A Comparative Evaluation of Threshold Segmentation and LiDAR for Sawmill Residue Volume Estimation

Abstract

The sawn timber production process generates up to 63% of residues during primary processing in sawmills. For this industry, the devaluation and disposal of these residues remain significant challenges; proper management requires a more accurate quantification of the volume. This study evaluates and compares two indirect methods for estimating the volume of stacked residues: one based on image processing and the other on terrestrial LiDAR technology. Residues of Pinus spp. from a sawmill were used, with their actual volume determined using a xylometer. The image-based method, which uses threshold-based segmentation, achieved a R² = 0.64 and RMSE = 0.006 m³. In contrast, the LiDAR-based method, which derives measurements directly from 3D reconstruction, obtained an R² = 0.506 and RMSE = 0.009 m³. Despite these differences, ANOVA testing (p > 0.05) indicated no statistically significant differences between the methods. The results suggest that both approaches may serve as preliminary tools for forest residue quantification and provide a solid foundation for future research aimed at developing field-applicable technological solutions.

1. Introduction

The world has 4.1 billion of hectares of the land area dedicated to forest, representing 31% of the worldwide area, equivalent to 0.52 ha·person⁻¹ [1]. More than half of this area is concentrated in just five countries: Russia, Brazil, Canada, the United States, and China [2]. In Mexico, forested land accounts for 66.6 million hectares [3], positioning it as the country with the 12th highest land area. In the country, the states of Chihuahua, Sonora, Coahuila, Durango, and Baja California Sur account for 49% of the national area altogether [4]. Particularly, the state of Durango encompasses 5.8 million hectares of forested area [5]. These forests host rich biodiversity, with three dominant genera: Pinus, with 22 species; Quercus, with 44 species; and Arbutus, with 7 species [6]. The structure of these forests is heterogeneous, both in terms of the spatial distribution of trees (vertical and horizontal irregularity) and in the variation of tree and stand ages [7]. The importance of these ecosystems extends beyond their environmental dimension; they also support a forest industry that is vital for economic development. Despite facing challenges such as illegal logging and the need for sustainable management, this activity remains a strategic pillar of local economies. In this context, the wood production chain relies heavily on sawmills, which operate under various forest management schemes aimed at improving efficiency and minimizing environmental impact.

The sawmilling agroindustry, present throughout Mexico, has a strong concentration in states with extensive forest cover, such as the state of Durango. At the national level, sawn wood production reaches 3.362 million cubic meters, representing 0.7% of global output, and it serves as a vital source of income for rural and indigenous communities that depend on forest management. However, despite accounting for more than two-thirds of total timber production, the sector has shown a slightly negative trend in both output and economic value, indicating a deterioration that requires urgent attention [8].

The sawn timber process generates over 63% residues, with 10 to 15% produced at logging sites and 40 to 50% during primary processing. Sawmill residues can account for nearly 55% of the log feedstock volume, depending on the log diameter and sawing pattern [9]. In Mexico, timber harvesting and processing operations, such as felling, transportation, and sawing, generate significant volumes of residues that often hold minimal commercial value [10,11].

Sawmill residues that are not collected or utilized can generate several problems, such as social, economic, and environmental ones [12,13]. For instance, they are a potential source of fire generation, and their decomposition produces odors that affect the population as well as greenhouse gases that exacerbate global warming [14,15]. Therefore, one of the ways to utilize these residues is in the power industry, where they account for the largest portion of solid biomass used in combined heat and power plants and heat-generating facilities, primarily in the form of chips. This results in an increasing global demand for this material, highlighting the importance of understanding the characteristics and variability of sawmill biofuels [9], and requires a detailed knowledge of the volumes and qualities of these residues [11,16].

The estimation of solid wood volume from stacked wood is a common practice across global forestry sectors, where conversion or stacking factors are applied to relate gross (apparent) volume to solid content. These factors, extensively used in countries such as Romania, Norway, Spain, and Canada, are essential for biomass estimation, transportation logistics, and commercial transactions involving pulpwood and firewood [17].

The estimation of firewood, chips, and sawdust volume in the forest industry traditionally relies on stacking methods [18]. Approaches such as stere-based estimation or xylometry, although accurate, are time-consuming and operationally restrictive. These conditions highlight the need for practical, efficient, and scalable alternatives that can be adapted to diverse forest contexts, especially in developing regions where infrastructure is limited. In this sense, the development of visual and digital tools to estimate residue volume more reliably contributes to standardizing metrics and improving the economic valuation of forest by-products across the value chain [17,19].

