Skip to content
Frontiers
Search

Original Research

Frontiers in Forests and Global Change ·

Volume 9 - 2026 | https://doi.org/10.3389/ffgc.2026.1886496

Estimation of stem volume and tree structural attributes using backpack LiDAR: effects of leaf-on and leaf-off conditions

  • Can Vatandaslar1,2,3*corresponding author
  • Ivan Sačkov2
  • Vlastimil Murgaš2
  • Jozef Pajtík2
  • Mustafa Zeybek3
  1. 1Faculty of Forestry, Artvin Coruh University, Artvin, Türkiye
  2. 2Department of Forest Policy, Forest and Game Management, National Forest Centre, Zvolen, Slovakia
  3. 3Guneysinir Vocational School of Higher Education, Selcuk University, Konya, Türkiye
Article metrics0 viewsView details
PDFCiteMetrics

Abstract

Mobile light detection and ranging (LiDAR) technologies offer a promising alternative to traditional forest inventory methods by enabling rapid acquisition of three-dimensional point cloud data. This study evaluates the performance of a backpack laser scanning system in terms of stem volume, diameter at breast height (dbh), tree height, and stem taper estimation in deciduous forests under contrasting phenological conditions (leaf-off and leaf-on). Two sample plots dominated by Betula pendula and Populus tremula in Tatra National Park (Slovakia) were scanned using a LiBackpack DGC-50 system, and LiDAR-derived estimates were compared with reference measurements from destructively sampled trees. The results showed that stem volume can be estimated directly from point clouds with acceptable accuracy for most forest inventory applications, without relying on allometric models. The highest accuracy was achieved under leaf-off conditions (RMSE = 0.03 m 3), whereas errors increased under leaf-on conditions, particularly in the Populus plot (up to 0.06 m 3). Despite this decline, no significant differences were detected between LiDAR-derived and reference stem volumes across seasons (p > 0.05). dbh and tree height were also estimated more accurately under leaf-off conditions (RMSE = 0.97–1.87 cm and 0.91–0.93 m, respectively), although tree height was consistently underestimated (bias = −0.65 to −1.53 m) due to incomplete canopy representation. Overall, backpack laser scanning can be considered as a viable tool for estimating certain structural attributes in young- to mid-aged Betula and Populus stands; however, reliable stem taper reconstruction remains challenging in structurally complex trees because of occlusion and limited point density in upper stem sections. Another challenge is its limited feasibility for large-scale forest inventory applications, as operational forest inventories often require highly automated workflows.

Keywords

1 Introduction

Forest inventory information is still commonly collected through traditional field measurements and ecological observations; however, these approaches are labor-intensive, time-consuming, and costly. Alternatively, tree-, plot-, and stand-level attributes can be derived from three-dimensional point clouds acquired using close-range light detection and ranging (LiDAR) systems, a topic that has received considerable attention in forestry research over the past decade (Cabo et al., 2018; Chen et al., 2019; Hyyppä et al., 2020a; Liang et al., 2022; Öztürk et al., 2026). LiDAR is an active remote sensing technology that employs millions of laser pulses to measure the shape and structure of the surrounding environment. By recording the time each pulse takes to travel to an object and return to the sensor, the system can calculate the distance to that object, enabling users to reconstruct the environment from a dense collection of millions of point records known as a point cloud (OLAF, 2024).

As LiDAR sensors continue to decrease in size and weight, mobile systems (i.e., mobile laser scanning) have emerged as viable field-based alternatives among LiDAR platforms (Holvoet et al., 2025; Kükenbrink et al., 2025). With the advancement of simultaneous localization and mapping (SLAM) technology, the adoption of these systems has further accelerated. SLAM-based scanners enable continuous data acquisition within a local coordinate framework without relying on Global Navigation Satellite Systems (GNSS), which are sometimes unreliable under dense forest canopies (Lee et al., 2023; Wu et al., 2025). In addition, SLAM streamlines field and processing workflows by eliminating the need for ground control targets and post-processing registration, which are typically required in multi-scan scenarios.

Mobile LiDAR systems can be broadly categorized into the five following groups: (i) handheld, (ii) backpack, (iii) smartphone/tablet-based, (iv) drone-based, and (v) vehicle-based laser scanning systems. The first three categories are often collectively referred to as personal laser scanning (Wu et al., 2025). These systems are increasingly being adopted not only by researchers (Bobrowski et al., 2023; Čeňava et al., 2024) but also by operational field crews in countries such as France, Switzerland, and Finland to enhance the efficiency and comprehensiveness of national forest inventories (Holvoet et al., 2025; Kükenbrink et al., 2025). However, smartphone/tablet-based systems remain limited for professional applications due to range and data quality constraints (Oikawa et al., 2025; Výbošťok et al., 2026), while vehicle-based systems are often impractical in forest environments due to accessibility and regulatory limitations. Furthermore, although drone-based LiDAR can generate dense point clouds, its top-down acquisition perspective limits the accurate detection of lower stem sections and key attributes such as diameter at breast height (dbh) and stem position (Ko et al., 2021).

Among these platforms, backpack laser scanning is considered one of the most promising approaches for LiDAR-assisted forest inventories, owing to its relatively high data quality and operational efficiency in sample plots and small- to medium-scale forest areas (Ko et al., 2021). Unlike handheld systems, backpack scanners enable hands-free operation, improving mobility and safety, under uneven terrain and dense understory conditions. Moreover, mounting the system on the operator’s back provides greater stability during data acquisition, potentially reducing motion-induced noise. In contrast, handheld systems are more susceptible to drift and reduced data quality due to abrupt arm movements, inconsistent walking speeds, and prolonged static positioning (Chen et al., 2019; Wu et al., 2025). Studies by Ko et al. (2021) and Xie et al. (2022) demonstrated that backpack scanners can produce complete point clouds using appropriate walking patterns and support operational forest inventories with acceptable accuracy across diverse forest conditions.

While numerous studies have focused on dbh, tree height, and stem density, relatively few have addressed stem volume estimation using backpack laser scanners. In traditional forest inventories, stem volume is typically estimated indirectly using species-specific allometric models based on dbh alone or in combination with tree height measured from standing trees (Vidal et al., 2016; Hyyppä et al., 2020a). More accurate volume estimates can be obtained from felled trees using precise methods such as fluid displacement or stem dissection (Kershaw et al., 2017); however, these approaches require destructive sampling and are often impractical and ecologically undesirable. Therefore, direct estimation of stem volume from standing tree point clouds, without reliance on allometric models, represents a significant opportunity for forest inventory and precision forestry applications.

In direct estimation of stem volume from point clouds, occlusion remains a critical factor affecting point cloud quality. Occlusion refers to the obstruction of an object or surface by another object located between it and the laser sensor, potentially preventing laser pulses from capturing complete structural information along the line of sight. Earlier backpack laser scanning studies highlighted the effects of occlusion, particularly in the reconstruction of Picea stems (Hyyppä et al., 2020a) and upper sections of Pinus stems (Hartley et al., 2022). One potential mitigation strategy is conducting data acquisition under leaf-off conditions, when reduced foliage improves visibility. Although several studies have compared leaf-on and leaf-off conditions, most have focused on LiDAR platforms other than backpack systems (e.g., Davison et al., 2020; Imangholiloo et al., 2020). To our knowledge, the only comparative study using a backpack system to collect data from the same trees under different phenological conditions was conducted by Ko et al. (2024). The authors found that leaf-on conditions reduced tree detection and reconstruction accuracy due to increased occlusion, whereas leaf-off conditions improved overall point data quality. However, they did not evaluate stem volume, and the inclusion of evergreen species may have reduced differences between phenological conditions. Therefore, further research in fully deciduous forests is needed to determine the optimal timing for LiDAR data acquisition while considering its effects on point cloud quality.

The main objective of this study was to investigate the effects of data acquisition timing (i.e., leaf-off versus leaf-on conditions) on the estimation accuracy of key forest inventory attributes—with a particular emphasis on stem volume—using backpack laser scanning in deciduous forests. Plots dominated by Betula pendula and Populus tremula were scanned under both conditions, and estimates of dbh, tree height, and stem volume were evaluated against reference measurements obtained from destructively sampled trees. Using this reference measurements, this study also compared stem volume predictions derived from national allometric models using traditional measurements of dbh and tree height obtained from standing trees.

2 Materials and methods

2.1 Study area

The study area comprises two deciduous forest sample plots located in Tatra National Park, the first national park in Slovakia, established in 1949 (Figure 1). The Tatra Mountains form the highest mountain range within the Carpathians, and the park has been designated a UNESCO Biosphere Reserve since 1993. Its geological structure is primarily composed of granites and crystalline schists, particularly in the High Tatras. Owing to highly heterogeneous topography over a relatively short spatial scale, the region exhibits substantial diversity in both biotic and abiotic components. Five vegetation zones are recognized—submontane, mountain, subalpine, alpine, and nival—largely determined by elevation (TANAP, 2026). Dominant tree species include P. abies, Fagus sylvatica, Abies alba, Acer pseudoplatanus, and Sorbus aucuparia. The deadwood volume is also considerable; a recent inventory reported that it accounts for nearly half of the total standing volume (Bodziarczyk et al., 2020).

Location of the study area and forest sample plots (Basemap source: Esri).
Figure 1. Location of the study area and forest sample plots (Basemap source: Esri).

