Methodology

1. National Weather Service alerts

Every 15 minutes, a scheduled job on Floodi's servers calls the public NWS API at api.weather.gov for all active alerts in the State of Florida. We ingest:

• Flash Flood Warnings & Flood Warnings
• Flood Watches & Flash Flood Watches
• Coastal Flood Advisories, Warnings & Watches
• Storm Surge Warnings
• Hurricane / Tropical Storm warnings tied to flooding risk

Alert polygons are drawn on the map exactly as published by the NWS — we do not modify the geometry, the severity classification, or the expiration time. When the NWS cancels or updates an alert, the next 15-minute poll picks up the change. If our scheduler misses a cycle, the map automatically catches up on the following run.

2. USGS stream gauges & flood stages

Real-time water level readings are pulled from the USGS Water Services API. A gauge is plotted on the map when its current reading reaches or exceeds its official action stage. Each stage is defined by the National Weather Service in coordination with USGS:

Stage thresholds are gauge-specific, resolved from NOAA's official flood-category data for that location — Floodi never invents or adjusts them. For the small number of gauges NOAA hasn't assigned an official category to, we fall back to a clearly conservative estimate and label it as such rather than presenting it as an official threshold.

3. Community reports & vote-based verification

Anyone in Florida can submit a flood report in about 30 seconds — location, severity, an optional photo, and an optional comment. To keep the map trustworthy, every community report passes through a multi-stage verification pipeline:

• Confirm votes: any signed-in user (other than the original reporter) can tap "Confirm" if they see the same flooding. Each confirm increments a public counter shown on the report.
• Dispute votes: if a user believes a report is inaccurate, they can tap "Dispute." Once a report receives 3 or more dispute votes, its status is automatically downgraded to clearing and the pin is de-emphasized on the map.
• "Likely cleared" status: when 2 or more independent users mark a report as cleared (water has receded), the report transitions to likely_cleared.
• Auto-expiration: any report still active after 6 hours is automatically transitioned tolikely_cleared. Floods rarely persist on a single street longer than that without a fresh observation.
• Rate limit: a single account may submit at most5 reports per hour. This is enforced at the database level to block spam.
• Flag-based hiding: any user can flag a report. A single flag immediately hides the photo; 3+ unique flags hide the entire report pending admin review.
• Photo moderation: all uploaded photos are scanned for unsafe content before becoming visible on the map.
• Self-vote prevention: users cannot confirm or dispute their own reports, and each user can vote on a given report only once.

3.5 AI-powered photo severity classification

When a community report includes a photo, Floodi's AI vision system independently analyzes the image before it becomes publicly visible. This analysis runs on Google Gemini Vision through a secured AI gateway and produces two outputs:

• Content safety check: the image is screened for nudity, graphic violence, hate content, and non-flood imagery. Only confirmed flood-condition photos are approved for public display.
• Flood severity classification: approved photos are classified into one of four severity levels:
  • minor — puddles, wet pavement, nuisance flooding
  • moderate — significant standing water, partially flooded roads or yards
  • severe — major flooding, structures inundated, vehicles submerged
  • unknown — severity cannot be determined from the image alone

If the AI classification service is unavailable for any reason, the photo is held for manual admin review — photos are never auto-approved without a verification signal.

4. The 2-mile deduplication radius

During a major event — a hurricane, a king tide, an afternoon thunderstorm sitting over a single neighborhood — dozens of people will report the same flooding within minutes. To keep the map readable, Floodi groups those reports using a 2-mile deduplication radius:

• When a new report is submitted, Floodi checks for existing active reports within a 2-mile radius on the same street segment.
• If a match is found, the new report is treated as an additional confirmation of the existing event rather than a separate incident — the existing pin gains a confirm vote and stays in place.
• If no match is found, a new pin is created. This is what triggers the "X new nearby" badge for users already viewing the map.

The result: during widespread events the map shows real density (many pins across a region) rather than dozens of overlapping pins on a single block.

5. What "NWS-confirmed" actually means

A community report is tagged with the green NWS-confirmed badge when, at the moment of submission, its coordinates fall inside an active National Weather Service flood warning polygon.

Reports without the badge are still useful neighbor observations — they simply haven't been corroborated by an active federal warning at that exact location and time. Vote counts, photos, and timestamps still apply.