To overcome the disadvantages of manual measurement, researchers have proposed automated measurement methods, which employ photoelectric, optical, or laser-based technologies to calculate the volume [20]. Modern aerial and terrestrial remote sensing platforms equipped with camera and/or light detection and ranging (LiDAR) units have been increasingly popular for conducting high-fidelity geometric measurements [21]. Although these methods achieve satisfactory detection accuracy, their widespread adoption is limited due to the lack of portability of the equipment [20].

Image segmentation plays a critical role in computer vision and pattern recognition. As a core component of modern image processing systems, computer vision-based object detection and recognition depend heavily on accurate segmentation to ensure reliable and effective analysis of visual data [22,23,24]. To improve the portability and efficiency of the measurement, researchers utilize computer vision to automatically detect wood and measure it [20]. Thresholding is the simplest method of image segmentation. The input to a thresholding operation is typically a grayscale or color image. In the simplest implementation, the output is a binary image representing the segmentation. This method of segmentation applies a single fixed criterion to all pixels in the image simultaneously. The advantage of obtaining a binary image first is that it reduces the complexity of the data and simplifies the process of recognition and classification [25]. Threshold segmentation is an efficient method for estimating the volume of a residue pile from the stacking coefficient from images [16].

Given the limited literature on standardized methods for quantifying sawmill residues (particularly in regions such as Latin America), this study adopts an exploratory approach to compare the accuracy of two indirect methods for estimating the volume of stacked forest residues: one based on image segmentation and the other employing terrestrial LiDAR scanning for residues 3D reconstruction. To evaluate the performance of both methods, reference volumes were obtained using a xylometer.

2. Materials and Methods

2.1. Study Area

The study was conducted at a private sawmill located in the municipality of Durango, Mexico, a region with a temperate climate, average annual temperatures ranging from 16 to 20 °C, and precipitation between 500 and 800 mm [26] (Figure 1a). The sawmill produces a variety of wood products, including boards, planks, and beams, in a wide range of dimensions. These products are obtained through mechanical wood processing under controlled conditions, ensuring compliance with technical quality standards. The resulting physical and dimensional properties make them suitable for both structural and non-structural applications, particularly in the fields of construction, carpentry, and manufacturing. During the sawing process, wood residues such as sawdust, bark, and offcuts are also generated (Figure 1b).

2.2. Materials

The residues utilized in this study, derived from Pinus spp., were generated during the primary sawing process using a main bandsaw. These materials, which are often discarded or incinerated without energy recovery, reflect common waste management practices in sawmills across the region (Figure 2a). To quantify the volume of these residues, twelve control piles were constructed, arranged in three groups of four piles each. The piles were trapezoidal in shape, with a uniform base length of 1.2 m and a width of 0.5 m. However, the height of the piles varied (Figure 2b).

2.3. Workflow Diagram

A workflow diagram (Figure 3) was developed to summarize the main stages involved in estimating the volume of sawmill residues using two indirect methods: 2D image analysis and LiDAR technology. The process begins with the arrangement of Pinus spp. residues in trapezoidal piles. The real volume is first determined using the xylometer method, which serves as a reference for validating the indirect estimations.

In the first approach, 2D images are captured using Nikon and iPhone cameras. The apparent volume is calculated using the trapezoidal volume formula. A threshold-based segmentation is then applied to distinguish pixels corresponding to solid wood and voids. This information is used to compute a stacking coefficient, representing the proportion of solid wood within the total pile volume. The estimated volume is then obtained by applying this coefficient to the apparent volume.

In the second approach, a LiDAR scanner is used to generate 3D point clouds. From these, a rasterization process is performed to derive the stacking coefficient. The apparent volume is calculated from the point cloud and adjusted using the stacking coefficient to estimate the volume.

Finally, both indirect estimations are compared against the reference volume to assess their accuracy and effectiveness.

2.4. Xylometer Method and Real Volume

The reference volume of residue piles was determined using the water displacement method (xylometer), a commonly used and reliable approach for irregularly shaped materials [27]. This method involves submerging a porous solid in water within a graduated cylinder and measuring the difference between initial and final water levels to calculate the bulk volume (Figure 4). Although it requires careful execution, it is considered one of the most accurate methods for approximating real volume [28]. The obtained results were used as a control in this study.

