No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Modelling growing stock volume of forest stands with various ALS area-based approaches
Abstract
Airborne laser scanning (ALS) is one of the most innovative remote sensing tools with a recognized important utility for characterizing forest stands. Currently, the most common ALS-based method applied in the estimation of forest stand characteristics is the area-based approach (ABA).
The aim of this study was to analyse how three ABA methods affect growing stock volume (GSV) estimates at the sample plot and forest stand levels. We examined (1) an ABA with point cloud metrics, (2) an ABA with canopy height model (CHM) metrics and (3) an ABA with aggregated individual tree CHM-based metrics. What is more, three different modelling techniques: multiple linear regression, boosted regression trees and random forest, were applied to all ABA methods, which yielded a total of nine combinations to report. An important element of this work is also the empirical verification of the methods for estimating the GSV error for individual forest stand. All nine combinations of the ABA methods and different modelling techniques yielded very similar predictions of GSV for both sample plots and forest stands. The root mean squared error (RMSE) of estimated GSV ranged from 75 to 85 m3 ha−1 (RMSE% = 20.5–23.4 per cent) and from 57 to 64 m3 ha−1 (RMSE% = 16.4–18.3 per cent) for plots and stands, respectively. As a result of the research, it can be concluded that GSV modelling with the use of different ALS processing approaches and statistical methods leads to very similar results. Therefore, the choice of a GSV prediction method may be more determined by the availability of data and competences than by the requirement to use a particular method.
... As summarized by Maltamo et al. (2021) and Naesset (2015), ALS-based forest management inventories are now the operational standard in several Nordic countries and have almost entirely replaced partly subjective stand-level assessment in the field. The area-based approach is also increasingly used for operational forest management in North America , White et al. 2021 and Poland (Parkitna et al. 2021). ...
... In the context of operational forest management, the error associated with the estimate of an individual pixel for which biomass or wood volume is predicted may be notably larger than for a forest stand comprising multiple pixels, because local over-and underestimations may cancel each other out. Such observations have been, e.g. reported in studies examining plot size effects on biomass estimates (Fassnacht et al. 2018;Parkitna et al. 2021). Hence, the way the question was posed in the survey (with no indication of spatial scale) may be inadequate to truly understand the needs of practitioners. ...
Despite decades of development, the uptake of remote sensing-based information products in the forestry sector is still lagging behind in central and southern Europe. This may partly relate to a mismatch of the developed remote sensing products and the requirements of potential users. Here, we present the results of a questionnaire survey in which we questioned 355 forest practitioners from eight central and southern European countries. We aimed to learn about forest practitioners' technical requirements for four remote sensing-based information products, including information on tree species, canopy height, wood volume/biomass, and forest disturbances. We asked for practitioners’ preferences with respect to thematic and spatial detail as well as the maximal acceptable error and the temporal frequency with which the information layers would be needed. We then examined whether the education, age, and professional background affect the requirements. Preferences with respect to spatial and thematic detail were comparably diverse while more homogenous patterns could be observed for demands with respect to errors and temporal frequency. Our results indicate that for some information products such as canopy height maps, existing remote sensing technology, and workflows can match all demands of practitioners. Remotely sensed information on forest disturbances partly fulfils the demands of the practitioners while for products related to tree species and wood volume/biomass the level of thematic detail and the accuracy of the products demanded by practitioners in central and southern Europe is not yet fully matched. We found no statistically significant differences between the demographic groups examined. The findings of this study improve our understanding of matches and mismatches of the technical requirements of practitioners for remote sensing-based information products.
... An alternative is the utilization of virtual sample plots created by an expert based on a set of various remote sensing materials, such as a CHM or orthophoto maps. The expert can establish and outline sample plots along with tree crown segments manually [45,46]. ...
The accurate detection of individual tree crowns and estimation of tree density is essential for effective forest management, biodiversity assessment, and ecological monitoring. The precision of tree crown detection algorithms plays a critical role in providing reliable data for these applications, where even slight inaccuracies can lead to significant deviations in tree population estimates and ecological indicators. Various algorithmic parameters, such as pixel size and crown segmentation thresholds, can substantially impact tree crown detection accuracy. This study aims to explore the influence of tree stand features and parameters on the effectiveness of the individual tree crown detection method based on a watershed algorithm, leading to identifying optimal configurations that enhance the reliability of forest inventories and support sustainable management practices. Our analysis of the algorithm results shows that the features of the tree stand, such as tree height variance and tree crown size variance, significantly impact the algorithm’s output in precisely estimating tree count. Consequently, adjusting the pixel size of a canopy height model in the context of tree stand features is necessary to minimize error. Additionally, our findings show that there is a need to carefully assess the criterion of membership of a detected tree crown in a circular sample plot, which we based on the point cloud.
