# Flood Property Risk — National Harbor, Maryland # # "The hotel you're standing in right now has a flood risk score." # # The Gaylord National Resort & Convention Center — the venue for # ClimateTech Connect 2026 — sits on the Potomac River waterfront # in National Harbor, MD. It is in a FEMA flood zone. The parking # garage has flooded. The Potomac rises. # # This demo computes flood risk for any US address from raw elevation # data, satellite-derived land cover, FEMA regulatory zones, and # physics-based flood models. Full provenance. Every layer auditable. # # ─── WHY THIS LOCATION ─────────────────────────────────────────── # # National Harbor (Oxon Hill, MD) sits at the confluence of the # Potomac River and Broad Creek, 10 miles south of DC. The venue is # built on reclaimed waterfront at ~3m above mean sea level. Across # the river, Alexandria's Old Town floods regularly during king tides # and nor'easters. The Anacostia waterfront in DC is one of the most # flood-vulnerable urban corridors on the East Coast. # # FEMA Zone AE (1% annual chance flood) covers the Potomac # waterfront. Climate change projections show 0.3-1.0m of sea level # rise in the Chesapeake Bay by 2050, which would put the convention # center's ground floor at risk during storm surge events. # # Everyone in this room insures properties like this one. # # ─── THE PITCH ──────────────────────────────────────────────────── # # "Right now, you're standing in a FEMA flood zone. The Potomac is # 300 meters from this room. Let me show you your flood risk." # # [types venue address — map zooms to National Harbor] # [blue flood risk overlay appears around the venue] # [clicks venue location — "Flood risk: 67/100"] # # "Every layer you see traces back to source data — FEMA, USGS, # ESA, Deltares. Full provenance. No black box. # # Now give me YOUR address." # # ─── PIPELINE ───────────────────────────────────────────────────── # # source/dem ──────────── terrain/slope ──────────────────┐ # │ └── terrain/twi ──────────┤ # └───── terrain/hillshade │ # └───── hydro/breached-dem │ # └── hydro/flow-direction │ # └── hydro/flow-accum ──┤ # └───── sea-level/inundation-1m │ # └───── sea-level/inundation-2m │ # │ # source/landcover ──────── hydro/impervious ────────────┤ # │ # source/fema-zones ─────────────────────────────────────┤ # │ # source/flood-depth ────────────────────────────────────┤ # │ # source/daymet ──────── hydro/precipitation ────────────┤ # ▼ # risk/flood # │ # risk/flood-classified # │ # source/population ─── exposure/population ─── analysis/exposure # # ─── DATA SOURCES ───────────────────────────────────────────────── # # Copernicus DEM GLO-30 ESA / Airbus Open License # ESA WorldCover 2021 ESA / Copernicus CC-BY-4.0 # FEMA NFHL FEMA Public Domain # Deltares Global Flood Maps Deltares CC-BY-4.0 # Daymet Daily NA ORNL DAAC Public Domain # Census ACS Tracts US Census Public Domain # # Usage: # folia validate demos/flood-property-risk/folia.yaml # folia run demos/flood-property-risk # # See also: # - docs/use_cases/climate-risk-data-supplier.md (UC-033) # - docs/contacts/climatech-connect-2026/ (buyer contacts) name: flood-property-risk version: "1.0" description: > Property-level flood risk assessment from raw elevation data, FEMA zones, physics-based flood models, and satellite land cover. Type an address — see terrain drainage, impervious runoff, regulatory flood zones, and a composite risk score. Full provenance. National Harbor, MD — the ClimateTech Connect 2026 venue. settings: # National Harbor → Alexandria → DC waterfront # Captures: Gaylord National (venue), Old Town Alexandria, Anacostia, # Reagan Airport, Navy Yard — all flood-vulnerable areas # Bbox stays within one Copernicus DEM tile (N38W078: -78 to -77) # to avoid multi-tile merge alignment issues. # Centered on National Harbor / Oxon Hill / Alexandria waterfront. default_bbox: [-77.15, 38.73, -77.01, 38.88] default_crs: EPSG:4326 address_search: true layers: # ================================================================ # SOURCE DATA # # Six federal/open datasets. Everything you need to assess flood # risk at the property level. No proprietary data. No special # access. No license fees. Just public infrastructure