Laser Scanning Data Research Articles

Exploring the Scientific Mechanism of Tree Structure Network based on LiDAR Point Cloud Data

Abstract. To explore how trees optimize their structure, we developed a method based on Pareto optimality theory. This method consists of the following operations. Firstly, we utilize Quantitative Structure Models for Single Trees from Laser Scanner Data (TreeQSM) to extract tree structures from point clouds acquired through Light Detection and Ranging (LiDAR). Subsequently, we utilize a graph-theoretical model to characterize the natural tree structure networks and implement a greedy algorithm to generate Pareto optimal tree structure networks. Finally, based on the Pareto optimality theory, we explore whether tree structures are multi-objective optimized. This paper demonstrates that tree structures lie along the Pareto front between minimizing "transport distance" and minimizing "total length". The growth pattern of trees, which produces multi-objective optimized structures, is likely an intrinsic mechanism in the generation of tree structure networks. The location of tree structures along the Pareto front varies under different environmental conditions, reflecting their diverse survival strategies.

Tree growth potential and its relationship with soil moisture conditions across a heterogeneous boreal forest landscape

Forest growth varies across landscapes due to the intricate relationships between various environmental drivers and forest management. In this study, we analysed the variation of tree growth potential across a landscape scale and its relation to soil moisture. We hypothesised that soil moisture conditions drive landscape-level variation in site quality and that intermediate soil moisture conditions demonstrate the highest potential forest production. We used an age-independent difference model to estimate site quality in terms of maximum achievable tree height by measuring the relative change in Lorey’s mean height for a five year period across 337 plots within a 68 km2 boreal landscape. We achieved wall-to-wall estimates of site quality by extrapolating the modelled relationship using repeated airborne laser scanning data collected in connection to the field surveys. We found a clear decrease in site quality under the highest soil moisture conditions. However, intermediate soil moisture conditions did not demonstrate clear site quality differences; this is most likely a result of the nature of the modelled soil moisture conditions and limitations connected to the site quality estimation. There was considerable unexplained variation in the modelled site quality both on the plot and landscape levels. We successfully demonstrated that there is a significant relationship between soil moisture conditions and site quality despite limitations associated with a short study period in a low productive region and the precision of airborne laser scanning measurements of mean height.

Combining terrestrial lidar with single line transects to investigate geomorphic change: A case study on the Upper Verde River, Arizona

The Upper Verde River in northern Arizona, USA is a vital resource for the wildlife and humans that rely on its waters. We characterize the riparian corridor topography using terrestrial laser scanner (TLS) data from 2021 to 2022. We also quantify geomorphic changes associated with human and climate-driven alterations in river flow and vegetation changes by combining the contemporary lidar surveys with legacy measurements from single line geomorphology transects measured by the United States Forest Service (USFS) in 2009. Seventeen plots along the Upper Verde River were surveyed with the TLS and the data were coregistered within individual plots with a Root Mean Square Error of <0.03 m among scan positions. Digital Elevation Models (DEM) were derived for each plot from the TLS data at 10 cm resolution and compared to the 2009 USFS cross-section data to quantify elevation changes. In areas with statistically significant change, we detected maximum changes in elevation due to erosion and deposition of −0.37 m and + 0.97 m, respectively. Topographic changes over the 13-year period were predominately aggradation and associated with sediment deposition, which we hypothesize might have resulted from altered river flow and vegetation encroachment. This study also demonstrates a quantitative and statistical methodology to fuse traditional single line cross-section data with contemporary lidar data to quantify geomorphic change. The novel approach demonstrated here is broadly applicable to natural resource managers for integrating and contextualizing legacy topographic data for understanding past, present, and future landscape and habitat changes.

Spatio-Temporal Morphodynamics of a Nourished Sandy Shore Based on LiDAR Measurements

Coastal erosion is a pervasive global phenomenon, exemplified by the Hel Peninsula situated in the Gulf of Gdańsk, the southern Baltic Sea. The geological constitution of the Hel Peninsula, characterized by sandy and loosely consolidated material, predisposes its coastal zones to continual morphological changes. The peninsula’s limited width and elevation exacerbate shoreline erosion, particularly during periods of heightened storm activity. This study scrutinizes the effectiveness of coastal nourishment interventions, with a specific focus on segments influenced by the Władysławowo port and the Kuźnica vicinity, over several years. This specific section of the coast serves as a significant case study due to its role as a transit zone for sand transport along the whole peninsula. Protective measures, including shore nourishments and coastal groynes, aim to mitigate erosion impacts. Utilizing Airborne Laser Scanning (ALS) data spanning from 2008 to 2022, erosion dynamics were analyzed. The analysis reveals significant erosion patterns coinciding with the frequency and volume of nourishment material deposition, particularly evident in heavily nourished areas proximate to Władysławowo and Kuźnica. Despite persistent monitoring endeavors, persistent erosive trends pose imminent threats to Kuźnica’s infrastructure, necessitating further research into the efficacy of implemented coastal protection measures.