... Polish experience with ALS-based two-phase forest inventory shows GSV estimation error rates between 16 and 25% RMSE [47,49]. In other countries, reported errors in ABA-ALS forest stock inventories range between 10 and 30% RMSE [50]. ...
Remote sensing (RS)-based forest inventories are becoming increasingly common in forest management. However, practical applications often require subsequent optimisation steps. One of the most popular RS-based forest inventory methods is the two-phase inventory with regression estimator, commonly referred to as the area-based approach (ABA). There are many sources of variation that contribute to the overall performance of this method. One of them, which is related to the core aspect of this method, is the spatial co-registration error between ground measurements and RS data. This error arises mainly from the imperfection of the methods for positioning the sample plots under the forest canopy. In this study, we investigated how this positioning accuracy affects the area-based growing stock volume (GSV) estimation under different forest conditions and sample plot radii. In order to analyse this relationship, an artificial co-registration error was induced in a series of simulations and various scenarios. The results showed that there were minimal differences in ABA inventory performance for displacements below 4 m for all stratification groups except for deciduous sites, where sub-metre plot positioning accuracy was justified, as site- and terrain-related factors had some influence on GSV estimation error (r up to 0.4). On the other hand, denser canopy and spatially homogeneous stands mitigated the negative aspects of weaker GNSS positioning capabilities under broadleaved forest types. In the case of RMSE, the results for plots smaller than 400 m² were visibly inferior. The BIAS behaviour was less strict in this regard. Knowledge of the actual positioning accuracy as well as the co-registration threshold required for a particular stand type could help manage and optimise fieldwork, as well as better distinguish sources of statistical uncertainty.
... These efforts can be categorized into two approaches: 1) Adding individual tree features into ABA (ABAITD): This approach combines individual tree detection (ITD, explained in Section 1.4.2) with ABAOrdinary methods by averaging the features extracted from ITD segments within plots and using them for statistical modeling of forest attributes. This approach was initially introduced by Hyyppä et al. (2012), then further investigated and verified in other studies in mature forests (Breidenbach et al. 2012;Vastaranta et al. 2012;Shinzato et al. 2017;Parkitna et al. 2021;Kelley et al. 2022). ...
Seedling stands are areas in forest landscapes where young trees, typically from newly planted or naturally regenerated seedlings, grow. These stands are in the early stages of forest development and are crucial for the renewal and future growth of the forest. They represent a vital phase in the forest's lifecycle, for which careful management is often employed to ensure the successful establishment and growth of young crop trees. To address the data-gathering requirements of forest management, seedling stands are typically assessed through field visits, a process that is considered time-consuming, expensive, and labor-intensive. As trees in the seedling stands are small and often densely stocked, they are difficult to assess in operational remote sensing-based forest inventories. However, recent developments in remote sensing, especially in laser scanning and the use of drones, could open new pathways to developing methods for the spatially explicit and timely inventorying of seedling stands; such methods could complement or even replace field visits. The aim here was to develop and assess remote sensing methods of estimating the tree density, mean tree height, and species of seedling stands, which are the key characteristics supporting forest management. For this purpose, new remote sensing techniques–namely drone photogrammetric point clouds, hyper- and multi-spectral imagery (studies I and IV), and multi-spectral and single-photon airborne laser scanning (ALS; studies II and III) data–were investigated over seedling stands located in three study sites in the boreal forests of Finland. Performance of leaf-off and leaf-on hyper-spectral drone imagery and multi-spectral ALS data was explored in seedling stands in studies I and II. A canopy-thresholding method (Cth) was also optimized to minimize the interference of understory vegetation (study II), and the performance of single-photon ALS was examined in study III. In that study, an area-based approach (ABA) that included single-tree features and corrected the effect of edge trees (ABAEdgeITD) was developed and compared to conventional ABA. In study IV, a new approach for feeding multispectral drone images to convolutional neural networks was proposed and validated for the classification of seedling tree species. The findings of this thesis demonstrated that drone imagery yielded more accurate tree density estimates, while dense multispectral ALS data outperformed other tested methods of tree height estimation (both when using leaf-on data). The use of ABAEdgeITD improved the tree density and height estimates compared to conventional ABA, although it was less accurate than the individual tree-based methods used in studies I and II. Characterization of advanced seedling stands was more accurate than that of early-growth stage stands (mean height < 1.3 m), which remained challenging. Finally, the image pre-processing approach, together with the convolutional neural network, used in study IV improved the species classification accuracy of seedlings. This thesis shows that the remote sensing methods used can be applied in operational forest inventories to complement or replace field visits. These new technologies are valuable approaches to increasing the efficiency and sustainability of forest management.