To determine the volume, a 220 L cylinder with a diameter of 0.568 m was used. An internal tape measure was attached to the cylinder to measure the displacement of water during immersion. The twelve piles were separated into samples that allowed immersion in the cylinder. The volume of the samples was calculated using Equation (1).

$$V=\pi \times r^2\times h$$

(1)

where

• V = volume (m³);
• r² = radius squared (m);
• h = height (m).

After obtaining the volume of each individual sample through water displacement, the values were summed to determine the total reference volume of each pile.

2.5. Volume Estimation from 2D Images

To estimate the volume of the piles, a comprehensive image processing pipeline was developed. This pipeline consists of several steps: image acquisition, calculation of apparent volume in piles, color space conversion, semantic segmentation, calculation of the stacking coefficient, and the subsequent estimation of the volume. The entire process was implemented using Python version 3.11.11, with image processing by the OpenCV library version 4.10.0 and NumPy version 1.26.4.

2.5.1. Image Acquisition

Images from two sides (front and rear faces) of the stacks were acquired with a Nikon HD digital SLR camera (Nikon, Tokyo, Japan) at a resolution of 6000 × 4000 pixels. Additional images were also captured with an iPhone SE 2022 (Apple Inc., Cupertino, CA, USA), which has a 12 MP wide-angle rear camera and ƒ/1.8 aperture (Figure 5). All images were captured under natural daylight to reduce potential lighting-induced bias and to ensure consistency across data. To reduce perspective distortion and optimize image resolution and feature detail, the camera was positioned perpendicular to the pile face, with the field of view adjusted to encompass the largest possible portion of the stack within the frame.

2.5.2. Apparent Volume in Piles

The piles were assumed to have a trapezoidal shape, and their volume was calculated using the formula for a regular prism (Equation (2)).

$$Vol={\displaystyle \frac{\left(b_1+b_2\right)\times h}{2}}\times H$$

(2)

where

• $Vol$ = apparent volume (m³);
• $b_1$ = length of the first base (m);
• $b_2$ = length of the second base (m);
• $h$ = height (perpendicular distance) between the two parallel bases of the trapezoid (m);
• $H$ = length of the prism (m).

2.5.3. Threshold Segmentation

Semantic segmentation involves classifying each pixel in an image into predefined categories [29]. Among various segmentation methods, thresholding is the simplest. It transforms grayscale or color images into binary outputs by comparing the intensity of each pixel to a defined threshold value [25,30,31]. This operation is defined in the following Equation (3):

$$I{m}^{\prime }\left(x,y\right)=\left\{\begin{array}{cc}255& Im\left(x,y\right)>threshold\\ 0& Im\left(x,y\right)<threshold\end{array}\right.$$

(3)

where

• Im′ (x, y) = new image with the classified pixels;
• Im (x, y) = original image;
• threshold = value that allows to differentiate the objects present in the image.

In this study, the pixel values in Im′ (x, y) were used to represent semantic classes, where 255 (white) corresponds to wood, and 0 (black) indicates holes or absence of wood. Figure 6 illustrates a representative result of the threshold segmentation applied to a 2D image, highlighting the distinction between solid wood and void areas. Threshold segmentation was applied to both image sets to compare the accuracy of a smartphone camera with a professional camera.

To improve the robustness of the threshold segmentation method against variations in lighting conditions, image processing was performed in the CIELAB color space. This color space is designed to be perceptually uniform and device-independent, making it more stable under different illumination scenarios compared to RGB [32]. Developed by the International Commission on Illumination (CIE) in 1976, it represents all human-perceptible colors within a three-dimensional space consisting of three channels: “L” for luminance and “a” and “b” for chromaticity. Unlike RGB and CMYK color models, CIE Lab is designed to approximate human vision, achieving perceptual uniformity by closely aligning the “L” component with human perception of lightness. This allows for precise color balance adjustments and lightness contrast corrections. Additionally, its ability to perform segmentation independently of brightness fluctuations makes it particularly useful in variable environmental conditions during image capture [33,34].

2.5.4. Stacking Coefficient Images

To obtain the stacking coefficient, the segmented images Im′ (x, y) are used, in which the pixels corresponding to wood (value 255) are counted and divided by the total number of pixels in the image. In other words, the percentage of solid wood is calculated. This procedure is summarized in Equation (4).

$$S{c}_{im}={\displaystyle \frac{Wp }{Tp}}$$

(4)

where

• $S{c}_{im}$ = stacking coefficient;
• $Wp$ = wood pixels;
• $Tp$ = total pixels.

