Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Forest inventory based on canopy height model derived from airborne laser scanning data

Forest inventory based on canopy height model derived from airborne laser scanning data Airborne laser scanning (ALS) has emerged as a remote sensing technology capable of providing data suitable for deriving all types of elevation models. A canopy height model (CHM), which represents absolute height of objects above the ground in metres (e.g., trees), is the one most commonly used within the forest inventory. The aim of this study was to assess the accuracy of forest inventory performed for forest unit covered 17,583 ha (Slovakia, Central Europe) using the CHM derived from ALS data. This objective also included demonstrating the applicability of freely available data and software. Specifically, ALS data acquired during regular airborne survey, QGIS software, and packages for R environment were used for purpose of this study. A total of 180 testing plots (5.6 ha) were used for accuracy assessment. The differences between CHM-predicted and ground-observed forest stand attributes reached a relative root mean square error at 10.9%, 23.1%, and 34.5% for the mean height, mean diameter, and volume, respectively. Moreover, all predictions were unbiased (p-value < 0.05) and the strength of the relationships between CHM-predicted and ground-observed forest stand attributes were relative high (R = 0.7 – 0.8). Key words: airborne LiDAR; natural resources; area-based approach; freeware Editor: Peter Surový resolution (Polidori & El Hage 2020), the main advan- 1. Introduction tages of CHM compared to the point cloud are higher Remote sensing (RS) has become a reliable source of availability and applicability. The reasons are as follows geospatial data and techniques for forest inventory (1) most countries in Central Europe acquire and pro- (Surový & Kuželka 2019). Especially, airborne laser vide CHM from routine ALS campaigns (Ginzler & Hobi scanning (ALS), as an active RS technology, represents 2015), (2) once a terrain model is available, CHM can an effective solution to acquire relevant information about also by generated from aerial images acquired by directly/ forests at tree or stand level (Goodbody et al. 2021). This remotely piloted aircraft system (Liu et al. 2020), and is primarily because this technology can be used at local (3) commonly available hardware and software, includ- or regional level, sensors are not affected by any geometri- ing free available applications, is usable for CHM-based cal distortions, and laser pulses can penetrate through the forest inventory (Wang et al. 2021). On the other hand, gaps of the canopy, even in dense forests, to the ground. the CHM (1) is generalized by interpolation or smoothing In this context, related ALS data can achieve high accu- procedures, (2) may contain unnatural black holes, and racy as well as density and allow automatic classification (3) does not contain the most of relevant information for of terrain, building, and vegetation objects directly from forest inventory related to the understory (Jamru 2018; point cloud (Díaz-Varela & González-Ferreiro 2021). Zhang et al. 2020). Since ALS can provide data not only from the surface The overall objective of this study was to assess the but also from the ground, this technology has been proven accuracy of forest inventory performed for forest unit to provide data suitable for derivation of all types of ele- covered 17,583 ha using the CHM derived from ALS vation models. A canopy height model (CHM), which data. This objective also included demonstrating the represents absolute height of objects above the ground applicability of freely available data and software. Spe- in metres (e.g., trees), is the one most commonly used cifically, ALS data from regular airborne survey, QGIS within the forest inventory (Vauhkonen et al. 2014; Zhen software (QGIS Development Team 2022), and packages et al. 2016). In addition to the sufficient accuracy and for R environment (R Core Team 2021) were used for *Corresponding author. Ivan Sačkov, e-mail: sackov@nlcsk.org, phone: +421 949 381 250 © 2022 Authors. This is an open access article under the CC BY 4.0 license. I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 purpose of this study. The CHM-based forest inventory of point cloud reached 26 points/m . The ALS-derived was focused on three characteristics that have a particu- digital terrain model (DTM) and digital surface model lar importance in forest management and represent the (DSM) were processed in SCOP++ Software environ- main forest stand attributes, such as mean height, mean ment (Trimble) with vertical accuracy of ±0.2 m. The diameter, and volume per hectare. CHM of 0.5 m resolution was generated as a result of subtraction of these DSM and DTM. 2. Material and methods 2.3. Ground Data 2.1. Study Area Ground data were obtained during the leaf-on season in The study was conducted in the territory of the forest 2021 within 360 circular sample plots with radius of 10 management unit University Forest Enterprise of the m (11.25 ha). The position of the plot center was located Technical University in Zvolen located in central Slovakia using the Global Navigation Satellite System (GNSS) (approx. 48°37´N, 19°05´E). The total area is 17,583 ha and a positional error from 1.44 m to 6.25 m was expected and forests occupy 7,483 ha from this area. The eleva- for all plots (Murgaš et al. 2018). A total of 12,176 trees tion reaches interval of 261 – 1,069 meters above sea with Diameter at Breast Height (DBH) higher than 8 cm level. Dominant species include European beech (Fagus were measured for stem position, species, height, and sylvatica L.), Sessile oak (Quercus petraea Matt.), and diameter. Tree height was measured using the Vertex European silver fir (Abies alba Mill.) with 70% coverage IV hypsometer (Haglöf) with total accuracy of ±1.0 m. (Fig. 1). Measurement of DBH was performed with a tree caliper at the millimeter scale. Tree volume was calculated based on tree height and DBH using allometric model with total accuracy of ±7–12% at tree level (Petráš & Pajtík 1991). 2.2. Remote Sensing Data Sample plots were classified into nine strata. The main Aerial survey was conducted as a part of a routine and reg- criteria for classification were forest types (B: Beech ular flight campaign in order to create a forest manage- forests, O: Oak-hornbeam forests, F: Beech-fir forests) ment plan. Since this survey is funded by state resources, and age classes (1: 11 – 40 years, 2: 41 – 80 years, 3: related ALS data and derivatives are provided free of >80 years). In this context, Beech stratum (B1–3), Oak charge based on the order addressed to the National stratum (O1–3), and a Fir stratum (F1–3) was created. Forest Centre. Furthermore, sample plots were divided into two groups. ALS data acquisition was performed in June 2021 The main criteria for dividing was proportionality of over the entire study area (17,583 ha). This area was these both datasets related to the number of plots, area scanned using a Leica ALS 70 CM scanner from average of plots and stand attributes. In this context, group of 180 altitude of 1,290 m with field of view of 43° and 282 kHz training plots (20 plots per stratum) and group of 180 laser pulse repetition rate. The resulting average density testing plots (20 plots per stratum) was created (Table 1). Fig. 1. The study area located in Slovakia, Central Europe. 