LiDAR for Forestry

Archaeological applications using airborne laser scanning (ALS) are increasing in number. Since the production of ALS-derived digital terrain models (DTM) involves a considerable amount of money, most applications use general purpose ALS data, which are usually cheaper and sometimes even provided for free for scientific applications. The main problem that comes with this kind of data is the frequent lack of meta-information. The archaeologist often does not get the information about original point density, time of flight, instrument used, type of flying platform, filter and DTM generation procedure etc. Therefore, ALS becomes a kind of “black box”, where the derived DTM is used without further knowledge about underlying technology, algorithms, and metadata. Consequently, there is a certain risk that the data used will not be suitable for the archaeological application.
Based on the experience of a two-year project “LiDAR-Supported Archaeological Prospection in Woodland”, the paper will give a review on archaeological ALS, explain its the basic process, demonstrate its potential for landscape archaeology especially in densely forested areas, and draw the attention to some critical parameters of ALS, which should be known to the user. Finally, further issues, which need to be solved in near future, are discussed.

Background and Purpose: Unmanned aerial vehicles (UAVs) are flexible to solve various surveying tasks which make them useful in many disciplines, including forestry. The main goal of this research is to evaluate the quality of photogrammetry-based digital surface model (DSM) from low-cost UAV’s images collected in non-optimal weather (windy and cloudy weather) and environmental (inaccessibility for regular spatial distribution of ground control points - GCPs) conditions. Materials and Methods: The UAV-based DSMs without (DSMP) and with using GCPs (DSMP-GCP) were generated. The vertical agreement assessment of the UAV-based DSMs was conducted by comparing elevations of 60 checkpoints of a regular 100 m sampling grid obtained from LiDAR-based DSM (DSML) with the elevations of planimetrically corresponding points obtained from DSMP and DSMP-GCP. Due to the non-normal distribution of residuals (vertical differences between UAV- and LiDAR-based DSMs), a vertical agreement was assessed by using robust measures: median, normalised median absolute deviation (NMAD), 68.3% quantile and 95% quantile. Results: As expected, DSMP-GCP shows higher accuracy, i.e. higher vertical agreement with DSML than DSMP. The median, NMAD, 68.3% quantile, 95% quantile and RMSE* (without outliers) values for DSMP are 2.23 m, 3.22 m, 4.34 m, 15.04 m and 5.10 m, respectively, whereas for DSMP-GCP amount to -1.33 m, 2.77 m, 0.11 m, 8.15 m and 3.54 m, respectively. Conclusions: The obtained results confirmed great potential of images obtained by low-cost UAV for forestry applications, even if they are surveyed in non-optimal weather and environmental conditions. This could be of importance for cases when urgent UAV surveys are needed (e.g. detection and estimation of forest damage) which do not allow careful and longer survey planning. The vertical agreement assessment of UAV-based DSMs with LiDAR-based DSM confirmed the importance of GCPs for image orientation and DSM generation. Namely, a considerable improvement in vertical accuracy of UAV-based DSMs was observed when GCPs were used.

Forests are one of the most important sinks for carbon. Estimating the amount of carbon stored in forests is a major task for understanding the global carbon cycle. From local to global scales, remote sensing has been extensively used for forest biomass estimation. With the availability of multisensor image data, fusion has become a valuable method in remote sensing applications. Light detection and ranging (LiDAR) can provide information on the vertical structure of forests, whereas hyperspectral images can provide detailed spectral information of forests. Effective fusion of LiDAR and hyperspectral data is expected to help extract important biophysical parameters of forests. However, it is still unclear as to how forest biophysical and biochemical attributes derived from hyperspectral data relate to structural attributes derived from LiDAR data. A summary of previous research on LiDAR-hyperspectral fusion for forest biomass estimation is valuable for further improvement of biomass estimation methods. A review on the status of hyperspectral data, LiDAR data, and the fusion of these two data sources for forest biomass estimation in the last decade is provided. Some future research topics and major challenges are also discussed.