The first sample plot (hereafter referred to as the Betula plot) consists of 22 B. pendula trees and is located at an elevation of approximately 980 m above sea level. The stand represents a young- to mid-developmental stage, with two individuals classified as standing deadwood at the time of the field survey. The second sample plot (hereafter referred to as the Populus plot) is located approximately 3.5 km west of the first, at an elevation of 1,000 m above sea level. It consists of 22 P. tremula trees, 2 of which were also identified as standing deadwood. This plot exhibits a similar developmental stage; however, the trees were slightly smaller in terms of dbh compared to those in the Betula plot. The locations of both sample plots are shown in Figure 1. Video recordings of the plots, acquired using the backpack laser scanning system’s integrated spherical camera under both phenological conditions, are provided in the Supplementary materials.

2.2 Field measurements

Field measurements were conducted during the winter and summer seasons of 2025, corresponding to leaf-off and leaf-on conditions in deciduous vegetation, respectively. Fundamental tree attributes were collected during the leaf-off period, when tree crowns and tops were more visible. Specifically, the dbh of all standing trees with a dbh > 7 cm was measured using a caliper in two perpendicular directions, and the values were averaged to reduce errors associated with stem eccentricity. Total tree height, crown base height (i.e., bole height), and tree positions were measured using the Field-Map system (IFER, Prague, Czech Republic), an integrated forest data collection system comprising a laser rangefinder, electronic compass, inclinometer, and field computer (Sačkov et al., 2016). Tree positions recorded in a local Cartesian coordinate system were subsequently transformed into the World Geodetic System 1984 (WGS84). In addition, tree species, crown condition, and stem quality were recorded for each individual.

Based on species identity, dbh, and total height measured on standing trees, stem volume over bark (hereafter referred to as stem volume) was estimated using species-specific allometric models developed by Petráš and Pajtík (1991), which are also applied in the Slovak National Forest Inventory for the studied species.

Following these measurements, all trees within the plots were destructively sampled during the leaf-on season. This procedure involved felling the trees, removing branches and tops, and sectioning the stems into 1 m segments for direct volume measurement. For each segment, diameters were measured at the base, middle, and upper positions using a caliper. The remaining top section (shorter than 1 m) was also measured and included in the total stem volume. Segment volumes were computed using Newton’s formula (Equation 1), which integrates cross-sectional areas at the base, middle, and upper positions, with greater weight assigned to the middle section (Kershaw et al., 2017). The total stem volume was obtained by summing all segment volumes, and is expressed in cubic meters (m3).Where V,L, A b, A m, and A u denote the segment volume, segment length (1 m in this study), and cross-sectional areas at the base, middle, and upper ends of the segment, respectively.

V = L 6 ( A b + 4 A m + A u ) (1)

Consequently, two types of field-based stem volume estimates were obtained: (i) estimates derived from species-specific allometric models based on dbh and total height of standing trees (non-destructive sampling), and (ii) estimates derived from Newton’s formula applied to measurements of felled trees (destructive sampling). The latter approach, known as the stem dissection method, is generally considered more accurate (Laar and Akça, 2007; Kershaw et al., 2017).

2.3 Backpack laser scanning

Two successive scans were conducted by the same operator over the same forest sample plots under different phenological conditions. The first scan was performed immediately after the leaf-off field measurements of standing trees, while the second scan was conducted just prior to the leaf-on field measurements of felled trees. A LiBackpack DGC-50 backpack laser scanning system (GreenValley International, CA, USA) was used. The system is equipped with two Velodyne VLP-16 LiDAR sensors, a GNSS receiver, an inertial measurement unit, a spherical camera, and an onboard processing unit. It employs a SLAM algorithm for real-time localization and mapping during data acquisition, while GNSS integration and point cloud colorization are performed during post-processing (Bartholomeus et al., 2021).

Rather than following a predefined trajectory, the operator returned to the starting point to close the loop (see Figure 2), a recommended practice in SLAM-based mobile LiDAR surveys to ensure trajectory consistency and improve registration accuracy (Hartley et al., 2022; Vatandaşlar et al., 2023). In addition, individual scanning sessions were limited to less than 10 min to reduce potential drifts that may arise when the SLAM algorithm fails to maintain accurate frame-to-frame alignment (Liang et al., 2016; Chen et al., 2019; Kükenbrink et al., 2025).

Backpack laser scanning trajectories with field photographs of the *Populus* **(a)** and *Betula* plots **(b)**. White dots on the trajectory maps indicate tree positions.
Figure 2. Backpack laser scanning trajectories with field photographs of the Populus (a) and Betula plots (b). White dots on the trajectory maps indicate tree positions.

The scanning duration was approximately 7 min in a single session in the Populus plot. The Betula plot, being larger and bisected by a road (see Figure 2b), was scanned in two sessions lasting approximately 7 and 8 min, respectively. The scanning duration was similar under both leaf-off and leaf-on conditions. Trajectory files from the leaf-off scans are presented in Figure 2, along with stem maps of the surrounding area. As seen in Figure 2a, the presence of a creek along one edge of the Populus plot resulted in one-sided scanning for some trees. The creek and the affected area are shown in the Supplementary materials (leaf-off video, 00:04:00–00:04:20 and 00:05:03–00:05:30).

2.4 Point cloud data

2.4.1 Routine analysis

Raw point cloud data and supplementary files were imported into LiFuser-BP to align the acquired point clouds with trajectory data recorded during field surveys. LiFuser-BP is a proprietary software developed specifically for processing, visualizing, editing, and refining data obtained from GreenValley International LiBackpack systems. It supports differential GNSS corrections, optimization of SLAM-based point cloud registration, and integration of 2D panoramic imagery with 3D point clouds (GreenValley International, 2021).

The pre-processed data were subsequently imported into LiDAR360 (GreenValley International, CA, USA), a commercial software platform for large-scale forestry point cloud analysis. For height normalization, the elevation derived from the digital elevation model (DEM) was subtracted from the z-coordinate of each point. Individual trees within the sample plots were segmented using an individual tree detection approach (Marcello et al., 2024). The dbh and tree height values were then automatically extracted from the segmented point clouds and exported in spreadsheet format for comparison with field measurements from both standing and felled trees. LiDAR360 calculates tree height as the difference between the highest and lowest z values within a tree point cloud. No independent tree-top detection algorithm was applied at this stage. Stem volume was also computed using a semi-automated workflow described in the following subsection.

In addition, ArcGIS Pro (Esri, CA, USA) was used to generate location maps and to match field-recorded trees with their corresponding point cloud representation. Tree positions collected using the Field-Map system (IFER, Prague, Czech Republic) were linked to segmented trees by overlaying LiBackpack trajectory data with vector point features representing individual trees within the sample plots. This procedure enabled a one-to-one comparison of structural attributes, including dbh, tree height, and stem volume. As mentioned earlier, a few Populus trees were scanned from one side only due to the blockage caused by a creek along the plot edge. Although the incomplete point clouds of these trees produced crescent-shaped cross-sections rather than the full circular profiles shown in Figure 3b, we included all trees in the analysis to assess the potential effect of low point cloud quality on estimation accuracy.

General workflow of the developed algorithms. The first algorithm normalizes the point cloud and isolates individual stems **(a)**, while the second algorithm segments stems, fits best-fitting cylinders to each segment, and generates a summary report **(b)**.
Figure 3. General workflow of the developed algorithms. The first algorithm normalizes the point cloud and isolates individual stems (a), while the second algorithm segments stems, fits best-fitting cylinders to each segment, and generates a summary report (b).

2.4.2 Stem volume computation

Two JavaScript-based algorithms were developed to estimate stem volume semi-automatically from individual tree point clouds in Cyclone 3dr (Leica Geosystems AG, Switzerland) (Figure 3). The first algorithm handled pre-processing and stem extraction (see Supplementary material). Ground and non-ground points were first separated based on their height relative to a locally approximated ground surface, derived using a grid-based minimum elevation model (Axelsson, 2000; Zhang et al., 2016) with a 0.50 m resolution. Points located within 0.20 m above this surface were classified as ground, while the remaining points were considered non-ground. Non-ground points were then partitioned into consecutive 1 m vertical slices according to their height above ground level while preserving their original x, y, and z coordinates (Raumonen et al., 2013; Hackenberg et al., 2015). This procedure enabled height normalization (Liang et al., 2016) and generated segment-wise point clouds for subsequent analysis.

Within each slice, stem points were manually identified through visual inspection in Cyclone 3dr, and non-stem elements (e.g., branches, foliage, and neighboring vegetation) were removed. The remaining stem points from all slices were then merged to generate a complete stem point cloud, which served as input for the second algorithm (see Supplementary material).

The second algorithm performed section-based geometric fitting (Gander et al., 1994) to estimate stem volume. For each 1 m segment, the “best-fit cylinder” tool in Cyclone 3dr was applied to derive the optimal cylindrical representation based on the point distribution, including axis orientation, centerline, and radius (Schnabel et al., 2007). The volume of each segment was then computed using the standard cylinder volume equation (Equation 2):Where V, A b, and H denote the segment volume, cross-sectional area of the fitted cylinder base, and segment height (1 m in this study), respectively. Total stem volume was obtained by summing all segment volumes along the stem and is expressed in cubic meters (m3). Pseudocode for both algorithms is provided in the Supplementary materials.