Tree-Related Microhabitats and Multi-Taxon Biodiversity Quantification Exploiting ALS Data

The quantification of tree-related microhabitats (TreMs) and multi-taxon biodiversity is pivotal to the implementation of forest conservation policies, which are crucial under the current climate change scenarios. We assessed the capacity of Airborne Laser Scanning (ALS) data to quantify biodiversity indices related to both forest beetle and bird communities and TreMs, calculating the species richness and types of saproxylic and epixylic TreMs using the Shannon index. As biodiversity predictors, 240 ALS-derived metrics were calculated: 214 were point-cloud based, 14 were pixel-level from the canopy height model, and 12 were RGB spectral statistics. We used the random forests algorithm to predict species richness and the Shannon diversity index, using the field plot measures as dependent variables and the ALS-derived metrics as predictors for each taxon and TreMs type. The final models were used to produce wall-to-wall maps of biodiversity indices. The Shannon index produced the best performance for each group considered, with a mean difference of −6.7%. Likewise, the highest R2 was for the Shannon index (0.17, against 0.14 for richness). Our results confirm the importance of ALS data in assessing forest biodiversity indicators that are relevant for monitoring forest habitats. The proposed method supports the quantification and monitoring of the measures needed to implement better forest stands and multi-taxon biodiversity conservation.

Modeling Geospatial Distribution of Peat Layer Thickness Using Machine Learning and Aerial Laser Scanning Data

Organic horizons including peat deposits are important terrestrial carbon pools, and various chemical, biological, and water exchange processes take place within them. Accurate information on the spatial distribution of organic soils and their properties is important for decision-making and land management. In this study, we present a machine learning approach for mapping the distribution of organic soils and determining the thickness of the peat layer using more than 24,000 peat layer thickness measurements obtained from field data, airborne laser scanning (ALS) data and various indices obtained from therein, as well as other cartographic materials. Our objectives encompassed two primary aims. Firstly, we endeavored to develop updated cartographic materials depicting the spatial distribution of peat layers. Secondly, we aimed to predict the depth of peat layers, thereby enhancing our understanding of soil organic carbon content. Continentality, a wet area map, latitude, a depth to water map with catchment area of 10 ha, and a digital elevation model were the most important covariates for the machine learning model. As a result, we obtained a map with three peat layer thickness classes, an overall classification accuracy of 0.88, and a kappa value of 0.74. This research contributes to a better understanding of organic soil dynamics and facilitates improved assessments of soil organic carbon stocks.

Branch architecture quantification of large-scale coniferous forest plots using UAV-LiDAR data

Branch architecture plays an important role in forest physical and ecological processes, greatly affecting forest spatial structure and the accumulation, production and distribution of organic carbon. Very few studies have quantified the branch architecture of large-scale forest plots, due to the lack of high-resolution large-scale three-dimensional data. The emergence of unmanned aerial vehicle-light detecting and ranging (UAV-LiDAR) provides great potential to overcome this limitation. However, due to insufficient quality of UAV-LiDAR data for branch reconstruction, existing algorithms remain tremendously challenging. We propose an effective branch reconstruction algorithm for UAV-LiDAR data to quantify branch architecture. First, individual branches are extracted using a region growth method that is guided by the branch growth direction and local smoothness constraint. Second, incorrect branches are eliminated based on three pieces of branch features, i.e., specific angles between branches at the same whorl, specific height differences between branches at different whorls, and the growth pattern of branches from the stem outward, reaching maximum distances at tips. Finally, branches are reconstructed using polynomial fitting, and branch architecture parameters are extracted based on branch models. The proposed algorithm simplifies branch reconstruction by skipping the separation of photosynthetic and non-photosynthetic components. It also adapts well to UAV-LiDAR point clouds, generating more realistic reconstructions. The algorithm successfully identified non-photosynthetic components in experiments involving 240 trees, including Scots pine, Norway spruce, and Radiata pine. The average overall accuracy for these three species was 0.88, 0.76, and 0.74, respectively. The proposed algorithm was tested for branch reconstruction using UAV-LiDAR data from a two-hectare Larix principis-rupprechtii plot. The accuracy of branch identification and parameter extraction accuracy were evaluated branch by branch based on the individual branches manually identified in terrestrial laser scanning data. Results showed the F-score of branch and stem identification was 0.58 and 1. The relative RMSE of branch length and angle was 36.87% and 18.3%, and ones of stem length and diameter were 1.46% and 6.16%. The proposed algorithm outperformed the well-known reconstruction algorithms, including TreeQSM, AdTree and Laplacian-based algorithms.