225 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Table 1. Statistics of ground data from 180 training plots and 180 testing plots (20 plots per stratum). Training Data Testing Data Stratum Height Diameter Volume Height Diameter Volume 3 −1 3 −1 [m] [cm] [m ha ] [m] [cm] [m ha ] All 18.9 ±4.5 23.0 ±6.9 375.9 ±195.2 18.7 ±4.9 22.2 ±7.3 383.5 ±205.0 B1 15.5 ±3.8 15.5 ±2.6 233.2 ±140.2 15.4 ±4.0 15.4 ±2.9 230.8 ±139.8 B2 20.1 ±2.9 22.1 ±3.7 407.2 ±112.6 20.3 ±3.2 22.3 ±3.5 413.4 ±124.5 B3 22.8 ±2.8 31.0 ±6.6 522.5 ±153.0 22.4 ±3.5 30.5 ±7.2 542.8 ±191.4 O1 15.0 ±3.8 15.2 ±2.7 185.8 ±107.5 15.2 ±3.5 15.4 ±2.3 197.6 ±99.2 O2 19.0 ±2.9 21.9 ±2.9 342.1 ±117.3 19.1 ±3.0 22.5 ±2.9 363.5 ±122.9 O3 21.2 ±2.2 30.4 ±6.1 402.0 ±126.5 20.6 ±3.1 28.2 ±6.5 421.8 ±130.2 F1 13.6 ±3.7 14.8 ±2.6 234.3 ±166.1 13.2 ±3.2 14.4 ±2.7 218.3 ±185.0 F2 19.8 ±4.0 23.7 ±3.2 513.7 ±146.6 20.0 ±3.7 23.7 ±3.1 499.5 ±117.9 F3 23.3 ±4.9 34.7 ±7.0 600.0 ±235.8 23.1 ±4.8 34.5 ±7.5 620.7 ±281.8 Note: All: All sample plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum; Height: arithmetic mean height; Diameter: arithmetic mean diameter; Volume: volume per hectare. the forest area include predicted mean height, mean 2.4. CHM-based Forest Inventory diameter and volume per hectare. The workflow for CHM-based forest inventory is shown in Fig. 2 and described in detail in the following points: – The height metrics (i.e., mean, median, standard deviation, minimum, maximum, range, minority, majority) were computed from CHM for specific areas of training plots (n = 180) using zonal statis- tic tool implemented in the QGIS software (Madry 2021). Specic fi ally, the CHM-based mean height was assessed as most highly correlated height metric to all related forest stand attributes (i.e., mean height, mean diameter, and volume per hectare) using Pear- son’s correlation analysis (Voght & Johnson 2012). – The simple nonlinear models for prediction of forest stand attributes were computed from training data using least-squares regression implemented in the R package NLStools (Baty et al. 2015). These models express the relationship between ground-observed forest stand attributes and CHM-based mean height. Here, only statistically significant and best perform- ing models for prediction of specific forest stand Fig. 2. Flowchart for CHM-based forest inventory. Note: GR: attribute (i.e., mean height, mean diameter, and ground; ALS: airborne laser scanning; SA: forest stand at- volume per hectare) were selected and used. The tribute. statistical significance was assessed using Fisher’s test (F-test). The goodness of fit was assessed using error metrics of regression model. 2.5. Accuracy Assessment – The mean heights were computed from CHM for The accuracy of the CHM-based forest inventory was specific areas of testing plots (n = 180) using zonal assessed by comparing the observed and predicted for- statistic tool function implemented in the QGIS soft- est stand attributes from 180 testing plots. The value of ware. Subsequently, the forest stand attributes were observed forest stand attributes for each sample plot was predicted for testing plots through the developed obtained by ground measurement (Section 2.3). The regression models. In this way, each of testing plots value of predicted forest stand attributes for each sample include observed as well as predicted mean height, plot was obtained by CHM-forest inventory (Section 2.4). mean diameter and volume per hectare. Finally, the The standard error (Bias, Equation (1)) and root accuracy of the CHM-based forest inventory was mean square error (RMSE, Equation (2)) were used to assessed by comparing the observed and predicted assess the model’s performance. The relative %Bias and forest stand attributes. %RMSE were calculated as the ratios of their absolute – The regular 20 × 20 m grid covering the forest area values and the arithmetic average of the observed data. was created and the CHM-based mean heights were The regression function, coefc fi ient of determination computed for all grid cells. The size of the grid cells (R ), and F-test of statistical signic fi ance of the regression (400 m ) was set to reflect the size of the ground plots model (p-value) were calculated to assess the strength of (314 m ). Subsequently, the forest stand attributes the relationship between predicted and observed forest were predicted for all grid cells through the developed stand attributes (Voght & Johnson 2012). regression models. In this way, each grid cell covering 226 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 The t-test, when a normal distribution of mean differ- area (Fig. 3). In this context, the RMSE of CHM-based ences was confirmed, or Wilcoxon test, when a normal prediction of forest stand attributes did not exceed 2.0 m distribution of mean differences was not confirmed (Fay (10.9%) for mean height, 5.1 cm (23.1%) for mean 3 −1 & Proschan 2010), were used to assess the significance diameter and 132.1 m ha (34.5%) for volume (Table of differences (p-value < 0.05). 3). These predictions were not statistically significantly different relative to the ground-observed data (p-value < 0.05). The relationships (R ) between CHM-predicted L% DV \Ö  \ ¦ L L [1] L  and ground-observed forest stand attributes fluctuated from 0.7 to 0.8 (Fig. 4). 05 6( \  \  Q [2] L L Table 2. Regression models for CHM-based forest inventory. Attribute Model Form N R SEE SEE% p-value Height 2.57 × (H^0.69) + E 180 0.81 2.05 10.97 <0.001 where n is the number of plots, y is the observed forest Diameter 9.84 × EXP(0.04 × H) + E 180 0.74 5.11 22.98 <0.001 stand attribute for plot i, and ݕො is the predicted forest Volume 1.99 × (H^1.80) + E 180 0.70 127.80 33.33 <0.001 Note: H: CHM-based mean height; E: error of prediction; N: number of training plots; R : co- stand attribute for plot i. efficient of determination; SEE: standard error of estimates; SEE%: relative standard error of estimates; p-value of F-test, null hypothesis is rejected at α = 0.05. 3. results Table 3. Accuracy of CHM-based forest inventory. Height Diameter Volume Stratum The form and performance of regression models for %Bias %RMSE %Bias %RMSE %Bias %RMSE CHM-based forest inventory are presented in Table 2. All −0.60 10.89 −2.72 23.07 −0.62 34.45 B1 3.88 11.67 19.60 26.83 13.44 41.64 We found the power function to be most suitable for B2 0.22 9.62 5.93 16.48 9.80 31.04 prediction of height as well as volume and exponential B3 −2.68 11.07 −14.44 21.83 0.45 24.42 O1 2.18 11.13 15.76 23.14 22.54 48.33 function for prediction of diameter. All predictions were O2 0.14 12.20 −2.92 15.30 7.37 29.63 unbiased (p-value < 0.05) and the strength of the rela- O3 −6.47 11.66 −21.33 27.78 −4.54 34.78 F1 1.52 11.75 8.38 14.34 −30.10 55.93 tionships between CHM-predicted and ground-observed F2 −0.01 9.82 −0.99 17.39 −10.46 31.20 forest stand attributes were relative high (R = 0.7–0.8). F3 −5.77 10.63 −24.43 30.51 −12.90 34.22 The CHM-based forest inventory evaluated the mean Note: All: All testing plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum; %Bias: relative standard error; %RMSE: relative root mean square error; Height: arithmetic mean height at 18.6 ±4.0 m, mean diameter at 21.6 ±5.0 cm height; Diameter: arithmetic mean diameter; Volume: volume per hectare. 