Robust hyperspectral estimation of eight leaf functional traits across different species and canopy layers in a subtropical evergreen broad-leaf forest

Abstract

Accurately estimating leaf functional traits across different species and canopy layers in subtropical evergreen broad-leaf forests remains a significant challenge due to the complexity of canopy structures and spectral noise. Although hyperspectral remote sensing holds substantial promise, existing methods struggle to deliver robust models capable of generalizing across diverse species and environmental conditions. This study aimed to develop a robust hyperspectral estimation approach for eight leaf traits across six species and three canopy layers, integrating successive projections algorithm (SPA) and random forest (RF) modeling. Utilizing 267 leaf samples and hyperspectral reflectance data acquired via a tower crane in Dinghushan National Nature Reserve, Guangdong Province, China, we demonstrated that the SPA-RF model, when applied to first derivative reflectance (FDR) data, significantly enhanced the accuracy and transferability of leaf trait estimations. The integration of SPA for wavelength selection and RF for modeling represented a robust approach, effectively mitigating the complexities introduced by species diversity and canopy heterogeneity. Leaf trait estimations derived from upper canopy layer samples generally yielded greater precision than those from lower and middle layers. Furthermore, species adapted to high-light environments (sun-tolerant) offered more accurate estimations than those adapted to low-light conditions (shade-tolerant). Among the eight leaf traits studied, flavonoid content, nitrogen balance index, and SPAD (relative leaf chlorophyll content) values emerged as more reliably estimated compared to carbon, nitrogen, phosphorus, equivalent water thickness, and specific leaf area. These findings illuminate the influence of canopy layer and species-specific traits on the precision of leaf trait estimations using hyperspectral remote sensing. The study’s insights emphasize the need for species- and canopy layer-specific approaches in ecological monitoring and conservation efforts.