... Regarding AGC mapping, we basically used the LRM because the error metrics of the two models were similar, and use of the NLRM in GIS would be challenging due to the complex structure of nonlinear equations compared to simple LRM. Although LRMs have been extensively used in remote sensing of forest attributes (Parkitna et al. 2021, Narin et al. 2022, the models' abilities to explain observed data patterns are rarely examined. Among a few studies, Fleming et al. (2015) have cautioned researchers about the potential for LRMs to yield negative values in relation to the estimation of forest attributes through remote sensing. ...
Flooded forests are rare and highly dynamic ecosystems, yet they can store a significant amount of carbon because of their ability to produce biomass rapidly. Estimation and mapping of the carbon that is stored in flooded forests are challenging tasks through the use of optical remote sensing because these ecosystems are often located in moist regions where clouds can interfere with data acquisition and image interpretation. This study models the aboveground carbon (AGC) stocks of a flooded forest in Turkish Thrace with synthetic aperture radar (SAR) data, which are less affected by weather and illumination conditions compared to optical imagery. Forest management plan data, including inventory records of 229 sample plots, a detailed forest cover map, and stand tables of the 2,119-ha Igneada Longoz Forest, were used to calculate AGC and to develop spatially explicit models based on ALOS/PALSAR-2 (Advanced Land Observing Satellite/Phased Array L-band Synthetic Aperture Radar) and Landsat-8 images. The results indicated that the horizontally transmitted and horizontally received (HH) and cross-polarization ratio (CPR) bands of ALOS/PALSAR were the most influential variables in the linear and nonlinear regression models. The models did not include any variables from either radar- or optical-based vegetation indices. While the estimation accuracies of the two models were similar (root mean square percentage error ≈ 26%), the linear model yielded negative estimations in several land cover classes (e.g., dune, forest opening, degraded forest). AGC stock was estimated and mapped using the nonlinear model in these cases. The density map revealed that Igneada Longoz Forest stored 279,258.9 t AGC, with a mean and standard deviation of 124 ± 115.4 t C ha-1. AGC density varied significantly depending on stand types and management units across the forest, and carbon hotspots accumulated in the northern and southern sites of the study area, primarily composed of ash and alder seed stands. The models and maps that this study developed are expected to help in the rapid and cost-effective assessment of AGC stored in flooded forest ecosystems across the temperate climate zone.
... By modeling drone-derived photogrammetric point cloud data, they were able to obtain a stand-level rRMSE of 13.1% for locally calibrated models and 20% when a nationwide model was applied. Parkitna et al. [83] and Lisańczuk et al. [21] reported rRMSE ranges of 16%-23% and 12%-20%, respectively, for Scots pine-dominated stands using the area-based ALS method. On the other hand, in our study, some estimation errors were as low as 0.6%. ...
Low-altitude aerial photogrammetry can be an alternative source of forest inventory data and a practical tool for rapid forest attribute updates. The availability of low-cost unmanned aerial systems (UASs) and continuous technological advances in terms of their flight duration and automation capabilities makes these solutions interesting tools for supporting various forest management needs. However, any practical application requires a priori empirical validation and optimization steps, especially if it is to be used under different forest conditions. This study investigates the influence of the main flight parameters, i.e., ground sampling distance and photo overlap, on the performance of individual tree detection (ITD) stand-level forest inventories, based on photogrammetric data obtained from budget unmanned aerial systems. The investigated sites represented the most common forest conditions in the Polish lowlands. The results showed no direct influence of the investigated factors on growing stock volume predictions within the analyzed range, i.e., overlap from 80 × 80 to 90 × 90% and GSD from 2 to 6 cm. However, we found that the tree detection ratio had an influence on estimation errors, which ranged from 0.6 to 15.3%. The estimates were generally coherent across repeated flights and were not susceptible to the weather conditions encountered. The study demonstrates the suitability of the ITD method for small-area forest inventories using photogrammetric UAV data, as well as its potential optimization for larger-scale surveys.
