, cook_county_cama, nsi_bldgtype. |
100%
|
PROVENANCE
Derived
|
Use this field to filter analyses to high-confidence records. Filter to fdor_nal or county_assessor for underwriting decisions that require permit-grade construction data. |
| year_built_assessor | Year structure was built or permitted, from tax assessor records. State coverage varies dramatically — see map. |
~15% national avg
≥80%
40–79%
<40%
0–10%
|
HIGH where present
County CAMAFDOR NAL
|
State coverage: FL 95%, CT 94%, AZ 61%, UT 47%, OR 33%, CO 31%, MN 27%, NE 29%, NY 21%, WA 19%, VA 16%, NV 15%, NC 15%, PA 13%, IL 12%, KS 12%, OK 14%, OH 7%, TX 6%, CA 4% — most other states 0–5%. Where absent, med_yr_blt (NSI tract median) substitutes but is LOW confidence. Drives wind-code vintage assignment. |
| effective_roof_year | Year of the most recent roof installation. Priority: roofing permit records (actual replacement year) > county assessor records. This is the best available indicator of current roof age — distinct from year_built. |
92% national avg
≥80%
40–79%
<20%
|
HIGH (permit)MED (CAMA)
Building PermitsCounty CAMA
|
Key states: IL 98%, IA/NE/OK/AR/RI 99–100%, MN 97%, OH/PA/NY 95–96%, FL 92%, MA 92%, AL 96%, TN 91%. Low coverage: WA 11%, CA 23%, NV 56%, HI 0%. A 20-year-old shingle vs. a 3-year-old one has dramatically different hail and wind loss performance. |
| roof_cover_assessor | Roof covering material from FDOR NAL. Highest-confidence roof material field when present. FL only. |
7% (FL only)
|
HIGH
FDOR NAL
|
FL only. Where present, this is permit-recorded roof material — more reliable than modeled roof_material (Tier 5). For all other states, use Tier 5 roof_material. |
| enclosure_class | Building enclosure per ASCE 7 / Florida Building Code. Values: Enclosed, Partially Enclosed, Open. FL only (13%). |
13% (FL only)
|
HIGH
FDOR NAL
|
Partially enclosed structures face significantly higher internal wind pressure under ASCE 7. Critical hurricane vulnerability variable. Not available from assessor records outside FL. |
| sqft_assessor | Conditioned floor area in square feet from FL FDOR assessor records. FL only. |
7% (FL only)
|
HIGH
FDOR NAL
|
FL only. Higher precision than NSI sqft. Nationally, use sqft (Tier 3, NSI) as the primary area field. |
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| ground_elv | Ground surface elevation in feet at building centroid. Measured by USGS 3DEP 1-meter lidar/photogrammetry — a direct instrument reading, not a model estimate. |
100%
|
HIGH
USGS 3DEP
|
Basis for all flood stage calculations and freeboard determinations. The most accurate nationally available digital elevation model — 1-meter horizontal resolution lidar. |
| first_floor_elv | Estimated first-floor elevation in feet AMSL = ground_elv + foundation height offset (varies by found_type). |
100%
|
HIGH
USGS 3DEPHB Derived
|
Flood damage is a function of water depth above first floor, not ground elevation. The difference between first_floor_elv and bfe_ft (freeboard) determines FEMA flood zone compliance. |
| ss | Mapped Maximum Considered Earthquake (MCE) spectral response acceleration at short periods (g), ASCE 7-22. |
100%
|
HIGH
USGS NSHM
|
Derived from decades of seismograph observations at site-specific locations. Input to seismic design category determination. |
| s1 | MCE spectral response acceleration at 1-second period (g). |
100%
|
HIGH
USGS NSHM
|
1-second period is critical for mid-rise and taller structures. Pair with num_story to determine dominant vibration period. |
| pgam | Geometric mean peak ground acceleration (g) for MCE ground motion. |
100%
|
HIGH
USGS NSHM
|
Used for nonlinear soil response and liquefaction hazard. Critical for Pacific Coast portfolios. |
| sds | Design spectral acceleration at short periods (g) — primary seismic demand parameter for low-rise residential. |
100%
|
HIGH
USGS NSHM
|
Used directly in seismic vulnerability functions for residential structures. |
| sd1 | Design spectral acceleration at 1-second period (g). |
100%
|
HIGH
USGS NSHM
|
Pair with num_story to determine dominant vibration period for mid-rise structures. |
Federal agencies publish these as official standards or regulatory designations — the same maps that govern building codes, mandatory flood insurance, and wildfire land management. High confidence for what they measure, but area-based (not property-specific instrument readings). Every building gets a value for every field in this tier.