Determining the trend of geometrical changes of a hydrotechnical object based on data in the form of LiDAR point clouds

Monitoring the technical condition of hydrotechnical facilities is crucial for ensuring their safe usage. This process typically involves tracking environmental variables (e.g., concrete damming levels, temperatures, piezometer readings) as well as geometric and physical variables (deformation, cracking, filtration, pore pressure, etc.), whose long-term trends provide valuable information for facility managers. Research on the methods of analyzing geodetic monitoring data (manual and automatic) and sensor data is vital for assessing the technical condition and safety of facilities, particularly when utilizing new measurement technologies. Emerging technologies for obtaining data on the changes in the surface of objects employ laser scanning techniques (such as LiDAR, Light Detection, and Ranging) from various heights: terrestrial, unmanned aerial vehicles (UAVs, drones), and satellites using sensors that record geospatial and multispectral data. This article introduces an algorithm to determine geometric change trends using terrestrial laser scanning data for both concrete and earth surfaces. In the consecutive steps of the algorithm, normal vectors were utilized to analyze changes, calculate local surface deflection angles, and determine object alterations. These normal vectors were derived by fitting local planes to the point cloud using the least squares method. In most applications, surface strain and deformation analyses based on laser scanning point clouds primarily involve direct comparisons using the Cloud to Cloud (C2C) method, resulting in complex, difficult-to-interpret deformation maps. In contrast, preliminary trend analysis using local normal vectors allows for rapid threat detection. This approach significantly reduces calculations, with detailed point cloud interpretation commencing only after detecting a change on the object indicated by normal vectors in the form of an increasing deflection trend. Referred to as the cluster algorithm by the authors of this paper, this method can be applied to monitor both concrete and earth objects, with examples of analyses for different object types presented in the article.

Digital 3D Reconstruction of Ancient Chinese Great Wild Goose Pagoda by TLS Point Cloud Hierarchical Registration

Digital reconstruction of ancient buildings is very challenging due to its architectural complexity and structural delicacy. Therefore, how to apply new Earth Observation (EO) technology to digitally reconstruct the complete and real model of ancient buildings without damage has become an urgent issue. This article proposes a multi-scan point cloud hierarchical registration method for 3D reconstruction of the ancient Chinese Great Wild Goose Pagoda based on Terrestrial Laser Scanner (TLS) data. First, a multi-scan point cloud hierarchical registration method is elaborately designed for 3D reconstruction. Then, the outline feature of the ancient Chinese Great Wild Goose Pagoda is automatically extracted by point cloud slicing. Finally, fine modeling is carried out according to the outline feature extracted from the reconstructed point cloud, and after that block combination and texture mapping are conducted for 3D digitalization and documentation. The experiments indicate that the Root Mean Square Error (RMSE) of multi-scan hierarchical registration of the translation and rotation are less than 3 cm and 0.862 ∘ , respectively, and the complete and real model of the Great Wild Goose Pagoda is obtained by fine modeling and texture mapping, which provides an important reference for 3D digital protection of ancient buildings.

