1. Introduction

Objective and scope of the wiki / how to use it

The goal of this document is to be the basis of a wiki elaborated within the framework of the Cost Action 3DForEcoTech. This document is a living document with a collection of current state of knowledge on point cloud collection of trees and forests to develop allometries. It aims at providing an overview in form of text and links to existing information sources on this subject.

2. Application

2.1 Objectives of CRS campaign

The objectives are the basis for all further steps to be taken within a CRS campaign to improve allometries (or generally spoken forest resources estimates, e.g. in forest inventories).

The objectives should cover the following points:

• Level of CRS-allometry application of the study: individual tree level, stool level and stand level. Even if all the levels are possible to be studied within CRS campaigns, we are going to focus on the individual tree level.
• Type of target variable of the individual tree to be estimated by the allometry: e.g. Biomass, wood volume, carbon content of the woody biomass. Also other tree dimensions such as diameter, tree height, crown shape could be target variables but are not part of this wiki.
• Tree compartments to be estimated: 
  • Examples of tree compartments: Stem wood (from ground to tip), merchantable stem wood, thick branches, medium branches, thin branches, bark, leaves/needles, stumps. Diameter thresholds defining the compartmens should be considered, as they may vary between countries, project etc.
• Tree status: living, dead trees, standing, lying (note, that lying dead wood is often not in form of trees, but only in heterogeneously formed tree parts, see above under “tree parts of fragmented trees”).
• Ground related location: above, below ground (not accessible to CRS measurements)
• Improvement of the current estimation process:
  • Are there existing allometric models? If yes, what is the quality of these models and do they satisfy all assumptions for an unbiased estimation (e.g. is the ground truth sample representative for the population to be estimated). 
  • What is the required precision and accuracy for the allometric models to be developed.
  • Which assumptions for an unbiased estimation have to be satisfied?
• Data related requirements: 
  • Additivity of allometric models of tree compartments 
  • Convertibility to different types of forest resources: e.g. conversion models to convert small branch volume to small branch biomass.

2.2 Restrictions

Restrictions play an uncomfortable, but important role in the planning of a CRS campaign. Mostly the restrictions are concerning the budget and also the time available for measurement (e.g. if CRS measurements are included in an existing forest inventory, it has to be clear, how much time field crews can spend with CRS measurements alongside other measurements).

Ground truth data for the calibration of new allometric models, can be of different kinds. Weighed trees might be the silver bullet, but is very labour and cost intensive. Section wise measurement of harvested trees might provide similar data quality (depending on the targeted tree compartment) to a much lesser cost. Density measurements, sometimes included in ground truth campaigns, might not always be necessary, due to the very strong correlation between (dry) weight and volume of trees.

In some cases, valuable ground truth samples are already available, but need just some completion of under-represented individuals.

3. Allometries

The development of allometric models is based on the following pillars in terms of closeness to the true value of the targeted variable:

• Truth (target variable): the real but unknown value of the target variable (e.g. true tree biomass or true tree volume). This is the value that should be predicted by the allometric models.
• Ground truth (of target variable): metric used to approximate the truth. Depending on the quality requirements, it could be tree weight, sectionwise measurement (mostly destructive measurement). Also CRS features, such as QSM can serve as ground truth. The ground truth is the best value one can get of the target variable.
• Allometry: A mathematical function to derive the target variable based on easy acquireable measurements. The allometry is developed based on ground truth data where both, the target variable is measured as well as the “easyly acquireable measurements”. Certain forest inventory apply a two stage estimation procedure, where an additional allometric model ist used, which is one step simpler than the ground truth based allometric model. These simple models help to reduce labour cost in the field. 

3.1 Objectives

The general objective of an allometric model is to predict the variable as accurate as possible and using the simplest model as possible.

3.2 Target variables

Definition of the target variable according to section 2.1., depending also on the restrictions (s. section 2.2).

3.3 Common aspects related to data collection

Data are commonly collected from plots that must be established throughout the area of distribution of the studied species. Plots are subjectively selected to represent the existing range of ages/diameters, stand densities/structures and sites of the species.

3.4 Common statistical aspects between traditional and new CRS procedure

Linear and non linear models have been widely used in forest biomass estimation (Cunia and Briggs, 1984; Reed and Green, 1985; Reed et al., 1996). The most common mathematical model used for biomass prediction takes the form of Snell (1892) power function  (Kaitaniemi, 2004; Zianis and Mencuccini, 2004; Zianis et al., 2005). However, nonlinear models allow maintain the additive structure of the residuals and avoid the inherent bias associated with logarithmic-transformed linear models. The general mathematical formulations of these models are:

Linear model:	(1)
Non lineal model: +…	(2)

where AGB is the aboveground biomass or aerial biomass (kg), ß_i are the parameters of the model, X_j are the independent variable at each level (diameter at breast height, root-collar diameter, total height….) and ℇ is the error. 

Model fitting is carried out in two stages. First, each biomass component is fitted individually. Initial parameters for running the iteration process when fitting allometric models were obtained from the linearized form of a previous linear fit. The best equation for each tree component is selected based on goodness of fit statistics. The most common ones are coefficient of determination (R²), adjusted coefficient of determination (R_adj²) and root mean square error (RMSE). In the second step, the best equation for each biomass component are fitted simultaneously by nonlinear seemingly unrelated regression (NSUR). This method fits an apparently non related equation system formed by the equations of the biomass components considered and the total biomass equation. The NSUR method takes into account that equation errors are correlated (Borders, 1989; Parresol 1999, 2001), and it presents the best fitting solution that minimizes the global errors associated with these equations, although the solution for each biomasscomponent is not necessarily the best. This important feature of biomass systems refers to the fact that estimates of total biomass equation must be equal to the sum of the estimates of the equations of each biomass component. To ensure the additivity of the system using NSUR procedure, the total biomass equation must be expressed as the sum of the equations of each component. 

Biomass equations have two common problems: heteroscedasticity and multicollinearity. Although the least squares estimates the regression coefficients remain unbiased and consistent under the presence of multicollinearity and heteroscedasticity, they are not necessarily the most efficient (Myers, 1990). Multicollinearity refers to the existence of strong intercorrelations among the independent variables, mainly due to the use of complicated models with several polynomial terms (Kozak, 1997). The presence of multicollinearity is usually evaluated by the condition number (CN), which is defined as the square root ratio of the largest (λ_max) to the smallest eigenvalue (λ_min). According to Belsey (1991), if the condition number ranges from 5 to 10, collinearity is not a major problem, if it is the range 30-100, then there are problems associated with multicollinearity, and if it is the range 1000-3000 there are severe problems associated with collinearity. 