Why this matters: Regulatory spatial data is the foundation of catastrophe risk zoning. FEMA flood zones determine mandatory flood insurance purchase. ASCE 7 wind speeds set the design standard builders were required to meet. USGS seismic parameters define required building code seismic design category. WUI classifications identify structures in ember cast risk zones. These fields directly enable zone-based accumulation, treaty line definition, and regulatory compliance checking.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| asce7_wind_speed_mph | ASCE 7-22 design wind speed (3-sec gust at 10m) at building location, in mph. This is the current design standard — not the standard in effect at year of construction. |
100%
|
HIGH
ASCE 7-22
|
FL coastal 165–200+ mph; Gulf barrier islands similar; Midwest interior ~115 mph. Based on long-term weather station observations. Use wind_code_vintage for the applicable standard at construction time. |
| wind_code_vintage | Wind design code edition in effect when the building was constructed. Derived from year_built and ASCE 7 / state building code adoption history. |
100%
|
HIGH
HB Derived
|
Pre-1993 FL structures were built to pre-Andrew codes — significantly weaker than post-2002 Florida Building Code. Key signal for FL wind vulnerability stratification. |
| sdc | Seismic Design Category (ASCE 41). Values A–E: A/B = low seismic risk (most of eastern US); D/E = high (Pacific Coast, New Madrid Zone). |
100%
|
HIGH
USGS NSHM
|
SDC D/E = seismic-resistant structural systems required by code. Used to flag seismic vulnerability and determine required construction methodology at time of build. |
| nehrp_site_class | NEHRP soil site amplification class. A = hard rock; B = rock; C = dense soil; D = stiff soil; E = soft clay / Bay Mud. |
100%
|
HIGH
USGS NSHM
|
Class E (SF Bay Mud, Seattle fill areas) can amplify ground motion 3–5× relative to Class A rock at the same location. Critical for CA and PNW earthquake accumulation analysis. |
| firmzone | FEMA NFIP Flood Insurance Rate Map zone at building location. AE = 1% annual flood risk (riverine); VE = 1% flood + wave action (coastal); X = outside 500-yr boundary; AO = sheet flow. |
100% joined
16% of buildings in a mapped flood zone
|
HIGH
FEMA NFIP FIRM
|
VE zones = highest flood damage potential (wave action + inundation). Mandatory flood insurance purchase trigger for federally backed mortgages in AE/VE zones. |
| bfe_ft | FEMA Base Flood Elevation in feet AMSL — the regulatory 1% annual chance flood level. Only exists where FEMA has published detailed depth grids. |
3%
Major rivers, coastal SFHA zones only
|
HIGH where present
FEMA FIRM Depth Grids
|
Where present, this is the most authoritative available flood elevation benchmark. Low national coverage because FEMA only produces depth grids for recently updated, detailed study FIRMs. |
| freeboard_ft | First floor elevation minus BFE, in feet. Positive = elevated above BFE (lower flood risk). Negative = below BFE (high flood damage probability). |
3%
|
HIGH
HB Derived
|
Buildings with freeboard < −2 ft historically experience catastrophic losses even in moderate flood events. Available only where BFE data exists. Strong predictor of NFIP claim frequency. |
| in_wui_ignition | TRUE if building is inside FEMA 2020 WUI Intermix zone — wildlands physically surround the structure. Highest direct fire risk. FALSE or null = not in Intermix. |
100% joined
1–14% TRUE by state
|
HIGH
FEMA WUI 2020
|
Intermix WUI = structures scattered among dense vegetation. Direct flame contact from adjacent vegetation is plausible. State distribution: ME 13.4%, VT 13.3%, KY 11.6%, WV 11.4% highest; AZ 0.9%, NV 1.0%, FL 1.4% lowest. |
| in_wui_ember | TRUE if building is in FEMA 2020 WUI Interface zone — within ~2km ember cast distance of wildlands. Broader than Intermix — most US buildings qualify. |
100% joined
51–93% TRUE by state
|
HIGH
FEMA WUI 2020
|
HI 92.5%, DE 89.5%, MD 83.7% highest; VT 51.4%, NV 57.7%, NH 62% lowest. Ember ignition is the primary cause of structure loss in wildfire interface events. Pair with wrc_bp_rank for actual community fire probability. |
