Estimation of soil organic carbon in tropical rainforest regions by combining UAV hyperspectral and LiDAR data

Abstract

Accurate acquisition of soil organic carbon (SOC) is important for the stability of ecosystems and the global climate. Hyperspectral remote sensing has emerged as a significant data source for digital soil mapping. However, in ecosystems with dense canopy cover, optical sensors face challenges in directly capturing soil surface spectra due to canopy obstruction. Consequently, estimating SOC in forest ecosystems remains a difficult task. LiDAR can capture forest structural attributes and soil surface information, providing possibility for SOC estimation in forest ecosystems. To date, the potential of LiDAR data in forest SOC estimation has not been fully explored. Here, we conducted SOC investigations in two tropical rainforests. Firstly, hyperspectral and LiDAR data were collected for the study areas using unmanned aerial vehicle (UAV) platforms. Subsequently, 40 vegetation indices (VI) and 101 LiDAR-derived variables were extracted from the hyperspectral images and LiDAR data, respectively. Furthermore, the RRelieff algorithm was employed to select the most robust variables. Finally, a deep neural network (DNN) model was utilized to establish the relationship between these variables and measured SOC, and to map the spatial distribution of SOC in the rainforests. It stems from the results that carotenoid reflectance index 2 (CRI2), non-linear index (NLI), and carotenoid reflectance index 1 (CRI1) were the best VIs for forest SOC estimation. Among the LiDAR variables, canopy height model (CHM) and digital elevation model (DEM) were the most important for model building. The performance of SOC estimation model based solely on LiDAR features (R² = 0.61) surpassed that of the model using only optical VI (R² = 0.59). Additionally, the combination of VI + LiDAR feature variables achieved optimal estimation accuracy (R² = 0.76) among the tested models. This study provided strong evidence for the potential use of LiDAR data in digital soil mapping in forest ecosystems.