3 −1 and volume at 381.1 ±183.8 m ha within the study Fig. 3. Ground-observed and CHM-predicted forest stand attributes including standard deviation (whiskers). Note: All: All test- ing plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum. 227 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Fig. 4. Relationship between ground-observed and CHM-predicted forest stand attributes from all testing plots. The dashed line represents a 1:1 correspondence. The n fi al result of CHM-based forest inventory repre - 4. Discussion sents a raster layer with resolution of 20 × 20 m consisting Forest inventory is strictly required by relevant stake- values of forest stand attributes (i.e., mean height, mean holders (e.g., forest managers, environmentalist, and diameter, and volume per hectare). This result is provided policy-makers). However, ground-based measurements to the stakeholders directly as a product (i.e., raster or are expensive and time consuming. The RS technologies vector layer) or indirectly through a web-map service has become a reliable source of geospatial data and tech- (i.e., website link). The illustrative example of 3D visu- niques for forest inventory. Especially, ALS-based CHM alization representing CHM-predicted volume for part of represents an available and usable data source (Kakou- study area over grid cell 20 × 20 m is displayed in Fig. 5. laki et al. 2021) for prediction of forest stand attributes and related techniques provide an opportunity to comple- ment ground-based measurements. Fig. 5. Visualization of 3D map representing CHM-predicted forest stand attribute for part of study area over grid cell 20 × 20 m (illustrative example for volume per hectare). 228 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 The objective of this study was, therefore, to assess Although the CHM-based predictive models are not the accuracy of forest inventory performed using ALS- sufficiently able to describe the full range of variability based CHM from regular aerial campaign and free avail- of forest stand attributes, there are several options to able software tools, such as QGIS and packages for R improve the performance of CHM-based forest inven- environment. This CHM-based forest inventory was tory. For example, (1) higher position accuracy of sample conducted in Central European heterogeneous forests plots (Zhang et al. 2017; Novo-Fernández et al. 2019), covering 7,483 ha and related accuracy was assessed on (2) higher vertical accuracy and resolution of CHM 180 testing plots covering 5.6 ha. (Deluzet et al. 2022; Lisiewicz et al. 2022), or (3) appli- In this study, the bias of CHM-based forest inventory cation of multiple nonlinear regression or nonparametric for all strata was −0.6% in the case of mean height as well machine learning methods (Jiang et al. 2020; Zhao et al. as volume and −2.7% in the case of mean diameter. These 2022) could contribute to develop predictive models with results were mainly related to the underestimation of all higher accuracy. forest stand attributes within the strata representing old Alternatively, the point cloud-based forest inventory forest (B3, O3, F3) and relative high underestimation of is more suitable solution in heterogeneous forest stands. volume in strata representing young fir forest (F1). The For example, the point cloud-based forest inventory per- prediction of other forest stand attributes within the strata formed for the same study area achieved the %RMSE representing young forest were overestimated (B1, O1). which did not exceed 7.2% for mean height, 8.6% for The RMSE of CHM-based forest inventory reached the mean diameter, and 21.4% for total volume (Sačkov et al. value of 10.9%, 23.1%, and 34.5% for the mean height, 2019a; Sačkov et al. 2019b). Moreover, similar or better mean diameter, and volume, respectively. While the pre- results have also been provided by other authors, such as diction of mean height achieved homogenous accuracy study from Brazilian forests (Martins-Neto et al. 2021), within the strata (10–12%), range of errors related to the Canadian forests (Lamb et al. 2018), or Italian forests prediction of mean diameter and volume was consider- (Versace et al. 2020). These and other studies provided able. Specic fi ally, the overall error of mean diameter pre - %RMSE at 6%–14%, 9%–20%, and 17%–23% for the diction u fl ctuated from 14% (F1) to 31% (F3) and overall mean height, mean diameter, and volume, respectively. error of volume prediction fluctuated from 24% (B3) to 56% (F1). Thus, the largest prediction error was found in the youngest (esp., volume) and oldest (esp., diam- 5. Conclusions eter) forest stands (Fig. 2). Although this trend has been found also in similar studies (Hill et al. 2014; Cosenza et This study assessed the accuracy of forest inventory al. 2018; Mielcarek et al. 2018), statistical significance performed using ALS-based CHM and free available of increasing or decreasing the development stage on software tools, such as QGIS and packages for R envi- prediction accuracy was not confirmed. However, it is ronment. Our results confirmed that CHM-based for - evident that level of RMSE was related to the variability est inventory represents a practical solution that allows of forest stand characteristics. The overall accuracy of make a general overview of natural resources for the CHM-forest inventory decreased with increasing vari- whole forest unit. Moreover, this solution can be per- ability of forest stand characteristics. For example, the formed directly and repeatedly by stakeholders without variability of volume is signic fi antly higher than variabil - the need for additional investments focused on data, soft- ity of mean height (Fig. 6) and consequently the RMSE ware and hardware. On the other hand, overall accuracy of CHM-predicted volume is significantly higher than of CHM-based forest inventory in this study did not meet RMSE of CHM-predicted mean height (Table 3). the requirements of forest management from some coun- tries in Central Europe (e.g., Slovakia). Consequently, continuous research and development of CHM-based forest inventory seems to be valid and necessary. Acknowledgments This research was supported by the Slovak Research and Devel- opment Agency in the framework of the project “Development of advanced geospatial technologies for multiscale monitoring of forest ecosystems” (APVV-19-0257) as well as by the Euro- pean Regional Development Fund in the framework of the project “Research and development of contactless methods for acquisi- tion of geospatial data usable in the forest monitoring in order Fig. 6. Relative variability of ground-observed forest stand to streamline forest management and increase forest protection” attributes. Note: All: All sample plots; B1–3: Beech stratum; (NFP313011V465). O1–3: Oak stratum; F1–3: Fir stratum. 