data # that every insurance company ingests independently. # ================================================================ source/dem: uri: stac://microsoft-pc/cop-dem-glo-30 type: raster description: > Copernicus DEM GLO-30 — 30m global elevation. The foundation of every flood model. Water flows downhill. Elevation determines where it goes, how fast, and how deep it gets. National Harbor sits at ~3m above mean sea level on the Potomac waterfront. Old Town Alexandria across the river is even lower. The Anacostia River corridor in DC is a natural floodplain that the city built on anyway. For US properties, USGS 3DEP (1-3m LiDAR) provides higher resolution. We use Copernicus for global coverage. Same pipeline, swap one URI for another. metadata: provider: ESA / Airbus license: Copernicus DEM License concepts: [elevation, terrain] source/landcover: uri: stac://microsoft-pc/esa-worldcover type: raster description: > ESA WorldCover 2021 — 10m land cover. Class 50 = built-up (impervious surface). Impervious surface is the #1 driver of urban flooding — rain that would soak into soil instead runs off concrete and asphalt into storm drains, channels, and your basement. The DC metro area is heavily developed. National Harbor is a concrete-and-glass development on filled waterfront land. Impervious fraction here approaches 90%. metadata: provider: ESA / Copernicus license: CC-BY-4.0 concepts: [land-cover, impervious-surface] source/fema-zones: uri: https://hazards.fema.gov/gis/nfhl/rest/services/public/NFHL/MapServer/28/query?where=1%3D1&geometry=-77.15,38.73,-77.01,38.88&geometryType=esriGeometryEnvelope&inSR=4326&outFields=FLD_ZONE,ZONE_SUBTY,SFHA_TF,STATIC_BFE&outSR=4326&f=geojson&resultRecordCount=5000 type: vector description: > FEMA National Flood Hazard Layer (NFHL) — the regulatory standard for flood risk in the United States. Zone designations: AE — 1% annual chance flood (100-year), with base flood elevation determined. Mandatory flood insurance. A — 1% annual chance, no BFE determined. VE — Coastal high hazard (wave action). Strictest building codes. Mandatory insurance. X — 0.2% annual chance (500-year) or minimal risk. The Potomac waterfront at National Harbor is Zone AE. Old Town Alexandria is Zone AE. The Anacostia corridor has both AE and VE designations. FEMA maps are the floor, not the ceiling. They're based on historical data and don't account for climate change. The real risk is worse than what these maps show. style: renderer: maplibre fill-opacity: 0.4 stroke-width: 0.8 categorize: field: FLD_ZONE values: VE: "#08306b" AE: "#2171b5" A: "#6baed6" AO: "#9ecae1" AH: "#9ecae1" X: "#c6dbef" default: "#deebf7" info: fields: [FLD_ZONE, ZONE_SUBTY, STATIC_BFE] format: "FEMA Zone {FLD_ZONE} ({ZONE_SUBTY}) — BFE: {STATIC_BFE} ft" section: Regulatory legend: title: "FEMA Flood Zones" items: - { label: "VE — Coastal High Hazard", color: "#08306b" } - { label: "AE — 1% Annual Chance", color: "#2171b5" } - { label: "A — 1% (No BFE)", color: "#6baed6" } - { label: "X — Minimal Risk", color: "#c6dbef" } source/flood-depth: uri: stac://microsoft-pc/deltares-floods type: raster description: > Deltares Global Flood Maps — physics-based modeled flood extent and depth for multiple return periods. Uses MERIT DEM, NASADEM, and LiDAR-derived elevation with GTSMip6 storm surge forcing. This is what FEMA maps SHOULD look like — continuous depth estimates, not just zone boundaries. A house in Zone AE might get 0.3m or 3.0m of water depending on its exact elevation. Deltares shows the difference. FEMA doesn't. We use the 100-year return period to match FEMA's 1% annual chance standard, but Deltares provides 2yr through 1000yr. metadata: provider: Deltares license: CC-BY-4.0 params: return_period: 100 concepts: [flood, hydrology, inundation] source/daymet: uri: stac://microsoft-pc/daymet-daily-na type: raster description: > Daymet daily surface weather — 1km gridded for North America. Precipitation totals show how much rain a location receives. The DC metro area averages ~1000mm/year, but intense summer thunderstorms can drop 50-100mm in an hour. That's when urban flooding happens — storm drains can't keep up. metadata: provider: ORNL DAAC license: Public Domain