1. Light detection and ranging (LiDAR) technology provides ecologists with high-resolution data on three-dimensional vegetation structure. Large LiDAR datasets challenge predictive ecologists, who commonly simplify point clouds into structural attributes (namely LiDAR-based metrics such as canopy height), which are used as predictors in ecological models, potentially with loss of relevant information. 2. We illustrate an efficient alternative approach to reduce the dimensionality of LiDAR data that aims at minimal data filtering with no a priori assumptions on the ecology of the target species. We first fit the ecological model exploiting the full variability in the LiDAR point cloud, then we explain the results using post-modelling LiDAR-data classification for ecological interpretation only. This is the classical logic of explorative, hypothesis generating and predictive statistics, rather than testing specific vegetation-structural hypotheses. 3. First, we reduce the dimensionality of the LiDAR point cloud by principal component analysis (PCA) to fewer predictors. Second, we show that LiDAR-PCs are capable to outperforming commonly used environmental predictors in ecological modelling, including LiDAR-based metrics. We exemplify this by modelling red deer (Cervus elaphus) and roe deer (Capreolus capreolus) resource selection in the Bavarian Forest National Park, Germany. After fitting the ecological model, we provide an interpretation of the information included in LiDAR-PCs, which allows users to draw conclusions whenever using them as predictors. We make use of the PCA rotation matrix and post-modelling data classification, and document deer selection for understorey vegetation at unprecedented fine scale. 4. Our approach is the first attempt in animal ecology to avoid the use of LiDAR-based metrics as model predictors, but rather generate principal components able to capture most of the LiDAR point cloud variability. Our study demonstrates that LiDAR-PCs can boost ecological models. We envision a potential use of LiDAR-PCs in several applications, particularly species distribution and habitat suitability models. We demonstrate an application of our approach by building suitability maps for both deer species, which can be used by practitioners to visualize model spatial predictions and understand the type of forest structures selected by deer.

Inventories of temperate forests of Central Europe mainly rely on terrestrial measurements. Rapid alterations of forests by disturbances and multilayer silvicultural systems increasingly challenge the use of conventional plot based inventories, particularly in protected areas. Airborne LiDAR offers an alternative or supplement to conventional inventories, but despite the possibility of obtaining such remote sensing
data, its operational use for broader areas in Central Europe remains experimental. We evaluated two methods of forest inventory that use LiDAR data at the landscape level: the single tree segment-based method and an area-based method. We compared a set of structural forest attributes modeled by these
methods with a conventional forest inventory of the highly heterogeneous forest of the Bavarian Forest National Park (Germany), which partially includes stands affected by severe natural disturbances. Area-based models were accurate for all structural attributes, with cross-validated average root mean
squared error ranging from ∼3.4 to ∼13.4 in the best modeling case. The coefficients of variation for the mapped area-based estimations were mostly minor. The area-based estimations were varied but highly correlated (Pearson’s correlations between∼ 0.56 and 0.85) with single tree segmentation stimations; undetected trees in the single tree segmentat-based method were the main sources of inconsistency. The single tree segment-based method was highly correlated (∼ 0.54 to 0.90) with data from ground-based
forest inventories. The single tree-based algorithm delivered highly reliable estimates for a set of forest structural attributes that are of interest in forest inventories at the landscape scale. We recommend LiDAR forest inventories at the landscape scale in both heterogeneous commercial forests and large protected
areas in the central European temperate sites.

inventory
by measuring systematically distributed sample plots and mapping forest stands. In recent years, remote-sensing
methods have been developed that show a high potential for forest applications and are capable of registering
area-wide forest structure data. Algorithms have been developed that automatically extract relevant
information from these data. We used a fully automated analysis system developed over a ten-year period to
detect and measure single trees and estimate area-based forest structures of the entire area of the Bavarian
Forest National Park at a measurement density of 30 – 40 per m². More than 13 million single trees were detected
and their properties of height, crown volume, starting point of the crown, species group, dead wood, wood
volume, and diameter at breast height were determined. Both single tree detection and area-based forest structure
estimates yielded good results for important structural variables, with coeffi cients of determination between
0.6 and 0.9. Two drawbacks of the remote-sensing methods were that only coniferous and deciduous trees could
be distinguished and trees in old deciduous stands were over- segmented. Based on the results of this study,
the national park administration has developed a new concept for forest inventories, which combines remotesens
ing and ground-based inventories of reference stands, regeneration surveys, and inventories of biodiversity

