Evaluating the potential of LiDAR data for fire damage assessment: A radiative transfer model approach

Abstract

Providing accurate information on fire effects is critical to understanding post-fire ecological processes and to design appropriate land management strategies. Multispectral imagery from optical passive sensors is commonly used to estimate fire damage, yet this type of data is only sensitive to the effects in the upper canopy. This paper evaluates the sensitivity of full waveform LiDAR data to estimate the severity of wildfires using a 3D radiative transfer model approach. The approach represents the first attempt to evaluate the effect of different fire impacts, i.e. changes in vegetation structure as well as soil and leaf color, on the LiDAR signal. The FLIGHT 3D radiative transfer model was employed to simulate full waveform data for 10 plots representative of Mediterranean ecosystems along with a wide range of post-fire scenarios characterized by different severity levels, as defined by the composite burn index (CBI). A new metric is proposed, the waveform area relative change (WARC), which provides a comprehensive severity assessment considering all strata and accounting for changes in structure and leaf and soil color. It showed a strong correlation with CBI values (Spearman's Rho = 0.9 ± 0.02), outperforming the relative change of LiDAR metrics commonly applied for vegetation modeling, such as the relative height of energy quantiles (Spearman's Rho = 0.56 ± 0.07, for the relative change of RH60, the second strongest correlation). Logarithmic models fitted for each plot based on the WARC yielded very good performance with R2 (± standard deviation) and RMSE (± standard deviation) of 0.8 (± 0.05) and 0.22 (± 0.03), respectively. LiDAR metrics were evaluated over the King Fire, California, U.S., for which pre- and post-fire discrete return airborne LiDAR data were available. Pseudo-waveforms were computed after radiometric normalization of the intensity data. The WARC showed again the strongest correlation with field measures of GeoCBI values (Spearman's Rho = 0.91), closely followed by the relative change of RH40 (Spearman's Rho = 0.89). The logarithmic model fitted using WARC offered an R2 of 0.78 and a RMSE of 0.37. The accurate results obtained for the King Fire, with very different vegetation characteristics compared to our simulated data, demonstrate the robustness of the new metric proposed and its generalization capabilities to estimate the severity of fires.