Application of Landsat ETM+ and OLI Data for Foliage Fuel Load Monitoring Using Radiative Transfer Model and Machine Learning Method

Foliage fuel load (FFL) is a critical factor affecting crown fire intensity and rate of spread. Satellite observations provide the potential for monitoring FFL dynamics across large areas. Previous studies commonly used empirical methods to estimate FFL, which potentially lacks reproducibility. This study applied Landsat 7 ETM+ and 8 OLI data for FFL retrieval using radiative transfer model (RTM) and machine learning method. To this end, the GeoSail, SAIL, and PROSPECT RTMs were first coupled together to model the near-realistic scenario of a two-layered forest structure. Second, available ecological information was applied to constrain the coupled RTM modeling phases in order to decrease the probability of generating unrealistic simulations. Third, the coupled RTMs were linked to three machine learning models-random forest, support vector machine, and multilayer perceptron-as well as the traditional lookup table. Finally, the performance of each method was validated by FFL measurements from Southwest China and Sweden. The resulting multilayer perceptron (R2 = 0.77, RMSE = 0.13, and rRMSE = 0.43) outperformed the other three methods. The evaluation of the applicability of the FFL estimates was conducted in a southwest China forest where two occurred in 2014 and 2020. The FFL dynamics from 2013 through 2020 showed that the fire was likely to occur when the FFL accumulated to a critical point (around 27 × 106 kg), highlighting the relevance of remote sensing derived FFL estimates for understanding potential fire occurrence.