Heteroscedasticity often occurs in biomass data, that is, the error variance is not constant over all observations (Parresol, 1999). The easiest way to detect heteroscedasticity is by plotting the studentized residuals against the predicted values. The lack of homogeneity in the error variance is corrected by weighting each observation during the fitting process by the inverse of its variance (σ_i²). Although the variance is unknown, it is often assumed that the variance of the error of the i_th individual can be modelled as a power function of the independent variables X_i (Furnival, 1961), i.e. . The most reasonable value of the exponential term k is obtained by optimizing the method proposed by Harvey (1976), which consists of using the estimated errors of the unweighted equation () as the dependent variable in the error model, or taking the natural logarithm of the equation. All the weighting equations selected individually for the biomass equations are later used in the simultaneous fitting. 

3.5 Case studies of traditional allometries procedure          

 With open structure within the examples (also from other countries).

3.5.1 Data collection for biomass and volume in Central Europe: the case of Switzerland

3.5.2 Data collection for biomass and volume in Southern Europe: the case of Spain

Data are commonly collected from networks of experimental plots that must be established throughout the area of distribution of the studied species. Plots are subjectively selected to represent the existing range of ages/diameters, stand densities/structures and sites of the species. Commonly, all the trees included in the plot are identified, and the diameter at breast height (diameter at 1.3 m) (d, cm)  are measured with a calliper, total height (h, m), height to the base of the crown (considered as the lower insertion point of at least three consecutive live branches in a tree) are measured with a digital hypsometer . These measures would allow the estimation at stand level e.g. of the following variables: dominant diameter (d₀, cm) as the average diameter of the 100 thickest trees per hectare, dominant height or average height of the 100 thickest trees per hectare (H₀, m), basal area (G, m²) and number of trees per hectare (N). 

• Data collection for biomass purposes

Individual trees with non-anomalous development (e.g., no bifurcation, stem breakage) and average growth conditions are selected for destructive sampling in medium site qualities and distributed according to age/diameter classes. Before tree felling, diameter at breast height, total height, crown diameters (dc, cm) are measured in all the sampled trees. When working with small trees, coming for afforestation or reforestation, sometimesthe diameter at breast height  is still not reached. In these cases, root-collar diameter (RCD, cm) is also measured. After felling, trees are destructively sampled to separate aerial biomass (aboveground biomass) into the different components. Depending on working with small or big trees, these components could be (Montero et al. 1999; Ruiz-Peinado et al. 2011, 2012; Menéndez-Miguélez et al. 2013, 2021, 2022):

• Small trees: leaves/needles, branches with diameter smaller than 2cm, branches with diameter bigger than 2 cm, stem.
• Big trees: leaves/needles, branches with diameter smaller than 2 cm, branches with diameter between 2 and 7 cm, branches with diameter bigger than 7 cm, stem (wood+bark). In the case of the stem, it is cut into logs of approximately 1 m length with a thin-end diameter of 7 cm).

The total fresh weight of each component is measured in the field with a portable weighing scale. For big trees, three disks of wood including bark are cut in each stem (from the bottom, middle and the top). The disks together with a representative composite sample of each tree component, are sampled at the same time as bulk weighting is carried out, and they are transported to the laboratory and weighed on a digital scale. Once in the laboratory, the sample of branches less than 2 cm is later subdivided into twigs (diameter less than 0.5 cm), thin branches (diameter 0.5-2 cm) and leaves/needles. The three disks are separated into wood and bark. All samples are oven-dried at 102ºC to determine sample humidity and the dry weight of the component. The dried disks are also used to calculate the dry weight ratios of wood and bark.  

• Data collection for volume purposes

The same as for biomass models, data collected for the fitting of volume models should cover the existing range of ages, stand densities, and sites of the species in the distribution area. Trees had to be healthy and of a standard shape (i.e., not forked nor excessively branched), ensuring also a representative distribution of diameter and height classes. 

Before felling, diameter at breast height (d, cm) is measured with a calliper for each tree. The trees are then felled and the total bole lenght, that is, total height (h, m), measured. Trees are cut into 1-m logs, up to a top diameter of 7 cm, and measured to the nearest centimeter. Two perpendicular overbark diameters and two perpendicular bark thickness are measured in each cross section (at height h_i, in m). Over bark log volumes are calculated in cubic meters using Smalian’s formula, and the top section is usually treated as a cone. Over bark total volume is obtained by summing the over bark log volumes and the volume of the top section. Finally, a certain pairs of diameter (d) at certain height (h_i) measurements are used for the fitting data set. 

• Fitting biomass models

Depending on the type of the stand, the level of detail to be reached, and the available data, fitting process can be carried out at three different levels: at tree level, stool level or stand level. Based on it, the explanatory variables that could be used in any of these levels, slightly variate. 

Considering standing tree variables, diameter at breast height (d, cm) is the most common explanatory variable since it is most closely correlated to the biomass. However, the accuracy of the biomass estimates is usually increased by inclusion of the tree height (h, m) as the second predictor and development of combined d-h models (Wang, 2006). When working with small trees, they have not already reached 1.3 m height. Therefore, it is necessary to consider alternative explanatory variables like root-collar diameter (RCD, cm). Live crown variables such as length or the diameter at the base of the crown have improved estimates of the branch biomass or total crown biomass (Satoo and Madgwick, 1982; Clark, 1982). Crown projection area (CPA) and biomass-packing (BP, crown projection area * height) can also be considered as descriptive variables for aboveground biomass, especially for small trees. 

The case of coppice stands, also allows the possibility of fitting models at stool level. It seems reasonable to estimate biomass at this level as the product of the biomass of a representative tree of the stool (given by mean diameter and/or height) and the number of stems per stool. In this case, some of the explanatory variables can be diameter of the thickest tree and its height, number of stems per stool, arithmetic mean stool diameter, quadratic mean stool diameter or stool basal area.

Considering stand variables, some of the following variables could be used to predict aboveground biomass: mean stand height, an indicator of stand density, basal area and site index (Clutter et al., 1983).  

The crown compartment is one of the most difficult ones to which fit an accurate model to estimate the biomass. The analysis of the published biomass models in Spain, allowed the identification of certain problems with estimations for some biomass components, for instance the thick-branch biomass models for some species give unrealistic values when applied to the tallest trees. Severe deviations could occur in the biomass estimations for trees with diameters above or below the range of diameters used in the fitting process. Caution is required when applying biomass models outside the range for which they were fitted. 

To solve these problems, a new methodology was proposed based on: (i) the increase in sample size of the biomass dataset and the use of new datasets (volume) even those lacking crown biomass information; (ii) crown biomass ratio (crown biomass divided by total aboveground biomass); and (iii) simultaneous fitting of the system of the stem biomass, crown biomass ratio, and total aboveground biomass models. 