V = A b xH (2)

The second algorithm is based on several assumptions and rule-based constraints. Specifically, each stem segment is assumed to be cylindrical with circular cross-sections at both ends. Segments not satisfying predefined quality criteria were excluded, and the fitting process was terminated to avoid unrealistic dbh and volume estimates. The applied criteria are summarized in Table 1, and are as follows:

  • Only segments containing a sufficient number of points were considered for cylinder fitting;
  • Fitted cylinders were accepted only if they met threshold values for radius, standard deviation, form error, and cylinder length;
  • If the radius difference between two consecutive segments exceeded 3 cm, the segment was refitted using either the previously accepted segment or a lower 0.5 m seed band.

Table 1. Rule-based decision criteria applied in Algorithm #2 (Supplementary materials).

Slice settingsCylinder initial guess / limits
Slice thickness (m)1.00Default radius (m)0.12
Min. points per slice25Min. radius (m)0.06
Max. points for fit5,000Max. radius (m)0.80
Max. slice count200Best cyl. Elim. points0
Radius continuity controlCylinder fit quality filters
Rad. cont. Threshold (m)0.03Max. cyl. Standard dev. (m)0.05
Inner region rad. Margin (m)0.02Max. cyl. Form error (m)0.10
Inner region extract tolerance (m)0.03Min. cyl. Length (m)0.20
Min. points inner region12Circle band setting
Lower refit band (m)0.50Circle band thickness (m)0.20
Min. points for lower refit band12Min. points per circle band12
Max. circle residual (m)0.03

The threshold values for the cylinder-fitting process (Table 1) were determined empirically through pilot testing and visual inspection to avoid undesired fittings to non-stem elements (e.g., branch or crown) and abrupt changes in stem taper. The maximum and minimum radius limits, on the other hand, were defined based on the lower dbh threshold used in the national forest inventory and the dbh of the largest tree in the scanned area, respectively. As previously mentioned, some Populus trees with crescent-shaped cross-sections resulting from incomplete point clouds included in the analysis. The rules specified in Table 1 enabled our algorithm to extract dbh and stem volume information from these trees. For example, a minimum of 12 points was required in the inner regions of crescent-shaped stem profiles to fit a reasonable circle or cylinder to these data. Further algorithmic details can be found in the pseudocode provided in the Supplementary material.

2.5 Accuracy assessment

Measurements obtained from felled trees on the ground are generally considered more accurate and precise than those obtained from standing trees (Laar and Akça, 2007; Kershaw et al., 2017). Total tree height, for example, is often measured indirectly using hypsometers, which rely on geometric or trigonometric principles and are susceptible to errors arising from instrument calibration and other sources of uncertainty. In some cases, accurately identifying tree tips can also be challenging due to crown complexity and occlusion (Kershaw et al., 2017). In contrast, measuring the length of a felled tree using a measuring tape or a sliding pole enables direct and more precise assessments. For this reason, measurements obtained from felled trees were adopted as the reference in this study. Measurements from standing trees, as well as estimates derived from LiDAR point clouds collected under leaf-on and leaf-off conditions, were subsequently evaluated against this reference.

Estimation accuracy for dbh, height, and stem volume was assessed using bias (Bias), relative bias (Bias%), standard error (SE), relative standard error (RSE%), root mean squared error (RMSE), and relative RMSE (RMSE%), as defined in Equations 3–8. Estimated stem volumes obtained from different methods—namely, backpack laser scanning under leaf-on and leaf-off conditions, and national allometric models based on field-measured dbh and height from standing trees—were further compared with reference volumes using paired-samples t-tests and Wilcoxon signed-rank tests to assess the presence of statistically significant differences. However, some authors note that a non-significant test result does not necessarily demonstrate equivalence between methods, but rather indicates insufficient evidence to reject the null hypothesis (Altman and Bland, 1983; Florea et al., 2024; Mahmud et al., 2025). Therefore, an agreement analysis was also performed to provide additional support for assessing the interchangeability of the measurement methods using the Bland–Altman approach. This approach evaluates the agreement between two quantitative measurement methods by plotting the differences between paired measurements against their means. By doing so, it allows the assessment of systematic bias and defines the 95% limits of agreement (LoA) to determine whether the two methods are interchangeable (Altman and Bland, 1983).

The Shapiro–Wilk test indicated that the data followed a normal distribution. However, the sample size decreased to fewer than 30 trees when the data were stratified by tree species and phenological condition. Therefore, both parametric (t-test) and non-parametric (Wilcoxon) tests were applied and reported for the species-specific and aggregated datasets. Finally, scatter plots were generated to visually assess the relationship between estimated and observed stem volumes based on the aggregated data. The R2 was calculated and included in the plots, in order to quantify the strength of the potential relationships between estimated and reference values. All statistical tests and analyses were performed at a significance level of 0.05.Where y i, y ̂ i and y i ¯ denote the observed, estimated and mean values of a given tree attribute, respectively; n is the number of observations; s is the standard deviation; and x̄ is the sample mean.

Bias = ∑ i = 1 n ( y ̂ i − y i ) n (3)

Bias % = Bias y i ¯ x 100 % (4)

SE = s n (5)

RSE % = SE x ¯ x 100 % (6)

RMSE = ∑ i = 1 n ( y i − y ̂ i ) 2 n (7)

RMSE % = RMSE y i ¯ x 100 % (8)

3 Results

3.1 Descriptive analysis of structural attributes

Summary statistics for dbh, tree height, and stem volume estimated using different methods are presented in Table 2 by sample plot. While mean tree heights were similar across plots, mean dbh and mean stem volume were higher in Plot #1 than in Plot #2. In Plot #1, mean stem volume was estimated at 0.183 m3 and 0.181 m3 under leaf-off and leaf-on conditions, respectively. Using felled tree measurements as a reference, LiDAR-derived stem volume estimates were generally consistent, with minor deviations. In Plot #2, mean stem volume was estimated at 0.124 m3 and 0.119 m3 under leaf-off and leaf-on conditions, respectively, indicating slightly larger overestimations relative to Plot #1.

Table 2. Summary statistics of tree attributes measured using different methods.

MethodDescriptive statisticsPlot #1 (Betula pendula)Plot #2 (Populus tremula)
dbh (cm)H (m)V (m3)dbh (cm)H (m)V (m3)
Laser scanning (leaf-off condition)Min.10.4111.290.0538.999.820.034
Mean18.5114.170.18313.6614.640.124
Max.25.2517.230.33119.9720.480.301
SD4.601.530.0812.843.090.076
Laser scanning (leaf-on condition)Min.10.5010.040.0157.807.540.016
Mean19.2913.280.18114.1914.100.119
Max.25.9017.110.35220.9021.530.308
SD4.202.260.0993.753.040.077
Traditional measurements of standing trees (non-destructive sampling)Min.9.5511.300.0407.3510.400.020
Mean18.0114.790.15813.4515.080.096
Max.25.7518.500.35020.9520.300.270
SD4.691.830.0853.463.070.064
Reference measurements of felled trees (destructive sampling)Min.10.3511.950.0428.7010.600.031
Mean18.4215.070.18013.9614.950.117
Max.26.3518.360.40519.8518.400.238
SD4.831.640.0943.042.810.061

In contrast, mean tree height was underestimated in both plots regardless of phenological conditions (Table 2). This underestimation was more pronounced in Plot #1, potentially due to differences in stand density and/or species-specific crown structure. Indeed, several tree tops were occluded in the point clouds for both Betula and Populus plots, particularly under leaf-on conditions (Figure 4). According to the reference measurements, mean and dominant heights were similar across plots, at approximately 15.0 m and 18.4 m, respectively.

Point clouds of *Populus* and *Betula* trees, colored by height above ground level. *Populus* tree #2 under leaf-on **(a)** and leaf-off **(b)** conditions, *Betula* tree #2 under leaf-on **(c)** and leaf-off **(d)** conditions. Note that missing crown points are more pronounced for both species under leaf-on conditions.
Figure 4. Point clouds of Populus and Betula trees, colored by height above ground level. Populus tree #2 under leaf-on (a) and leaf-off (b) conditions, Betula tree #2 under leaf-on (c) and leaf-off (d) conditions. Note that missing crown points are more pronounced for both species under leaf-on conditions.

The mean dbh in Plot #1 was slightly overestimated with the backpack laser scanning method under both phenological conditions; however, leaf-off estimates were closer to reference values (Table 2). In Plot #2, the mean dbh was overestimated under leaf-on conditions but underestimated under leaf-off conditions. Reference measurements indicated mean and maximum dbh values of 18.42 cm and 26.35 cm in Plot #1, respectively, confirming larger tree dimensions compared to Plot #2.

3.2 Accuracy assessment

At the plot level, the most accurate stem volume estimates were obtained for Plot #1 under leaf-off conditions, with an RMSE of 0.03 m3 (15.48%) and negligible bias (Table 3). This was followed by Plot #2 under leaf-off conditions, which showed a comparable RMSE but a positive bias of 4.07%. In contrast, leaf-on estimates were less accurate in both plots, with RMSE values increasing to 0.05 m3 (27.72%) for Plot #1 and 0.06 m3 (54.76%) for Plot #2.

Table 3. Accuracy assessment of tree attributes measured using different methods, with precise field measurements of felled trees used as the reference.

