An internship labeling one-way-in, one-way-out roads became a repeatable statewide assessment of evacuation congestion, shared exits, and how difficult it may be for a home to reach a major road.
Wind, temperature, humidity, vegetation, terrain, ignition conditions, and historical fires were brought together to show what is happening around each home—not just on its route out.
The evacuation and wildfire evidence became one interface. Getting there required reconciling statewide building sources, using AI to audit remaining gaps, and developing a directional fire-exposure method from the available weather and landscape data.
During my HWMO internship, I drew polygons around Kauaʻi neighborhoods with limited road access. Those labels showed where one road failure could leave many homes depending on the same exit.
I expanded that work into a repeatable analysis of road access, distance, nearby homes, escape options, and shared downstream bottlenecks. The map shows the original human-reviewed layer that started the project.
In addition to downsteam road bottlenecks, what other factors impact evacuation burden?
Haleilio Road contains 800 mapped buildings. The highest-burden home has 795 buildings ahead of it on the route to the zone’s one usable exit.
In this remote Kōkeʻe-area example, the analyzed road route is 24.944 km from the zone to its primary highway connection.
I brought wind, temperature, humidity, vegetation, terrain, ignition conditions, and past fires into the same view.
The University of Hawaiʻi at Mānoa’s Hawaiʻi Climate Data Portal (HCDP), which has built a map to visualize ignition probabilities across Hawaiʻi, belongs here as the official daily modeled layer. Its authenticated product details are required before real values can be displayed.
The Hawaiʻi Climate Data Portal is a University of Hawaiʻi research platform built to make reliable, high-resolution climate information easier to use. Its daily ignition-probability maps combine current and recent weather, vegetation condition, land cover, and other modeled fire factors to estimate where a large wildfire is more likely to start. This is a planning indicator—not a report that a fire has already begun.
The temperature map and arrows use NWS (National Weather Service) hourly forecasts sampled across every island. Arrow direction and length show wind flow and speed; color and weight encode relative humidity.
The National Weather Service, an agency within the National Oceanic and Atmospheric Administration, produces the nation’s official weather observations and forecasts. Hourly temperature, relative humidity, wind direction, and wind speed are useful here because they describe the near-term atmospheric conditions that can dry fuels and support fire movement; they remain forecasts and can change as new observations arrive.
Reported Wildland Fire Interagency Geospatial Services (WFIGS) incidents and NASA Fire Information for Resource Management System (FIRMS) heat detections are separate evidence layers and must never be treated as the same thing.
WFIGS publishes authoritative interagency incident locations and mapped perimeters assembled through federal wildfire reporting systems such as IRWIN. NASA FIRMS distributes near-real-time MODIS and VIIRS satellite thermal detections. WFIGS describes reported incidents; a FIRMS dot marks satellite-observed heat and is not, by itself, confirmation of a wildfire.
Explore Pacific Fire Exchange’s mapped large-fire perimeters year by year from 1999 through 2022.
Pacific Fire Exchange’s Hawaiʻi Large Fire Perimeter dataset combines records from county fire departments, the Hawaiʻi Department of Land and Natural Resources, the National Park Service, and mapping led by the University of Hawaiʻi at Mānoa’s Department of Natural Resources and Environmental Management. It tracks mapped perimeters from 1999–2022, generally emphasizing fires of at least 50 acres, so it is a strong statewide history—not a record of every small ignition.
Statewide LANDFIRE vegetation cover and 3DEP shaded relief show how the surrounding landscape changes across every island.
The U.S. Geological Survey’s 3D Elevation Program (3DEP) supplies the national elevation surface used to derive terrain and slope. The interagency LANDFIRE program’s Existing Vegetation Cover product estimates live canopy cover by life form for each 30-meter cell. Together they provide consistent statewide landscape context; they describe mapped terrain and vegetation, not real-time fuel moisture or a fire-spread forecast.
Kauaʻi began with a trusted internship building layer. Going statewide meant reconciling county footprints with FEMA, Microsoft, and OpenStreetMap — then finding a routable road network and public address points that could connect buildings to exits and searches.
The audit exposed the hard part: no single building source was complete everywhere. Select any of the eight islands, compare its inventories, and toggle every available source directly on the map.
Kauaʻi county-building sample. The internship reference is the benchmark for the first island audit.
Step 1 shows why the AI check still matters: even the strongest county datasets are not perfect. On Lānaʻi, for example, the detector finds clusters of roof-like shapes beyond the county footprints.
This audit now runs across all eight islands. It starts with a scientific queue of 518 strategically selected checks, then keeps the full 84,425 unresolved candidates visible as the larger review backlog. Orange shapes are AI guesses; green outlines show nearby FEMA or Microsoft footprints.
Loading the gap audit. Orange candidates are unconfirmed; green outlines show the selected reference baseline without labeling individual footprints by source.
Build one playful version of the Haleilio Road case study, one evidence layer at a time.
Start with the road network, then add each source used by the screening tool.
Read the source-by-source methodology, validation gates, directional pathway logic, known limitations, and reproducibility notes behind the prototype.