The crown biomass ratio was defined as the crown proportion of the total aboveground biomass, which is calculated as crown biomass divided by total aboveground biomass. The graphical study of the crown biomass ratio against d²h (d = diameter at breast height, h = tree height), a variable that shows a direct relationship with the total volume of the tree, allowed the distinction of three different patterns: an increasing pattern, a constant one, and a decreasing pattern with tree size. The crown biomass ratio was modelled based on these observed patterns. 

• Fitting volume models

There are numbers of ways to estimate volume stocks as accurate as possible, the two most important of which are volume-ratio equations that predict merchantable volume as a percentage of total volume (Burkhart, 1977; Clutter, 1980; Reed and Greeen, 1984) or using taper functions. Taper functions describe stem taper (Brink and Gadow, 1986; Kozak, 1988; Riemer et al., 1995) and provide forest managers with estimates of (a) diameter at any point along the stem, (b) total stem volume, (c) merchantable volume and merchantable height to any top diameter and from any stump height, and (d) individual volumes for logs of any length at any height above the ground (Kozak, 2004). Ideally, a taper equation should be compatible, meaning that the volume computed by integration of the taper function should be equal to that calculated by a total volume equation (Clutter, 1980; Demaerschalk, 1972; Fang et al., 2000). The most common taper functions are those of Fang et al. (2000), Bi (2000), Kozak (2004), Demaerschalk (1972), and Thomas and Parresol (1991). 

When fitting a taper function, the process begins with the priorization of the taper function in a first step, and subsequently performing the predicted volume calculation from the estimation parameters obtained. 

There are several problems associated with stem taper and volume equation analyses that violate the fundamental least squares assumption of independence and equal distribution of errors with zero mean and constant variance. One of the most common is the presence of autocorrelation in the data as a result of working with multiple observations on each tree. To resolve this problem, the error term is modeled using a continuous autoregressive error structure (CAR(x)), which allows the model to be applied to irregularly spaced, unbalanced data (Zimmerman and Nuñez-Antón 2001). Another problem in taper functions, shared with biomass equations, is multicollinearity, which refers to the existence of high intercorrelations among the independent variables in multiple linear or non-linear regression analyses. To evaluate the presence of multicollinearity, it is commonly used the condition number (CN). According to Belsey (1991), if the condition number is between 5 and 10, collinearity is not a major problem; if it is in the range of 30–100, then there are problems associated with collinearity; and if it is in the range of 1,000–3,000, the problems are severe.

3.6 New CRS input data (explanatory variables)

CRS can be a very powerful tool to both apply allometric equations on a much larger scale as include predictor variables in allometric equations that were previously infeasible to measure in the field in routine/ large scale surveys. Traditionally, predictor variables for allometric equations are manually measured in the field, often in routine surveys. These manual measurements are time consuming and often limited to relatively simple metrics. On top of that, these measurements are prone to biases and errors (sources needed + more explanation).

CRS facilitates the collection of input variable data on a much larger scale and allows the measurement of more complex metrics. An overview (table?) of what structural metrics can be measured with what CRS platform can be found in Chapter 4.

The following metrics can be measured using CRS:

Figure 1: Illustration of metrics that can be derived from a single tree point cloud. 

1. Stem diameters

Feature description:

The diameter of the tree stem at a specific height. Can be measured once (usually DBH) or in short height intervals along the (possibly whole) stem to derive a stem curve. Stem volume can be directly estimated from the stem curve, but diameters at specific heights (at 1.3 m and also e.g. 7 or 10 m) are also used in allometric models.

Methods for calculation:

• Fitting a circle to a 2D projection of the points in a stem slice
• Fitting a cylinder to a 3D stem slice
• Mean of furthest distance of points in multiple directions
• 2D Convex hull around a 2D projection of the points in a stem slice

Considerations:

• Horizontal coverage: when using TLS, multiple scans from different directions (around the stem) provide more reliable diameter estimations.
• Vertical coverage: occlusion by branches can cause difficulties, especially in the crown area. Thus, diameter estimation may not be possible above a certain height.
• Circle or cylinder fitting assumes a circular stem cross-section. This is often not the case, especially with broadleaf trees. 

Suitable instruments/technology:

• TLS: Suitable for diameter estimation. Sensitive to the number of scan positions around the tree.
• MLS: Ddepending on the range of the instruments, diameter estimation may only be possible up to a certain height.
• UAVLS: Only suitable if lower parts of crown and stems are not occluded
• ALS: Mostly unsuitable

b) Tree height

Feature description:

The vertical distance between the base of the tree and the tip of the highest branch on the tree.

Methods for calculation:

• Distance between highest point belonging to a tree (or e.g. 95^th quantile) and the ground (DTM) or lowest point, in case of a segmented tree point cloud.

Considerations:

• Tree height is often measured more accurately using CRS technologies than when measuring with a hypsometer in 2-points mode.

Suitable instruments/technology:

• TLS: Depending on range, only if treetops are not occluded
• MLS: Depending on range, only if tree tops are not occluded
• UAVLS: Suitable
• ALS: Suitable

c) Crown base height and crown length

Feature description:

The crown base height is the distance from the tree root at which crown starts (considered as the lower insertion point of at least three consecutive live branches in a tree) and the crown length is its distance to the tip of the highest branch.

Methods for calculation:

• Separate points belonging to the stem from point belonging to crown. Use lowest crown point or layer.
• Use tree shape (vertical profile)
• Use vertical distribution of point density

Considerations:

• There is no clear definition of crown base height and, depending on tree species, it can be very challenging to assess visually.

Suitable instruments /technology:

• TLS: Suitable
• MLS: Suitable
• UAVLS: Only suitable if lower parts of crown and stems are not occluded
• ALS: Mostly unsuitable

d) Crown diameters

Feature description:

The horizontal extent of a tree. The crown diameter can be measured at the maximal extent of the crown, or for more detailed information, measured in multiple directions and at multiple heights.

Methods for calculation:

• Mean of the crown extent in x and y direction
• Based on a 2D convex hull (crown projected area), the crown diameter is computed as the mean distance of each point to the furthest point.
• Fitting a circle to the 2D projection 

Considerations:

• Highly influenced by the quality of single tree segmentation
• Crown diameter can be computed as the maximal extent or for any horizontal slice

Suitable instruments /technology:

• TLS: Suitable if crowns are not occluded and in range
• MLS: Suitable if crowns are not occluded and in range
• UAVLS: Suitable, if single tree segmentation is reliable
• ALS: Suitable, if single tree segmentation is reliable

e) Projected crown area

Feature description:

The crown projection area is the proportion of the forest floor that is covered by the vertical projection of the tree crowns.