Assessing the Performance of Handheld Laser Scanning for Individual Tree Mapping in an Urban Area

Precise individual tree or sample-based inventories derived from 3D point cloud data of mobile laser scanning can improve our comprehensive understanding of the structure, function, resilience, biodiversity, and ecosystem services of urban forests. This study assessed the performance of a handheld laser scanning system (HLS) for the extraction of tree position, diameter at breast height (DBH), and tree height (H) in an urban area. A total of 2083 trees of 13 species from 34 plots were analyzed. The results showed that the registration of tree positions using ground control points (GCPs) demonstrated high accuracy, with errors consistently below 0.4 m, except for a few instances. The extraction accuracy of DBH for all trees and individual species remained consistently high, with a total root mean square error (RMSE) of 2.06 cm (6.89%) and a bias of 0.62 cm (2.07%). Notably, broad-leaved trees outperformed coniferous trees, with RMSE and bias values of 1.86 cm (6%) and 0.76 cm (2.46%), respectively, compared to 2.54 cm (9.46%) and 0.23 cm (0.84%), respectively. The accuracy of H extraction varied significantly among different species, with R2 values ranging from 0.65 to 0.92. Generally, both DBH and H were underestimated compared to ground measurements. Linear mixed-effects models (LMEs) were applied to evaluate factors affecting the performance of HLS with the plot as a random factor. LME analysis revealed that plant type and terrain significantly influenced the accuracy of DBH and H derived from HLS data, while other fixed factors such as plot area, tree density, and trajectory length showed no significance. With a large sample size, we concluded that the HLS demonstrated sufficient accuracy in extracting individual tree parameters in urban forests.

Characterizing post-fire northern boreal forest height dynamics

ABSTRACT The forest structure of Canada’s northern unmanaged boreal forests are not well characterized despite their critical importance for ecosystem services such as carbon storage and caribou habitat, as well as their vulnerability to climate change and increasing disturbance rates, especially wildfire. Spaceborne lidar observations acquired from the Advanced Topographic Laser Altimeter System (ATLAS) instrument onboard the Ice, Cloud and Elevation Satellite-2 (ICESat-2), enables monitoring of forest structure in these remote northern ecosystems and characterization of long-term post-fire recovery; however, ICESat-2 is a sampling instrument, acquiring data transects rather than wall-to-wall data coverage. By imputing ICESat-2 estimates of canopy height using annual time-series (1984–2021) of spectral indices, change metrics, and land cover metrics from Landsat surface reflectance composites, we derived annual, spatially explicit, wall-to-wall estimates of canopy height for the period 1984–2021 across a 19.6 Mha area of boreal forests in northwestern Ontario and a 630,000 ha area of managed boreal forest further south that had coincident airborne laser scanning (ALS) data for validation. The accuracy of the imputation of derived canopy height estimates in the northwestern site was assessed using a reserved set of ICESat-2 observations (r = 0.76; RMSD = 3.04 m, pRMSD = 33.9%, MD = −0.10 m, and pMD = −1.0%). Using coincident ALS data, the accuracy of the imputed canopy heights in the southern study site were also assessed (r = 0.81; RMSD = 3.65 m, pRMSD = 25.87%), MD = −0.70 m, pMD = 4.96%). Examination of height dynamics post fire indicated that canopy height decreased 8–10 years post fire and recovered between 60% and 85% of pre-fire canopy height within 30 years post fire. The approach presented could readily be extended to similar northern boreal forest areas to provide broad-scale characterizations of fire impacts on forest structure and subsequent recovery.

Automatic measurement of grid structures displacement through fusion of panoramic camera and laser scanning data

Measuring a large grid structure is always based on the traditional total station, which requires a significant workforce. Aiming at the ring structure's characteristics, this paper fully uses multi-sensor advantages and proposes an automatic identification method of grid structure nodes. This method combines deep learning, panoramic camera, and laser scanning and improves the existing method of measuring the lifting process of the grid structure. The main contributions of this paper are as follows: (1) Aiming at the issue with existing measurement methods relying on complex multi-camera systems, a method based on a single panoramic camera and deep learning is proposed to quickly complete the full-field image target points detection, which provides a basis for the identification of target areas in the point clouds; (2) For the challenge posed by a large number of point clouds models and complex processing, a method of automatically capturing point clouds nodes by integrating panoramic camera and laser scanning was proposed. The projection points of the interest region of the image corresponding to the target region in the point clouds and precise identification of nodes were completed; (3) A node classification method and optimization method based on density peak clustering was proposed to solve the problem of node aliasing and processing node center in some areas of interest, for the classification and displacement calculation of grid structure nodes. The proposed method is validated on the scale frame model of a gymnasium, demonstrating the practicability of the proposed approach.