A B S T R A C T Lidar transforms how we map ecosystems, but its prospect for measuring ecosystem dynamics is limited by practical factors, such as variation in lidar acquisition and lack of ground data. To address practical use of multitemporal lidar for forest and carbon monitoring, we conducted airborne lidar surveys four times from 2002 to 2012 over a region in Scotland, and combined the repeat lidar data with field inventories to map tree growth, biomass dynamics, and carbon change. Our analyses emphasized both individual tree detection and area-based, grid-level approaches. Lidar-detected heights of individual trees correlated well with field values, but with noticeable underestimation biases (r = 0.94, bias = −1.5 m, n = 598) due to the increased probability of missing treetops as pulse density decreases. If not corrected for such biases, lidar provided unrealistic or wrong estimates of tree growth unless laser sampling rates were high enough (e.g., > 7 points/m 2). Upon correction, lidar could detect sub-annual tree growth (p-value < 0.05). At grid levels, forest biomass density was reliably estimated from area-based lidar metrics by both Random Forests (RF) and a linear functional model (r > 0.86, RMSEcv < 21 Mg/ha), irrespective of laser sampling rates. But RF constantly overfit the data, often with poorer predictions. The better generality of the linear model was further confirmed by its transferability—fitted for one year but applicable to other years—a strength not possessed by RF but desired to alleviate the reliance on ground biomass data for model calibration. Resultant lidar maps of forest structure captured canopy dynamics and carbon flux at fine scales, consistent with growth histories and known disturbances. The entire 20-km 2 study area sequestered carbon at a rate of 0.59 ± 0.4 Mg C/ha/year. Overall, our study describes robust techniques well suited for multitemporal lidar analysis and affirms the utility and potential of repeat lidar data for resource monitoring and carbon management; however, the full potential cannot be attained without the support of accompanying field surveys or modeling efforts in enhancing stakeholders' trustworthiness of lidar-based inference .

Downed dead wood is regarded as an important part of forest ecosystems from an ecological perspective, which drives the need for investigating its spatial distribution. Based on several studies, Airborne Laser Scanning (ALS) has proven to be a valuable remote sensing technique for obtaining such  nformation. This paper describes a unified approach to the detection of fallen trees from ALS point clouds based on merging short segments into whole stems using the Normalized Cut algorithm. We introduce a new method of
defining the segment similarity function for the clustering procedure, where the attribute weights are learned from labeled data. Based on a relationship between Normalized Cut’s similarity function and a class of regression models, we show how to learn the similarity function by training a classifier. Furthermore, we propose using an appearance-based stopping criterion for the graph cut algorithm as an alternative to the standard Normalized Cut threshold approach. We set up a virtual fallen tree generation
scheme to simulate complex forest scenarios with multiple overlapping fallen stems. This simulated data is then used as a basis to learn both the similarity function and the stopping criterion for Normalized Cut. We evaluate our approach on 5 plots from the strictly protected mixed mountain forest within the Bavarian Forest National Park using reference data obtained via a manual field inventory. The experimental
results show that our method is able to detect up to 90% of fallen stems in plots having 30–40% overstory cover with a correctness exceeding 80%, even in quite complex forest scenes. Moreover, the performance for feature weights trained on simulated data is competitive with the case when the weights are calculated using a grid search on the test data, which indicates that the learned similarity function
and stopping criterion can generalize well on new plots.

The main purpose of the present research is to verify whether medium resolution LiDAR data, not forest-specific, can be used to carry out statistical models for estimating timber volume.
LiDAR data with different characteristics, and taken in different seasons (winter and summer) on the same forest area (Foresta di Paneveggio, Trentino, Italy), were processed, and an estimation model using winter low-resolution data was performed. Such model was then applied to a different territory, having similar forest characteristics. Moreover, LiDAR and ground data, surveyed during the management plan inventory, were integrated.
The final model allows to obtain good volume estimates with fair precision if applied at forest compartments level, and can produce detailed timber volume maps, very useful when planning forestry operations.

This study proposes a new metric called canopy geometric
volume G, which is derived from small-footprint lidar data,
for estimating individual-tree basal area and stem volume.
Based on the plant allometry relationship, we found that
basal area B is exponentially related to G (B  1G3⁄4, where
1 is a constant) and stem volume V is proportional to
G (V  2G, where 2 is a constant). The models based on
these relationships were compared with a number of models
based on tree height and/or crown diameter. The models
were tested over individual trees in a deciduous oak woodland
in California in the case that individual tree crowns are
either correctly or incorrectly segmented. When trees are
incorrectly segmented, the theoretical model B  1G3⁄4 has
the best performance (adjusted R2,  0.78) and the model
V  2G has the second to the best performance (  0.78).
When trees are correctly segmented, the theoretical models
are among the top three models for estimating basal area
(  0.77) and stem volume (  0.79). Overall, these
theoretical models are the best when considering a number
of factors such as the performance, the model parsimony,
and the sensitivity to errors in tree crown segmentation.
Further research is needed to test these models over sites
with multiple species.