Methods for calculation

• 2D Convex hull around single tree point cloud on the horizontal plane

Considerations:

• Results are highly dependent on the tree segmentation approach. A region growing or watershed approach can be used to directly derive 2D polygons of the crown area.

Suitable instruments/technology:

• TLS: Suitable if crowns are not occluded and in range
• MLS: Suitable if crowns are not occluded and in range
• UAVLS: Suitable, if single tree segmentation is reliable
• ALS: Suitable, if single tree segmentation is reliable

f) Crown (or full tree) 3D hulls

Feature description:

3D hull around the full segmented tree or only the points belonging to the crown.

Methods for calculation:

• Convex hull
• Alpha-shape: define degree of convexity by the value of alpha

Considerations:

• Results are highly dependent on the tree segmentation approach. 
• Can be a useful parameter to approximate the biomass stored in smaller branches.

Suitable instruments /technology:

• TLS: Suitable if crowns are not occluded
• MLS: Suitable if crowns are not occluded
• UAVLS: Suitable if lower parts of the tree are not occluded
• ALS: Suitable, if single tree segmentation is possible and lower parts of the tree are not occluded

• Variables derived from QSM
• Further CRS features 
• Available software for extracting metrics from CRS data: 
  • R packages: ITSMe, TreeLS
  • (link to software database by other working group)

3.7 CRS as ground truth (training data)

Many of the issues that have been raised regarding traditional allometries can only be addressed by increasing the amount of input data. The best ground truth data for this comes from destructive sampling, where trees are cut down and weighed. However, this is labour heavy, time consuming and has a considerable scaling cost for larger tree sizes, which causes a bias towards smaller trees in many such datasets. 

CRS techniques have therefore been proposed as a method to estimate the required ground truth data in a non destructive way. TLS specifically has shown a lot of promise and many studies have reported a high level of agreement between TLS derived biomass estimates and reference biomass data, in many different forest types (Calders et al., 2015, Demol et al., 2022, Disney et al., 2019, …). 

• destructive sampling
• AGB/V estimate from QSMs      
• voxels

Some text about validation is still to be added.

4. CRS data collection

In this section different platforms and their properties are described.

4.1 TLS

Terrestrial laser scanning (TLS) is a technique in which the scanner is deployed from stationary positions from the ground. Because the scanner is stationary, data results in less noisy point clouds compared to mobile platforms. Forests are usually scanned in a grid, in which the scanner is constantly moved between acquisitions. Eventually, all the single scans are co-registered into one point cloud. However, single scan applications also exist.

Advantages: 

• Highly Accurate 3D Data: TLS provides highly accurate and detailed 3D data, allowing for precise measurements of tree height, canopy structure, and vegetation density. Not only general size and shape of trees is captured, but it provides a clean representation of stem and (large) branches. Depending on the scanner and the distance to the scanner, leaves can also be captured.
• Non-destructive: TLS is a non-destructive technique (you still have to walk through the forest with a scanner so the soil can get compacted and small vegetation can be trampled), however it could be some disturbances in the field when campaigns are repeated over time.
• Relatively easy to scan: You just need a scanner, no extra permissions that are e.g. needed for drone flights, no pilot license needed, risk of breaking the equipment is small.

Disadvantages:

• Limited Field of View: The range of the laser scanner and its limited field of view may result in incomplete data capture, particularly in dense forests with overlapping canopies.
• Relatively High Cost: Terrestrial laser scanning equipment can be expensive to purchase and maintain. Additionally, data processing and analysis can require specialized software and expertise, adding to the overall cost.
• Time-Consuming Data Collection and Processing: TLS is a slow method. Forest plots are usually scanned in a grid of 10×10 m (tropical), 15×15 m (moderate) or larger, in which after each scan the equipment has to be moved. Depending on the complexity of the forest, it can take days to scan a hectare. The raw point cloud data generated by TLS can also be vast and require significant time and computational resources to process, register, and convert into usable forest information.
• Challenging Terrain: Forests can have uneven terrain and obstacles, making it difficult to position the scanner optimally for data collection, and creating risks for the field researchers. 
• Weather Conditions: Adverse weather conditions such as wind, rain, fog, or snow can affect the accuracy of TLS data acquisition and limit its practicality in certain situations.
• Closed source software: Majority, if not all, TLS processing software is closed source and typically requires an expensive license to be used.  

4.1.1 Scanner types

There is a wide range of TLS scanners. Depending on the price range and the application, an optimal device can be chosen. In general, we can split TLS scanners into 5 major instrument categories (Calders, 2022);

• Short-range TOF + large beam divergence (A)
• Mid-range TOF + medium beam divergence (B)
• Long-range PS + small beam divergence (C)
• Long-range TOP + medium beam divergence + low noise (D)
• Mid-range, dual wavelength + medium beam divergence (E)

Table X presents a comparison of the different categories:

Major Instrument Categories	Short-range TOF + large beam divergence	Mid-range TOF + medium beam divergence	Long-range PS + small beam divergence	Long-range TOF + medium beam divergence + low noise	Mid-range Dual Wavelength + medium beam divergence
Cost	$	$	$$	$$$	$$$
Ideal Forest condition	+ Sparse/simple forests + Remote areas	Sparse/simple forests + Remote areas	+ Leaf-off or structurally simple forest stands + Remote areas	+ Best in tall/dense forests	+ Accessible, structurally simple forest stands
Optimal Forestry Applications	+ Rapid assessment + Robust + Cost-effective forest structural metrics	+ Rapid assessment + Cost-effective forest structural metrics	+ Finely resolving small branches	+ Finely resolving small branches + Potential for full waveform applications	+ Leaf-wood separation+ Biochemical properties + Improved vertical foliage distribution+ Potential for full waveform applications
Example Instrument	UMB CBL (SICK Lidar; non-commercial)	Leica BLK360	FARO Focus3D X 330	RIEGL-VZ400i	SALCA (non-commercial)
Ranging method	TOF	TOF	PS	TOF	TOF
# returns	1st + 2nd	Single	Single	Multiple	Full waveform
Wavelength [nm]	905	830	1550	1550	1545.4 & 1063.4
Maximum Range [m]	40	0.6 60	0.6 330	1.5 – 250 (high speed) 0.5-800 (long range)	100 m
Samples/sec	11000	360000	122,000-976,000	42,000-500,000	5000
Beam Divergence [mrad]	15	0.4	0.19	0.35	0.56
Weight [kg]	3.9	1	5.2	9.7	17
Temperature range [deg C]	-30 to 50	5 to 40	5 to 40	0 to 40	5 to 30
References	Paynter et al. (2018, 2016)	Disney et al. (2019)	Liang et al. (2015); Pyörälä et al. (2018)	Bienert et al. (2018); Tian et al. (2019)	Danson et al. (2018); Schofield et al. (2016)