229 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Liu, Q., Fu, L., Chen, Q., Wang, G., Luo, P., Sharma, R. references P. et al., 2020: Analysis of the Spatial Differences in Baty, F., Ritz, Ch., Charles, S., Brutsche, M., Flandrois, Canopy Height Models from UAV LiDAR and Pho- J. P., Delignette-Muller, M., 2015: A Toolbox for togrammetry. Remote Sensing, 12:2884. Nonlinear Regression in R – The Package nlstools. Madry, S., 2021: Introduction to QGIS: Open Source Journal of Statistical Software, 66:1–21. Geographic Information System. USA, Locate Press Cosenza, D., Soares, V., Leite, H., Gleriani, J., Amaral, C., LLC, 219 p. Junior, J. et al., 2018: Airborne laser scanning applied Martins-Neto, R. P., Tommaselli, A. M. G., Imai, N. N., to eucalyptus stand inventory at individual tree level. David, H. C., Miltiadou, M., Honkavaara, E., 2021: Pesquisa Agropecuaria Brasileira, 53:1373–1382. Identification of Significative LiDAR Metrics and Deluzet, M., Erudel, T., Briottet, X., Sheeren, D., Fabre, Comparison of Machine Learning Approaches for S., 2022: Individual Tree Crown Delineation Method Estimating Stand and Diversity Variables in Hetero- Based on Multi-Criteria Graph Using Geometric and geneous Brazilian Atlantic Forest. Remote Sensing, Spectral Information: Application to Several Temper- 13:2444. ate Forest Sites. Remote Sensing, 14:1083. Mielcarek, M., Stereńczak, K., Khosravipour, A., 2018: Díaz-Varela, R. A., González-Ferreiro, E., 2021: 3D Point Testing and evaluating different LiDAR-derived can- Clouds in Forest Remote Sensing. Remote Sensing, opy height model generation methods for tree height 13:2999. estimation. International Journal of Applied Earth Fay, M. P., Proschan, M. A., 2010: Wilcoxon-Mann- Observation and Geoinformation, 71:132–143. Whitney or t-test? On assumptions for hypoth - Murgaš, V., Sačkov, I., Sedliak, M., Tunák, D., Chudý esis tests and multiple interpretations of decision F., 2018: Assessing horizontal accuracy of inven- rules. Statistics Survey, 4:1–39. tory plots in forests with different mix of tree species Ginzler, C., Hobi, M. L., 2015: Countrywide Stereo- composition and development stage. Journal of For- Image Matching for Updating Digital Surface Mod- est Science, 64:478–485. els in the Framework of the Swiss National Forest Novo-Fernández, A., Barrio-Anta, M., Recondo, C., Inventory. Remote Sensing, 7:4343–4370. Cámara-Obregón, A., López-Sánchez, C. A., 2019: Goodbody, T. R. H., Coops, N. C., Luther, J. E., Tom- Integration of National Forest Inventory and Nation- palski, P., Mulverhill, C. F., Fournier, R. et al., 2021: wide Airborne Laser Scanning Data to Improve Airborne laser scanning for quantifying criteria and Forest Yield Predictions in North-Western Spain. indicators of sustainable forest management in Can- Remote Sensing, 11:1693. ada. Canadian Journal of Forest Research, 51:14. Petráš, R., Pajtík, J., 1991: Sústava česko-slovenských Hill, A., Breschan, J., Mandallaz, D., 2014: Accuracy objemových tabuliek drevín. Lesnícky časopis, Assessment of Timber Volume Maps Using Forest 37:49–56. Inventory Data and LiDAR Canopy Height Models. Polidori, L., El Hage, M., 2020: Digital Elevation Model Forests, 5:2253–2275. Quality Assessment Methods – A Critical Review. Jamru, L. R., 2018: Correction pit free canopy height Remote Sensing, 12:3522. model derived from LiDAR data for the broad leaf Sačkov, I., Kulla, L., Bucha, T., 2019a. A Comparison of tropical forest. IOP Conference Series: Earth and Two Tree Detection Methods for Estimation of For- Environmental Science, 169:012113. est Stand and Ecological Variables from Airborne Jiang, X., Li, G., Lu, D., Chen, E., Wei, X., 2020: Strati- LiDAR Data in Central European Forests. Remote c fi ation-Based Forest Aboveground Biomass Estima - Sensing, 11:1431. tion in a Subtropical Region Using Airborne Lidar Sačkov, I., Scheer, L., Bucha, T., 2019b. Predicting for- Data. Remote Sensing, 12:1101. est stand variables from airborne LiDAR data using Kakoulaki, G., Martinez, A., Florio, P., 2021: Non-com- a tree detection method in Central European forests. mercial Light Detection and Ranging (LiDAR) data Central European Forestry Journal, 65:191–197. in Europe. Luxembourg, Publications Office of the Surový, P., Kuželka, K., 2019: Acquisition of Forest European Union, 35 p. Attributes for Decision Support at the Forest Enter- Lamb, S. M., MacLean, D.A., Hennigar, C.R., Pitt, D.G, prise Level Using Remote-Sensing Techniques - A 2018: Forecasting Forest Inventory Using Imputed Review. Forests, 10:273. Tree Lists for LiDAR Grid Cells and a Tree-List Vauhkonen, J., Maltamo, M., McRoberts, R. E., Næs- Growth Model. Forests, 9: 167. set, E., 2014: Introduction to forestry applications Lisiewicz, M., Kamińska, A., Kraszewski, B., Stereńczak, of airborne laser scanning. In: Maltamo, M., Næs- K., 2022: Correcting the Results of CHM-Based Indi- set, E., Vauhkonen, J. (eds.): Forestry Application of vidual Tree Detection Algorithms to Improve Their Airborne Laser Scanning: Concept and Case Studies. Accuracy and Reliability. Remote Sensing, 14: 1822. Springer Netherlands: Dordrecht, The Netherlands, p. 1–16. 230 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Versace, S., Gianelle, D., Frizzera, L., Tognetti, R., Garfì, Zhao, J., Zhao, L., Chen, E., Li, Z., Xu, K., Ding, X., V., Dalponte, M., 2019: Prediction of Competition 2022: An Improved Generalized Hierarchical Esti- Indices in a Norway Spruce and Silver Fir-Dominated mation Framework with Geostatistics for Mapping Forest Using Lidar Data. Remote Sensing, 11:2734. Forest Parameters and Its Uncertainty: A Case Study Voght, W. P., Johnson, R. B., 2012: Correlation and of Forest Canopy Height. Remote Sensing, 14:568. Regression Analysis. USA, SAGE Publications Ltd, Zhen, Z., Quackenbush, L. J., Zhang, L., 2016: Trends 1632 p. in Automatic Individual Tree Crown Detection and Wang, H., Seaborn, T., Wang, Z., Caudill, Ch., Link, T., Delineation – Evolution of LiDAR Data. Remote 2021: Modeling tree canopy height using machine Sensing, 8:333. learning over mixed vegetation landscapes. Inter- Other sources national Journal of Applied Earth Observation and QGIS Development Team, 2022: QGIS Geographic Geoinformation, 101:102353. Information System. Open Source Geospatial Foun- Zhang, Z., Cao, L., She, G., 2017: Estimating For- dation Project. http://qgis.osgeo.org. est Structural Parameters Using Canopy Metrics R Core Team, 2021: R: A language and environment for Derived from Airborne LiDAR Data in Subtropical statistical computing. R Foundation for Statistical Forests. Remote Sensing, 9:940. Computing. https://www.R-project.org. Zhang, W., Cai, S., Liang, X., Shao, J., Hu, R., Yu, S. et al., 2020: Cloth simulation-based construction of pit-free canopy height models from airborne LiDAR data. Forest Ecosystems, 7:1–13. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Forestry Journal de Gruyter

Forest inventory based on canopy height model derived from airborne laser scanning data

Sačkov, Ivan
Forestry Journal , Volume 68 (4): 8 – Dec 1, 2022
Download PDF
8 pages

Loading next page...