Computation of prediction intervals for forest aboveground biomass predictions using generalized linear models in a large-extent boreal forest region

Abstract Remotely sensed data have an important application for estimation of forest variables, e.g. height, volume, and aboveground biomass (AGB). The increased use of remotely sensed data implemented along with model-based inference has shown improved efficiency in prediction and mapping of such forest variables. In this study, plot-level airborne laser scanning data and Swedish National Forest Inventory field reference data were used to predict AGB using generalized linear models (GLMs) assuming Gamma and Tweedie distributions for the field observed AGB. The GLMs were selected considering the convenience of not correcting transformation bias as it is required in other regression models with transformed response variable. To overcome the challenge in providing reliable uncertainty estimates for the estimated forest variable map products at individual pixel-scale, we focused on computing 95% prediction intervals (PIs) for Gamma and Tweedie GLMs with a square root link function. The relative uncertainties were computed as the ratio between the half-width of the PIs and the predicted AGBs. The AGB-airborne laser scanning models were developed with root mean square error values of 22.6 Mgha−1 (26%) and 21.7 Mgha−1 (25%), respectively, for the Gamma and Tweedie GLMs. Two methods were applied to compute PIs for the Gamma GLM, one using the R package ‘ciTools’ and another derived through asymptotic theory. It was found that the 95% PIs computed using ‘ciTools’ had the most accurate coverage probability in comparison to the other method. An extended version of these PIs was also utilized for the Tweedie GLMs. The range of PIs associated with the prediction of AGB were narrower for lower predicted AGB values compared with the length of higher predicted AGB values. Comparing the two fitted models, the Gamma GLM showed lower relative uncertainties for the lower range of predicted AGBs, whereas the Tweedie GLM showed lower relative uncertainties for the higher range of predicted AGBs. Overall, the Tweedie GLM provided a better model fit for AGB predictions.

ENHANCING URBAN PATHFINDING FOR PEDESTRIANS THROUGH FUSION OF MLS AND HMLS DATA

Abstract. Pedestrian pathfinding is crucial for enhancing pedestrian mobility in urban environments. In this research a method to generate navigation graphs based on Mobile Laser Scanning (MLS) and Handheld Laser Scanning (HMLS) data fusion is developed. The input data comprises a 2-kilometer urban street network that integrates both MLS and HMLS data, effectively mitigating sidewalk occlusions caused mainly by parked vehicles. The proposed method encompasses the following steps: (1) Deep Learning semantic segmentation of urban ground elements and navigation-related obstacles, (2) Static vs. Dynamic object differentiation to replicate conditions of unobstructed passage. (3) Sidewalk enrichment with inclination, preservation status, and width. (4) Physical accessibility estimation between sidewalks and crosswalks incorporating curb information. And (5) navigation graph computation based on the enhanced sidewalk data, crosswalks, with accurate node location and connections. Routes were calculated within a Geographical Information System, and results ensure that pedestrians can navigate in urban environments with precision and efficiency.

Semantic segmentation of raw multispectral laser scanning data from urban environments with deep neural networks

Real-time semantic segmentation of point clouds has increasing importance in applications related to 3D city modelling and mapping, automated inventory of forests, autonomous driving and mobile robotics. Current state-of-the-art point cloud semantic segmentation methods rely heavily on the availability of 3D laser scanning data. This is problematic in regards of low-latency, real-time applications that use data from high-precision mobile laser scanners, as those are typically 2D line scanning devices. In this study, we experiment with real-time semantic segmentation of high-density multispectral point clouds collected from 2D line scanners in urban environments using encoder - decoder convolutional neural network architectures. We introduce a rasterized multi-scan input format that can be constructed exclusively from the raw (non-georeferenced profiles) 2D laser scanner measurement stream without odometry information. In addition, we investigate the impact of multispectral data on the segmentation accuracy. The dataset used for training, validation and testing was collected with multispectral FGI AkhkaR4-DW backpack laser scanning system operating at the wavelengths of 905 nm and 1550 nm, and consists in total of 228 million points (39 583 scans). The data was divided into 13 classes that represent various targets in urban environments. The results show that the increased spatial context of the multi-scan format improves the segmentation performance on the single-wavelength lidar dataset from 45.4 mIoU (a single scan) to 62.1 mIoU (24 consecutive scans). In the multispectral point cloud experiments we achieved a 71 % and 28 % relative increase in the segmentation mIoU (43.5 mIoU) as compared to the purely single-wavelength reference experiments, in which we achieved 25.4 mIoU (905 nm) and 34.1 mIoU (1550 nm). Our findings show that it is possible to semantically segment 2D line scanner data with good results by combining consecutive scans without the need for odometry information. The results also serve as motivation for developing multispectral mobile laser scanning systems that can be used in challenging urban surveys.