... The primary goal of forest management is to obtain precise individual tree metrics, including tree species, location, height, diameter at breast height (DBH), and crown dimensions (González-Ferreiro et al. 2012;Hu et al. 2014;Koch et al. 2006;Maltamo and Gobakken 2014). These metrics are crucial for estimating forest characteristics, such as tree species composition, growing stock volume, canopy density, and mean basal area Lee et al. 2013;Parkitna et al. 2021;Unger et al. 2014). However, the accuracy of these parameters is influenced by errors in various individual tree detection methods, including treetop detection, crown delineation, and tree segmentation. ...
In the past decade, the use of three-dimensional forest information from airborne Light Detection and Ranging (LiDAR) has become widespread in forest inventories. Accurate Individual Treetop Detection (ITD) and crown boundary delineation using LiDAR data are critical for obtaining precise inventory metrics. To address this need, we introduced a novel growing tree region (GTR)-driven ITD method that utilizes canopy height models (CHM) derived from very low-resolution airborne LiDAR data. The GTR algorithm consists of three key stages: (i) preserving all height layers through incremental cutting and stacking of CHM; (ii) employing a three-layer concept to identify individual treetops; and (iii) refining the detected treetops using a distance-based filter. Our method was tested in five temperate forests across Central Europe and was compared against the widely-used local maxima (LM) search combined with an optimized variable window filtering (VWF) technique. Our results showed that the GTR method outperformed LM with VWF, particularly in forests with high canopy density. The achieved root mean square accuracies were 74% for the matching rate, 19% for commission errors, and 27% for omission errors. In comparison, the LM with the VWF method resulted in a matching rate of 71%, commission errors of 20%, and omission errors of 31%. To facilitate the application of our algorithm, we developed an R package called TREETOPS, which seamlessly integrates with the lidR package, ensuring compatibility with existing treetop-based segmentation methods. By introducing TREETOPS, we provide the most accurate open-source tool for detecting treetops using low-resolution LiDAR-derived CHM.
... Recently, LiDAR technology has been successfully used in forestry surveys in many of its forms-airborne laser scanning (ALS), including unmanaged aerial vehicles (UAV), terrestrial laser scanning (TLS), mobile laser scanning (MLS), and handheld or personal laser scanning (PLS) [1][2][3], each producing specific results in aspects of the precision, scale, and parameters measured. The two main approaches [4,5] to derive forest information from laser scanning data are the statistical area-based approach (ABA) [6][7][8] and individual tree detection (ITD) [9][10][11][12], including their combination. ...
GNSS/INS-based positioning must be revised for forest mapping, especially inside the forest. This study deals with the issue of the processability of GNSS/INS-positioned MLS data collected in the forest environment. GNSS time-based point clustering processed the misaligned MLS point clouds collected from skid trails under a forest canopy. The points of a point cloud with two misaligned copies of the forest scene were manually clustered iteratively until two partial point clouds with the single forest scene were generated using a histogram of GNSS time. The histogram’s optimal bin width was the maximum bin width used to create the two correct point clouds. The influence of GNSS outage durations, signal strength statistics, and point cloud parameters on the optimal bin width were then analyzed using correlation and regression analyses. The results showed no significant influence of GNSS outage duration or GNSS signal strength from the time range of scanning the two copies of the forest scene on the optimal width. The optimal bin width was strongly related to the point distribution in time, especially by the duration of the scanned plot’s occlusion from reviewing when the maximum occlusion period influenced the optimal bin width the most (R² = 0.913). Thus, occlusion of the sub-plot scanning of tree trunks and the terrain outside it improved the processability of the MLS data. Therefore, higher stem density of a forest stand is an advantage in mapping as it increases the duration of the occlusions for a point cloud after it is spatially tiled.
... Lidar can obtain detailed structural information of forest areas. However, it is expensive and unsuitable for estimating large areas such as the whole country or the world [8], [9], [10]. Active microwave remote sensing represented by Synthetic Aperture Radar (SAR) can penetrate clouds and fog without being limited by time and space, and has received increasing attention. ...