 
/lp/de-gruyter/forest-inventory-based-on-canopy-height-model-derived-from-airborne-4kLjZ64s6n
Publisher
de Gruyter
Copyright
© 2022 Ivan Sačkov, published by Sciendo
ISSN
0323-1046
eISSN
2454-0358
DOI
10.2478/forj-2022-0013
Publisher site
See Article on Publisher Site

Abstract

Airborne laser scanning (ALS) has emerged as a remote sensing technology capable of providing data suitable for deriving all types of elevation models. A canopy height model (CHM), which represents absolute height of objects above the ground in metres (e.g., trees), is the one most commonly used within the forest inventory. The aim of this study was to assess the accuracy of forest inventory performed for forest unit covered 17,583 ha (Slovakia, Central Europe) using the CHM derived from ALS data. This objective also included demonstrating the applicability of freely available data and software. Specifically, ALS data acquired during regular airborne survey, QGIS software, and packages for R environment were used for purpose of this study. A total of 180 testing plots (5.6 ha) were used for accuracy assessment. The differences between CHM-predicted and ground-observed forest stand attributes reached a relative root mean square error at 10.9%, 23.1%, and 34.5% for the mean height, mean diameter, and volume, respectively. Moreover, all predictions were unbiased (p-value < 0.05) and the strength of the relationships between CHM-predicted and ground-observed forest stand attributes were relative high (R = 0.7 – 0.8). Key words: airborne LiDAR; natural resources; area-based approach; freeware Editor: Peter Surový resolution (Polidori & El Hage 2020), the main advan- 1. Introduction tages of CHM compared to the point cloud are higher Remote sensing (RS) has become a reliable source of availability and applicability. The reasons are as follows geospatial data and techniques for forest inventory (1) most countries in Central Europe acquire and pro- (Surový & Kuželka 2019). Especially, airborne laser vide CHM from routine ALS campaigns (Ginzler & Hobi scanning (ALS), as an active RS technology, represents 2015), (2) once a terrain model is available, CHM can an effective solution to acquire relevant information about also by generated from aerial images acquired by directly/ forests at tree or stand level (Goodbody et al. 2021). This remotely piloted aircraft system (Liu et al. 2020), and is primarily because this technology can be used at local (3) commonly available hardware and software, includ- or regional level, sensors are not affected by any geometri- ing free available applications, is usable for CHM-based cal distortions, and laser pulses can penetrate through the forest inventory (Wang et al. 2021). On the other hand, gaps of the canopy, even in dense forests, to the ground. the CHM (1) is generalized by interpolation or smoothing In this context, related ALS data can achieve high accu- procedures, (2) may contain unnatural black holes, and racy as well as density and allow automatic classification (3) does not contain the most of relevant information for of terrain, building, and vegetation objects directly from forest inventory related to the understory (Jamru 2018; point cloud (Díaz-Varela & González-Ferreiro 2021). Zhang et al. 2020). Since ALS can provide data not only from the surface The overall objective of this study was to assess the but also from the ground, this technology has been proven accuracy of forest inventory performed for forest unit to provide data suitable for derivation of all types of ele- covered 17,583 ha using the CHM derived from ALS vation models. A canopy height model (CHM), which data. This objective also included demonstrating the represents absolute height of objects above the ground applicability of freely available data and software. Spe- in metres (e.g., trees), is the one most commonly used cifically, ALS data from regular airborne survey, QGIS within the forest inventory (Vauhkonen et al. 2014; Zhen software (QGIS Development Team 2022), and packages et al. 2016). In addition to the sufficient accuracy and for R environment (R Core Team 2021) were used for *Corresponding author. Ivan Sačkov, e-mail: sackov@nlcsk.org, phone: +421 949 381 250 © 2022 Authors. This is an open access article under the CC BY 4.0 license. I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 purpose of this study. The CHM-based forest inventory of point cloud reached 26 points/m . The ALS-derived was focused on three characteristics that have a particu- digital terrain model (DTM) and digital surface model lar importance in forest management and represent the (DSM) were processed in SCOP++ Software environ- main forest stand attributes, such as mean height, mean ment (Trimble) with vertical accuracy of ±0.2 m. The diameter, and volume per hectare. CHM of 0.5 m resolution was generated as a result of subtraction of these DSM and DTM. 2. Material and methods 2.3. Ground Data 2.1. Study Area Ground data were obtained during the leaf-on season in The study was conducted in the territory of the forest 2021 within 360 circular sample plots with radius of 10 management unit University Forest Enterprise of the m (11.25 ha). The position of the plot center was located Technical University in Zvolen located in central Slovakia using the Global Navigation Satellite System (GNSS) (approx. 48°37´N, 19°05´E). The total area is 17,583 ha and a positional error from 1.44 m to 6.25 m was expected and forests occupy 7,483 ha from this area. The eleva- for all plots (Murgaš et al. 2018). A total of 12,176 trees tion reaches interval of 261 – 1,069 meters above sea with Diameter at Breast Height (DBH) higher than 8 cm level. Dominant species include European beech (Fagus were measured for stem position, species, height, and sylvatica L.), Sessile oak (Quercus petraea Matt.), and diameter. Tree height was measured using the Vertex European silver fir (Abies alba Mill.) with 70% coverage IV hypsometer (Haglöf) with total accuracy of ±1.0 m. (Fig. 1). Measurement of DBH was performed with a tree caliper at the millimeter scale. Tree volume was calculated based on tree height and DBH using allometric model with total accuracy of ±7–12% at tree level (Petráš & Pajtík 1991). 2.2. Remote Sensing Data Sample plots were classified into nine strata. The main Aerial survey was conducted as a part of a routine and reg- criteria for classification were forest types (B: Beech ular flight campaign in order to create a forest manage- forests, O: Oak-hornbeam forests, F: Beech-fir forests) ment plan. Since this survey is funded by state resources, and age classes (1: 11 – 40 years, 2: 41 – 80 years, 3: related ALS data and derivatives are provided free of >80 years). In this context, Beech stratum (B1–3), Oak charge based on the order addressed to the National stratum (O1–3), and a Fir stratum (F1–3) was created. Forest Centre. Furthermore, sample plots were divided into two groups. ALS data acquisition was performed in June 2021 The main criteria for dividing was proportionality of over the entire study area (17,583 ha). This area was these both datasets related to the number of plots, area scanned using a Leica ALS 70 CM scanner from average of plots and stand attributes. In this context, group of 180 altitude of 1,290 m with field of view of 43° and 282 kHz training plots (20 plots per stratum) and group of 180 laser pulse repetition rate. The resulting average density testing plots (20 plots per stratum) was created (Table 1). Fig. 1. The study area located in Slovakia, Central Europe. 