The stacking coefficient for each pile is then determined as the average of the coefficients for the front and rear faces.

2.5.5. Estimated Volume Images

Finally, the estimated volume from each pile is obtained by multiplying the apparent volume by the stacking coefficient according to Equation (5) [35].

$$EVol=Vol·\overline{Sc}$$

(5)

where

• $EVol$ = estimated volume (m³);
• $Vol$ = apparent volume (m³);
• $\overline{Sc}$ = average stacking coefficient.

2.6. Volume Estimatation Using Light Detection and Ranging (LiDAR)

For volume estimation using a terrestrial laser scanning (TLS), the pipeline involved several key steps. First, the concave hull was used to obtain the areas of the slices from the point cloud piles, which were then summed to estimate the apparent volume of each pile. Next, the stackings coefficients were computed through rasterization, which involved converting the point cloud sides into a raster format to estimate the percentage of solid wood. Finally, the estimated volume was calculated by applying the stacking coefficient to the apparent volume derived from the LiDAR data. The entire process was implemented using R software ver. 2023.12.1 Build 402 [36], which facilitated the processing and analysis of the LiDAR point cloud data.

2.6.1. Point Cloud Acquisition

In this study, TLS data were acquired using a FARO Focus Laser Scanner M70 (FARO, Lake Mary, FL, USA), which has a maximum range of 70 m and a measurement accuracy of ±3 mm. The scanner was configured with the “interior” profile, using a 1/4 resolution and 4× quality settings. This configuration resulted in an average point density of 234,679 points per square meter and a resolution of 10,310 × 4268 points.

To obtain a point cloud of the piles, a LiDAR scanner was positioned within a designated area of 4 × 4.5 m. Five scans were performed: four from each corner of the area and one from the center. This scanning arrangement was designed to capture a comprehensive dataset from multiple angles, ensuring an accurate representation of the pile dimensions and volume. The scanning angles were set to 300 degrees vertically and 360 degrees horizontally (Figure 7).

After scanning, the five scans were merged with eight reference targets distributed throughout the site, with the merging process completed using FARO^® SCENE software ver. 5.5.3.16. and for segmentation of each pile. We used Cloud Compare software ver. 2.12.4.

2.6.2. Apparent Volume in LiDAR Data

For apparent volume, each point cloud was divided into 100 uniform levels along the z-axis, generating cross-sections with quasi-quadrangular geometries.

The areas of these sections were calculated using the Concaveman package [37], which constructs a concave hull fitted to the distribution of points in each plane, ensuring accuracy in surface representation. The volume between consecutive levels was estimated by multiplying the calculated area by the height between the levels, and the total volume of each pile was obtained by summing the volumes of all sections. This procedure ensured accuracy in representing the surface of each section (Figure 8).

2.6.3. Projected Area Calculation

For each pile, both of its sides were segmented using CloudCompare software ver. 2.13.0, and the x coordinate was discarded to project the points onto a (y, z) plane. The Concaveman [37] and Geometry [38] packages were then reapplied to calculate the area enclosed by the contour generated by concave hull.

Subsequently, a gridded mesh with a resolution of 0.00001225 m² was generated to cover the spatial boundary of the point cloud. Grid cells were classified as either occupied or empty based on their intersection with the point cloud. The adjusted area was calculated by multiplying the number of occupied cells by the area of each cell [39,40]. This process is illustrated in Figure 9.

The accuracy of this method depends on the grid resolution: finer resolutions improve void detection but may approach the resolution limit of the laser scanner; in contrast, coarser resolutions reduce processing time but may underestimate voids. This combined geometric and rasterization approach enables accurate and reproducible surface estimations by incorporating both pile contours and internal voids (Figure 9).

2.6.4. Stacking Coefficient LiDAR Data

The stacking coefficient was calculated by dividing the area occupied by the wood by the total projected area of the plane.

$$S{c}_{LiDAR}={\displaystyle \frac{Wa}{Ta}}$$

(6)

where

• $S{c}_{LiDAR}$ = stacking coefficient;
• $Wa$ = wood area;
• $Ta$ = total area.

