There is increasing evidence that stands structural variables determine forest functioning, yet its link with species diversity remains poorly understood. Specifically, we still have a limited understanding of how tree species richness affects within- and between-plot structural variability. Moreover, the impact of species identity and species’ interactions on the variation of forest stand structure at the plot level remains unclear. To answer these two questions, we performed terrestrial laser scanning (TLS) scanning in the FORBIO tree species diversity experiment at three experimental sites in Belgium in the summer of 2020. One site was 9 years old (Hechtel-Eksel) and the other two sites were 11 years old (Zedelgem and Gedinne). Each site included five different sets of one, two, three, and four species mixtures with one replica. At each plot, four single scan locations were used. Six stand structural variables were quantified based on gap fraction analysis and vertical plant profiles from TLS: total plant area index, canopy top height, foliage height diversity, coefficient of variation of vertical plant profiles, the peak of vertical plant profiles, and the height of the peak. The stand structural variability refers to the within-plot and between-plot standard deviation of each structural variable on species richness level, the stand structure was calculated by the average of four scan locations for each structural variable in each plot. We found that within-plot structural variability increased with species richness and that the between-plot structural variability decreased with species richness. By adopting diversity interaction modeling, a model framework that can predict the contribution of species identity and interactions between each pair of two species, we found that models were significantly improved when species identity and interaction effects rather than only species richness were considered.

Tree species (TS) composition is essential information for forest management, and remotely sensed data have proven to be useful for its prediction. In many countries, TS is interpreted manually from aerial images, which is a time and resource consuming process that entails substantial uncertainty. This study presents the results from eight large-area forest inventories in which species composition was predicted following an area-based approach using parametric Dirichlet models. The study assessed and compared combinations of three data sources: airborne laser scanning (ALS), spectral information from aerial images and multispectral satellite information from Sentinel-2, respectively. Field data from 1235 sample plots of 250 m² were used to classify TS in eight regions in Norway. Ten-fold cross validation was performed for all regions and independent validations were performed for five regions. The most accurate prediction of TS composition was obtained by combining ALS and multidate imagery from Sentinel-2. The independent validation showed larger accuracies than obtained when using cross validation. The root mean square differences (RMSD) for the independent validations were 65.5, 50.4 and 21.1 m³/ha (relative RMSD were 40%, 109% and 183%) for Norway spruce, Scots pine and deciduous species, respectively. In one of the regions with independent validation plots of 3700 m², predictions of the dominant species were compared to results obtained through manual photointerpretation. For these plots, greater accuracies were obtained for model predictions compared to the manual photo interpretation. This study highlights the utility of remotely sensed data for prediction of tree species compositions and their use in operational forest inventories.

Laser scanners have been successfully employed for forest variables estimation. Nevertheless, the regular monochromatic LiDAR systems cannot capture enough information for tree species classification. Therefore, optical multispectral (MS) imagery or their integration with airborne LiDAR has been studied for classifying forest tree species. Modern multispectral airborne LiDAR has improved tree species identification accuracy compared to single-channel systems, by incorporating both structural and spectral information. Hence, the characterization of tree species is one of the primary and most popular potential applications of multi-wavelength laser scanning attracted attention in recent years. State-of-the-art deep learning (DL) algorithms have recently demonstrated promising potential in boosting the automation and accuracy of geospatial data processing. However, the effectiveness of DL models has never been explored in tree species classification using MS-LiDAR data, and the conducted studies have relied only on conventional handcrafted features. Therefore, the aim of this paper is to report our investigations on the capability of automatic DL-based features for unsupervised classification of tree species utilizing MS airborne LiDAR data. First, the point clouds of three missions with Green, NIR, and SWIR wavelengths were pre-processed and merged to create MS LiDAR datasets. Subsequently, individual trees were delineated using SLIC superpixel segmentation. A Convolutional Auto-Encoder (CAE) was used for extracting deep features from the MS LiDAR data. Then, several statistical features were extracted for each tree segment from deep ones. Finally, the Mini Batch K-means was used to cluster the resulting features and identify the tree species. The achieved results indicate the superiority of DL-based features over the manual ones for tree species classification, improving the overall accuracy and kappa coefficient by 5.63% and 8.32%, respectively. To the best of our knowledge, this is the first study exploring unsupervised DL for forest mapping using MS LiDAR and confirms its potential as a single-data source solution.

Person-carried laser-scanning has recently become an efficient tool to survey forest stands, and with appropriate software routines individual tree measurements can be generated from the three-dimensional point clouds. Diameter at breast height (DBH), tree height and crown dimensions can already be precisely measured. However, it remains still a challenge to correctly predict tree species labels – due to missing spectral data and the noise inherent in the point cloud.

This study used a fully automatic workflow to segment individual tree point clouds from data acquired with the GeoSLAM ZEB Horizon person-carried laser scanning system. A total of 4.800 trees were extracted and the tree species labels were recorded in the field as reference. Two approaches were tested to predict the labels of 8 common tree species. To represent the physical structure of different trees, two-dimensional images were created in different projection angles, and a convolutional neural network was trained to classify the tree species. The second approach was based on non-linear regression using Gaussian kernels with individual tree measurement ratios and the distribution of intensity values of the tree crown as input variables. Both approaches were tested using k-fold cross validation.

Predicting tree species from point clouds is an essential step towards using person-carried laser scanning in forest inventory. The total growing stock of a forest stand can be reported according to different assortments and tree species, which improves the base for forest operation planning. Further uses are competition analysis and relating the spatial distribution of tree species to forest regeneration.

Forest productivity, the accumulation of biomass due to photosynthetic activity, and natural degradation, due to mortality, are key components of land-atmosphere interactions that contribute to and respond to changing climate. Productivity and degradation are dependent on diverse factors including plant physiology, disturbance, environmental conditions, geography, and soil attributes. We are developing a multi-path fusion Machine Learning (ML) system to predict forest productivity and degradation using training inputs from remote sensing, climatology, and soil data. We use increasing and decreasing canopy height, derived from multi-year airborne lidar mapping, as surrogates for productivity and degradation. The ML development is done at National Ecological Observatory Network (NEON) forest sites, where NEON’s airborne remote sensing data for model development includes: [1] lidar point clouds and gridded products, used to characterize canopy structure and topographic attributes, and [2] hyperspectral imaging and derived vegetation indices, used to characterize canopy composition and function. Additional data sets used for the ML training are NASA MERRA-2 climate time series, NOAA NIDIS drought time series, USDA NATSGO soil mapping and large-footprint lidar waveforms we simulate from the point clouds. Our ML method makes use of a hierarchical collection of deep neural networks (NN) with NN layers optimized for our disparate types of input data. We will illustrate our methods and present initial results at the Mountain Lake Biological Station, VA NEON site. Canopy height increases due to growth and decreases due to canker worm defoliation and individual tree mortality, and associated spectral changes, are documented using airborne acquisitions in 2015, 2017, 2018, 2021 and 2022. Our ML architecture will contribute to the design and operation of New Observing Systems, NASA’s next generation approach for Earth remote sensing, providing a dynamic and more complete picture of natural phenomena utilizing optimized measurements from multiple vantage points across spectral, spatial, and temporal dimensions.