Tabel 1. Table summarizing different TLS instrument categories with examples. Derived from Calders et al  (2022) 

Examples commercial scanners:

RIEGL vz-400i (D)

datasheet: http://www.riegl.com/uploads/tx_pxpriegldownloads/RIEGL_VZ-400i_Datasheet_2022-09-27.pdf

Heavier option

RIEGL vz-600i (D)

datasheet: http://www.riegl.com/uploads/tx_pxpriegldownloads/RIEGL_VZ-600i_Preliminary-Datasheet_2023-05-24.pdf

RIEGL vz-2000i (D?)

datasheet:  http://www.riegl.com/uploads/tx_pxpriegldownloads/RIEGL_VZ-2000i_Datasheet_2022-09-27.pdf

Faro Focus 3D x330 (C)

datasheet: https://pdf.directindustry.com/pdf/faro-europe/tech-sheet-faro-laser-scanner-focus3d-x-330/21421-459177.html

Leica BLK 360 (B)

datasheet: https://shop.leica-geosystems.com/sites/default/files/2019-04/blk360_spec_sheet_2_0.pdf

Leica RTC 360

datasheet: https://leica-geosystems.com/-/media/files/leicageosystems/products/datasheets/leica-rtc360-lt-ds-897298-0821-en.ashx?la=de-ch&hash=D040AAFC2A5CF72279AF297A6BEFCAF4

TRIMBLE DEVICES (Daniel’s Proposal)

4.1.2 Specifications of the scanner types

Example instrument	RIEGL vz-400i	RIEGL vz-600i	RIEGL vz-2000i	LEICA BLK360	FARO FOCUS 3D x330	FARO FOCUS 3D 120
Ranging accuracy (cm)	5mm (1 sigma @ 100 m)	5mm (1 sigma @ 100 m)	5mm (1 sigma @ 100 m	6mm @ 10m / 8mm @ 20m
Wavelength (nm)	1550	NIR	NIR	830	1550
Minimum range (m)	1.5/1.2/0.5/0.5	1/0.5/0.5/0.5	2/1.5/1.5/1/1		0.6
Maximum range (m)	800/480/350/250 (rho>90%)	1000/420/320/220 (rho>90%)	2500/1850/1100/800/600 (rho>90%)		330
Beam Diameter				3.80		4.24
Beam divergence	0.25 mrad @ 1/e, 0.35 mrad @ 1/e2	0.25 mrad @ 1/e, 0.35 mrad @ 1/e2	0.19 mrad @ 1/e , 0.27 mrad @ 1/e2 1	0.68	0.19 mrad @ 1/e	0.27
Pulse repitition rate (kHz)	100/300/600/1200	140/600/1200/2200	50/100/300/600/1200
Effective measurement rate (meas.sec^-1)	42,000/ 125,000/ 250,000/ 500,000		21,000/ 42,000/ 125,000/ 250,000/ 500,000		122.000/ 244.000/ 488.000/ 976.000
max nr. of targets per pulse	15/15/8/4	15/15/10/5	15/15/15/8/4
Weight (kg)	9.7	6	9.8	1	5.2
FOV	vertical: 100°, horizontal: 360°	vertical: 105°, horizontal: 360°	vertical: 100°, horizontal: 360°	360?	vertical:300°,horizontal:360°
Cost	$$$	$$$	$$$	$	$$	$$?
References

4.1.3 Protocol how to scan

(also depends on type of forest, weather conditions, what you want to achieve – scanning is different if biomass or just dbh and tree height)

• Generic considerations
• How to scan at: tree level, plot level, stand level

Measurement protocol highly depends on the target variable of the measurement campaign. A good guideline on how to plan and execute a TLS campaign can be found in Wilkes et al. (2017).

4.2 MLS

(zeb horizon, zeb revo, greenvalley backpack, stonex)

Mobile Mapping System (MMS) dates to the 1980s when the Centre for Mapping at the Ohio State University developed GPSVan, the first MMS. The concept of mobile laser scanning systems (MLS) regards the direct georeferencing integrating the laser technology together with GNSS/ISN. 

Mobile LiDAR term is widely used for a laser scanner deployed on any mobile platform. In fact, MMS can be classified according to the platform onboard such as vehicle, van, boat. MLS systems can also be carried personally, either in hand or in a backpack. The major drawback of using handheld, backpack, and ATV-based MLS is the low quality of the Global Navigation Satellite System (GNSS) signals inside the forest canopies causing georeferencing errors and thus additional noise to the point cloud data. Therefore, simultaneous localization and mapping (SLAM) technology has been applied in inside-canopy MLS systems. Commercial systems applying SLAM include, e.g., HERON Lite (Gexcel, Brescia, Italy), Frontier (NavLive, UK), Stencil-1/2 (Kaarta, Pittsburgh, USA), GeoSLAM Zeb-Revo and -Horizon and Zebedee systems (GeoSLAM, Brisbane, Australia).

MLS point cloud quality is highly dependent on the quality of the used SLAM solution, which can vary a lot depdending on the used algorithms and situation where SLAM is applied (e.g. forest structure). Another drawback of using MLS is the lower density and accuracy of point-cloud data compared to static TLS data. Therefore, characterizing different tree compartments, such as stem, branches and leaves is not as accurate as with TLS . Nevertheless, key forest inventory measures including tree DBH, total height, the number of stem, basal area, stem volume, and canopy cover can be successfully derived from MLS data often at the plot level even in natural forest setting. Good quality MLS data can also be applied to similar applications as TLS data, but with caution.

Advantages

• Fast data acquisition: current MLS devices have a fast measuring rate with up to 600’000 pts/second. In combination with a moving acquisition approach, larger areas can be acquired in a much faster way than e.g. TLS. A 1/4 ha plot can typically be acquired within 10-15 minutes. 
• Easy data acquisition: Data acquisition is very easy and does not require a lot of experience or know-how to perform. However, a few guid-lines should be followed, such as: no sudden eratic movements, try to make a closed loop etc.
• Non-destructive: As with all lidar devices, the mere lidar acquisition is non-destructive. However, when walking through the forest with the MLS device, damage to the environment (trampling of ground vegetation, broken branches etc.) can still occur.

Disadvantages

• Noisy data: Currently, MLS point clouds are still relatively noisy and have a lower precision than TLS point clouds. This can result in errors for DBH extraction or that the diameters of smaller trees or branches cannot be extracted.
• Shifts/Drifts in point-clouds: especially handheld and Backpack MLS systems rely on the principle of simultaneous localization and mapping (SLAM), which relies on clear distinguishable geometric features (i.e. planes, edges). As forests naturally have a limited amount of such features, the SLAM approach can fail. This often manifests in shifts in the point cloud, which can be recognized by duplicated trees or missaligned ground points. So far, no clear rules of when these shifts occure were recognized. Mokros et al. (2021) stated, that overly complicated acquisition paths with multiple crossings in trajectory could actually promote such drifts, even though, intuitively, one would think that this would actually prevent such drifts. 
• Black-box data processing: SLAM processing of the mostly propriatary processing software is often a bit of a black-box and therefore hard to control.
• Expensive: Even-though prices of new MLS devices came down a bit in recent years, they are still a quite expensive investment, often in the same area as TLS.