225 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Table 1. Statistics of ground data from 180 training plots and 180 testing plots (20 plots per stratum). Training Data Testing Data Stratum Height Diameter Volume Height Diameter Volume 3 −1 3 −1 [m] [cm] [m ha ] [m] [cm] [m ha ] All 18.9 ±4.5 23.0 ±6.9 375.9 ±195.2 18.7 ±4.9 22.2 ±7.3 383.5 ±205.0 B1 15.5 ±3.8 15.5 ±2.6 233.2 ±140.2 15.4 ±4.0 15.4 ±2.9 230.8 ±139.8 B2 20.1 ±2.9 22.1 ±3.7 407.2 ±112.6 20.3 ±3.2 22.3 ±3.5 413.4 ±124.5 B3 22.8 ±2.8 31.0 ±6.6 522.5 ±153.0 22.4 ±3.5 30.5 ±7.2 542.8 ±191.4 O1 15.0 ±3.8 15.2 ±2.7 185.8 ±107.5 15.2 ±3.5 15.4 ±2.3 197.6 ±99.2 O2 19.0 ±2.9 21.9 ±2.9 342.1 ±117.3 19.1 ±3.0 22.5 ±2.9 363.5 ±122.9 O3 21.2 ±2.2 30.4 ±6.1 402.0 ±126.5 20.6 ±3.1 28.2 ±6.5 421.8 ±130.2 F1 13.6 ±3.7 14.8 ±2.6 234.3 ±166.1 13.2 ±3.2 14.4 ±2.7 218.3 ±185.0 F2 19.8 ±4.0 23.7 ±3.2 513.7 ±146.6 20.0 ±3.7 23.7 ±3.1 499.5 ±117.9 F3 23.3 ±4.9 34.7 ±7.0 600.0 ±235.8 23.1 ±4.8 34.5 ±7.5 620.7 ±281.8 Note: All: All sample plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum; Height: arithmetic mean height; Diameter: arithmetic mean diameter; Volume: volume per hectare. the forest area include predicted mean height, mean 2.4. CHM-based Forest Inventory diameter and volume per hectare. The workflow for CHM-based forest inventory is shown in Fig. 2 and described in detail in the following points: – The height metrics (i.e., mean, median, standard deviation, minimum, maximum, range, minority, majority) were computed from CHM for specific areas of training plots (n = 180) using zonal statis- tic tool implemented in the QGIS software (Madry 2021). Specic fi ally, the CHM-based mean height was assessed as most highly correlated height metric to all related forest stand attributes (i.e., mean height, mean diameter, and volume per hectare) using Pear- son’s correlation analysis (Voght & Johnson 2012). – The simple nonlinear models for prediction of forest stand attributes were computed from training data using least-squares regression implemented in the R package NLStools (Baty et al. 2015). These models express the relationship between ground-observed forest stand attributes and CHM-based mean height. Here, only statistically significant and best perform- ing models for prediction of specific forest stand Fig. 2. Flowchart for CHM-based forest inventory. Note: GR: attribute (i.e., mean height, mean diameter, and ground; ALS: airborne laser scanning; SA: forest stand at- volume per hectare) were selected and used. The tribute. statistical significance was assessed using Fisher’s test (F-test). The goodness of fit was assessed using error metrics of regression model. 2.5. Accuracy Assessment – The mean heights were computed from CHM for The accuracy of the CHM-based forest inventory was specific areas of testing plots (n = 180) using zonal assessed by comparing the observed and predicted for- statistic tool function implemented in the QGIS soft- est stand attributes from 180 testing plots. The value of ware. Subsequently, the forest stand attributes were observed forest stand attributes for each sample plot was predicted for testing plots through the developed obtained by ground measurement (Section 2.3). The regression models. In this way, each of testing plots value of predicted forest stand attributes for each sample include observed as well as predicted mean height, plot was obtained by CHM-forest inventory (Section 2.4). mean diameter and volume per hectare. Finally, the The standard error (Bias, Equation (1)) and root accuracy of the CHM-based forest inventory was mean square error (RMSE, Equation (2)) were used to assessed by comparing the observed and predicted assess the model’s performance. The relative %Bias and forest stand attributes. %RMSE were calculated as the ratios of their absolute – The regular 20 × 20 m grid covering the forest area values and the arithmetic average of the observed data. was created and the CHM-based mean heights were The regression function, coefc fi ient of determination computed for all grid cells. The size of the grid cells (R ), and F-test of statistical signic fi ance of the regression (400 m ) was set to reflect the size of the ground plots model (p-value) were calculated to assess the strength of (314 m ). Subsequently, the forest stand attributes the relationship between predicted and observed forest were predicted for all grid cells through the developed stand attributes (Voght & Johnson 2012). regression models. In this way, each grid cell covering 226 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 The t-test, when a normal distribution of mean differ- area (Fig. 3). In this context, the RMSE of CHM-based ences was confirmed, or Wilcoxon test, when a normal prediction of forest stand attributes did not exceed 2.0 m distribution of mean differences was not confirmed (Fay (10.9%) for mean height, 5.1 cm (23.1%) for mean 3 −1 & Proschan 2010), were used to assess the significance diameter and 132.1 m ha (34.5%) for volume (Table of differences (p-value < 0.05). 3). These predictions were not statistically significantly different relative to the ground-observed data (p-value < 0.05). The relationships (R ) between CHM-predicted L% DV \Ö  \ ¦ L L [1] L  and ground-observed forest stand attributes fluctuated from 0.7 to 0.8 (Fig. 4). 05 6( \  \  Q [2] L L Table 2. Regression models for CHM-based forest inventory. Attribute Model Form N R SEE SEE% p-value Height 2.57 × (H^0.69) + E 180 0.81 2.05 10.97 <0.001 where n is the number of plots, y is the observed forest Diameter 9.84 × EXP(0.04 × H) + E 180 0.74 5.11 22.98 <0.001 stand attribute for plot i, and ݕො is the predicted forest Volume 1.99 × (H^1.80) + E 180 0.70 127.80 33.33 <0.001 Note: H: CHM-based mean height; E: error of prediction; N: number of training plots; R : co- stand attribute for plot i. efficient of determination; SEE: standard error of estimates; SEE%: relative standard error of estimates; p-value of F-test, null hypothesis is rejected at α = 0.05. 3. results Table 3. Accuracy of CHM-based forest inventory. Height Diameter Volume Stratum The form and performance of regression models for %Bias %RMSE %Bias %RMSE %Bias %RMSE CHM-based forest inventory are presented in Table 2. All −0.60 10.89 −2.72 23.07 −0.62 34.45 B1 3.88 11.67 19.60 26.83 13.44 41.64 We found the power function to be most suitable for B2 0.22 9.62 5.93 16.48 9.80 31.04 prediction of height as well as volume and exponential B3 −2.68 11.07 −14.44 21.83 0.45 24.42 O1 2.18 11.13 15.76 23.14 22.54 48.33 function for prediction of diameter. All predictions were O2 0.14 12.20 −2.92 15.30 7.37 29.63 unbiased (p-value < 0.05) and the strength of the rela- O3 −6.47 11.66 −21.33 27.78 −4.54 34.78 F1 1.52 11.75 8.38 14.34 −30.10 55.93 tionships between CHM-predicted and ground-observed F2 −0.01 9.82 −0.99 17.39 −10.46 31.20 forest stand attributes were relative high (R = 0.7–0.8). F3 −5.77 10.63 −24.43 30.51 −12.90 34.22 The CHM-based forest inventory evaluated the mean Note: All: All testing plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum; %Bias: relative standard error; %RMSE: relative root mean square error; Height: arithmetic mean height at 18.6 ±4.0 m, mean diameter at 21.6 ±5.0 cm height; Diameter: arithmetic mean diameter; Volume: volume per hectare. 