The forest is an important part of carbon resources. Forest growing stock volume (GSV) is an important parameter of forest. The Water Cloud Model (WCM) is a simple equation that describes the interaction between ground objects and electromagnetic waves. It has also been applied in the estimation of forest GSV. When estimating GSV, the WCM equation parameters are usually calculated using least squares, but the least squares method relies on field reference data. The subsequent WCM development algorithm BIOMASAR uses a sliding window method that does not rely on measured data. However, the sliding window method is inefficient and can easily lead to missing pixels. We designed the backscatter/FVC feature space based on WCM and BIOMASAR to estimate forest GSV. Comparing with the NFI reference data set and the BIOMASAR algorithm results in the study area, the method is evaluated from three aspects: accuracy, efficiency, and texture. The results show that this method does not rely on actual reference data, and the efficiency is increased from 1661s in the sliding window to 663s. The correlation with the NFI reference data is 0.45, the RMSE is 116.2327 m³/ha, and the RRMSE is 64.86%. The accuracy is better than the BIOMASAR sliding window GSV results in this study area, and compared with Google Earth images of the same period, it is also more consistent with the field texture. In short, the backscatter/FVC feature space can efficiently obtain forest GSV estimates more consistent with field conditions without relying on measured data.
... Such results could be of high interest for the development of improved area-based models [53], that are used to predict forest stand attributes with variables derived from the analysis of ALS point clouds at the level of traditional field inventory, i.e. for forest plots with diameters ranging from about 20 to 30 m. Indeed, simulations could be used to better analyze, understand and model impacts of acquisition conditions, e.g. pulse density [54]- [56] or scan angles [57]. ...
With the recent progresses in lidar technology for Earth remote sensing, the development of a reliable lidar simulator is becoming central in order to define specifications for new sensors, perform inter-comparisons, train machine learning algorithms and help transferring information from one scale to another. The Discrete Anisotropic Radiative Transfer (DART) model includes such a lidar simulator. Although already tested on several virtual scenes, the DART outputs still need to be rigorously evaluated against actual sensor acquisitions, especially on real complex scenes of various forest types, such as dense tropical forests. That is the purpose of the present study. A real Airborne Laser Scanner (ALS) with full-waveform capacity was first radiometrically calibrated on targets of measured reflectance. The properties of the ALS system were then introduced in the DART model, along with a 3D virtual scene built from terrestrial laser scans and spectroscopic measurements acquired on a forest plot near the calibration site. Finally, an ALS acquisition was simulated and the shape and magnitude of the waveforms were compared with real acquisitions. The comparison between measured and simulated data was performed at different scales by aggregating waveform samples into a 3D grid with a vertical resolution of 1m and an horizontal resolution ranging from 2m to 80m. Results showed a high similarity between simulated and measured waveforms at all scales with R²>0.9 and NRMSE<10%. These promising results open up numerous perspectives for improved spaceborne and airborne lidar data processing and for the development of new systems.
... Active optical remote sensing mainly uses LiDAR for GSV estimation. LiDAR is mainly used to extract the vertical forest structure, distribution density, and other information through the point cloud and is combined with a digital elevation model (DEM) to obtain the GSV of a sample through multiple linear regression or machine learning [10,[16][17][18][19]. Due to the effects of weather and the expensive data acquisition over a large area, LiDAR technology is also limited in application. ...
Forest growing stock volume (GSV) is an essential aspect of ecological carbon stock monitoring. The successive launches of spaceborne microwave satellites have provided a broader way to use microwave remote sensing to monitor forest accumulation. Currently, the inversion parameterization models of active microwave remote sensing stock volume mainly include the interferometric water cloud (IWCM), BIOMASAR, and Siberia. Among them, the IWCM introduces backscattering and coherence, the BIOMASAR model only introduces backscattering, and the Siberia model only introduces coherence. Although these three models combine the backscatter coefficient and coherence of SAR to estimate volume accumulation, the performance of the models has not been evaluated at the same time in the same area. Therefore, this article starts from the perspective of the three combinations of coherence and backscattering, relies on three models that do not require measured data, and evaluates the accuracy of the models’ overall inversion of GSV. In addition, we combine precipitation meteorological information, vegetation types, and seasonal variation to separately explore model performance. The comparison results show that the IWCM model is relatively stable in the process of stock volume inversion and is more sensitive to the vegetation types of coniferous and deciduous forests. The influence of seasons and precipitation on the model is weak, and the accuracy of the multi-time-series model is slightly improved. The Siberia model has a good storage volume inversion effect in this study area, but the multiple time series did not improve the model accuracy. The BIOMASAR model is simple, and its performance was slightly inferior in this study area. Precipitation can negatively affect BIOMASAR. The model results for multiple time series outperform those for single time. In summary, the stability of IWCM is more suitable for research with unknown information. The BIOMASAR model is simple, does not require coherence calculations, and is ideal for the estimation of large-scale national or world-level storage distributions. The Siberian model performs better in small regions and smaller spatiotemporal baselines.