2.6.5. Estimated Volume of LiDAR Data

Finally, the estimated volume from each pile is obtained by multiplying the apparent volume by the stacking coefficient according to Equation (7) [35].

$$E{V}_{LiDAR}=V_{LiDAR}{\overline{Sc}}_{LiDAR}$$

(7)

where

• $E{V}_{LiDAR}$ = estimated volume (m³);
• $V_{LiDAR}$ = apparent volume in LiDAR data (m³);
• ${\overline{Sc}}_{LiDAR}$ = average stacking coefficient of two sides.

2.7. Data Analysis

The methods were statistically compared using a completely randomized experimental design, the normality of the data was evaluated with Shapiro–Wilk test, and variance homogeneity was assessed with Levene’s test. When the assumptions were met (p > 0.05), an analysis of variance (ANOVA) was performed, followed by Tukey’s test; otherwise (p < 0.05), a Kruskal–Wallis test with Dunn’s post hoc test was applied. Additionally, to assess the accuracy and the relationship between the studied methods and the reference method, the root mean square error (RMSE) and the coefficient of determination (R²) were calculated with Equations (8) and (9), respectively:

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n{\left(y_i-\widehat{y_i}\right)}^2}$$

(8)

where

• n = number of observations;
• y_i = observed values;
• ŷ_i = predicted values.

$$R^2={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{\left(y_i-\widehat{y_i}\right)}^2}{{{\displaystyle \sum }}_{i=1}^n{\left(y_i-\overline{y}\right)}^2}}$$

(9)

where

• y_i = observed values;
• ŷ_i = predicted values;
• $\overline{y}$ = mean of observed values.

These metrics are useful for identifying the magnitude of the error and the proportion of variance explained, respectively. All the analyses were conducted using R software ver. 2023.12.1 Build 402 [36].

3. Results

3.1. Apparent Volume

The analysis of apparent volumes obtained using LiDAR and the trapezoidal formula demonstrated high consistency across both methods, as shown in

Table 1. Descriptive statistics of apparent volume.

Method	Minimun	Median	Mean	Maximum	SD
LiDAR	0.0848	0.0987	0.0985	0.1070	0.0056
Trapezoidal formula	0.0879	0.1005	0.0990	0.1031	0.0044

SD = standard deviation.

. While the Shapiro–Wilk test indicated that the LiDAR data followed a normal distribution (p > 0.05), the trapezoidal formula data did not (p < 0.05). Kruskal–Wallis test (p = 0.7727) revealed no statistically significant differences between the methods. Furthermore, statistical grouping with Bonferroni correction [41] confirmed that both methods belonged to the same group, labeled as “a”.

3.2. Estimated Volume

The statistical analyses of volume obtained via iPhone (Apple Inc., Cupertino, CA, USA), Focus M70 (FARO, Lake Mary, FL, USA), Nikon (Nikon, Tokyo, Japan), and xylometer (Figure 10) showed no significant deviations from normality (Shapiro–Wilk, p > 0.05) or variances (Levene’s, p = 0.3403). ANOVA found no significant differences among methods (p = 0.0977), and Tukey’s test grouped all methods in the same category (“a”).

3.3. Evaluation of R² and RMSE

The relationship between the control method (xylometer) and the iPhone, LiDAR, and Nikon methods was evaluated by calculating the coefficients of determination (R²) and the root mean square error (RMSE), as presented in Figure 11.

4. Discussion

These results suggest that, although all methods demonstrate a moderate ability to model the observed data, the Nikon method stands out for its accuracy, while the iPhone method offers an acceptable balance between R² and RMSE. In contrast, the LiDAR method showed the lowest performance in both metrics (R² = 0.506, RMSE = 0.009), which could be attributed to methodological differences or limitations in the processed data. LiDAR is a highly versatile tool that, based on the information obtained, enables the generation of diverse representations and the application of techniques to ensure reliable results, even when faced with limitations in data acquisition, such as unfavorable angles or objects obstructing the complete capture of information. In this case, the LiDAR allowed us to approximate the reference volume using the same principle as the threshold segmentation method through the use of the stacking coefficient.

