Assessing wildfire spatial variability and hyperspectral data in debris-flow modeling

Abstract Rising wildfire activity and stronger rainfall in the western US have increased Post-Fire Debris Flow hazards (PFDF). The United States Geological Survey (USGS) employs the M1 model to evaluate PFDF likelihood, with basin-averaged differenced Normalized Burn Ratio (dNBR) as a primary input. The Thomas Fire reports non-normal dNBR distributions in 81% of basins, reducing the means’ capacity to capture burn heterogeneity. Here, we modify the M1 model to incorporate alternative dNBR metrics (median, peak frequency, and first quartile) that better reflect the burn distribution. We compare multispectral satellite and hyperspectral aircraft-derived dNBR metrics to assess sensor type and optimized dNBR index wavelengths in PFDF prediction. We validate against a Thomas Fire PFDF inventory, classifying initiation as probabilities exceeding defined thresholds (50%–90%), using Montecito, California, as a case study. The first quartile was the most effective metric at a 50% threshold (OA = 61.33%), followed by peak frequency, median, and mean. Hyperspectral imagery was more sensitive to vegetation health, reporting higher dNBR and PFDF risk than multispectral data, but performance remained similar because smoke limited basin coverage. The M1 model’s structure limits performance at higher rainfall intensities, where false positives sharply rise at 24–32 mm/hr, likely due to the multiplicative rainfall term. The first quartile addresses this multiplicative effect and should not be interpreted as a better indicator of burn heterogeneity or PFDF initiation conditions. Future work should balance rainfall intensity in the M1 model, incorporate more spatially representative inputs, and shift from single-band indices to methods leveraging hyperspectral dimensionality.