This study scrutinises the use of terrestrial laser scanning (TLS) to measure diameter at breast height (DBH) and tree height at individual tree species level. LiDAR point cloud scans are collected from uniformly defined control points. The result of processed TLS data demonstrates the precise measurements of tree height and DBH by comparing it with field data (DBH, tree height, tree species and location). The average tree height and DBH obtained through TLS measurements were 9.44 m and 43.30 cm, respectively. A linear equation between TLS derived parameters and field measured values were established, which gave the coefficient of determination (r2) of 0.79 and 0.96 for tree height and DBH, respectively. Further, these parameters were used to calculate above ground biomass (AGB) for individual tree species by considering a non-destructive approach. The total AGB and carbon stock from 80 different trees are computed to be 49.601 and 22.320 tonnes, respectively.

The purpose of this study was to test a method for delineating individual tree crowns based on a fully automated recognition methodology. The study material included small-footprint time-of-flight laser scanner data acquired in the spring and summer of 2002. The data were collected with a Toposys II airborne laser system flown over the Norway spruce (Picea abies) and European beech (Fagus sylvatica) dominated forests of the Bavarian Forest National Park, Germany. The applied algorithm, which earlier had been validated for Swedish forest conditions, is a watershed algorithm that is based on the use of laser scanning data. 2584 trees in a total of 28 representative reference stands, each 0.1–0.25 ha in area, were included in the investigation. With the algorithm, 76.9% of the trees in the upper layer could be recognised. This corresponds to 85.2% of the timber volume determined by ground measurements. The results for conifers were more accurate in this respect than for deciduous trees. A negative aspect was the number of falsely identified trees, the percentage of which was 5.4%.Based on the values for tree height and crown radius for trees delineated through laser scanning, multiple regression equations were used to determine tree height, crown diameter, diameter at breast height and single tree volume. The results for the determination of single tree parameter were, again, more accurate for conifers than they were for deciduous trees. Using the resulting regression equations, it was possible to identify 93.3% of the wood volume of all of the trees measured on the ground. Based on these results, it is possible to automatically detect most of the economically interesting wood volume and to classify it by diameter at breast height.

A B S T R A C T Increasing tree canopy cover has led to increasing wildfire activity in conifer dominated areas of the southwestern United States. Estimating historical changes in the spatial distribution of tree canopy cover can provide further insights into the dynamics of forest and fuel conditions in these landscapes and help prioritize areas for restoration to mitigate wildfire risks and restore biological functioning. In this study, we explored the relationship between LiDAR derived canopy cover data and Landsat reflectance values, and derived a model to estimate percent canopy cover (PCC) on historical Landsat data from 1987 to 2015 for the Valles Caldera National Preserve (VCNP), located in the southwest Jemez Mountains of New Mexico. We developed a regression model between LiDAR generated canopy cover collected in June 2010 and Landsat Thematic Mapper (TM) reflectance values (bands 1–7 except band 6) and vegetation indices collected for the same date. About 5% (17,000) of the total LiDAR points (329,102) were used as training points and a separate, non-overlapping set of 17,000 points as test points to validate the regression model. A simple linear model with the red band (band 3; R 2 = 0.70) was selected as the best model to predict PCC in the rest of the images for 1987–2015. In general, we found a strong consistency between the spatial dynamics of modelled tree canopy cover based on historical Landsat data, wildfire events and forest management practices that occurred during the same period. Results showed that about 11% of the study area experienced an increase in PCC for the period of 1987–2015 while 41% of the study area experienced a reduction in PCC during the same time period, mostly in the areas which were affected by stand replacing wildfires in 2011 and 2013. The results indicate an overall increase in medium and high canopy cover classes in specific regions of the study area, which could lead to hazardous wildfires such as those in 2011 and 2013. In the context of ongoing ecological restoration of these montane forests, predicted PCC of contemporary forests could help local managers to identify the areas in the need of immediate restoration efforts by focusing management practices on the areas with closed canopy.

Vengono illustrati modalità ed esiti di una sperimentazione avente l’obiettivo di verificare l’idoneità di riprese LiDAR invernali e a bassa risoluzione, non espressamente realizzate per scopi forestali, nella messa a punto di modelli di stima dei principali parametri dendrometrici forestali (volume legnoso, area basimetrica). A tal fine sono stati elaborati dati provenienti da due riprese LiDAR, con caratteristiche tecniche molto diverse, realizzate su una stessa area forestale del Trentino (Foresta di Paneveggio). Il modello messo a punto è stato poi applicato in un altro scenario forestale (Foresta Demaniale di Cadino) e integrato con dati al suolo provenienti da inventario assestamentale per campionamento. I modelli di stima prodotti, sebbene caratterizzati da performance statistiche apparentemente non elevate, hanno consentito di ottenere stime con errori contenuti, applicabili alle tipiche compartimentazioni in uso nella pianificazione forestale (particelle forestali, comprese...