4.2.1 Scanner types

• Terrestrial Mobile Laser Scanners (MLS): These scanners are vehicle-mounted systems composed of GPS and IMUs to capture highly accurate 3D data on the surrounding environment while the vehicle is in motion. MLS allows rapid data acquisition over large areas. Such as, for example:

• Lynx Mobile Mapper- Optech: The system has a maximum range of 200 m, a full circle 360º angular coverage, 500 kHz of pulser measurement rate and a scan rate of 100 Hz. In addition, the system control software enables the camera image frame size selection for highly efficient image capture. This system provides one of the most accurate commercially available. 
• VMX-250- RIEGL: The system integrates two RIEGL VQ-250 scanners, inertial and satellite navigation hardware, and mounting points for digital cameras or video equipment. Each scanner provides a full circle of 360º scan within its 2D scanning plane and can measure ranges of 200 m (with 80% reflectivity) with a 300 kHz for a PRR (Pulse Repetition Rate) and 100 scan lines per second. 

• Backpack Mobile Laser Scanners: These types of scanners are designed to be carried by an operator. These MLS are lightweight and portable, allowing the operator to capture 3D data while walking through the forest. These scanners are commonly used for mapping environments where traditional vehicle-mounted scanners are impractical. Some examples of the most common scanners:

• Greenvalley backpack D50: This system has two laser scanners mounted on the backpack vertically and horizontally, the main scanner collects data horizontally while the secondary scanner collects data vertically. The laser sensor is a velodyne VLP-16 x 2 with a LiDAR accuracy of  3cm, 100 m of scan range and a scan rate of 600,00 pts/s.
  • Akhaka-R3 Backpack: The backpack laser scanner is equipped with a Riegl VUX-1HA 2D laser scanner and miniVUX-1UAV 2D, a LITEF UIMU-LCI inertial measurement unit, and a positioning system consisting of a NovAtel Felxpak6 GNESS receiver and GGG-703 antenna. The ranging accuracy is 1-1,5 cm with a scan frequency of 250&100 Hz. 

• Handheld Mobile Laser Scanners: These scanners can be operated by hand, are compact and lightweight. Generally, they are used for smaller-scale scanning tasks or in areas where mobility is essential. 

• GeoSLAM Zeb horizon: This system is a lightweight and versatile scanner that incorporates a rotating LiDAR sensor, SLAM algorithms, and high-resolution cameras to create accurate 3D point clouds. The system provides 300,000 points per second with a maximal range of 100 m and accuracy of 1-3 cm. Thanks to the attached cameras (ZEB Vision) to the system, 4K video records or 360degree panaromic images of the environment can be captured during data collection.
  • Zeb revo: This system utilizes SLAM technology to capture 2D data in real-time. It features a rotating LiDAR sensor with 360º field of view, allowing for complete coverage of the surrounding environment. The system captures up to 40,000 points per second and offers a range of approximately 30 m and relative positional accuracy of 2-3 cm depending on the environment conditions. 

4.2.2 Aspects of the scanner types[M(40] 

Categories	MLS		Backpack		Handheld
Example instrument	Lynx laser scanner	VQ-250	Akhka R3	Greenvalley D50	Zeb Horizon	Zeb revo
System	Lynx mobile mapper	VMX-250- RIEGL	Riegl VUX-1HA NovAtel Felxpak6, LITEF UIMU-LCI	Velodyne VLP-16 x 2	Velodyne Punck VLP-16	Hokuyo UTM-30LX-F
Ranging accuracy (cm)	+- 1	+- 0,5	+-1	+-3	+-3	+-3
Wavelength (nm)	905	Near infrared	1550	905	903	905
Maximum range (m)	200 m (p80%)	200 m (p80%)		100	100	30
Data Acquisition rate	500 000 points / sec	1MPoints /sec	~10/min	600 000points /sec	300 000 points/sec	43 200 points /sec
Weight (kg)	78	11	50	8.8	2.4	4.1
FOV	360º without gaps	360º without gaps		360	130º	270º x 360º
Density of points (points/m2)	(Speed and sensor distance)	(Speed and sensor distance)	6 x 104	2x 104	2x 104-4x 104	2x 104-4x 104
Cost
References	PUENTE, Iván, et al. Review of mobile mapping and surveying technologies. Measurement, 2013, vol. 46, no 7, p. 2127-2145.	PUENTE, Iván, et al. Land-based mobile laser scanning systems: a review. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2012, vol. 38, p. 163-168.	HYYPPÄ, Eric, et al. Comparison of backpack, handheld, under-canopy UAV, and above-canopy UAV laser scanning for field reference data collection in boreal forests. Remote Sensing, 2020, vol. 12, no 20, p. 3327.	Levick, S. R., Whiteside, T., Loewensteiner, D. A., Rudge, M., & Bartolo, R. (2021). Leveraging TLS as a Calibration and Validation Tool for MLS and ULS Mapping of Savanna Structure and Biomass at Landscape-Scales. Remote Sensing, 13(2), Article 2. https://doi.org/10.3390/rs13020257	CHUDÁ, J., et al. The handheld mobile laser scanners as a tool for accurate positioning under forest canopy. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2020, vol. 43, p. 211-218.	CHUDÁ, J., et al. The handheld mobile laser scanners as a tool for accurate positioning under forest canopy. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2020, vol. 43, p. 211-218.

4.2.3 Protocol how to scan

• MLS

The data acquisition integrates the GPS/ISN system which calculates the trajectory and records the laser scanner input, the laser scanner which measures two-dimensional points, and the camera system which captures scenery images synchronously with the DMI displacement. 

The first step is the computation of the vehicle trajectory, so the MSL must remain totally stationary before and after the survey for INS alignment to match the position with GPS and to estimate the alignment variables. Following, software packages process GPS data of MLS which requires a reference station which should be located on a known vertex in a barycenter position with respect to the underground track. Being ETRS89 coordinates of this vertex belongs to the National Reference Network in Europe. Then, in the georeferencing phase, each laser point recorded is assigned an appropriate time stamp, trajectory information of the location, and altitude of the scanner converting in three dimensions. 