... Information on forest stand attributes is important for forest inventory, management and conser− vation. Reliable information on characteristics such as tree density (Błaszczak−Bąk et al., 2022), average height (Wang et al., 2019) and growing stock volume (Parkitna et al., 2021) is crucial for improving operational decision−making for sustainable forest management. Estimates of these attributes are fundamental to forest management and silviculture planning (Stereńczak, 2010;Socha et al., 2019). ...
Information on stand characteristics is of great importance for forest inventory, management and conservation. For more than two decades, Airborne Laser Scanning (ALS) data have enabled remotely sensed estimation of forest stand attributes. Two main approaches are used to estimate biometric forest attributes using Airborne Laser Scanning (ALS): the area−based approach (ABA) and individual tree detection (ITD). So far, the ABA method has been much more commonly used in forestry, as it requires only point cloud metrics. However, with the requirement for precise height information and the development of ITD methods, it is increasingly used to estimate tree biometric characteristics and stand attributes. With this in mind, this study assessed the impact of an ITD correction method based on the Canopy Height Model (CHM) on the estimation of forest characteristics such as tree density and average tree height. The three−step correction method first classifies erroneous segments from ITD methods, which are then refined. In this study, two ITD methods were tested and their results subsequently corrected on a diverse forest area within Białowieża Forest in Poland. In general, more accurate estimates of stand attributes were obtained using the Local ITD method developed in this study area, while correction procedure produced greater improvement using the basic ITD method, which is a marker−controlled watershed with a kernel size of five pixels (MCWS 5×5). Both ITD methods were reliable for estimating tree density for deciduous trees. The correction worked most reliably for estimating tree density with both methods for the area consisted of deciduous trees, while it was most reliable for estimating average tree height with the Local method for the deciduous trees and with the MCWS 5×5 for the conifers. The results indicate that correction improved ITD estimates of stand characteristics, but this varied with species groups, tree height and amount of height variation. Therefore, further development of ITD methods is advisable, as estimating stand attributes using ALS at the individual tree level offers possibilities for improved forest management.
... 1. Adding ITD-derived features to ABA: This approach is based on averaging the ITD-derived features of segments located inside plots and using them in accordance with the ABA Ordinary -derived features to improve estimation of forest attributes compared to ABA Ordinary . This approach was introduced by Hyyppä et al. (2012), then applied and proved in other studies as well (Kelley et al., 2022;Parkitna et al., 2021;Shinzato et al., 2016;Breidenbach et al., 2012;Vastaranta et al., 2012). 2. Correcting the effect of edge (border) trees in a plot boundary: This approach is based on adjusting plot boundaries by extending or shrinking based on the boundary of edge trees falling inside or outside the plot to solve the problem of a discrepancy between ALSderived features and corresponding field measurements at plot-level caused by the canopy of the edge trees. ...
Silvicultural tending of seedling stands is important to producing quality timber. However, it is challenging to allocate where and when to apply these silvicultural tending actions. Here, we tested and evaluated two methodological modifications of the ordinary area-based approach (ABAOrdinary) that could be utilized in the airborne laser scanning-based forest inventories and especially seedling stand characterization. We hypothesize that ABA with added individual tree detection-derived features (ABAITD) or correcting edge-tree effects (ABAEdge) would display improved performance in estimating the tree density and mean tree height of seedling stands. We tested this hypothesis using single-photon laser (SPL) and linear-mode laser (LML) scanning data covering 89 sample plots.
The obtained results supported the hypothesis as the methodological modifications improved seedling stand characterization. Compared to the performance of ABAordinary, relative bias in tree density estimation decreased from 17.2% to 10.1% when we applied ABAITD. In the case of mean height estimation, the relative root mean square error decreased from 19.5% to 16.3% when we applied ABAEdgeITD. The SPL technology provided practically comparable or, in some cases, enhanced performance in seedling stand characterization when compared to conventional LML technology. Based on the obtained findings, it seems that the tested methodological improvements should be carefully considered when ALS-based inventories supporting forest management and silvicultural decision-making are developed further.