Light detection and ranging (LiDAR) sampling or full-area coverage is deemed as favorable means to achieve timely and robust characterizations of forests. Recently, a 3D segmentation approach was developed for extracting single trees from LiDAR data. However, key parameters for modules used in the strategy had to be empirically determined. This paper highlights a comprehensive study for the sensitivity analysis of 3D single tree detection from airborne LiDAR data. By varying key parameters, their influences on results are to be quantified. The aim of the study is to enlighten the optimal combination of parameter values towards new applications. For the experiment, a number of sample plots from two temperate forest sites in Europe were selected. LiDAR data with a point density of 25 pts/m 2 over the first site in the Bavarian forest national park were captured with under both leaf-on and leaf-off conditions. Moreover, a Riegl scanner was used to acquire data over the Austrian Alps forest with four-fold point densities of 5 pts/m 2 , 10 pts/m 2 , 15 pts/m 2 and 20 pts/m 2 , respectively, under leaf-off conditions. The study results proved the robustness and efficiency of the 3D segmentation approach. Point densities larger than 10 pts/m 2 did not seem to significantly contribute to the improvement in the performance of 3D tree detection. The performance of the approach can be further examined and improved by optimizing the parameter settings with respect to different data properties and forest structures.

Airborne laser scanning (ALS) is the one of the most accurate remote sensing techniques for data acquisition where the terrain and its coverage is concerned. Modern scanners have been able to scan in two or more channels (frequencies of the laser) recently. This gives the rise to the possibility of obtaining diverse information about an area with the different spectral properties of objects. The paper presents an example of a multispectral ALS system-Titan by Optech-with the possibility of data including the analysis of digital elevation models accuracy and data density. As a result of the study, the high relative accuracy of LiDAR acquisition in three spectral bands was proven. The mean differences between digital terrain models (DTMs) were less than 0.03 m. The data density analysis showed the influence of the laser wavelength. The points clouds that were tested had average densities of 25, 23 and 20 points per square metre respectively for green (G), near-infrared (NIR) and shortwave-infrared (SWIR) lasers. In this paper, the possibility of the generation of colour composites using orthoimages of laser intensity reflectance and its classification capabilities using data from airborne multispectral laser scanning for land cover mapping are also discussed and compared with conventional photogrammetric techniques.

This article presents an airborne Light Detection and Ranging (LiDAR)-based method to extract interesting stand attributes for forest management in high-density Eucalyptus globulus Labill. plantations. An adaptive morphological filter (AMF) for classifying terrain LiDAR points in forested areas is used to classify LiDAR points; canopy cover (CC), number of LiDAR-detected trees per hectare (N LD) and individual tree height (h tree) were calculated using the canopy height model (CHM); and several statistics and metrics extracted from the CHM and the normalized height of the LiDAR data cloud (NHD) were incorporated into the linear and multiplicative models for estimating mean height (H m), dominant height (H d), mean diameter (d m), quadratic mean diameter (d g), number of stems per hectare (N), basal area (G) and volume (V). The height accuracy results of the LiDAR-derived digital terrain model (DTM), root mean square error (RMSE) = 0.303 m, revealed that the developed filter behaved well. The values of the RMSE for CC, N LD and h tree were 13.2%, 733.3 stems ha–1 and 1.91 m, respectively. The regressions explained 78% of the variance in ground-truth values for H m (RMSE = 1.33 m); 92% for H d (RMSE = 1.18 m); 71% for d m (RMSE = 1.68 cm); 73% for d g (RMSE = 1.66 cm); 49% for N (RMSE = 667 stems ha–1); 78% for G (RMSE = 5.30 m2 ha–1); and 81% for V (RMSE = 53.6 m3 ha–1).

The study evaluates five existing segmentation algorithms to determine the method most suitable for individual tree detection across a species-diverse forest: raster-based region growing, local maxima centroidal Voronoi tessellation, point-cloud level region growing, marker controlled watershed and continuously adaptive mean shift. Each of the methods has been tested twice over one mixed and five single species plots: with their parameters set as constant and with the parameters calibrated for every plot. Overall, continuous adaptive mean shift performs best across all the plots with average F-score of 0.9 with fine-tuned parameters and 0.802 with parameters held at constant. Raster-based algorithms tend to achieve higher scores in coniferous plots, due to the clearly discernible tops, which significantly aid the detection of local maxima. Their performance is also highly dependent on the moving size window used to detect the local maxima, which ideally should be readjusted for every plot. Crown overlap, suppressed and leaning trees are the most likely sources of error for all the algorithms tested.