concepts: [weather, precipitation] source/population: uri: https://tigerweb.geo.census.gov/arcgis/rest/services/TIGERweb/tigerWMS_Current/MapServer/8/query?where=STATE+IN+(%2724%27,%2711%27,%2751%27)&geometry=-77.15,38.73,-77.01,38.88&geometryType=esriGeometryEnvelope&inSR=4326&outFields=GEOID,BASENAME,POP100,AREALAND&outSR=4326&f=geojson&resultRecordCount=5000 type: vector description: > Census tract population — Maryland (24), DC (11), Virginia (51). How many people are exposed to flood risk? The Potomac waterfront corridor from Alexandria through National Harbor to DC houses hundreds of thousands of residents plus millions of daily commuters and visitors. style: renderer: maplibre fill: "#1565C0" fill-opacity: 0.2 stroke: "#0D47A1" stroke-width: 0.3 info: fields: [BASENAME, POP100] format: "{BASENAME}: {POP100:,} people" section: Demographics # ================================================================ # TERRAIN ANALYSIS # # Water flows downhill. Terrain tells you where. # ================================================================ terrain/slope: type: raster description: > Slope in degrees. The inverse of slope is flatness — flat areas accumulate water, steep areas shed it. The Potomac floodplain is nearly flat (<2°). National Harbor is built on flat, low-lying land at the river's edge. Water has nowhere to go. compute: op: terrain_slope params: units: degrees inputs: dem: { layer: source/dem } style: renderer: maplibre opacity: 0.7 colorRamp: - { value: 0, color: "#08306b", label: "Flat (flood risk)" } - { value: 5, color: "#2171b5" } - { value: 10, color: "#6baed6" } - { value: 20, color: "#c6dbef" } - { value: 30, color: "#f7fbff", label: "Steep (sheds water)" } info: fields: [value] format: "Slope: {value:.1f}°" section: Terrain terrain/hillshade: type: raster description: > Hillshade relief — shows the Potomac River valley, Broad Creek, and the low-lying areas where National Harbor was built. compute: op: terrain_hillshade params: azimuth: 315 altitude: 45 inputs: dem: { layer: source/dem } style: renderer: maplibre opacity: 0.4 colorRamp: - { value: 0, color: "#000000" } - { value: 255, color: "#ffffff" } # ================================================================ # HYDROLOGY # # Where does water flow? How fast does it run off? These layers # answer the physics questions that FEMA zone boundaries can't. # ================================================================ hydro/breached-dem: type: raster description: > Depression-breached DEM — artificial depressions in the elevation model are removed so water can flow continuously downhill to the river. Without this preprocessing, flow routing gets stuck in DEM artifacts (bridges, overpasses, data gaps). This is the standard first step in any hydrological analysis. compute: op: breach_depressions inputs: dem: { layer: source/dem } hydro/flow-direction: type: raster description: > D8 flow direction — for each cell, which of its 8 neighbors does water flow toward? Computed from the breached DEM using the steepest-descent algorithm. This is the routing grid that all downstream analysis depends on. compute: op: flow_direction_d8 inputs: dem: { layer: hydro/breached-dem } hydro/flow-accumulation: type: raster description: > Flow accumulation — how many upstream cells drain through each cell? High values = drainage channels where water concentrates. The Potomac River has the highest accumulation. Broad Creek (which flows past National Harbor) and Four Mile Run (Alexandria) show as bright drainage lines. This layer reveals the invisible hydrology underneath the pavement. Urban flooding happens where these natural drainage channels have been paved over but the water still comes. compute: op: flow_accumulation inputs: flow_direction: { layer: hydro/flow-direction } style: renderer: maplibre opacity: 0.8 colorRamp: - { value: 0, color: "#f7fbff" } - { value: 10, color: "#c6dbef" } - { value: 100, color: "#6baed6", label: "Minor channel" } - { value: 1000, color: "#2171b5", label: "Major channel" } - { value: 10000, color: "#08306b", label: "River" } info: fields: [value] format: "Upstream cells: {value:,.0f}" section: Hydrology terrain/twi: type: raster description: > Topographic Wetness