Estimation methodError metricsPlot #1 (Betula pendula)Plot #2 (Populus tremula)
dbh (cm)H (m)V (m3)dbh (cm)H (m)V (m3)
Laser scanning (leaf-off condition)Bias0.17 (0.94%)−0.80 (−5.31%)0.00 (0.58%)−0.29 (−2.04%)−0.65 (−4.34%)0.00 (4.07%)
SE1.86 (10.11%)0.47 (3.14%)0.03 (15.47%)0.92 (6.61%)0.64 (4.28%)0.02 (15.24%)
RMSE1.87 (10.15%)0.93 (6.17%)0.03 (15.48%)0.97 (6.92%)0.91 (6.10%)0.02 (15.77%)
Laser scanning (leaf-on condition)Bias0.83 (4.52%)−1.53 (−10.16%)−0.01 (−3.05%)−0.13 (−0.96%)−1.06 (−7.12%)0.00 (0.94%)
SE2.82 (15.33%)1.16 (7.73%)0.05 (27.55%)2.69 (19.27%)1.58 (10.58%)0.06 (54.75%)
RMSE2.94 (15.98%)1.92 (12.77%)0.05 (27.72%)2.69 (19.30%)1.91 (12.75%)0.06 (54.76%)
Traditional measurement of standing trees*Bias−0.38 (−2.05%)−0.16 (−1.04%)−0.02 (−11.34%)−0.40 (−2.84%)−0.15 (−1.03%)−0.03 (−22.62%)
SE1.16 (6.28%)0.32 (2.15%)0.01 (7.50%)0.23 (1.66%)0.28 (1.87%)0.01 (12.63%)
RMSE1.22 (6.61%)0.36 (2.39%)0.02 (13.59%)0.46 (3.29%)0.32 (2.14%)0.03 (25.90%)

*In this method, stem volume is estimated using national allometric models based on field-measured dbh and tree height (i.e., non-destructive sampling).

Values in parentheses indicate the percentage of the corresponding error metric.

In addition to LiDAR-based estimates, stem volumes derived from national allometric models using field-measured dbh and tree height of standing trees showed limited accuracy. The models (i.e., volume functions) yielded RMSEs of 13.59 and 25.90% for Plot #1 and Plot #2, respectively. Bias reached −0.03 m3 (−22.62%) in the Populus plot, while underestimation in the Betula plot remained above −0.02 m3 (−11.34%) (Table 3).

At the individual-tree level, stem volume estimates using the backpack laser scanning method under leaf-off conditions showed a strong agreement with reference values (R2 = 0.92). This relationship weakened under leaf-on conditions (R2 = 0.62). Estimated versus observed stem volumes for both conditions are plotted in Figure 5. A small number of Populus trees contributed disproportionately to the reduced explanatory power, particularly under leaf-on conditions (Figure 5b). Despite reduced performance in leaf-on conditions, statistical tests indicated no significant differences between LiDAR-derived and reference stem volumes (p > 0.05) (Table 4). In contrast, allometric model estimates differed significantly from reference values at the individual tree level (p < 0.05). This was confirmed by both parametric and non-parametric statistical tests at the species level and for aggregated data (Table 4).

Scatter plots of estimated versus reference stem volumes at the individual tree level. Estimates were derived from LiDAR point clouds under leaf-off **(a)** and leaf-on **(b)** conditions. Data points corresponding to *Populus* trees, which were primarily responsible for the lower *R*<sup>2</sup> under leaf-on conditions, are highlighted in black. The solid line represents the best-fit linear regression between the estimated and reference values, whereas the dashed line represents the 1:1 relationship.
Figure 5. Scatter plots of estimated versus reference stem volumes at the individual tree level. Estimates were derived from LiDAR point clouds under leaf-off (a) and leaf-on (b) conditions. Data points corresponding to Populus trees, which were primarily responsible for the lower R2 under leaf-on conditions, are highlighted in black. The solid line represents the best-fit linear regression between the estimated and reference values, whereas the dashed line represents the 1:1 relationship.

Table 4. p-values from statistical tests evaluating differences between estimated and reference stem volumes.

Estimation methodPlot #1 (Betula pendula)Plot #2 (Populus tremula)All trees (plots aggregated)
Shapiro–Wilkt-testWilcoxonShapiro–Wilkt-testWilcoxonShapiro–Wilkt-testWilcoxon
Laser scanning (leaf-off condition)0.5480.8690.8520.7390.3180.3940.7160.5180.755
Laser scanning (leaf-on condition)0.5140.6260.8520.9330.9480.7760.7100.7770.999
Traditional measurement of standing trees*0.0750.0010.0010.0920.0010.0010.0080.0010.001

*In this method, stem volume is estimated using national allometric models based on field-measured dbh and tree height (i.e., non-destructive sampling).

Bland–Altman plots also indicated reasonable agreements between LiDAR-derived and reference stem volumes at the individual-tree level, as most data points clustered around the centerline of the plots (Figure 6). Under leaf-off conditions, the mean difference between the LiDAR estimates and the reference was close to zero, with LoA ranging from −0.0495 m3 to 0.0442 m3 (Figure 6a). These limits define the range within which the differences between the methods (i.e., backpack laser scanning and destructive sampling) are expected to fall for 95% of measurements. Under leaf-on conditions, the mean difference was again close to zero, but the LoA were wider, ranging from −0.108 m3 to 0.107 m3 (Figure 6b). Overall, both plots showed some degree of bias and variability in stem volume estimation, but the agreement between the LiDAR-derived and reference values was stronger under leaf-off conditions than under leaf-on conditions.

Bland–Altman plots of estimated *versus* reference stem volumes at the individual tree level. Estimates were derived from LiDAR point clouds under leaf-off **(a)** and leaf-on **(b)** conditions. LoA refers to the limits of agreement, which define the range within which the differences between the methods are expected to fall for 95% of measurements.
Figure 6. Bland–Altman plots of estimated versus reference stem volumes at the individual tree level. Estimates were derived from LiDAR point clouds under leaf-off (a) and leaf-on (b) conditions. LoA refers to the limits of agreement, which define the range within which the differences between the methods are expected to fall for 95% of measurements.

dbh estimation under leaf-off conditions yielded RMSE values of 1.87 cm (10.15%) and 0.97 cm (6.92%) for Plots #1 and #2, respectively (Table 3). Accuracy declined under leaf-on conditions in both plots, with the most pronounced deterioration observed in Plot #2, where the RMSE increased by 12.38%. The dbh was consistently underestimated in all cases, indicating a slight but systematic negative bias in the backpack laser scanning measurements.

A similar pattern was observed for tree height estimation. Under leaf-off conditions, the RMSE values ranged from 0.93 m (6.17%) to 0.97 m (6.92%). They increased substantially under leaf-on conditions, reaching 1.92 m (12.77%) and 1.91 m (12.75%) for Plot #1 and #2, respectively. Tree height was consistently underestimated across both plots and phenological conditions, with the largest bias (−1.53 m, −10.16%) observed in Plot #1 under leaf-on conditions (Table 3). This increased underestimation is attributed to missing upper canopy returns.

3.3 Point cloud quality

Each tree was represented by tens of thousands of points in the point-cloud dataset (Table 5). Point counts at least doubled under leaf-on conditions across both species; however, this increase was primarily driven by noise from leaves, shoots, flowers, and surrounding vegetation. Consequently, despite the higher point density, overall data quality seemed lower in leaf-on point clouds (Figures 7, 8).

Table 5. Summary statistics of per-tree point clouds and fitted stem cylinders under leaf-on and leaf-off conditions.

SeasonDescriptive statisticsPlot #1 (Betula pendula)Plot #2 (Populus tremula)
Total # of points*# of tree points# of stem points# of points per m3 of wood**Sum of fitted cyl. H (m) ***Stem captured (%)Total # of points*# of tree points# of stem points# of points per m3 of wood**Sum of fitted cyl. H (m) ***Stem captured (%)
Leaf-offMin.20,24211,2014,71126,8855.0033.4726,06619,7696,52667,9857.0057.12
Mean323,174144,77412,707132,2228.0857.1791,07457,51412,323373,08910.2670.80
Max.957,252335,32123,906406,43612.0074.84202,884165,30422,4911,541,82815.0095.35
SD***261,725100,6226,06879,6292.0212.9844,88831,1624,189396,7682.3411.90
Leaf-onMin.16,7598,0364,15730,7175.0045.4363,30855,78910,187115,9275.0041.48
Mean901,925325,57344,219532,7797.9460.39830,151206,00562,5911,703,0018.9864.61
Max.3,149,822848,839103,5732,890,18611.0077.792,692,274459,135187,1276,627,57413.0080.88
SD***889,368246,87934,014653,0281.479.76664,784103,02840,9071,654,6091.9110.65

*The total # of points includes all tree points, as well as ground and understory vegetation points, located within the crown-projected area of each individual tree. See Figure 3a for visual representation.

**The number of stem points per cubic meter of stem wood.

***SD, standard deviation; H, height.

Characterization of *Betula* tree #10 under leaf-off conditions. Full point cloud **(a)**, segmented tree point cloud **(b)**, and stem point cloud **(c)**. The stem point density was 182,899 points per m<sup>3</sup> of stem wood for this sample.
Figure 7. Characterization of Betula tree #10 under leaf-off conditions. Full point cloud (a), segmented tree point cloud (b), and stem point cloud (c). The stem point density was 182,899 points per m3 of stem wood for this sample.
Characterization of *Populus* tree #10 under leaf-on conditions. Full point cloud **(a)**, segmented tree point cloud **(b)**, and stem point cloud **(c)**. The stem point density was 954,915 points per m<sup>3</sup> of stem wood for this sample. Noisy returns are visible in (a, b), commonly observed in small *Populus* trees under leaf-on conditions.
Figure 8. Characterization of Populus tree #10 under leaf-on conditions. Full point cloud (a), segmented tree point cloud (b), and stem point cloud (c). The stem point density was 954,915 points per m3 of stem wood for this sample. Noisy returns are visible in (a, b), commonly observed in small Populus trees under leaf-on conditions.