| wui_class | USFS legacy WUI density/vegetation category (e.g., Med_Dens_Interface, High_Dens_Intermix). Superseded by in_wui_ignition / in_wui_ember for modeling. |
100%
|
MEDIUM
USFS WUI
|
More granular than binary WUI flags but less current than FEMA 2020 layer. Useful for legacy analysis compatibility. Use in_wui_ignition and in_wui_ember for primary wildfire risk stratification. |
| wrc_bp_rank | USFS Wildfire Risk to Communities community burn probability rank, 0–1 scale. Higher = higher community fire probability nationally. Community-level (TIGER Places) — 31.5% of buildings in unincorporated areas are null. |
68.5%
31.5% null (unincorporated)
≥80%
40–79%
<40%
|
MEDIUM
USFS WRC
|
CA mean 0.93 (93rd pctile nationally), FL 0.81, VT 0.35. Most actionable wildfire risk stratification field in the SI for portfolio accumulation. Null for unincorporated rural areas (31.5% of buildings). |
The National Structures Inventory (NSI) is a federally maintained building database produced by the US Army Corps of Engineers and FHWA. It provides 100% national coverage for all fields in this tier. The tradeoff is granularity: NSI uses statistical inference at the census tract level, not per-building permit data. Confidence varies significantly by field within this tier — check the Confidence column for each.
Why this matters: NSI provides the foundational building skeleton — occupancy type, story count, foundation type, baseline replacement cost — for every structure in the country. Where assessor coverage is sparse (most of the US outside FL), NSI is the primary source for construction class. Hummingbird applies local Marshall & Swift cost calibration on top of NSI values for replacement cost (see Tier 5).
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| occtype | Occupancy classification. Common values: RES1-1SNB (SFR, no basement), RES1-1SWB (SFR, basement), RES2 (mobile home), RES3A (multi-family 3–4 units), COM1 (retail). |
100%
|
MED–HIGH
NSI
|
Reasonably accurate for basic residential type classification. Essential for occupancy-differentiated loss curves. |
| num_story | Number of above-grade stories. NSI uses census-calibrated story assignments by occupancy type and vintage. |
100%
|
MEDIUM
NSI
|
Mostly accurate but not permit-recorded. Influences wind pressure calculations (taller buildings face higher design wind loads) and replacement cost estimation. |
| found_type | Foundation type. Values: Slab, Crawlspace, Basement, Pier. Modeled from census tract data — not measured. |
100%
|
MEDIUM
NSI
|
Critical for flood loss estimation. Slab-on-grade floods at grade level; pier/crawlspace have elevated first floors. Drives first_floor_elv calculation. |
| sqft | Gross floor area in square feet. Primary area field for all states. NSI calibrates from ACS housing surveys at the tract level. |
100%
|
MEDIUM
NSI
|
Primary area basis for rcv_adj_struct_usd nationally. Less accurate than assessor sqft_assessor (FL). Use sqft_assessor where available. |
| med_yr_blt | Median year built for the NSI census block group — a tract-level proxy, not a building-specific value. |
100%
|
LOW
NSI
|
Use only when year_built_assessor is absent. A tract median applied to all buildings in the tract obscures within-tract vintage variation. Do not use for building-specific wind code vintage assignment. |
| val_struct | NSI baseline replacement cost in USD before Hummingbird calibration. Known to underestimate in high-cost coastal markets (FL, CA, TX) by 20–40%. |
100%
|
LOW
NSI
|
Do not use for loss modeling — use rcv_adj_struct_usd (Tier 5) instead. Retained here for benchmarking against the NSI baseline only. |
| bldgtype | NSI construction class inference: W (wood / light frame) or M (masonry). Inferred from census tract housing survey data. |
100%
|
LOW–MED
NSI
|
Superseded by construction_type where CAMA is available. Used as the national fallback (fills ~8.5% of buildings where no assessor record exists). Statistically unbiased at portfolio level but not reliable at individual building level. |
Overture Maps provides GPS location and building geometry from satellite and aerial imagery. High confidence for spatial position — sub-5-meter accuracy for 97% of structures — but lower confidence for physical attributes like height that require dense lidar coverage.
