Resilience Analytics Lab, LLC
AHI is a calibrated, multi-hazard risk prediction system for the contiguous United States. It produces daily county-level probabilities for four natural hazards — wildfire, flood, wind, and winter storm — using a hybrid engine (spatial attention + gradient boosting) trained on 25 years of weather, satellite, and hazard event data across 3,109 counties.
The system is designed for emergency management decision-making at the county and regional level, providing calibrated risk tiers, audit-ready outputs, and structured decision support. Probabilities are isotonic-calibrated so they mean what they say — not just relative rankings.
Key properties:
- Calibrated probabilities (not just rankings) via per-hazard isotonic regression, verified to reproduce evaluated AUCs within 0.001
- 62-feature input vector spanning weather, reanalysis, vegetation, terrain, flood zones, wildland-urban interface, satellite fire detections, streamflow, and severe wind reports
- Single CONUS-wide hybrid engine: spatial attention for wind/winter, XGBoost for fire/flood — engine selection by paired bootstrap on 2.95M test county-days
- FIRMS/SPC-validated labels: satellite fire detections and severe wind reports filter training labels to remove noise
- Mean AUC 0.898 across all four hazards (fire 0.878, flood 0.903, wind 0.869, winter 0.942)
- Runs on CPU via ONNX Runtime + XGBoost — no GPU required
- Live at ahi.run
AHI v5.0 uses a hybrid engine that combines two model families, with the best engine per hazard selected by paired bootstrap on 2,950,441 identical test county-days:
- CONUS-wide spatial attention model with learned gated coupling between temporal and spatial signal
- Baked-in scaler, k-NN adjacency, and county/state embeddings
- Single ONNX graph processes all 3,109 counties in one forward pass
- Grounded in heat kernel attention, where attention weights follow the heat equation solution
- Gradient-boosted trees on the same 62-feature matrix
- One model per hazard, no scaler needed (tree-based)
- Per-hazard isotonic regression maps raw probabilities to calibrated outputs
- Portable JSON lookup tables:
np.interp(p_raw, x_thresholds, y_thresholds) - Verified to reproduce evaluated AUCs within 0.001 tolerance
| Hazard | Engine | AUC |
|---|---|---|
| Fire | XGBoost | 0.878 |
| Flood | XGBoost | 0.903 |
| Wind | Attention | 0.869 |
| Winter | Attention | 0.942 |
| Mean | Hybrid | 0.898 |
| Source | Features | Coverage |
|---|---|---|
| NOAA GridMET | Temperature, humidity, wind, precipitation, ERC, VPD, fire weather | Daily, 2000-2025 |
| ECMWF ERA5 | Integrated vapor transport, total column water vapor, surface pressure, wind gusts, 2m temperature | Daily, 2000-2025 |
| NASA MODIS (Terra/Aqua) | NDVI, EVI, vegetation anomalies | Monthly, 2000-2025 |
| NASA FIRMS | Satellite fire detections — trailing 3-day and 7-day fire count, FRP (lagged features + label validation) | Daily, 2000-2025 |
| USGS NWIS | Daily streamflow discharge — trailing 3-day mean, delta, max (lagged features) | Daily, 2000-2025 |
| NOAA SPC | Severe wind reports — trailing 3-day severe days, max wind speed (lagged features + label validation) | Daily, 2000-2025 |
| FEMA NFHL | Special flood hazard areas, V-zones, X-zones | Static |
| SILVIS WUI | Wildland-urban interface fractions, vegetation cover, housing density | Static |
| NOAA Storm Events | Flood, wind, winter storm, tornado records | 2000-2025 |
| WFIGS | Historical wildfire locations and perimeters (10+ acre threshold) | All CONUS states |
| FEMA Disaster Declarations | County-level disaster records | 2000-2025 |
| USGS Earthquake Catalog | Seismic events (model trained but hidden from UI) | 2000-2025 |
This repository contains the deployed dashboard and inference pipeline. The model architecture, training pipeline, and training data preparation are maintained as proprietary assets.
| Path | Description |
|---|---|
app.py |
Streamlit dashboard application |
inference_v5.py |
v5 hybrid inference engine (ONNX + XGBoost + isotonic calibration) |
inference_onnx.py |
Legacy v4 ONNX inference engine (per-state calibration) |
state_context.py |
Per-state configuration loader |
models/v5/ |
CONUS-wide v5 model artifacts (ONNX, XGBoost, calibrators, county order) |
states/<XX>/ |
Per-state inference data, GeoJSON, and config |
states/registry.yaml |
State deployment status and region mapping |
data/ |
National precomputed prediction CSVs |
scripts/ |
Utility scripts (precompute_v5, state onboarding) |
- Python 3.11
- Streamlit — dashboard framework
- ONNX Runtime — attention model inference (CPU-only, no GPU required)
- XGBoost — gradient boosting inference
- Plotly — interactive mapping and visualization
- Pandas / NumPy — data processing
- Deployed on Render
AHI's architecture is grounded in published research on attention mechanisms and topological computation by Joshua D. Curry, available on SSRN:
- Diffusion Attention: Replacing Softmax with Heat Kernel Dynamics
- Heat Kernel Attention: Provable Sparsity via Diffusion Dynamics
- Simplicial Computation: Topology as a Control Variable
MIT License. See LICENSE for details.
The model architecture, training pipeline, and trained weights are proprietary assets of Resilience Analytics Lab, LLC and are not included in this repository.