FireRisk-Multi

This study integrates multi-source geospatial data to build a dynamic wildfire risk assessment framework. The datasets were systematically compiled and processed within Google Earth Engine (GEE) to ensure spatial and temporal consistency. The primary data sources include:

High-Resolution Aerial Imagery: The National Agriculture Imagery Program (NAIP) provides 10-meter RGB imagery for detailed land cover analysis. This dataset was fused with Wildfire Hazard Potential (WHP) data to establish a baseline risk assessment layer (NAIP-WHP).

Historical Fire Data: Near-real-time active fire detections from the FIRMS dataset (MODIS/VIIRS, 375m resolution) and burned area records from MODIS MCD64A1 (500m) were used for model validation and hotspot identification.

Topographic Data: Elevation and slope metrics were derived from the Shuttle Radar Topography Mission (SRTM) at 30-meter resolution to analyze terrain influences on fire spread.

Meteorological Data: ERA5 reanalysis data (1km resolution) provided temperature and precipitation records, critical for assessing weather-driven fire risks.

Vegetation Indices: MODIS NDVI/EVI (250m, 16-day intervals) quantified vegetation moisture and fuel availability.

Data Fusion Framework
The framework employs three progressive fusion levels:

Single-Modality Baseline (NAIP-WHP): Combines NAIP imagery with static WHP risk labels at 10m resolution.

Fixed-Weight Fusion (FIXED): Integrates terrain, weather, and vegetation data with predefined weights (e.g., WHP: 50%, other factors: 10% each).

Dynamic-Weight Fusion (FUSED): Adjusts feature contributions based on regional characteristics (e.g., amplifying slope weights in mountainous areas).