After cleaning branches, crowns, and other non-stem elements, the number of points decreased substantially. For example, a Betula tree was represented by an average of 144,774 points under leaf-off conditions, of which only 12,707 corresponded to the stem (Table 5). This density was generally sufficient for visual assessment and stem volume estimation. However, upper stem sections above approximately 9 m were not adequately captured (Figure 7c), hindering cylinder fitting and likely contributing to stem volume underestimation.

Under leaf-on conditions, the mean cumulative height of fitted cylinders was 7.94 m and 8.98 m for the Betula and Populus plots, respectively. These values increased to 8.08 m and 10.26 m under leaf-off conditions. Overall, this corresponds to capturing approximately 60 and 70% of the total stem length for typical Betula and Populus trees. Further details on cylinder fitting are provided in Table 5.

To account for differences in stem size, point counts were normalized by stem volume for each tree (Table 5). The average number of points per cubic meter (of stem wood) was 132,222 for Betula and 373,089 for Populus plots under leaf-off conditions. These values increased to 532,779 and 1,703,001 under leaf-on conditions, respectively. Although point density was higher in the Populus plot, no substantial differences were observed in the visual characterization of tree structure between species. Higher values in Populus trees were mainly attributed to noise effects (see scattered returns along stems in Figures 8A,B), partly associated with smaller tree size, twigs, and branch bifurcations.

Underrepresentation of upper stem sections was also evident in Populus trees, where these parts were largely occluded by crown elements. As shown in Figure 8, stem sections above 5.5 m were not detected in a 7 m tree. Indeed, a general decrease in stem point density with height was observed across all samples (Figures 7c, 8c).

Despite these issues, stem taper curves were derived to some extent. Figure 9 illustrates stem profiles obtained under both leaf-off and leaf-on conditions using diameter estimates at multiple heights. In general, diameters decreased toward the treetop, consistent with expected tapering behavior (Figure 9a). An exception occurred at the fifth height interval under leaf-off conditions, where a local diameter overestimation was observed. Visual inspection indicated that whorls and large branches at this height caused this anomaly (see Figure 9b). Interestingly, these whorls’ non-stem elements did not produce a similarly pronounced anomaly in the leaf-on dataset for the same tree, likely due to increased noise and/or reduced geometric fidelity in mobile laser scanning compared with static terrestrial scanning.

Reference stem curves for *Betula* #14 derived from leaf-off and leaf-on point clouds **(a)** and an example of diameter overestimation at 5 m above ground level, potentially caused by whorls and large branches at that height **(b)**. In **(b)**, 1-m tree sections are shown in purple and light blue. Fitted cylinders along the stem are shown in brown, while the cylinder representing the stem section between 4 and 5 m is highlighted in gray. The horizontal lines at the base and top of the highlighted cylinder denote the lower and upper limits of the circle bands, whose thickness was set according to the thresholds in Table 1.
Figure 9. Reference stem curves for Betula #14 derived from leaf-off and leaf-on point clouds (a) and an example of diameter overestimation at 5 m above ground level, potentially caused by whorls and large branches at that height (b). In (b), 1-m tree sections are shown in purple and light blue. Fitted cylinders along the stem are shown in brown, while the cylinder representing the stem section between 4 and 5 m is highlighted in gray. The horizontal lines at the base and top of the highlighted cylinder denote the lower and upper limits of the circle bands, whose thickness was set according to the thresholds in Table 1.

4 Discussion

4.1 Stem volume

The largest discrepancy in stem volume estimates between leaf-off and leaf-on conditions was observed in Populus trees. Estimation errors increased markedly under leaf-on conditions, whereas the corresponding increase in the Betula plot was relatively limited. Several factors likely contributed to this pronounced decline in accuracy. First, the smaller stem dimensions of Populus trees compared to Betula likely reduced detectability by the backpack laser scanner. This aligns with previous findings indicating that mobile LiDAR systems perform better in mature stands, while thinner stems are often underrepresented due to limited return density (Ryding et al., 2015; Vatandaşlar et al., 2023). Second, species-specific leaf morphology may have exacerbated noise levels. The flattened petiole structure of Populus leaves causes continuous movement even under low wind conditions (von Wühlisch, 2009)—a trait reflected in the species name Populus tremula. This movement can introduce SLAM-related artifacts (“ghost objects”), as the same elements are recorded at different spatial positions during scanning (Liang et al., 2022). Third, the Populus plot contained denser understory vegetation than the Betula plot. The combined effects of canopy foliage, understory plants, and herbaceous layers—along with the tree canopy itself—likely increased occlusion and noise under leaf-on conditions. Understory vegetation is a well-documented source of error in close-range forest LiDAR applications (Liang et al., 2022; Výbošťok et al., 2026).

Despite the reduced accuracy under leaf-on conditions—particularly in the Populus plot—the statistical tests revealed no significant differences in stem volume estimates across seasons or species. This suggests that backpack laser scanning remains a viable data acquisition method for estimating over-bark stem volume in young- to mid-aged Betula and Populus stands with low to moderate understory, and that acquisition timing may be less critical when high precision is not required. This finding is mostly consistent with previous studies. For example, Tupinambá-Simões et al. (2024) reported no significant differences between leaf-on and leaf-off volume estimates using a Zeb-Horizon handheld scanner in mixed-species stands. However, they observed significant discrepancies in a second leaf-on scan (Scan02on), despite the implementation of increased trajectory complexity to reduce occlusion. This observation is interesting, and together with the results presented in Table 5, suggests that increased point cloud density does not necessarily translate into improved data quality or higher estimation accuracy. In the present study, point density at least doubled under leaf-on conditions, yet estimation accuracy generally declined.

Tupinambá-Simões et al. (2024) also reported wide RMSE variations across species, ranging from 20.1% (P. pinaster) to 84.4% (Q. pyrenaica). In contrast, the present study found a narrower RMSE range, from 15.48% (B. pendula) to 54.76% (P. tremula). The comparatively lower error margins observed here may be attributed to differences in both reference data and LiDAR platforms. In their study, reference values were derived from allometric models applied to field data from standing trees, potentially introducing additional uncertainty. In contrast, the present study employed destructive sampling, with stem volumes measured directly from felled trees. Such direct methods are generally considered more accurate and precise (Chave et al., 2014; Schettini et al., 2022), likely reducing discrepancies between LiDAR-derived and reference values.

Aside from LiDAR-based estimates, this study showed that stem volumes derived from national allometric models (Petráš and Pajtík, 1991) differed significantly from the reference measurements. This finding has important implications for forest management, as these models are currently used in Slovakia to assess wood availability, update forest plans, and report biomass and carbon stocks. Although uncertainties of up to 25% per hectare may be considered acceptable in some contexts (Mäkelä and Pekkarinen, 2004; Qiu et al., 2018), the systematic bias observed in this study suggests that these models may no longer be adequate. Replacing outdated allometries with models developed using modern measurement technologies and advanced modeling approaches could improve estimation accuracy (Yang et al., 2023; Ercanli, 2025). Alternatively, developing eco-region-specific models may provide further improvements (Seki and Sakici, 2022). Given the widespread reliance on non-destructive measurements and allometric equations globally (Vidal et al., 2016; Pitkänen et al., 2021), further research in this area is warranted.

4.2 Dbh, tree height, and stem curve

Similarly to stem volume, dbh and tree height were estimated more accurately under leaf-off conditions, with the largest discrepancies observed for dbh. In the Populus plot, the RMSE nearly tripled when the same trees were scanned under leaf-on conditions. In contrast, RMSE values in the Betula plot were more stable. The error rates observed in the present study are comparable to those reported in recent close-range mobile LiDAR studies. Using a similar system, Ko et al. (2024) reported RMSEs ranging from 4.57% to 22.71% under leaf-off conditions and from 10.71% to 43.23% under leaf-on conditions across six coniferous plots in Korea. The reduced performance under leaf-on conditions was attributed to increased foliage and understory density. In another study, Kükenbrink et al. (2025) used a handheld laser scanner across 25 Swiss National Forest Inventory plots and reported an RMSE of 2.8 cm (7.7%) for dbh estimates of correctly matched trees. Although their study did not compare leaf-on and leaf-off conditions, it highlighted the influence of factors such as occlusion and terrain slope on estimation accuracy.

As for tree height, the results were comparatively stable, with no substantial differences between plots. Sub-meter RMSEs were observed under leaf-off conditions, but these errors nearly doubled under leaf-on conditions. This finding supports the recommendation of Holvoet et al. (2025) to conduct scanning during leaf-off periods to improve height estimation accuracy. Despite the relatively low mean stand heights, tree height was consistently underestimated across plots and seasons in the present study. Such underestimation is widely reported in mobile LiDAR studies (Holvoet et al., 2025), and its magnitude may increase in mature or old-growth forests, where trees can be substantially taller. Consistent with this expectation, Chiappini et al. (2022) documented decreasing accuracy with increasing tree height, which aligns with the pattern observed in this study. While recent studies have noted that smartphone apps can also provide highly accurate height estimates based on geometric principles and image processing (Mahmud et al., 2025; Zeybek, 2025), these low-cost methods seem unlikely to meet the requirements of operational forest inventories because they lack the mapping features needed to match individual trees. Moreover, these studies often analyze trees outside of forests typically in urban areas which have substantially different conditions from those found in forests. For example, smartphone GNSSs often hinder the attainment of sub-meter level positional accuracy under forest conditions unless an external antenna is attached (Merry et al., 2026).