Index — TWI = ln(a / tan β), where a is upstream contributing area (proxied by flow accumulation × cell width) and β is slope. The canonical hydrological "where does water sit" derivative (Beven & Kirkby 1979). High TWI = flat land downstream of a large contributing area = wet. Low TWI = ridges or steep terrain = dry. Typical range: ~1 (ridges) to ~25 (river floodplains). National Harbor and Old Town Alexandria sit on the wet end of this scale; the Maryland ridges to the north show low TWI. compute: op: raster_calc params: expression: > log( (max(flow_accum, 1.0) * 30.0) / max(tan(slope * 0.017453292519943295), 0.01) ) output_band: twi inputs: flow_accum: { layer: hydro/flow-accumulation } slope: { layer: terrain/slope } style: renderer: maplibre opacity: 0.7 colorRamp: - { value: 1, color: "#fff7bc", label: "Dry (ridges)" } - { value: 8, color: "#fec44f" } - { value: 12, color: "#41b6c4" } - { value: 18, color: "#225ea8", label: "Wet (floodplain)" } info: fields: [value] format: "TWI: {value:.2f}" section: Hydrology hydro/impervious: type: raster description: > Impervious surface — built-up areas from ESA WorldCover (class 50) used as a runoff proxy. compute: op: raster_calc params: expression: "where(landcover == 50, 100.0, 0.0)" output_band: impervious inputs: landcover: { layer: source/landcover } style: renderer: maplibre opacity: 0.65 colorRamp: - { value: 0, color: "#00000000", label: "Pervious" } - { value: 100, color: "#455A64", label: "Impervious" } info: fields: [value] format: "Impervious: {value:.0f}%" section: Hydrology hydro/precipitation: type: raster description: > Annual total precipitation from Daymet (2024). 1km resolution. The DC metro area receives ~1000mm/year with a summer maximum. But annual totals mask the real driver of urban flooding: intensity. A 100mm storm in 1 hour causes flooding. The same 100mm over a week does not. Future versions will include precipitation frequency (IDF curves) from NOAA Atlas 14. compute: op: temporal_reduce params: method: sum variable: prcp inputs: timeseries: { layer: source/daymet } style: renderer: maplibre opacity: 0.65 colorRamp: - { value: 600, color: "#ffffcc" } - { value: 800, color: "#a1dab4" } - { value: 1000, color: "#41b6c4" } - { value: 1200, color: "#225ea8" } info: fields: [value] format: "Annual precipitation: {value:.0f} mm" section: Hydrology # ================================================================ # SEA LEVEL RISE # # What happens when the Potomac rises? Simple elevation thresholds # show which areas go underwater at +1m and +2m above current # mean sea level. NOAA projects 0.3-1.0m of rise in the # Chesapeake Bay by 2050. +2m is a reasonable 2100 scenario. # ================================================================ sea-level/inundation: type: raster description: > Areas below the sea level rise threshold — controlled by slider. Blue = literally underwater at the selected rise level. Drag the "Sea Level Rise" slider to see the flood line advance. compute: op: raster_calc params: expression: "where(dem <= sea_level_m, 100.0, 0.0 / 0.0)" output_band: inundation inputs: dem: { layer: source/dem } style: renderer: maplibre opacity: 0.7 colorRamp: - { value: 0, color: "#00000000" } - { value: 100, color: "#1565F0" } info: fields: [value] format: "Inundated at +{value:.0f}m" section: Sea Level Rise form: sea_level_m: type: slider min: 0 max: 5 step: 0.5 default: 1 label: "Sea Level Rise (meters)" # ================================================================ # EXPOSURE # ================================================================ exposure/population: type: raster description: > Population density rasterized from Census tracts. How many people live in flood-prone areas? The Potomac waterfront corridor has seen massive development in the last 20 years — The Wharf, National Harbor, Pentagon City, Crystal City — putting hundreds of thousands of people closer to the water. compute: op: raster_rasterize params: attribute: POP100 background: 0.0 all_touched: true dtype: float32 inputs: features: { layer: source/population } reference: { layer: source/dem } style: renderer: maplibre opacity: 0.6 colorRamp: - { value: 0, color: "#f7fbff" } - { value: 1000, color: "#6baed6" } - { value: 5000, color: "#2171b5" } - { value: 10000, color: "#08306b" } info: fields: [value] format: "Tract population: {value:,.0f}" section: Exposure # ================================================================ # COMPOSITE FLOOD RISK # # Five inputs, weighted and combined into a single 0-100 score. # # This is not a hydrodynamic simulation (LISFLOOD-FP, HEC-RAS). # It's a screening model — fast, transparent, and good enough # for property-level risk ranking. The same approach that # Previsico and 7Analytics use for their initial screening, # before running full hydraulic models on flagged properties. # ================================================================ risk/flood: type: raster description: > Composite flood risk index (0-100). Weighted overlay of six conditioning factors: - Terrain flatness (inverse slope) — flat areas accumulate water - Impervious surface — built-up = runoff, not infiltration - Very low elevation (<5m) — coastal storm-surge proximity - Low elevation (<20m) — broader floodplain risk - Flow accumulation — drainage concentration channels - Topographic Wetness Index — hydrological wetness derivative Default weights are FIT, not hand-tuned. Logistic regression against Deltares 100yr depth pseudo-labels for the Mid-Atlantic region (HUC-4 0207 Potomac watershed, bbox -79.5,-76.5 × 38.0,39.8). AUC test = 0.999. Methodology: Khosravi et al. 2018, Sci. Total Environ. 627, 744-755. See `calibration/REPORT.md` for full fit details, including the honest caveat that AUC > 0.99 in this coastal-dominant bbox reflects elevation- dominance of Deltares, not exceptional model skill. Click any pixel to see its score. Drag sliders to override the fit. validation: method: logistic_regression region: HUC-4 0207 (Potomac, Mid-Atlantic) region_bbox: [-79.5, 38.0, -76.5, 39.8] auc_test: 0.9993 n_positive: 9290 n_negative: 10000 seed: 42 methodology: "Khosravi et al. 2018, Sci. Total Environ. 627, 744-755" label_source: dataset: deltares-floods version: v2021.06 return_period_years: 100 sea_level_year: 2018 license: CC-BY-4.0 calibration_artifact: calibration/mid-atlantic.json caveats: - > AUC > 0.99 reflects elevation-dominance of Deltares 100yr in coastal Mid-Atlantic, not exceptional model skill. The fit is calibration, not validation. For honest held-out validation use Sentinel-1 GFM detections or NFIP claims (Phase C ensemble labels). - > Train/test split is random; spatial autocorrelation likely inflates AUC. Block-aware split is a v2 improvement. - > drainage and twi conditioning factors used a local-relief proxy in the offline fit (no whitebox flow routing in fit env). The recipe itself uses the full hydrological flow_accum and TWI layers at run time — minor numerical drift between fit-time and run-time features is expected. compute: op: raster_calc params: expression: > min(max( ((1.0 - min(slope, 30.0) / 30.0) * flatness_w) + (impervious / 100.0 * impervious_w) + (where(dem <= 5.0, (5.0 - dem) / 5.0 * low_elev_w, 0.0)) + (where(dem <= 20.0, (20.0 - dem) / 20.0 * mid_elev_w, 0.0)) + (min(log(flow_accum + 1.0) / 10.0, 1.0) * drainage_w) + (min(max(twi, 0.0) / 25.0, 1.0) * twi_w), 0), 100) output_band: flood_risk inputs: slope: { layer: terrain/slope } impervious: { layer: hydro/impervious } dem: { layer: source/dem } flow_accum: { layer: hydro/flow-accumulation } twi: { layer: terrain/twi } style: renderer: maplibre opacity: 0.85 colorRamp: - { value: 0, color: "#1a985020" } - { value: 20, color: "#FFEB3B" } - { value: 40, color: "#FF9800" } - { value: 60, color: "#F44336" } - { value: 80, color: "#D32F2F" } - { value: 100, color: "#B71C1C" } info: fields: [value] format: "Flood risk: {value:.0f}/100" section: Risk Score form: # Default values are LR-fit weights for HUC-4 0207 Potomac # (Mid-Atlantic). See calibration/mid-atlantic.json for raw # coefficients and slider rescaling. Drag to override. flatness_w: type: slider min: 0 max: 50 step: 5 default: 5 label: "Terrain Flatness" impervious_w: type: slider min: 0 max: 50 step: 5 default: 5 label: "Impervious Surface" low_elev_w: type: slider min: 0 max: 50 step: 5 default: 20 label: "Very Low Elevation (<5m)" mid_elev_w: type: slider min: 0 max: 40 step: 5 default: 30 label: "Low Elevation (<20m)" drainage_w: type: slider min: 0 max: 40 step: 5 default: 0 label: "Drainage Accumulation" twi_w: type: slider min: 0 max: 40 step: 5 default: 5 label: "Topographic Wetness Index" risk/flood-classified: type: raster description: > Flood risk classified into 5 tiers. Maps cleanly onto insurance pricing bands: 1 = Very Low — preferred rates 2 = Low — standard rates 3 = Moderate — surcharge zone 4 = High — mandatory flood insurance 5 = Extreme — coverage may be unavailable compute: op: raster_reclassify params: method: quantile n_classes: 5 class_values: [1.0, 2.0, 3.0, 4.0, 5.0] labels: ["Very Low", "Low", "Moderate", "High", "Extreme"] inputs: raster: { layer: risk/flood } style: renderer: maplibre opacity: 0.8 colorRamp: - { value: 1, color: "#f7fbff" } - { value: 2, color: "#c6dbef" } - { value: 3, color: "#6baed6" } - { value: 4, color: "#2171b5" } - { value: 5, color: "#08306b" } info: fields: [value] format: "Risk tier: {value:.0f}/5" section: Risk Score # ================================================================ # EXPOSURE ANALYSIS # ================================================================ analysis/exposure: type: table description: > Population exposed to flood risk by Census tract. Aggregates flood risk scores across tracts to show which neighborhoods have the most people at risk. National Harbor and Old Town Alexandria should rank highest. compute: op: analysis_zonal_stats params: stats: [mean, max, count] inputs: raster: { layer: risk/flood } zones: { layer: source/population } style: table: renderer: tanstack-table columns: - { field: zone_id, label: "Tract" } - { field: mean, label: "Mean Flood Risk", format: ".1f" } - { field: max, label: "Max Risk", format: ".1f" } - { field: count, label: "Cells", format: "d" } sortBy: mean sortOrder: desc # ================================================================ # VIEWS — Organized for the pitch # # View 1: The opener — "you're standing in a flood zone" # View 2: The science — where water flows # View 3: The future — sea level rise scenarios # View 4: FEMA vs reality — regulatory gaps # View 5: The differentiator — view the spec # ================================================================ views: - name: "Property Risk" layers: [risk/flood, sea-level/inundation] map: center: [-77.08, 38.80] zoom: 14 description: > Flood risk for the ClimateTech Connect venue. The Gaylord National Resort is at the center of the map. Blue = flood risk. FEMA zone boundaries overlaid for comparison. Click any point for a score. Give me your address — I'll show you yours. - name: "Terrain & Impervious" layers: [terrain/slope, hydro/impervious] map: center: [-77.08, 38.80] zoom: 13 description: > Where does water go? Flow accumulation shows drainage channels underneath the pavement. Broad Creek (past National Harbor) and Hunting Creek (Old Town Alexandria) are natural flood corridors that have been built over. The water still comes. - name: "Sea Level Rise" layers: [terrain/hillshade, sea-level/inundation] map: center: [-77.08, 38.80] zoom: 13 description: > What happens when the Potomac rises. Dark blue = underwater at +1m (NOAA 2050 mid-range projection). Light blue = underwater at +2m (2100 high-end). Reagan Airport runways. Old Town Alexandria. The Tidal Basin. National Harbor parking. All below the line. - name: "FEMA vs Physics" layers: [terrain/hillshade, source/fema-zones, source/flood-depth] map: center: [-77.08, 38.80] zoom: 13 description: > FEMA zones are regulatory boundaries — binary in/out. Deltares flood depth is continuous — how deep does the water get? A house 1 meter inside Zone AE might see 0.1m of water or 2.0m. FEMA can't tell you which. Physics-based models can. Both layers are here — compare them. - name: "The Spec" layers: [] description: > This is the entire flood risk pipeline. 18 layers. 6 source datasets. 12 compute operations including hydrological flow routing (WhiteboxTools), FEMA regulatory zones, physics-based flood modeling (Deltares), and sea level rise scenarios. Every output traces back to source. Every weight is adjustable. Every layer is auditable. No black box. "What does your geospatial data pipeline look like?" This is ours. One file.