3 −1 and volume at 381.1 ±183.8 m ha within the study Fig. 3. Ground-observed and CHM-predicted forest stand attributes including standard deviation (whiskers). Note: All: All test- ing plots; B1–3: Beech stratum; O1–3: Oak stratum; F1–3: Fir stratum. 227 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Fig. 4. Relationship between ground-observed and CHM-predicted forest stand attributes from all testing plots. The dashed line represents a 1:1 correspondence. The n fi al result of CHM-based forest inventory repre - 4. Discussion sents a raster layer with resolution of 20 × 20 m consisting Forest inventory is strictly required by relevant stake- values of forest stand attributes (i.e., mean height, mean holders (e.g., forest managers, environmentalist, and diameter, and volume per hectare). This result is provided policy-makers). However, ground-based measurements to the stakeholders directly as a product (i.e., raster or are expensive and time consuming. The RS technologies vector layer) or indirectly through a web-map service has become a reliable source of geospatial data and tech- (i.e., website link). The illustrative example of 3D visu- niques for forest inventory. Especially, ALS-based CHM alization representing CHM-predicted volume for part of represents an available and usable data source (Kakou- study area over grid cell 20 × 20 m is displayed in Fig. 5. laki et al. 2021) for prediction of forest stand attributes and related techniques provide an opportunity to comple- ment ground-based measurements. Fig. 5. Visualization of 3D map representing CHM-predicted forest stand attribute for part of study area over grid cell 20 × 20 m (illustrative example for volume per hectare). 228 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 The objective of this study was, therefore, to assess Although the CHM-based predictive models are not the accuracy of forest inventory performed using ALS- sufficiently able to describe the full range of variability based CHM from regular aerial campaign and free avail- of forest stand attributes, there are several options to able software tools, such as QGIS and packages for R improve the performance of CHM-based forest inven- environment. This CHM-based forest inventory was tory. For example, (1) higher position accuracy of sample conducted in Central European heterogeneous forests plots (Zhang et al. 2017; Novo-Fernández et al. 2019), covering 7,483 ha and related accuracy was assessed on (2) higher vertical accuracy and resolution of CHM 180 testing plots covering 5.6 ha. (Deluzet et al. 2022; Lisiewicz et al. 2022), or (3) appli- In this study, the bias of CHM-based forest inventory cation of multiple nonlinear regression or nonparametric for all strata was −0.6% in the case of mean height as well machine learning methods (Jiang et al. 2020; Zhao et al. as volume and −2.7% in the case of mean diameter. These 2022) could contribute to develop predictive models with results were mainly related to the underestimation of all higher accuracy. forest stand attributes within the strata representing old Alternatively, the point cloud-based forest inventory forest (B3, O3, F3) and relative high underestimation of is more suitable solution in heterogeneous forest stands. volume in strata representing young fir forest (F1). The For example, the point cloud-based forest inventory per- prediction of other forest stand attributes within the strata formed for the same study area achieved the %RMSE representing young forest were overestimated (B1, O1). which did not exceed 7.2% for mean height, 8.6% for The RMSE of CHM-based forest inventory reached the mean diameter, and 21.4% for total volume (Sačkov et al. value of 10.9%, 23.1%, and 34.5% for the mean height, 2019a; Sačkov et al. 2019b). Moreover, similar or better mean diameter, and volume, respectively. While the pre- results have also been provided by other authors, such as diction of mean height achieved homogenous accuracy study from Brazilian forests (Martins-Neto et al. 2021), within the strata (10–12%), range of errors related to the Canadian forests (Lamb et al. 2018), or Italian forests prediction of mean diameter and volume was consider- (Versace et al. 2020). These and other studies provided able. Specic fi ally, the overall error of mean diameter pre - %RMSE at 6%–14%, 9%–20%, and 17%–23% for the diction u fl ctuated from 14% (F1) to 31% (F3) and overall mean height, mean diameter, and volume, respectively. error of volume prediction fluctuated from 24% (B3) to 56% (F1). Thus, the largest prediction error was found in the youngest (esp., volume) and oldest (esp., diam- 5. Conclusions eter) forest stands (Fig. 2). Although this trend has been found also in similar studies (Hill et al. 2014; Cosenza et This study assessed the accuracy of forest inventory al. 2018; Mielcarek et al. 2018), statistical significance performed using ALS-based CHM and free available of increasing or decreasing the development stage on software tools, such as QGIS and packages for R envi- prediction accuracy was not confirmed. However, it is ronment. Our results confirmed that CHM-based for - evident that level of RMSE was related to the variability est inventory represents a practical solution that allows of forest stand characteristics. The overall accuracy of make a general overview of natural resources for the CHM-forest inventory decreased with increasing vari- whole forest unit. Moreover, this solution can be per- ability of forest stand characteristics. For example, the formed directly and repeatedly by stakeholders without variability of volume is signic fi antly higher than variabil - the need for additional investments focused on data, soft- ity of mean height (Fig. 6) and consequently the RMSE ware and hardware. On the other hand, overall accuracy of CHM-predicted volume is significantly higher than of CHM-based forest inventory in this study did not meet RMSE of CHM-predicted mean height (Table 3). the requirements of forest management from some coun- tries in Central Europe (e.g., Slovakia). Consequently, continuous research and development of CHM-based forest inventory seems to be valid and necessary. Acknowledgments This research was supported by the Slovak Research and Devel- opment Agency in the framework of the project “Development of advanced geospatial technologies for multiscale monitoring of forest ecosystems” (APVV-19-0257) as well as by the Euro- pean Regional Development Fund in the framework of the project “Research and development of contactless methods for acquisi- tion of geospatial data usable in the forest monitoring in order Fig. 6. Relative variability of ground-observed forest stand to streamline forest management and increase forest protection” attributes. Note: All: All sample plots; B1–3: Beech stratum; (NFP313011V465). O1–3: Oak stratum; F1–3: Fir stratum. 