A growing body of research suggests that traditional field-based height measurements from standing trees (e.g., using hypsometers, clinometers, or sliding poles) may be less reliable than commonly assumed (Sterenczak et al., 2019; Wang et al., 2019; Jurjevic et al., 2020). This raises concerns regarding the reliability of reference data, as the ground-truth used for validation may itself contain systematic errors. For example, Sterenczak et al. (2019) reported biases of up to 5 m when comparing traditional measurements with precise measurements from felled trees. Consequently, point cloud-based height estimates may, in some cases, be more reliable than indirect field measurement methods. This may also explain why some studies use static terrestrial laser scanning (TLS) data as a reference for evaluating mobile LiDAR outputs (e.g., Hyyppä et al., 2020b). The present results partially support this perspective; in particular, although traditional measurements from standing trees exhibited systematic bias and deviated from reference values (i.e., direct height measurements from felled trees), they were still closer to the reference (RMSEs < 3%) than LiDAR-derived estimates (best RMSE = 6.1%; Table 3).

Finally, this study explored the derivation of stem curves using automatically fitted cylinders along stem points from backpack laser scanning data. This attribute has received limited attention in the forest LiDAR literature (Hyyppä et al., 2020b; Pitkänen et al., 2021; Nurunnabi et al., 2024), with most previous work relying on static TLS data due to their higher point density and geometric precision. In our case, several anomalies inconsistent with biological expectations were observed (see Figure 9). Although similar issues have been reported in TLS-based studies (e.g., Nurunnabi et al., 2024), their higher-quality data and structurally simpler tree forms enabled stem curve reconstruction up to 11–32 m in tall conifers depending on the algorithm. In contrast, stem reconstruction in the present study was limited to approximately 12 m for trees around 16 m in height (Figure 10), with classification accuracy decreasing markedly with height. Fortunately, our algorithms (see Supplementary materials) were able to fit cylinder segments of reasonable size even under reduced classification quality, resulting in relatively accurate total stem volume estimates. Beyond a certain height, however, stem detection became unreliable in Betula and Populus, which often lack a well-defined single stem due to their growth form. Changing acquisition conditions (leaf-on versus leaf-off) did not mitigate this limitation. Overall, these findings indicate that mobile laser scanning has limited suitability for deriving stem taper functions in standing deciduous trees when full stem representation is required.

Fitted cylinder arcs along the stem of *Betula* tree #2. Diameters at the base and top of each segment were used to derive the stem taper curve.
Figure 10. Fitted cylinder arcs along the stem of Betula tree #2. Diameters at the base and top of each segment were used to derive the stem taper curve.

4.3 Limitations

While the proposed methodology enabled the estimation of stem volume, dbh, and tree height from point clouds acquired under contrasting phenological conditions, several limitations may restrict the generalizability of the results across different forest ecosystems and spatial scales. The primary limitation relates to sample size. At first glance, the number of sampled trees may appear insufficient for LiDAR-based forest inventory studies, where most research relies on non-destructive sampling (i.e., indirect measurements from standing trees). However, the present study included direct destructive sampling to obtain high-quality reference data. Measurements derived from felled trees are widely regarded as the most accurate form of field data collection (Lu et al., 2016; Kershaw et al., 2017; Schettini et al., 2022). This procedure involved felling all trees within the plots, removing branches and tops from the main stems, and sectioning the stems into 1 m segments, along with dendrometric measurements conducted on the ground. As illustrated in Figure 11, destructive sampling is labor-intensive and extremely time-consuming (Chave et al., 2014). Moreover, obtaining permission to fell selected trees is often challenging and may limit sampling opportunities for researchers. Consequently, studies employing such approaches, or similarly intensive measurement protocols such as whorl detection (Hartley et al., 2022), typically rely on relatively small datasets. Nevertheless, when measurements are conducted rigorously, reliable estimates can still be achieved with sample sizes as small as 28 trees (Calders et al., 2015).

Selected stages of the destructive sampling procedure, illustrating its time-intensive nature even for a single tree. A felled *Betula* tree with all branches removed **(a)**, and sectioning of the stem into 1-m segments for detailed volumetric measurements **(b)**. Figure contains images of the author(s) only.
Figure 11. Selected stages of the destructive sampling procedure, illustrating its time-intensive nature even for a single tree. A felled Betula tree with all branches removed (a), and sectioning of the stem into 1-m segments for detailed volumetric measurements (b). Figure contains images of the author(s) only.

The second limitation concerns user intervention during point cloud analysis, as forest inventory teams typically desire fully automated workflows. Specifically, visual inspection and noise removal were performed manually by the data analyst during the stem construction stage. Several software were used for this purpose, which may hinder the development of a faster and more streamlined data analysis pipeline. Although we did not conduct a detailed time-efficiency analysis comparing the traditional and backpack laser scanning methods, the complete stem construction and volume computation process took approximately 6 min per tree. While the algorithmic computations were relatively fast and the total processing time depended on the point density per tree (see Table 5), most of the time was spent on manual inspection, data editing, and output management. These requirements may limit the operational adoption of the method for large-scale inventory applications, which typically involve hundreds or thousands of sample plots containing dozens of trees each. Perhaps more importantly, this aspect of the workflow may introduce user bias and consequently limit the repeatability and consistency of the data analysis across different users.

Finally, the occlusion effect limited the ability to fit cylinders along the entire stem. Although most trees were scanned from multiple directions, 1 m cylinder fitting could not be consistently applied in upper stem sections due to occlusion by non-stem elements and reduced point density. This limitation was more pronounced in Populus but was observed in both species, even under leaf-off conditions, and likely contributed to a slight negative bias in stem volume estimates. The relatively low overall bias and RMSE values can be explained by the vertical distribution of stem volume, which is concentrated in the lower portions of the stem. For example, Hyyppä et al. (2020b) reported that more than 80% of the total stem volume is contained within the first 13 m for stems with an average height of about 20 m. Similarly, Hartley et al. (2022) found that 71.9% of the total volume is located within the first 10 m for stems averaging 23.9 m in height. Overall, terrestrial-based static or mobile LiDAR systems seem to be more effective at capturing lower stem sections with high precision than upper portions, which largely explains the relatively strong performance observed in stem volume estimation despite the aforementioned limitations.

5 Conclusion

  • 1 Stem volume, dbh, and tree height can be estimated directly from backpack laser scanning point clouds. The estimation accuracy is generally acceptable for forest inventory applications, with the exception of the Populus plot under leaf-on conditions. Nevertheless, the semi-automated workflow for point cloud analysis may limit its operational adoption, particularly for large-scale inventory applications.
  • 2 No significant differences were observed between stem volume estimates derived from leaf-off and leaf-on point clouds and the reference values (p > 0.05). However, point cloud quality and estimation accuracy were consistently higher under leaf-off conditions.
  • 3 dbh and tree height were also estimated more accurately under leaf-off conditions than under leaf-on conditions.
  • 4 The Populus plot was more challenging to reconstruct and analyze than the Betula plot using backpack laser scanning data, due to (i) smaller tree dimensions, (ii) species-specific morphological characteristics, (iii) denser stand and understory conditions, and (iv) limitations in scanning trajectories.
  • 5 Stem volume estimates derived from national allometric models using measurements from standing trees differed significantly from the reference values (p < 0.05) and systematically underestimated stem volume.
  • 6 The results and assessments presented in this study are applicable to young- to middle-aged deciduous stands dominated by B. pendula and P. tremula in temperate forest zones. Therefore, caution should be exercised when generalizing these findings to other forest types and site conditions.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/ffgc.2026.1886496/full#supplementary-material

Video recordings of the sample plots captured during scanning with the LiBackpack add-on 596 spherical camera. Available at: https://doi.org/10.5281/zenodo.19767390

Supplementary material Supplementary material at the source

Footnotes

Edited by: Ram P. Sharma, Tribhuvan University, Nepal

Reviewed by: Christine Swanson, University of Florida, United States

Mubarak Mahmud, National Research Institute of Science and Technology for Environment and Agriculture (IRSTEA), France

Statements

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work was primarily supported by the project “Upgrade of the Centre of Excellence -LignoSilva” (Grant agreement ID: 101059552), funded by the Horizon Europe Research & Innovation Programme. The work was further supported by the projects “High-tech National Forest Inventory based on Mobile Laser Scanning: from Research to Realization (APVV-24-0064)”, funded by the Slovak Research and Development Agency and “Digital Technologies in Increasing Added Value in Wood Assortment, Sorting and Processing (DigiWood)”, funded by the Ministry of Agriculture and Rural Development of the Slovak Republic.

Data availability

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Author contributions

CV: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Visualization, Writing – original draft, Writing – review & editing. IS: Conceptualization, Formal analysis, Methodology, Project administration, Resources, Supervision, Writing – review & editing. VM: Data curation, Writing – review & editing. JP: Data curation, Writing – review & editing. MZ: Formal analysis, Investigation, Methodology, Visualization, Writing – review & editing.

Acknowledgements

The authors would like to thank the National Forest Centre in Zvolen (Slovakia) for providing complementary resources.