Why this matters: Accurate building geometry is essential for catastrophe modeling. Footprint area (not floor area) determines roof exposure. Spatial precision affects whether a structure falls inside or outside a flood zone, WUI boundary, or wind speed contour. Overture's satellite-derived geometry is generally more accurate than parcel centroids for spatial intersection.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| lat | Building centroid latitude (WGS84 decimal degrees). Sub-5-meter accuracy for 97% of structures. |
100%
|
HIGH
Overture Maps
|
All spatial joins (FEMA flood zones, ASCE wind speeds, WUI layers, seismic parameters) are performed on these coordinates. Foundation of all risk assignment in the SI. |
| lon | Building centroid longitude (WGS84 decimal degrees). |
100%
|
HIGH
Overture Maps
|
See lat. |
| footprint_area_sqm | Building footprint area in square meters measured from the Overture polygon. Distinct from sqft (floor area) — for a 2-story building, sqft ≈ 2× footprint. |
90%
|
HIGH
Overture Maps
|
Roof exposure for hail and wind is proportional to footprint area, not total floor area. Especially useful for large commercial structures where satellite-derived footprint is more reliable than NSI estimates. |
| height | Building height in meters derived from satellite stereo imagery or photogrammetry. Sparse in rural areas. |
68%
|
MEDIUM
Overture Maps
|
Available for most urban/suburban areas; sparse in rural regions. Used to corroborate story count and calculate exposure-weighted wind pressure. Where absent, derive from num_story. |
| roof_shape | Roof geometry. Values: Gable, Hip, Flat, Complex. 8% national; FL better covered via FDOR supplemental data. |
8% national
|
MEDIUM where present
Overture MapsFDOR NAL
|
Hip roofs are significantly more wind-resistant than gable roofs under ASCE 7. Coverage too sparse to use nationally — where absent, roof shape is null (not inferred). Use only as a supplemental signal where available. |
| overture_id | Unique building identifier from the Overture Maps database. ~3% of SI buildings sourced from NSI geocoordinates without a matched Overture polygon. |
97%
|
IDENTIFIER
Overture Maps
|
Stable identifier for cross-referencing against future Overture releases. |
These fields do not come directly from any single government database. Hummingbird engineers and actuaries derive them by applying physical and actuarial models to the source data in Tiers 1–4. The inputs are all publicly sourced; the outputs are HB's best estimate of each building's engineering characteristics. Confidence depends on the quality of underlying inputs — higher for FL where Tier 1 data is complete, lower for NSI-only states.
Why this matters: The most important wind vulnerability variables — nail schedule, roof-to-wall connection, opening protection — are almost never recorded in any public database. They can only be inferred from building vintage, construction class, and the wind code in effect when the permit was filed. HB's inference logic is based on IBHS field research, Florida Building Code historical records, and ASCE 7 edition histories. The replacement cost field (rcv_adj_struct_usd) corrects for known NSI biases in high-cost coastal markets.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| rcv_adj_struct_usd | Hummingbird replacement cost in USD — locally calibrated per county using Marshall & Swift construction cost factors on sqft × construction_type. Replaces NSI val_struct. |
100%
|
MED–HIGH
HB Modeled
|
Use this field — not val_struct — for all exposure aggregation and loss modeling. Validated against FDOR improvements_value in FL. Significantly more accurate than NSI in coastal markets (Miami-Dade, Naples, coastal CA) where construction costs are 20–40% above national averages. |
| roof_material | Roof covering material. Values: Asphalt Shingle, Metal, Tile, Built-up, Single-ply Membrane. Priority: FDOR > Overture > NSI/inference. |
100%
|
MEDIUM
FDOR NALOverture MapsNSI fallback
|
Asphalt shingle dominates residential nationally (~75%). Metal and tile more prevalent in coastal FL. Tile is significantly more vulnerable to hail than metal. Critical for hail damage modeling. |