Improvement of photogrammetric joint roughness coefficient value by integrating automatic shooting parameter selection and composite error model

In order to improve the accuracy of the photogrammetric joint roughness coefficient (JRC) value, the present study proposed a novel method combining an autonomous shooting parameter selection algorithm with a composite error model. Firstly, according to the depth map-based photogrammetric theory, the estimation of JRC from a three-dimensional (3D) digital surface model of rock discontinuities was presented. Secondly, an automatic shooting parameter selection algorithm was novelly proposed to establish the 3D model dataset of rock discontinuities with varying shooting parameters and target sizes. Meanwhile, the photogrammetric tests were performed with custom-built equipment capable of adjusting baseline lengths, and a total of 36 sets of JRC data was gathered via a combination of laboratory and field tests. Then, by combining the theory of point cloud coordinate computation error with the equation of JRC calculation, a composite error model controlled by the shooting parameters was proposed. This newly proposed model was validated via the 3D model dataset, demonstrating the capability to correct initially obtained JRC values solely based on shooting parameters. Furthermore, the implementation of this correction can significantly reduce errors in JRC values obtained via photographic measurement. Subsequently, our proposed error model was integrated into the shooting parameter selection algorithm, thus improving the rationality and convenience of selecting suitable shooting parameter combinations when dealing with target rock masses with different sizes. Moreover, the optimal combination of three shooting parameters was offered. JRC values resulting from various combinations of shooting parameters were verified by comparing them with 3D laser scan data. Finally, the application scope and limitations of the newly proposed approach were further addressed.

Integrating Dendrochronological and LiDAR Data to Improve Management of Pinus canariensis Forests under Different Thinning and Climatic Scenarios

Thinning focused on achieving growth and diameter management objectives has typically led to stands with reduced climate sensitivity compared to unthinned stands. We integrated dendrochronological with Airborne Laser Scanner (LiDAR) data and growth models to assess the long-term impact of thinning intensity on Canary pine (Pinus canariensis) radial growth. In 1988, 18 permanent treatment units were established in 73-year-old Canary pine plantations and three thinning treatments were applied (C–control-unthinned; 0% basal area removal; MT–moderate thinning: 10% and 15% basal area removal, and HT–heavy thinning: 46% and 45% basal area removal on the windward and leeward slopes, respectively). Dendrochronological data were measured in 2022 and expressed as basal area increment (BAI). The impact of climate on growth was examined by fitting linear regression models considering two different Representative Concentration Pathway (RCP) climate scenarios, RCP 2.6 and RCP 4.5. Finally, LiDAR data were used for standing segmentation to evaluate changes in overall growth under different climatic scenarios. The LiDAR–stand attributes differed between aspects. The BAI of the most recent 20 years (BAI20) after thinning was significantly higher for the moderate and heavy treatments on the leeward plots (F = 47.31, p &lt; 0.001). On the windward plots, BAI decreased after moderate thinning. Considerable thinning treatments resulted in stronger changes in growth when compared to RCP climatic scenarios. From a silviculture perspective, the mapping of canopy structure and growth response to thinning under different climatic scenarios provides managers with opportunities to conduct thinning strategies for forest adaptation. Combining dendrochronological and LiDAR data at a landscape scale substantially improves the value of the separate datasets as forecasted growth response maps allow improving thinning management plans.

Mapping large European aspen (Populus tremula L.) in Finland using airborne lidar and image data

European aspen is a keystone species in boreal forests, which support numerous ecologically important and endangered species. As detection of those species by remote sensing is impossible, we instead investigated the detection of large aspen trees using airborne laser scanning and aerial image data. However, this is a challenge due to their low quantity and scattered occurrence. The performance was assessed with representative and unrepresentative (where aspens were over-represented) samples of the population. First, we detected individual trees and then the random forest (RF) classifier was used to identify large aspens. The RF classification was implemented with and without synthetic minority oversampling technique (SMOTE) to balance the training data due to the rarity of large aspens. At the tree-level, the best F1-score (0.44) was obtained when the unrepresentative plot data were used with SMOTE. However, the F1-score decreased to 0.21 when the representative data were used. The best plot-level (plots with at least one aspen tree) F1-score with the representative plot data was 0.41. We conclude that although data augmentation may improve the result, it is difficult to detect large aspen trees in genuine populations.