References

  1. Altman D. G., Bland J. M. (1983). Measurement in medicine: the analysis of method comparison studies. J. R. Stat. Soc. Ser. D Stat. 32, 307 – 317. doi: doi:10.2307/2987937
  2. Axelsson P. (2000). DEM generation from laser scanner data using adaptive TIN models International Archives of Photogrammetry and Remote Sensing 33 (B4/1). Available online at: (Accessed April 7, 2026).
  3. Bartholomeus H., Sarmiento A. L., Brede B., van Hoeve K., Koelewijn N. (2021). Comparing portable MLS to TLS and UAV-LS derived individual tree parameters. In: Proceedings of the SilviLaser Conference, 28–30 September 2021, Vienna, Austria, eds. Hollaus M., Pfeifer N.. Vienna, Austria: Technische Universität Wien. pp. 142 – 144. doi: doi:10.34726/wim.1961
  4. Bobrowski R., Winczek M., Zięba-Kulawik K., Wężyk P. (2023). Best practices to use the iPad pro LiDAR for some procedures of data acquisition in the urban forest. Urban For. Urban Green. 79: 127815. doi: doi:10.1016/j.ufug.2022.127815
  5. Bodziarczyk J., Szwagrzyk J., Zwijacz-Kozica T., Zięba A., Szewczyk J., Gazda A. (2020). The structure of forest stands in the Tatra National Park: the results of 2016–2017 inventory. For. Res. Pap. 80, 13 – 21. doi: doi:10.2478/frp-2019-0002
  6. Cabo C., Del Pozo S., Rodríguez-Gonzálvez P., Ordóñez C., González-Aguilera D. (2018). Comparing terrestrial laser scanning (TLS) and wearable laser scanning (WLS) for individual tree modeling at plot level. Remote Sens. 10: 540. doi: doi:10.3390/rs10040540
  7. Calders K., Newnham G., Burt A., Murphy S., Raumonen P., Herold M.. (2015). Nondestructive estimates of above-ground biomass using terrestrial laser scanning. Methods Ecol. Evol. 6, 198 – 208. doi: doi:10.1111/2041-210X.12301
  8. Čeňava J., Tuček J., Chudá J., Koreň M. (2024). Mobile laser scanning data collected under a forest canopy with GNSS/INS-positioned systems: possibilities of processability improvements. Remote Sens. 16: 1734. doi: doi:10.3390/rs16101734
  9. Chave J., Réjou-Méchain M., Búrquez A., Chidumayo E., Colgan M. S., Delitti W. B. C.. (2014). Improved allometric models to estimate the aboveground biomass of tropical trees. Glob. Change Biol. 20, 3177 – 3190. doi:, doi:10.1111/gcb.12629
  10. Chen S., Liu H., Feng Z., Shen C., Chen P. (2019). Applicability of personal laser scanning in forestry inventory. PLoS One 14: e0211392. doi:, doi:10.1371/journal.pone.0211392
  11. Chiappini S., Pierdicca R., Malandra F., Tonelli E., Malinverni E. S., Urbinati C.. (2022). Comparing mobile laser scanner and manual measurements for dendrometric variables estimation in a black pine (Pinus nigra Arn.) plantation. Comput. Electron. Agric. 198: 107069. doi: doi:10.1016/j.compag.2022.107069
  12. Davison S., Donoghue D., Galiatsatos N. (2020). The effect of leaf-on and leaf-off forest canopy conditions on LiDAR derived estimations of forest structural diversity. Int. J. Appl. Earth Obs. Geoinf. 92: 102160. doi: doi:10.1016/j.jag.2020.102160
  13. Ercanli I. (2025). “ An innovative technique for quantifying epistemic and aleatoric uncertainty in artificial intelligence–based modeling of forest attributes: evidential deep learning,” in International Compilation of Research and Studies in the Field of Agriculture, Forestry and Aquaculture Sciences. Chapter 13, eds. Özrenk K., Bolat A. (Ankara: Serüven Publishing), 232 – 242.
  14. Fassnacht F. E., White J. C., Wulder M. A., Næsset E. (2024). Remote sensing in forestry: current challenges, considerations and directions. Forestry 97, 11 – 37. doi: doi:10.1093/forestry/cpad024
  15. Florea S. C., Dutca I., Nita M. D. (2024). Tradeoffs and limitations in determining tree characteristics using 3D point clouds from terrestrial laser scanning: a comparison of reconstruction algorithms on European beech (Fagus sylvatica L.) trees. Ann. For. Res. 67, 185 – 199. doi: doi:10.15287/afr.2024.3943
  16. Gander W., Golub G. H., Strebel R. (1994). Least-squares fitting of circles and ellipses. BIT 34, 558 – 578. doi: doi:10.1007/BF01934268
  17. GreenValley International (2021). LiFuser-BP: Data Processing, Visualization, and Optimization Software for GVI LiBackpack. Available online at: (Accessed March 25, 2026).
  18. Hackenberg J., Spiecker H., Calders K., Disney M., Raumonen P. (2015). SimpleTree—an efficient open-source tool to build tree models from TLS clouds. Forests 6, 4245 – 4294. doi: doi:10.3390/f6114245
  19. Hartley R. J., Jayathunga S., Massam P. D., De Silva D., Estarija H. J., Davidson S. J.. (2022). Assessing the potential of backpack-mounted mobile laser scanning systems for tree phenotyping. Remote Sens. 14: 3344. doi: doi:10.3390/rs14143344
  20. Holvoet J., Eichhorn M. P., Giannetti F., Kükenbrink D., Liang X., Mokroš M.. (2025). Terrestrial and mobile laser scanning for national forest inventories: from theory to implementation. Remote Sens. Environ. 329: 114947. doi: doi:10.1016/j.rse.2025.114947
  21. Hyyppä E., Kukko A., Kaijaluoto R., White J. C., Wulder M. A., Pyörälä J.. (2020b). Accurate derivation of stem curve and volume using backpack mobile laser scanning. ISPRS J. Photogramm. Remote Sens. 161, 246 – 262. doi: doi:10.1016/j.isprsjprs.2020.01.018
  22. Hyyppä E., Yu X., Kaartinen H., Hakala T., Kukko A., Vastaranta M.. (2020a). Comparison of backpack, handheld, under-canopy UAV, and above-canopy UAV laser scanning for field reference data collection in boreal forests. Remote Sens. 12: 3327. doi: doi:10.3390/rs12203327
  23. Imangholiloo M., Saarinen N., Holopainen M., Yu X., Hyyppä J., Vastaranta M. (2020). Using leaf-off and leaf-on multispectral airborne laser scanning data to characterize seedling stands. Remote Sens. 12: 3328. doi: doi:10.3390/rs12203328
  24. Jurjevic L., Liang X., Gasparovic M., Balenovic I. (2020). Is field-measured tree height as reliable as believed–part II, a comparison study of tree height estimates from conventional field measurement and low-cost close-range remote sensing in a deciduous forest. ISPRS J. Photogramm. Remote Sens. 169, 227 – 241. doi: doi:10.1016/j.isprsjprs.2020.09.014
  25. Kershaw J. A., Ducey M. J., Beers T. W., Husch B. (2017). Forest mensuration. 5th Edn New York: Wiley Blackwell.
  26. Ko C., Kang J., Park J., Lee M. (2024). Assessing forest resources with terrestrial and backpack LiDAR: a case study on leaf-on and leaf-off conditions in Gari Mountain, Hongcheon, Republic of Korea. Forests 15: 2230. doi: doi:10.3390/f15122230
  27. Ko C., Lee S., Yim J., Kim D., Kang J. (2021). Comparison of forest inventory methods at plot-level between a backpack personal laser scanning (BPLS) and conventional equipment in Jeju Island, South Korea. Forests 12: 308. doi: doi:10.3390/f12030308
  28. Kükenbrink D., Marty M., Rehush N., Abegg M., Ginzler C. (2025). Evaluating the potential of handheld mobile laser scanning for an operational inclusion in a national forest inventory—a Swiss case study. Remote Sens. Environ. 321: 114685. doi: doi:10.1016/j.rse.2025.114685
  29. Laar A. V., Akça A. (2007). Forest Mensuration. Dordrecht: Springer.
  30. Lee T., Bettinger P., Merry K., Cieszewski C. (2023). The effects of nearby trees on the positional accuracy of GNSS receivers in a forest environment. PLoS One 18: e0283090. doi:, doi:10.1371/journal.pone.0283090
  31. Liang X., Hyyppä J., Kaartinen H., Lehtomäki M., Pyörälä J., Pfeifer N.. (2016). International benchmarking of terrestrial laser scanning approaches for forest inventories. ISPRS J. Photogramm. Remote Sens. 115, 63 – 77. doi: doi:10.1016/j.isprsjprs.2016.01.011
  32. Liang X., Kukko A., Balenović S. N. I, Junttila S., Kankare V. (2022). Close-range remote sensing of forests: the state of the art, challenges, and opportunities for systems and data acquisitions. IEEE Geosci. Remote Sens. Mag. 10, 32 – 71. doi: doi:10.1109/MGRS.2022.3168135
  33. Lu D., Chen Q., Wang G., Liu L., Li G., Moran E. (2016). A survey of remote sensing-based aboveground biomass estimation methods in forest ecosystems. Int. J. Digit. Earth 9, 63 – 105. doi: doi:10.1080/17538947.2014.990526
  34. Mahmud M., Lin J., Houballah M., Buba I. G., Barthes L. (2025). Development of mobile application for tree height measurement using geometric principle: establishing global database of tree height and data. Smart Agric. Technol. 10: 100846. doi: doi:10.1016/j.atech.2025.100846
  35. Mäkelä H., Pekkarinen A. (2004). Estimation of forest stand volumes by Landsat TM imagery and stand-level field-inventory data. For. Ecol. Manag. 196, 245 – 255. doi: doi:10.1016/j.foreco.2004.02.049
  36. Marcello J., Spínola M., Albors L., Marqués F., Rodríguez-Esparragón D., Eugenio F. (2024). Performance of individual tree segmentation algorithms in forest ecosystems using UAV LiDAR data. Drones 8: 772. doi: doi:10.3390/drones8120772
  37. Merry K., Bettinger P., Lowe R. C. (2026). Smartphone GNSS accuracy approaches mapping-grade receiver capabilities in forest conditions, and submeter accuracy may be possible using an external antenna. For. Sci. doi: doi:10.1007/s44391-026-00081-9
  38. Nurunnabi A., Teferle F., Novo A., Balado J., Ientilucci E. (2024). Derivation of tree stem curve and volume using point clouds. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.-ISPRS Arch. 48, 81 – 88. doi: doi:10.5194/ISPRS-ARCHIVES-XLVIII-4-W11-2024-81-2024
  39. Oikawa N., Nakagawa Y., Owari T., Tatsumi S., Suzuki S. N. (2025). Utilising LiDAR-equipped iPhone in forestry: constructing 3D models and measuring tree sizes in a planting site. Ecol. Solut. Evid. 6: e12399. doi: doi:10.1002/2688-8319.12399
  40. OLAF (2024). Remote sensing: LiDAR. Course Material in Online Learning in Applied Forestry (OLAF) Educational Tool. Available online at: (Accessed July 31, 2026).
  41. Öztürk A., Cabo C., Eichhorn M. P., Hollaus M., Iglseder A., Mokroš M.. (2026). Barriers and gaps in the implementation of close-range remote sensing technologies in forestry. J. For. Res. 37: 47. doi: doi:10.1007/s11676-026-01995-3
  42. Petráš R., Pajtík J. (1991). Sústava česko-slovenských objemových tabuliek drevín. Lesn. Cas. 37, 49 – 56. (in Slovak)
  43. Pitkänen T. P., Raumonen P., Liang X., Lehtomäki M., Kangas A. (2021). Improving TLS-based stem volume estimates by field measurements. Comput. Electron. Agric. 180: 105882. doi: doi:10.1016/j.compag.2020.105882
  44. Qiu Z., Feng Z., Jiang J., Lin Y., Xue S. (2018). Application of a continuous terrestrial photogrammetric measurement system for plot monitoring in the Beijing Songshan national nature reserve. Remote Sens. 10: 1080. doi: doi:10.3390/rs10071080
  45. Raumonen P., Kaasalainen M., Åkerblom M., Kaasalainen S., Kaartinen H., Vastaranta M.. (2013). Fast automatic precision tree models from terrestrial laser scanner data. Remote Sens. 5, 491 – 520. doi: doi:10.3390/rs5020491
  46. Ryding J., Williams E., Smith M. J., Eichhorn M. P. (2015). Assessing handheld mobile laser scanners for forest surveys. Remote Sens. 7, 1095 – 1111. doi: doi:10.3390/rs70101095
  47. Sačkov I., Santopuoli G., Bucha T., Lasserre B., Marchetti M. (2016). Forest inventory attribute prediction using lightweight aerial scanner data in a selected type of multilayered deciduous forest. Forests 7: 307. doi: doi:10.3390/f7120307
  48. Schettini B. L. S., Jacovine L. A. G., Torres C. M. M. E., de Carneiro A. C. O., Castro R. V. O., Villanova P. H.. (2022). Use of destructive and non-destructive methodologies to estimate stem biomass accumulation and carbon stock in an Eucalyptus forest. Rev. Arvore 46: e4611.
  49. Schnabel R., Wahl R., Klein R. (2007). Efficient RANSAC for point-cloud shape detection. Comput. Graph. Forum 26, 214 – 226. doi: doi:10.1111/j.1467-8659.2007.01016.x
  50. Seki M., Sakici O. E. (2022). Ecoregion-based height-diameter models for Crimean pine. J. For. Res. 27, 36 – 44. doi: doi:10.1080/13416979.2021.1972511
  51. Sterenczak K., Mielcarek M., Wertz B., Bronisz K., Zajączkowski G., Jagodzinski A. M.. (2019). Factors influencing the accuracy of ground-based tree-height measurements for major European tree species. J. Environ. Manag. 231, 1284 – 1292. doi: doi:10.1016/j.jenvman.2018.09.100
  52. TANAP (2026). Fauna and Flora in the official webpage of the Tatra National Park (Tatransky narodny park). Available online at: (Accessed March 23, 2026).
  53. Tupinambá-Simões F., Pascual A., Guerra-Hernández J., Ordóñez C., de Conto T., Bravo F. (2024). Accuracy of tree mapping based on hand-held laser scanning comparing leaf-on and leaf-off conditions in mixed forests. J. For. Res. 35: 93. doi: doi:10.1007/s11676-024-01747-1
  54. Vatandaşlar C., Seki M., Zeybek M. (2023). Assessing the potential of mobile laser scanning for stand-level forest inventories in near-natural forests. Forestry (Oxf) 96, 448 – 464. doi: doi:10.1093/forestry/cpad016
  55. Vidal C., Sallnas O., Redmond J., Alberdi I., Barreiro S., Hernández L.. (2016). “ Chapter 1: introduction,” in National Forest Inventories: Assessment of Wood Availability and Use, eds. Vidal C., Alberdi I., Hernandez L., Redmond J. (Cham: Springer), 1 – 25.
  56. von Wühlisch G. (2009). EUFORGEN Technical Guidelines for genetic Conservation and use of Eurasian aspen (Populus tremula). Rome: Bioversity International, 6.
  57. Výbošťok J., Chuda J., Skladan M., Tomčík D., Singh A., Mokros M. (2026). Assessing the influence of ground-based LiDAR devices, processing solutions, and forest stand characteristics on tree diameter and height estimation. Cent. Eur. For. J. 72, 3 – 26. doi: doi:10.2478/forj-2025-0018
  58. Wang Y., Lehtomäki M., Liang X., Pyörälä J., Kukko A., Jaakkola A.. (2019). Is field-measured tree height as reliable as believed—a comparison study of tree height estimates from field measurement, airborne laser scanning and terrestrial laser scanning in a boreal forest. ISPRS J. Photogramm. Remote Sens. 147, 132 – 145. doi: doi:10.1016/j.isprsjprs.2018.11.008
  59. Wu Y., Zhong S., Ma Y., Zhang Y., Liu M. (2025). Application of SLAM-based mobile laser scanning in forest inventory: methods, progress, challenges, and perspectives. Forests 16: 920. doi: doi:10.3390/f16060920
  60. Xie Y., Yang T., Wang X., Chen X., Pang S., Hu J.. (2022). Applying a portable backpack lidar to measure and locate trees in a nature forest plot: accuracy and error analyses. Remote Sens. 14: 1806. doi: doi:10.3390/rs14081806
  61. Yang S.-I., Burkhart H. E., Seki M. (2023). Evaluating semi- and nonparametric regression algorithms in quantifying stem taper and volume with alternative test data selection strategies. Forestry 96, 465 – 480. doi: doi:10.1093/forestry/cpad019
  62. Zeybek M. (2025). Measurement of tree height and DBH in urban areas using RTK-enabled smartphones. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. XLVIII-5/W3-2025, 155 – 162. doi: doi:10.5194/isprs-archives-XLVIII-5-W3-2025-155-2025
  63. Zhang W., Qi J., Wan P., Wang H., Xie D., Wang X.. (2016). An easy-to-use airborne LiDAR filtering method based on cloth simulation. Remote Sens. 8: 501. doi: doi:10.3390/rs8060501

