Predicting Fuel Moisture Codes Across Mountainous Forests of British Columbia Using L-Band ALOS-2 PALSAR-2 Imagery

Fuel moisture codes are numerical indices used to estimate the moisture content of different layers of forest fuels and are widely applied to assess the ignition potential and behavior of wildfires. Terrain-corrected L-band ALOS-2 PALSAR-2 synthetic aperture radar (SAR) imagery predicted the Drought Code (DC) and Duff Moisture Code (DMC) of the Canadian Forest Fire Danger Rating System in mountainous forests of British Columbia. Random forest regression models were developed using backscatter, decomposition parameters, and polarimetric discriminators obtained from forest (high-biomass, conifer-dominated), sparse (medium-biomass, open canopy), and unforested (low-biomass, herbaceous-dominated) fuel types near weather stations. Out-of-bag predicted values were compared against weather station-derived values. Models trained using SAR data predicted the DC with moderate accuracy in sparse (R2 = 0.49, RMSE = 165.65) and forest sites (R2 = 0.39, RMSE = 173.80), and with low accuracy in unforested sites (R2 = 0.15, RMSE = 205.96). Models trained using SAR data predicted the DMC with moderate accuracy in sparse (R2 = 0.64, RMSE = 26.07) and unforested sites (R2 = 0.47, RMSE = 31.52), and with low accuracy in forest sites (R2 = 0.12, RMSE = 40.71). Incorporating SAR data and day of year (DOY) improved the accuracy of DC models. Errors in both DC and DMC increased when models were evaluated with SAR data from weather stations not used in training, suggesting limited spatial transferability beyond training weather stations. Results suggest that terrain-corrected L-band polarimetric SAR has potential for predicting fuel moisture in mountainous and forested environments. Differences in surface and soil conditions across the fire environment must be accounted for to improve the spatial transferability of the models.