229 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Liu, Q., Fu, L., Chen, Q., Wang, G., Luo, P., Sharma, R. references P. et al., 2020: Analysis of the Spatial Differences in Baty, F., Ritz, Ch., Charles, S., Brutsche, M., Flandrois, Canopy Height Models from UAV LiDAR and Pho- J. P., Delignette-Muller, M., 2015: A Toolbox for togrammetry. Remote Sensing, 12:2884. Nonlinear Regression in R – The Package nlstools. Madry, S., 2021: Introduction to QGIS: Open Source Journal of Statistical Software, 66:1–21. Geographic Information System. USA, Locate Press Cosenza, D., Soares, V., Leite, H., Gleriani, J., Amaral, C., LLC, 219 p. Junior, J. et al., 2018: Airborne laser scanning applied Martins-Neto, R. P., Tommaselli, A. M. G., Imai, N. N., to eucalyptus stand inventory at individual tree level. David, H. C., Miltiadou, M., Honkavaara, E., 2021: Pesquisa Agropecuaria Brasileira, 53:1373–1382. Identification of Significative LiDAR Metrics and Deluzet, M., Erudel, T., Briottet, X., Sheeren, D., Fabre, Comparison of Machine Learning Approaches for S., 2022: Individual Tree Crown Delineation Method Estimating Stand and Diversity Variables in Hetero- Based on Multi-Criteria Graph Using Geometric and geneous Brazilian Atlantic Forest. Remote Sensing, Spectral Information: Application to Several Temper- 13:2444. ate Forest Sites. Remote Sensing, 14:1083. Mielcarek, M., Stereńczak, K., Khosravipour, A., 2018: Díaz-Varela, R. A., González-Ferreiro, E., 2021: 3D Point Testing and evaluating different LiDAR-derived can- Clouds in Forest Remote Sensing. Remote Sensing, opy height model generation methods for tree height 13:2999. estimation. International Journal of Applied Earth Fay, M. P., Proschan, M. A., 2010: Wilcoxon-Mann- Observation and Geoinformation, 71:132–143. Whitney or t-test? On assumptions for hypoth - Murgaš, V., Sačkov, I., Sedliak, M., Tunák, D., Chudý esis tests and multiple interpretations of decision F., 2018: Assessing horizontal accuracy of inven- rules. Statistics Survey, 4:1–39. tory plots in forests with different mix of tree species Ginzler, C., Hobi, M. L., 2015: Countrywide Stereo- composition and development stage. Journal of For- Image Matching for Updating Digital Surface Mod- est Science, 64:478–485. els in the Framework of the Swiss National Forest Novo-Fernández, A., Barrio-Anta, M., Recondo, C., Inventory. Remote Sensing, 7:4343–4370. Cámara-Obregón, A., López-Sánchez, C. A., 2019: Goodbody, T. R. H., Coops, N. C., Luther, J. E., Tom- Integration of National Forest Inventory and Nation- palski, P., Mulverhill, C. F., Fournier, R. et al., 2021: wide Airborne Laser Scanning Data to Improve Airborne laser scanning for quantifying criteria and Forest Yield Predictions in North-Western Spain. indicators of sustainable forest management in Can- Remote Sensing, 11:1693. ada. Canadian Journal of Forest Research, 51:14. Petráš, R., Pajtík, J., 1991: Sústava česko-slovenských Hill, A., Breschan, J., Mandallaz, D., 2014: Accuracy objemových tabuliek drevín. Lesnícky časopis, Assessment of Timber Volume Maps Using Forest 37:49–56. Inventory Data and LiDAR Canopy Height Models. Polidori, L., El Hage, M., 2020: Digital Elevation Model Forests, 5:2253–2275. Quality Assessment Methods – A Critical Review. Jamru, L. R., 2018: Correction pit free canopy height Remote Sensing, 12:3522. model derived from LiDAR data for the broad leaf Sačkov, I., Kulla, L., Bucha, T., 2019a. A Comparison of tropical forest. IOP Conference Series: Earth and Two Tree Detection Methods for Estimation of For- Environmental Science, 169:012113. est Stand and Ecological Variables from Airborne Jiang, X., Li, G., Lu, D., Chen, E., Wei, X., 2020: Strati- LiDAR Data in Central European Forests. Remote c fi ation-Based Forest Aboveground Biomass Estima - Sensing, 11:1431. tion in a Subtropical Region Using Airborne Lidar Sačkov, I., Scheer, L., Bucha, T., 2019b. Predicting for- Data. Remote Sensing, 12:1101. est stand variables from airborne LiDAR data using Kakoulaki, G., Martinez, A., Florio, P., 2021: Non-com- a tree detection method in Central European forests. mercial Light Detection and Ranging (LiDAR) data Central European Forestry Journal, 65:191–197. in Europe. Luxembourg, Publications Office of the Surový, P., Kuželka, K., 2019: Acquisition of Forest European Union, 35 p. Attributes for Decision Support at the Forest Enter- Lamb, S. M., MacLean, D.A., Hennigar, C.R., Pitt, D.G, prise Level Using Remote-Sensing Techniques - A 2018: Forecasting Forest Inventory Using Imputed Review. Forests, 10:273. Tree Lists for LiDAR Grid Cells and a Tree-List Vauhkonen, J., Maltamo, M., McRoberts, R. E., Næs- Growth Model. Forests, 9: 167. set, E., 2014: Introduction to forestry applications Lisiewicz, M., Kamińska, A., Kraszewski, B., Stereńczak, of airborne laser scanning. In: Maltamo, M., Næs- K., 2022: Correcting the Results of CHM-Based Indi- set, E., Vauhkonen, J. (eds.): Forestry Application of vidual Tree Detection Algorithms to Improve Their Airborne Laser Scanning: Concept and Case Studies. Accuracy and Reliability. Remote Sensing, 14: 1822. Springer Netherlands: Dordrecht, The Netherlands, p. 1–16. 230 I. Sackov / Cent. Eur. For. J. 68 (2022) 214–231 Versace, S., Gianelle, D., Frizzera, L., Tognetti, R., Garfì, Zhao, J., Zhao, L., Chen, E., Li, Z., Xu, K., Ding, X., V., Dalponte, M., 2019: Prediction of Competition 2022: An Improved Generalized Hierarchical Esti- Indices in a Norway Spruce and Silver Fir-Dominated mation Framework with Geostatistics for Mapping Forest Using Lidar Data. Remote Sensing, 11:2734. Forest Parameters and Its Uncertainty: A Case Study Voght, W. P., Johnson, R. B., 2012: Correlation and of Forest Canopy Height. Remote Sensing, 14:568. Regression Analysis. USA, SAGE Publications Ltd, Zhen, Z., Quackenbush, L. J., Zhang, L., 2016: Trends 1632 p. in Automatic Individual Tree Crown Detection and Wang, H., Seaborn, T., Wang, Z., Caudill, Ch., Link, T., Delineation – Evolution of LiDAR Data. Remote 2021: Modeling tree canopy height using machine Sensing, 8:333. learning over mixed vegetation landscapes. Inter- Other sources national Journal of Applied Earth Observation and QGIS Development Team, 2022: QGIS Geographic Geoinformation, 101:102353. Information System. Open Source Geospatial Foun- Zhang, Z., Cao, L., She, G., 2017: Estimating For- dation Project. http://qgis.osgeo.org. est Structural Parameters Using Canopy Metrics R Core Team, 2021: R: A language and environment for Derived from Airborne LiDAR Data in Subtropical statistical computing. R Foundation for Statistical Forests. Remote Sensing, 9:940. Computing. https://www.R-project.org. Zhang, W., Cai, S., Liang, X., Shao, J., Hu, R., Yu, S. et al., 2020: Cloth simulation-based construction of pit-free canopy height models from airborne LiDAR data. Forest Ecosystems, 7:1–13.

Journal

Forestry Journal – de Gruyter

Published: Dec 1, 2022

Keywords: airborne LiDAR; natural resources; area-based approach; freeware

There are no references for this article.