Citation

Vatandaslar, C., et al. (2026). Estimation of stem volume and tree structural attributes using backpack LiDAR: effects of leaf-on and leaf-off conditions. Frontiers in Forests and Global Change, Volume 9. https://doi.org/
Received
Accepted
Published
Volume
9 - 2026

Copyright

Copyright © 2026 Vatandaslar, Sačkov, Murgaš, Pajtík and Zeybek.

License: https://creativecommons.org/licenses/by/4.0/

*Correspondence: Can Vatandaslar

Article metrics

Views
0
Downloads
0
Citations
0
Shares
0

Related articles

Systematic ReviewEnvironmental, economic, and sociocultural outcomes of Community Resource Management Area in Ghana: a systematic reviewSamuel Adeyanju, Cornelius K. A. Pienaah, Janette Bulkan, Mirjam A. F. Ros-Tonen, Terry SunderlandOriginal ResearchMachine learning-based modeling of soil respiration components across four forest types in a subtropical mountainous forest with implications for modeling strategies and hydrothermal controlsNan Deng, Yaqin Xiao, Fengfeng Ma, Yuxin TianOriginal ResearchSoil metabarcoding reveals the distinct microbial signature of a remnant forest patch at a local restoration initiative in the Ecuadorian ChocóAlex Santillán-Sarmiento, Julio C. Carrión-Olmedo, Joseline Salinas-Tapia, Pamela Lojan-Cueva, Diana Flores, Daniela Reyes-Barriga, Liliana Jaramillo, Andrea E. Narváez, Pablo Jarrín-V