• Backpack and handheld

The protocol to scan with a handheld mobile laser scanner may consist of different steps depending on the type of forest inventory. The pattern to scan will strongly depend on the design of the plot (i.e. circular vs quadratic) as well as the size of the plot. For example, for areas covering more thatn 1/4 hectare one might consider multiple acquisition paths to cover the entire area while reducing the risk of adding drift to the data, due to a too long acquisition time. Then, a strategy to combine these individual scans needs to be defined (e.g. scan covering a mutual area). 

The acquisition pattern is dependent on the shape, dimension, terrain undulations and forest structure complexity of the plot. In Figure 2 some examples of different acquisition patterns for circular plot shapes are given.

Figure 2: Example of acquisition patterns mainly for circular plots as published in (a) Gollob et a. (2020), (b) Bauwens et al. (2016), and (c) Del Perugia et al (2019). For (c): left: tree centric acquisition path, where one walkes around each tree, middle: grid pattern with 10 m distance between traversing lines, right: grid pattern with 15 m distance between traversing lines.

In Figure XY some examples of different acquisition patterns for rectangular plot shapes are given. The firest example (b)  is taken from a relatively easy 50 x 50 m2 plot, where the targeted grid pattern could be followed quite closely. However, when plot conditions becamoe more difficult, e.g. due to dense understorey vegetation (example (c) in Figure 3) or the terrain is very steep (example (d) in Figure 3), following the targeted acquisition pattern becomes difficult. Also navigation within the forest with very dense understorey vegetation is not easy. For steep plots it is beneficial to align the grid pattern in a way to make walking in this terrain easier, i.e. by walking the traversing lines along the contour line of the terrain to minimise the elevation gain of the acquisition (see example (d) in Figure 3).

Figure 3: Exampl es of a gridded acquisition pattern as applied to a flat plot with no understorey vegetation (a), a plot with very dense understorey (c) and a very steep plot with gradients > 90% with cliffes within the plot. Figure taken from SilviLaser Contribution by Kükenbrink et al. (2023). The different colored trajectories come from different operators.  

Especially hand-held mobile laser scanning systems are often not equiped with a GPS receiver and hence, the acquired point cloud is only stored in local coordinates. If absolute coordinates are needed, additional efforts have to be taken. Certain Scanning systems, such as from the GeoSLAM family, offer the possibility to add a reference plate to the bottom of the handlebar. This reference plate is equiped with a marking, where the exact offset from the scanner is known. By holding this marking of the reference plate onto a point with known coordinates, the location is marked as a reference point  by the system. During processing of the point cloud, the absolute coordinates can be assigned to this reference point. By measuring multiple of such reference points, an absolute georeferencing can be achieved. Alternatively, reference targets, such as spheres or reflective cylinders could be setup throughout your study area, for which the exact locations have been measured with a GPS device. These reference targets can later be detected in the point cloud and through the known coordinates of the targets, a transformation matrix to transform the point cloud to absolute coordinates can be extracted. Tools for this are for example implemented in the freely available CloudCompare software toolbox. Another approach to absolutely reference your MLS point cloud would be by aligning the acquired point cloud to another point cloud with absolute coordinates and through this acquire the transformation matrix to rotate your point cloud.

4.3 Photogrammetric point clouds[DK(49] [DK(50] 

Photogrammetry can be an interesting and often cheaper alternative to LiDAR devices for the acquisition of 3D point cloud data. The principle of deriving 3D information from two or more overlapping images has already seen a multitude of applications over the last decades. The derivation of canopy heights from stereo-images is just one of these applications. While traditional photogrammetry applications focused on the derivation of surface models from large scale airborne image acquisitions, the advent of consumer grade drones made the acquisition of 3D point clouds and surface models from small and affordable drones popular in research as well as in industry. The derivation of 3D point clouds from drones relies on the principle of so-called structure from motion (SfM), where 3D information can be retrieved from multiple, overlapping images which will be aligned during the processing of the images. As the same object is observed from multiple angles, 3D information of this object can be derived. 

In recent years, the same principle has also been adapted to images taken with cameras on the ground, where an object is captured in 3D using e.g. a hand-held camera. This approach has especially received some attention in the research field of archeology or cultural heritage conservation and restoration, where ancient sculptures, buildings etc. were captured with thousands of images to extract a 3D model in order to conserve its historic and cultural value. But also in forestry, terrestrial photogrammetry received some attention in recent years, where consumer grade cameras were used to acquire 3D information of the lower part of the canopy. The main application were the detection of trees and the derivation of basic tree parameters as e.g. DBH. But also more small-scale applications can be found in literature, where micro-habitats on tree trunks (e.g. mosses, lichens, cavities etc.) were assessed using this measurment principle.

Advantages

• Relatively cheap: especially when compared to most LiDAR devices
• Lightweight: Depending on the setup, this approach is particularly lightweight. However, if you go for full-frame cameras and maybe even go for a multi-camera setup, the weight will become considerable.
• Broadly available: Commercial cameras are often easier to acquiere than LiDAR devices
• High density point clouds: terrestrial photogrammetry can deliver very high density and detailed point clouds (depending on the camera resolution and processing settings). 

Disadvantages

• High processing demands: Demanding in terms of processing power
• High costs go into processing: A performant processing computer is expensive
• Point cloud creation not always robust: Alignment of images is not always succesful, whereas the success can only be assured once you are back in the office
• Sensitive to environmental conditions: Quality of point cloud and success of image alignment is highly sensitive to environmental conditions (i.e. light conditions, see also Section 4.3.1)
• Problems with distortions: Cameras with a large field of view (i.e. fish-eye lenses) have the advantage of an increased overlap between neighboring images, however, the distortion in the image impair the image alignment.
• Raw point clouds are not scaled: Accurate scaling of the point clouds involves a bit of an effort. Either by placing markers with known distance or an actual scale bar into the scene or by using a multi-camera setup with known distance between the cameras (the so-called baseline).
• Arbitrary orientation (for terrestrial photogrammetry): For terrestrial photogrammetry, orientation of the camera during acquisition is not known. This results in an arbitrary orientation of the point cloud after processing. A way to orient the point cloud needs to be found, e.g. by placing a pole with knwon orientation in the scene, by using objects with known orientation (e.g. a tree that is growing exactly vertically) or by using an existing point-cloud with correct orientation as a reference.

4.3.1 Terrestrial photogrammetry

For the application of terrestrial photogrammetry, basically any camera can be used. Examples of used cameras systems for forestry applications range from full-frame cameras down to very small and affordable action cameras such as GoPros. However, the specifications of the cameras influence the quality and potentially the success in deriving point clouds from the images. For instance, cameras with higher resolution can produce more detailed point clouds. Cameras with large distortion (e.g. fisheye cameras) can potentially be more problematic during the processing of the images. 