| roof_deck_attachment | Roof deck nail specification. Values: 6d (6-penny common), 8d (8-penny common), 6d_ring_shank. Inferred from vintage + construction_type + state wind code history. |
100%
|
MEDIUM
HB Modeled
|
Nail schedule is the single most important predictor of roof deck failure in hurricanes (IBHS research). Pre-1994 FL = predominantly 6d; post-2002 = predominantly 8d ring-shank. Operationalizes a known carrier data bias: carriers claim 40% ring-shank; IBHS field surveys find 70% 6d in pre-1994 FL structures. |
| roof_wall_connection | Method connecting roof structure to top wall plate. Values: Toe-nail, Clip, Strap. Inferred from vintage + code uplift history. |
100%
|
MEDIUM
HB Modeled
|
Single strongest predictor of total roof loss in severe wind. Toe-nail connections (pre-1992 FL) fail at much lower wind speeds than hurricane straps. FL mandatory strap requirement phased in post-Andrew, fully enforced under 2002 Florida Building Code. |
| opening_protection | Window and door opening protection level. Values: None, Basic, Hurricane-rated. Inferred from age + wind zone. |
100%
|
MEDIUM
HB Modeled
|
Opening breach is the primary failure mode for interior hurricane damage — once a window fails, internal pressure spikes dramatically. Miami-Dade product approval (HVHZ) requirement in effect from 2002 FBC. Post-2002 FL = hurricane-rated; pre-1994 FL = None. |
| wind_vuln_source | Identifies which source combination drove the wind vulnerability inference for each building. Audit / provenance field. |
84%
|
PROVENANCE
Derived
|
Use to understand confidence of Tier 5 wind fields for any given building or portfolio segment. Filter to high-input-quality records for sensitivity testing. |
FEMA publishes historical flood insurance claims at the census tract level. These are real flood insurance payouts — not modeled risk. Medium-high confidence for flood frequency signal, but tract-level (not building-specific). This is the behavioral evidence layer: it tells you whether buildings in this neighborhood have actually flooded before, and at what severity.
Why this matters: Repeat flooding is highly predictable geographically. Tracts with high claim counts and high average losses are documented flood problem areas. The below-BFE percentage (nfip_below_bfe_pct_tract) is a particularly useful signal — it indicates whether the FIRM flood zone designation is consistent with observed losses or understates actual flood exposure in the area.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| nfip_claim_count_tract | Total NFIP flood insurance claims paid in the building's census tract, all years. National median: 14 claims/tract. Outlier tracts: 1,000+. |
97%
|
MED–HIGH
FEMA NFIP
|
Tracts with 1,000+ claims indicate chronic flood exposure regardless of current FIRM designation — the area has flooded repeatedly. Note: reflects NFIP-insured structures only; uninsured losses not captured. |
| nfip_avg_building_loss_tract | Average paid building loss per NFIP claim in the tract, nominal USD across all claim years. National median: ~$15k. |
93%
|
MED–HIGH
FEMA NFIP
|
High average loss + high count = persistent major flood risk. NFIP claim limits cap at $250K building / $100K contents — actual losses may exceed recorded amounts in high-value neighborhoods. |
| nfip_last_claim_year_tract | Most recent year a NFIP claim was paid in the census tract. |
97%
|
MEDIUM
FEMA NFIP
|
Recency signals ongoing vs. historical flood risk. Tracts with recent event claims (2017 Harvey, 2021 Ida, 2024 Helene) may have higher prevalence of flood-damaged structures with unresolved vulnerability. |
| nfip_below_bfe_pct_tract | % of NFIP claims in the tract where the insured property was reported below the Base Flood Elevation. |
97%
|
MEDIUM
FEMA NFIP
|
High below-BFE % = non-compliant construction prevalent in the area, or FIRM underestimates actual flood exposure. Key aggregate flood severity signal — buildings below BFE sustain dramatically higher losses. Also an adverse selection indicator. |
Census survey data and housing market indicators. All tract-level, 5-year rolling averages. Useful for behavioral and maintenance context but not primary physical risk indicators. These describe the neighborhood, not the individual building.
