Abstract

The novel hyperspectral LiDAR (HSL) system exhibits the aptitude to simultaneously capture both spectral and geometric information from the hyperspectral waveform data. However, conventional single-wavelength decomposition methods may not be compatible with HSL waveforms due to higher levels of unstable noise, more complex waveform shapes, and inconsistent time delay effects at different wavelengths within the hyperspectral waveforms. These limitations pose significant challenges for quantitative applications of the HSL system. To overcome these issues, an imperative and pressing need is to search for a suitable waveform processing algorithm for the HSL system. Therefore, we propose a novel method called Ranking Central Locations of Natural Target Echoes (Rclonte) to decompose full-waveform hyperspectral LiDAR data. The Rclonte introduces a new parameter initialization strategy that includes rough estimation and refined estimation steps, preventing the optimization process from being trapped in a local optimum state. Subsequently, a re-optimization step over ranking central locations of natural target echoes at different wavelengths compensates for the missing detection or false detection of hidden weak and overlapping components within the waveform at some wavelengths. Two data collections, including the synthetic and measured HSL waveform data, were employed in the decomposition. The results indicate that (1) Rclonte detected components and parameters much more accurately with the highest R2 and the lowest RMSE and rRMSE values, outperforming the Hofton GD and MSWD methods. (2) Both the synthetic and measured data decomposition results highlight the effectiveness and the apparent superiority of Rclonte over Hofton GD and MSWD regarding compensating for the hidden weak or overlapping components. (3) The ranging results indicate that Rclonte achieves the highest ranging precision with low relative neighbor distance error (RNDE) (0.026∼0.085) for the measured data. (4) The spectra derived from Rclonte are superior to Hofton GD and MSWD methods. The smoothed version of the retrieved spectrum using Rclonte decomposition results presents a spectral similarity to the HSL-measured reflectance spectrum of a single leaf. The proposed method comprehensively utilizes the invariance of the central location orders of multiple targets at different wavelengths to ensure accurate detection. It not only facilitates the development of decomposition algorithms for full-waveform hyperspectral LiDAR data but also holds promise for adoption in other full-waveform multispectral LiDAR (MSL) and HSL systems.