• data acquisition

            Data acquisition path generally depends on the main objective of the acquisition. If only a single tree should be covered, a or multiple circular patterns around the tree with different distances from the tree is often the best solution. For plot level acquisitions different approaches have been employed and various experiences from the different acquisition patterns have been reported. Some state to follow a regular grid pattern while others recommend to walk in circular pattern in order to promote more angles between matching image pairs. A thorough analysis on best practices regarding acquisition patterns have so far not been performed. However, such an analysis would also be difficult to be performed in a generic manner so that a final recommendation could be made, as different forest characteristics largely affect the quality of the derived point cloud.

A major point of consideration when designing a terrestrial photogrammetry campaign is the accurate scaling of the acquired point cloud. As accurate positioning with potentially built-in GPS receiver of the camera system is generally unavailable, a scaling through the exact image position as performed with e.g. drone-based acquisitions, is not possible in forests. Different alternatives for an accurate scaling of the point cloud need to be defined. The most-basic approach would be the setup of scale-bars visible throughout the acuqisition. This could be an actual scale-bar somewhere in the acquisition or also by using coded markers with known distance between two markers that can be automatically detected. From these markers an accurate scaling of the point cloud is possible. Alternatively, one could also use two cameras with an exactly known distance between each other that acquire images at the same time. Through the known distance between the cameras, an accurate scaling of the point cloud without the need of setting up markers on the plot, can be achieved. Further consideration needs to be taken towards the orientation of the point cloud. Contrary to drone-based photogrammetry, where images are often taken in nadir direction, the processing software does not know the exact orientation of the images, hence also the orientation of the final point cloud is not known. Orientation can be achieved also by using the setup markers, if exact positions of these markers are known (at least the position of 3 markers need to be known for that). Another approach would be, to have a reference for the vertical dimension somewhere in the acquisition. This could be for instance a pole which is perfectly levelled to point directly upward. The point cloud could then be rotated so that the z-axis matches the direction of this pole. If you know that a or multiple trees in your acquisitions grow perfectly vertical, you could also use these as your reference. Another approach would be to align the SfM point cloud to a reference point-cloud from e.g. TLS, MLS or UAVLS acquisitions which should be perfectly oriented.

Depending on the camera used for data acquisition, both images or also videos could be used. The latter could be of benefit when using two (or more) cameras simultaneously, as the synchronisation of image acquisition between the devices is easier for videos (e.g. through a visual or audio clue visible/audible in all videos) than for image based acquisition, where some sort of precise synchronisation scheme need to be employed (e.g. through an intervalometer as seen in e.g. Mokros et al. 2021).

Terrestrial photogrammetry is highly sensitive to light conditions, which can be especially difficult in forests, where generally very few light reaches the bottom of the canopy and light conditions can vary quickly and strongly over short periods of time. It is therefore recommendet to acquire images preferrably under overcast conditions, where lighting conditions are generally more stable. For the low light conditions, a longer exposure time or an increase in ISO can be an option to increase the contrast and therefore the success rate image alignment. However, when images are taken under movement, the exposure time should be low in order to reduce blurring effects. Also ISO should not be set too high or the noise level could be a problem. When images are taken under movement, an option could also be to use video acquisitions. 

Processing of the acquired images can be performed in many different commercial or freely available software solutions. Metashape is one of the most commonly used commercial solution and offers an intuitive and easy to use GUI, making the first steps in SfM point cloud creation easy. However, if hundreds or thousands of images should be processed, i relatively high performing computer should be used in order to keep the processing time to a minimum. 

4.4 UAVLS

UAV based operations are optimal for landscape to local scale mapping operations. That said, there are also UAVs that have the capability to be used in a similar manner to manned aircrafts, and carrying relatively large payloads. 

The used UAV system and it’s maximum takeoff weight (MTOW) largely controls what type of mapping payload can be caried onboard.  Typically the carried LiDAR sensors are same or similar to the sensors used in MLS applications.  Output data quality is highly dependent on the capabilities of the used sensor and the operating parameters, such as flight altitude,  pattern and speed.  In addition, an accurate trajectory is required for producing good quality point cloud. Trajectory quality depends on the used INS system and used algorithms. Some  providers use SLAM in trajectory creation, for example the ZebHorizon can be used in UAVs as well. 

Advantages

• Fast data acquisition and large area coverage: Larger areas can be covered in relatively short period of time
• Non-destructive: As with all lidar devices, the mere lidar acquisition is non-destructive. UAVLS are even less destructive than TLS or MLS as the observer does not have to go into the area of interest at all.
•  

Disadvantages

• Often Expensive: Depending on the features and sensors used, UAVLS systems can be very expensive. Some cheaper and more affordable solutions from e.g. DJI (L1 lidar sensor) or GeoSLAM (ZebHorizon with the UAV addon) are now available, however these often come at the cost of precission and accuracy.
• Relatively complex flight planning: Flight planning can be difficult, depending on the area where the drone is operated and should be done carefully to avoid an expensive crash.
• Legal restrictions: Depending on the country the UAVLS system is operated, there are more and more legal restrictions for flying drones, which need to be considered during flight and campaign planning.
• Heavy weight: especially the Riegl UAV lidars are quite heavy, needing a heavy-weight drone (e.g. DJI matrice M600pro) to lift them. This can make drone campaigns difficult to plan and execute
• Limited battery capacity: When flying with batteries, flight time can be quite restrictive, limiting the area covered as well as the accessibility of more remote areas with no access. For example, the maximum flight time of a Riegl miniVUX LiDAR mounted on a Matrice M600 pro is around 15-16 minutes. This shourt flight time mostly suffice for a detailed acquisition of a one hectare area.
• Occlusion in lower canopy: Above canopy UAVLS often encounter substantial amount of occlusion in the lower canopy part. This hinders certain applications such as tree base position or DBH estimation. However, UAVLS can be a great source to complement ground based acquisitions which often encounter occlusion in the upper part of the canopy (e.g. see Figure below or Schneider et al. (2019)

Illustration from Terryn et al, 2020. TLS data derived with a RIEGL vz-400, UAVLS data captured with a  RIEGL RiCOPTER equipped with the VUX-SYS system.

Examples commercial scanners:

RIEGL miniVUX series:

website: http://www.riegl.com/products/unmanned-scanning/riegl-minivux-3uav/

RiEGL VUX series:

Website: http://www.riegl.com/products/unmanned-scanning/new-riegl-vux-18024/

DJI Zenmuse L1:

Typically mounted on a DJI Matrice M300

Website: https://enterprise.dji.com/zenmuse-l1

GeoSLAM ZebHorizon with UAV addon:

Website: https://geoslam.com/solutions/zeb-horizon/

Emesent Hovermap series:

Website: https://emesent.com/hovermap-series/

---