Why this matters: Socioeconomic context influences both actual risk and behavioral response. Owner-occupancy rate and maintenance investment correlate with physical property condition relative to NSI estimates. Pre-1980 vintage percentage helps calibrate older construction prevalence. CBSA classification enables market-level exposure aggregation. FHFA HPI identifies markets where insured value inflation has diverged from reconstruction cost.
| Field | Description | Coverage | Data Quality | Underwriting Notes |
|---|---|---|---|---|
| acs_median_home_value_usd | ACS 5-year tract median owner-occupied home value, USD. |
66%
|
LOW–MED
ACS 5-Year
|
Proxy for local property values. Useful for identifying neighborhoods where NSI replacement cost may be understated. High-value coastal markets (Malibu, Naples FL) should be verified against rcv_adj_struct_usd. |
| acs_owner_occupancy_pct | % of housing units in the tract that are owner-occupied vs. renter or vacant. |
67%
|
LOW–MED
ACS 5-Year
|
Owner-occupied properties are generally better maintained. Lower occupancy correlates with higher wind event claim frequency. Used in behavioral model calibration as a maintenance investment proxy. |
| acs_pct_built_pre1980 | % of housing structures in the tract built before 1980. |
67%
|
LOW–MED
ACS 5-Year
|
Pre-1980 structures pre-date modern building codes in most states. High prevalence = elevated structural vulnerability. Important for seismic (unreinforced masonry) and wind (pre-Andrew code) portfolios. |
| hmda_median_loan_usd | Median home improvement loan amount originated in the tract from HMDA filings, USD. |
67%
|
LOW–MED
HMDA / CFPB
|
Proxy for neighborhood reinvestment and renovation activity. High HMDA improvement lending suggests more recent roof replacements and construction upgrades than assessor vintage alone indicates. |
| cbsa_code | Census Core-Based Statistical Area (CBSA) code for the building's metro or micropolitan area. |
98%
|
HIGH (classification)
Census TIGER
|
Standard geographic identifier for metro-level accumulation analysis. Also used for labor cost calibration in rcv_adj_struct_usd. 2% of buildings in rural non-CBSA areas. |
| cbsa_title | Human-readable CBSA name (e.g., "Miami-Fort Lauderdale-Pompano Beach, FL"). |
98%
|
HIGH (classification)
Census TIGER
|
See cbsa_code. |
| fhfa_hpi_2024 | FHFA House Price Index level for the CBSA as of 2024. Available for CBSAs with sufficient ARM transaction data. |
31%
|
MEDIUM
FHFA
|
Identifies markets where market value appreciation has significantly outpaced construction cost inflation — a key indicator of underinsurance risk as ACV policies diverge from current replacement cost. |
| fhfa_hpi_yoy_pct | FHFA HPI year-over-year % change for the CBSA. |
31%
|
MEDIUM
FHFA
|
Rapidly appreciating markets (Sun Belt metros 2020–2023) suggest increasing insurance-to-value gaps where policy limits have not kept pace with replacement cost increases. |
| acs_pct_single_unit | % of housing structures in the tract that are single-unit (detached or attached) buildings. |
67%
|
LOW–MED
ACS 5-Year
|
Used to calibrate NSI occupancy type distribution within each tract and to understand the concentration of residential vs. multifamily exposure. |
All data in the Hummingbird Structure Inventory is derived from publicly available government sources. No proprietary third-party data vendors are used. Complete source list: Florida Department of Revenue (FDOR NAL) · US Army Corps / FHWA National Structures Inventory (NSI) · Overture Maps Foundation · FEMA Flood Insurance Rate Maps (FIRM) and NFIP Claims · USGS 3D Elevation Program (3DEP) · USGS National Seismic Hazard Model (NSHM) · ASCE 7-22 design wind speed maps · US Census Bureau American Community Survey (ACS) · FHFA House Price Index · HMDA filings (CFPB) · USFS Wildfire Risk to Communities (WRC) · FEMA WUI boundary layers · County assessor / CAMA databases via public records requests or open data portals.
Storage: SI is stored in Apache Parquet format on Google Cloud Storage and Amazon S3. All source data is publicly licensed and suitable for storage under public data terms. No Fermat or other carrier proprietary data is incorporated into the SI.
Coverage exception: Alaska is excluded from the current version. All 49 continental states plus Hawaii are covered. Alaska is planned for a future release.
Currency: FDOR NAL, NSI, and FEMA regulatory layers refresh annually. Overture Maps refreshes quarterly. ACS 5-year estimates are on a rolling annual cycle. County CAMA data is ingested as publicly available, with most counties updating annually or quarterly.
Hummingbird Software, LLC — Confidential. This document contains proprietary methodology descriptions. Distribution restricted to authorized recipients.