Forest Canopy Height Retrieval and Analysis Using Random Forest Model with Multi-Source Remote Sensing Integration

Forest canopy height is an important indicator of the forest ecosystem, and an accurate assessment of forest canopy height on a large scale is of great significance for forest resource quantification and carbon sequestration. The retrieval of canopy height based on remote sensing provides a possibility for studying forest ecosystems. This study proposes a new method for estimating forest canopy height based on remote sensing. In this method, the GEDI satellite and ICESat-2 satellite, which are different types of space-borne lidar products, are used to cooperate with the Landsat 9 image and SRTM terrain data, respectively. Two forest canopy height-retrieval models based on multi-source remote sensing integration are obtained using a random forest regression (RFR) algorithm. The study, conducted at a forest site in the northeastern United States, synthesized various remote sensing data sets to produce a robust canopy height model. First, we extracted relative canopy height products, multispectral features, and topographic data from GEDI, ICESat-2, Landsat 9, and SRTM images, respectively. The importance of each variable was assessed, and the random forest algorithm was used to analyze each variable statistically. Then, the random forest regression algorithm was used to combine these variables and construct the forest canopy height model. Validation with airborne laser scanning (ALS) data shows that the GEDI and ICESat-2 models using a single data source achieve better accuracy than the Landsat 9 model. Notably, the combination of GEDI, Landsat 9, and SRTM data (R = 0.92, MAE = 1.91 m, RMSE = 2.78 m, and rRMSE = 12.64%) and a combination of ICESat-2, Landsat 9, and SRTM data (R = 0.89, MAE = 1.84 m, RMSE = 2.54 m, and rRMSE = 10.75%). Compared with the least accurate Landsat 9 model, R increased by 29.58%, 93.48%, MAE by 44.64%, 46.20%, RMSE by 42.80%, 49.40%, and the rRMSE was increased by 42.86% and 49.32%, respectively. These results fully evaluate and discuss the practical performance and benefits of multi-source data retrieval of forest canopy height by combining space-borne lidar data with Landsat 9 data, which is of great significance for understanding forest structure and dynamics. The study provides a reliable methodology for estimating forest canopy height and valuable insights into forest resource management and its contribution to global climate change.

Assessing the effect of ensemble learning algorithms and validation approach on estimating forest aboveground biomass: a case study of natural secondary forest in Northeast China

ABSTRACT Accurate estimation of forest aboveground biomass is essential for the assessment of regional carbon cycle and the climate change in the terrestrial ecosystem. Currently, ensemble learning algorithms and cross-validation methods have been widely applied to estimate regional forest Above Ground Biomass (AGB). However, the effects of ensemble learning algorithms, validation methods, and their interactions on forest AGB estimation were rarely investigated. Based on Landsat 8 Operational Land Imager (OLI) imagery, Airborne Laser Scanning (ALS) data and China’s National Forest Continuous Inventory data, this study explored the effects of five ensemble learning algorithms, including Simple Averaging (SA), Weighted Averaging (WA), Stacked Generalization (SG), Random Forest (RF) and Extreme Gradient Boosting (XGBoost), and two validation approaches (i.e. 10-fold and leave-one-out cross-validation) on the AGB estimation of the Natural Secondary Forests (NSFs) in northeast China. The results revealed that the ensemble learning algorithms that combine heterogenous-based models (i.e. SA, WA, SG) generally produced higher accuracy than the base models (i.e. Decision Tree (DT), K-Nearest Neighbor (KNN), Support Vector Regression (SVR), Convolutional Neural Network (CNN)). Among all ensemble learning algorithms, the SG algorithm has the highest accuracy whereas the XGBoost algorithm has the lowest accuracy. Although prediction models considerably impact the accuracy of AGB estimation, the validation approach also plays a non-negligible role in AGB estimation. The leave-one-out cross-validation produced much higher accuracy than the 10-fold cross-validation using the same prediction model and tends to generate over-optimistic AGB estimates compared to 10-fold cross-validation, especially for the averaging and stacking ensemble learning algorithms (i.e. SA, WA, SG). This study highlights the potential challenges of applying a leave-one-out cross-validation approach and provides a scientific foundation for the feasibility of different ensemble learning algorithms and cross-validation approaches for accurate AGB estimation.