Combining Field Observations and Remote Sensing to Forecast Fine Fuel Loads

Abstract

Effective wildfire management requires accurate information about the spatial distribution of fuels. While static maps work well for coarse forest fuels with slow turnover, fine fuels—the driver of rangeland fires—often vary dramatically from year to year. Our goal was to develop a fine fuel forecast to help managers optimally allocate fire suppression resources and funding before the start of the fire season and to identify critical gaps in understanding or data that limit forecast skill. We compiled a historical record of fine fuel loads collected in the Great Basin and combined it with remotely sensed data on herbaceous productivity. Using a Bayesian State-Space approach to account for both process and observation error, we built a “Fuels Model” in which the predicted fuel load at a location depends on the fuel load in the previous year and productivity of the current year. Next, we built a “productivity model” to predict current year production based on remotely sensed data available in early spring. Finally, we combined these two models to generate early spring forecasts for summer fuel loads from 1987 to 2020 and quantified the associated uncertainty. We found that current year productivity contributed twice as much as the previous year's fuel load to the current fuel load. Our productivity predictions reduced mean absolute predictive error by 11% compared with a strong null model without early-season weather covariates. However, when we fed the productivity predictions into our fuel model, the resulting fuel load forecasts had too much uncertainty to inform management decisions, with most uncertainty coming from the process error of the Fuels Model. Reducing this uncertainty will require higher-quality observations of fuel loads. Until those are available, our results suggest that managers could rely on productivity forecasts as a reasonable proxy for fuel load forecasts.