Leaf-level hyperspectral-based species identification has a long research history. However, unlike hyperspectral image-based species classification models, convolutional neural network (CNN) models are rarely used for the one-dimensional (1D) structured leaf-level spectrum. Our research focuses on hyperspectral data from five laboratories worldwide to test the general use of effective identification of the CNN model by reshaping 1D structure hyperspectral data into two-dimensional greyscale images without principal component analysis (PCA) or downscaling. We compared the performance of two-dimensional CNNs with the deep cross neural network (DCN), support vector machine, random forest, gradient boosting machine, and decision tree in individual tree species classification from leaf-level hyperspectral data. We tested the general performance of the models by simulating an application phase using data from different labs or years as the unseen data for prediction. The best-performing CNN model had validation accuracy of 98.6%, prediction accuracy of 91.6%, and precision of 74.9%, compared to the support vector machine, with 98.6%, 88.8%, and 66.4%, respectively, and DCN, with 94.0%, 85.7%, and 57.1%, respectively. Compared with the reference models, CNNs more efficiently recognized Fagus crenata, and had high accuracy in Quercus rubra identification. Our results provide a template for a species classification method based on hyperspectral data and point to a new way of reshaping 1D data into a two-dimensional image, as the key to better species prediction. This method may also be helpful for foliar trait estimation.

Hyperspectral spectroscopy based on partial least squares regression (PLSR) is an effective tool for monitoring plant photosynthesis. Despite their wide applications, the robustness of PLSR models on tracing photosynthetic capacity, which varies considerably among different species and at different times, have been far less explored, leading to doubt about whether hyperspectral information can accurately predict the capacity across different species and temporal changes. Ordinary applications of PLSR generally make use of original or integer-order derivative transformed reflected spectra, but recent advances in spectral analysis have revealed that fractional-order derivative transformed spectra could provide more details of spectral signals. In this study, PLSR models based on fractional-order derivatives coupled with different wavelength selection methods were developed to evaluate whether photosynthetic parameters (Vcmax and Jmax) could be correctly predicted from reflectance spectra. The result indicated that the best PLSR models for the Vcmax and Jmax were obtained based on the sensitive wavelengths selected by stepwise regression using the fractional orders of 1.25 and 1.60, respectively. The optimal PLSR models were able to capture the temporal variabilities of Vcmax and Jmax with the R ² of 0.62-0.94 and 0.65-0.85, for which the 1605-1845 nm region was consistently used. Meanwhile, these PLSR models have the ability to capture the variations in different species, plant functional types, and biomes. The findings of this study demonstrate that leaf spectra can be successfully used for the timely prediction of variable photosynthetic capacity and provide the fundamentals for monitoring and mapping plant functions from reflected information.

Detailed three-dimensional (3D) radiative transfer models (RTMs) enable a clear understanding of the interactions between light, biochemistry, and canopy structure, but they are rarely explicitly evaluated due to the availability of 3D canopy structure data, leading to a lack of knowledge on how canopy structure/leaf characteristics affect radiative transfer processes within forest ecosystems. In this study, the newly released 3D RTM Eradiate was extensively evaluated based on both virtual scenes reconstructed using the quantitative structure model (QSM) by adding leaves to point clouds generated from terrestrial laser scanning (TLS) data, and real scenes monitored by Sentinel-2 in a typical temperate deciduous forest. The effects of structural parameters on reflectance were investigated through sensitivity analysis, and the performance of the 3D model was compared with the 5-Scale and PROSAIL radiative transfer models. The results showed that the Eradiate-simulated reflectance achieved good agreement with the Sentinel-2 reflectance, especially in the visible and near-infrared spectral regions. Furthermore, the simulated reflectance, particularly in the blue and shortwave infrared spectral bands, was clearly shown to be influenced by canopy structure using the Eradiate model. This study demonstrated that the Eradiate RTM, based on the 3D explicit representation, is capable of providing accurate radiative transfer simulations in the temperate deciduous forest and hence provides a basis for understanding tree interactions and their effects on ecosystem structure and functions.

Timely acquisition of forest structure is crucial for understanding the dynamics of ecosystem functions. Despite the fact that the combination of different quantitative structure models (QSMs) and point cloud sources (ALS and DAP) has shown great potential to characterize tree structure, few studies have addressed their pros and cons in alpine temperate deciduous forests. In this study, different point clouds from UAV-mounted LiDAR and DAP under leaf-off conditions were first processed into individual tree point clouds, and then explicit 3D tree models of the forest were reconstructed using the TreeQSM and AdQSM methods. Structural metrics obtained from the two QSMs were evaluated based on terrestrial LiDAR (TLS)-based surveys. The results showed that ALS-based predictions of forest structure outperformed DAP-based predictions at both plot and tree levels. TreeQSM performed with comparable accuracy to AdQSM for estimating tree height, regardless of ALS (plot level: 0.93 vs. 0.94; tree level: 0.92 vs. 0.92) and DAP (plot level: 0.86 vs. 0.86; tree level: 0.89 vs. 0.90) point clouds. These results provide a robust and efficient workflow that takes advantage of UAV monitoring for estimating forest structural metrics and suggest the effectiveness of LiDAR in temperate deciduous forests.

The application of hyperspectral imagery coupled with deep learning shows vast promise in plant species discrimination. Reshaping one-dimensional (1D) leaf-level reflectance data (LLRD) into two-dimensional (2D) grayscale images as convolutional neural network (CNN) model input demonstrated marked effectiveness in plant species distinction. However, the impact of the image shape on CNN model performance remained unexplored. This study addressed this by reshaping data into fifteen distinct rectangular formats and creating nine CNN models to examine the effect of image structure. Results demonstrated that irrespective of CNN model structure, elongated narrow images yielded superior species identification results. The ‘l’-shaped images at 225 × 9 pixels outperformed other configurations based on 93.95% accuracy, 94.55% precision, and 0.94 F1 score. Furthermore, ‘l’-shaped hyperspectral images consistently produced high classification precision across species. The results suggest this image shape boosts robust predictive performance, paving the way for enhancing leaf trait estimation and proposing a practical solution for pixel-level categorization within hyperspectral imagery (HSIs).