The results of the comparative analysis indicate that the threshold segmentation method outperformed the LiDAR approach in estimating the volume of wood residue piles, achieving an R² of 0.64 with an RMSE of 0.006 when using smartphone images. This method’s portability and ability to analyze images captured with smartphones demonstrate its efficiency in extracting features for object recognition and classification, making it a practical and reliable approach. Moreover, the widespread availability and inherent portability of mobile devices such as the iPhone not only facilitate real-time decision making through rapid on-site data acquisition and processing but also demonstrate great potential for integration into low-cost, smartphone-based applications that provide accurate and reliable information for the sawmill industry. Regarding image acquisition, although the CIELAB color space helps reduce sensitivity to variations in lighting conditions, these can still affect the quality of volume estimations. It is recommended that users ensure consistent and neutral lighting, preferably close to the CIE D65 standard (midday illumination), to avoid systematic errors. The influence of lighting on the results remains an open research topic, which could focus on developing a fixed lighting protocol or using color calibration charts to improve the reliability of the proposed methods. In addition to lighting conditions, camera calibration during image acquisition plays a critical role, as it directly affects the overall performance of volume estimations. Proper calibration ensures geometrically accurate projections and improves the precision of spatial measurements [42].

Although LiDAR-based methods are gaining attention due to their ability to produce accurately scaled models and their integration into smartphones and tablets [43,44], in this study, the LiDAR method achieved a lower R² of 0.50 with an RMSE of 0.009. While LiDAR provides high precision, this method can significantly increase field measurement times (depending on the number of scans) and processing steps (depending on the number of targets and the method used for their detection in the point cloud) [45]. For example, Feifan et al. [46] reported a method using a mobile phone with a rangefinder that reached a measurement accuracy of 98.2%, highlighting the potential of emerging technologies. However, the effectiveness, processing time, and cost of these methods still depend on advancements in technology. An aspect that deserves attention in future work is the consideration of labor costs associated with the implementation of these methods, which was not addressed in this study. While equipment and accuracy are key factors, the time and effort required for preparing the residue piles, capturing consistent images, and processing data can represent significant costs. Regarding LiDAR, the time invested in each scan and the high computational cost impact its implementation. Evaluating these costs will provide a more comprehensive view of the economic feasibility and operational efficiency of each method in industrial contexts. Therefore, it is recommended that detailed analyses of labor costs be included in future research to strengthen the practical applicability of the proposed techniques. There will always be a need for faster, more accurate, and more cost-effective estimation methods, which highlights the importance of continuous improvement and ongoing testing. Accurately understanding sawmill residue volumes is essential for biomass assessment and optimizing resource management.

It is crucial to emphasize that using a xylometer as a reference to evaluate a LiDAR system is a paradoxical practice. The xylometer exhibits greater measurement uncertainty due to factors such as visual reading, limited resolution, and operator influence [47,48]. In contrast, LiDAR, as an advanced electronic device, offers significantly higher precision and stability in its measurements. Therefore, relying on a traditional method as a benchmark introduces an inherent discrepancy in the measurement process. This situation highlights the importance of carefully selecting an appropriate reference standard, especially in the estimation of forest residues. The results obtained may be biased due to this methodological choice, yet they still show great potential for improving accuracy in future studies.

Finally, this study shows that more accurately identifying the amount of sawmill residues allows not only an improvement of operational efficiency for the industry but also a better performance for determining circular indicators. Similar conclusions were reached by Krigstin et al. [49], who emphasized the importance of reliable data on sawmill residues for developing sustainable biomass supply chains. In this context, the sawmill industry plays a key role in achieving sustainable development, and technologies such as smartphones and pattern recognition are crucial for identifying and quantifying by-products. This contributes to the proposal of well-defined circular indicators to enhance the sustainable performance of the industry.

5. Conclusions

This study demonstrates that the threshold segmentation method is a practical and efficient approach for estimating the volume of residue piles from images, particularly those captured by smartphones. This method shows great potential for integration into low-cost, smartphone-based applications, providing accurate and reliable information for the sawmill industry. Additionally, Light Detection and Ranging (LiDAR)-based geospatial field data acquisition using smartphones and tablets can also produce accurate volume estimations, but its widespread adoption is limited because not all smartphones are equipped with this technology. Therefore, it is recommended to continue testing the threshold segmentation method against LiDAR-equipped smartphones and explore its application in low-cost smartphone-based solutions. Moreover, as smartphone camera technology continues to advance, it is possible that the accuracy of the threshold segmentation method could improve with higher-resolution sensors and enhanced image processing capabilities. The implementation of these smartphone-based solutions not only facilitates better decision making but also helps to promote more sustainable practices in the sawmill industry by reducing uncertainty regarding the amount and type of sawmill residues. Future work should focus on further refining the threshold segmentation method by testing and validating it in various operational environments to assess its robustness across different sawmill settings and residue types.