NB This is not the final version of the publication. For final published version go to: https://doi.org/10.1002/arp.1780

While airborne laser scanning (ALS) data is a key resource for studying afforested landscapes, other datasets are essential to understanding the character of surveyed areas, and for assessing the reliability of the remote sensed data interpretation. This paper presents results of such assessment based on the analysis of contemporary forestry data and historic maps for a study area located south of Polanów, Poland, which was subject of bespoke ALS survey. Forestry data and historic maps were initially used to plan airborne laser scanning data collection, and thereafter to assess data quality and bias. As a result, the longevity and patterning of afforestation were determined and these correlate with the outputs of interpretative mapping together offering a benchmark methodology for identifying survival zones for prehistoric and early medieval earthworks beyond the extents of early and late modern cultivation. In addition, the data characteristics discussed in this paper demonstrate the importance of data analysis and survey methodology to inform the understanding of potential bias in the outputs of ALS surveys of afforested landscapes. By understanding this phenomenon at a local scale, the reliability of the data can be estimated. Given the nation-wide availability of forestry data across Europe, as well as extensive coverage of historic maps, the results discussed in this paper have a general application. Focusing on areas with the highest survival potential can lead to immediate discovery of premodern monuments, and hence prevent them from extensive, forestry management related, destruction

Vengono illustrati modalità ed esiti di una sperimentazione avente l’obiettivo di verificare l’idoneità di riprese LiDAR invernali e a bassa risoluzione, non espressamente realizzate per scopi forestali, nella messa a punto di modelli di stima dei principali parametri dendrometrici forestali (volume legnoso, area basimetrica). A tal fine sono stati elaborati dati provenienti da due riprese LiDAR, con caratteristiche tecniche molto diverse, realizzate su una stessa area forestale del Trentino (Foresta di Paneveggio). Il modello messo a punto è stato poi applicato in un altro scenario forestale (Foresta Demaniale di Cadino) e integrato con dati al suolo provenienti da inventario assestamentale per campionamento. I modelli di stima prodotti, sebbene caratterizzati da performance statistiche apparentemente non elevate, hanno consentito di ottenere stime con errori contenuti, applicabili alle tipiche compartimentazioni in uso nella pianificazione forestale (particelle forestali, comprese, strati), con il vantaggio di ottenere anche una mappatura e una spazializzazione a livello sub-particellare del volume legnoso dei popolamenti, informazione molto importante per la pianificazione degli interventi selvicolturali da realizzare in foresta.

The main purpose of the present research is to verify whether medium resolution LiDAR data, not forest-specific, can be used to carry out statistical models for estimating timber volume. LiDAR data with different characteristics, and taken in different seasons (winter and summer) on the same forest area (Foresta di Paneveggio, Trentino, Italy), were processed, and an estimation model using winter low-resolution data was performed. Such model was then applied to a different territory, having similar forest characteristics. Moreover, LiDAR and ground data, surveyed during the management plan inventory, were integrated. The final model allows to obtain good volume estimates with fair precision if applied at forest compartments level, and can produce detailed timber volume maps, very useful when planning forestry operations.

In some cases a canopy height model (CHM) is the only available source of forest height information. For these cases it is important to understand the predictive power of CHM data for forest attributes. In this study we examined the use of lidar-derived CHM metrics to predict forest structure classes according to the amount of basal area present in understory, midstory, and overstory trees. We evaluated two approaches to predict size-based forest classifications: in the first, we attempted supervised classification with both linear discriminant analysis (LDA) and random forest (RF); in the second, we predicted basal areas of lower, mid, and upper canopy trees from CHM-derived variables by k-nearest neighbour imputation (k-NN) and parametric regression, and then classified observations based on their predicted basal areas. We used leave-one-out cross-validation to evaluate our ability to predict forest structure classes from CHM data and in the case of prediction-based classification approach we look at the performances in predicting basal area. The strategies proved moderately successful with a best overall classification accuracy of 41% in the case of LDA. In general, we were most successful in predicting the basal areas of small and large trees (R2 respectively of 71% and 69% in the case of k-NN imputation).