The need for objective methods to determine tree canopy cover (CC) across large numbers of stands has led to the development of techniques that utilise airborne laser scanning (ALS) data, which provides a reproducible and detailed representation of canopy geometry. We developed a method for determining CC area and evaluated the estimation accuracy for stands of different sizes, structure and composition. This method is based on tree crown geometries obtained from ALS data, and verified with field measurements using data for 3245 stands of the Białowieża Forest District in Poland. In relatively large stands (3–5 ha), the theoretical error of prediction decreased from 0.13 to 0.10 with increasing stand area. In stands larger than 10 ha, however, the error in estimating CC was less than ±0.10. Although every estimation method comes with its own assumptions and errors, the presented method eliminates the subjectivity in observer bias prevalent in traditional field-based ocular assessments and provides a more transparent and methodologically uniform approach for estimating CC in forests.
Mokradła (w tym torfowiska) i las współwystępują i przeplatają się ze sobą w Polsce od tysięcy lat, tj. od ustąpienia ostatniego lądolodu. Do wczesnego średniowiecza lasy i mokradła, w tym torfowiska, funkcjonowały w pewnej formie ciągłości. Wraz z rozwojem ekonomicznym rodzącej się Polski presja ze strony człowieka zaczęła rosnąć, najpierw zniknęły lasy naturalne, później zmeliorowane zostały bagna. Obecnie głównym wyzwaniem dla ochrony przyrody jest utrzymanie wartościowych ekosystemów w możliwie najbliższym naturalnemu stanie. Jednakże antropogeniczna zmiana klimatu powoduje, że cenne siedliska i gatunki są pod coraz większą presją oddziaływania człowieka. Pod szczególnie negatywnym wpływem zmian klimatycznych są mokradła, czyli tereny bagien, błot i torfowisk lub zbiorniki wodne, tak naturalne, jak i sztuczne, stałe i okresowe, o wodach stojących lub płynących, słodkich, słonawych lub słonych, łącznie z wodami morskimi, których głębokość podczas odpływu nie przekracza sześciu metrów. Losy mokradeł i ludzi są ze sobą wyjątkowo mocno splątane, choć taka świadomość nie jest szeroko rozpowszechniona w społeczeństwie. Mokradła dostarczają szeregu usług ekosystemowych (korzyści, które ludzie odnoszą z mokradeł) np. wpływają na stan wód powodziowych, uzupełniają wody gruntowe, oczyszczają wodę, determinują stan różnorodności biologicznej, dostarczają żywności, stanowią o wartościach kulturowych regionu, współkształtują warunki do rekreacji i turystyki, łagodzą zmianę klimatu i wpływają na adaptację do niej. Mokradła pełnią ważną rolę w lasach, szczególnie poprzez retencjonowanie wody niezbędnej dla roślin i innych organizmów. W ostatniej dekadzie na nowo odkrywamy mokradła w lasach, dostrzegając ich pozytywny wpływ na środowisko przyrodnicze, gospodarkę człowieka, a także na społeczeństwo. Bez wątpienia najwartościowsze w tym kontekście są torfowiska, które nie tylko są gąbkami nasączonymi wodą, ale także długoterminowymi zbiornikami węgla. Razem ze wzrostem świadomości i wiedzy na temat torfowisk i innych mokradeł w lasach dążymy w kierunku ich skuteczniejszej ochrony, wyższej jakości edukacji oraz zrównoważonego wykorzystania turystycznego. W niniejszej monografii prezentujemy nowe spojrzenie na walory przyrodnicze Borów Tucholskich, skupiając się głównie na ich historii, różnorodności biologicznej regionu oraz zasobach węgla zgromadzonego w torfowiskach. Naszym pragnieniem jest także zwrócenie szczególnej uwagi na aspekt odtwarzania mokradeł w sensie obecnych i przyszłych działań w tym zakresie. Współpraca Lasów Państwowych z uczelniami i lokalną społecznością może prowadzić do pozytywnych efektów w zakresie poprawy stanu ekosystemów mokradłowych na terenach leśnych oraz świadomości mieszkańców tego regionu. Wierzymy, że niniejsze opracowanie wywrze taki pozytywny wpływ na przyrodę Borów Tucholskich i zamieszkującą je społeczność.
Mariusz Lamentowicz, Stefan Konczal
Although aboveground biomass (AGB) estimation using area-based approaches (ABAs) and its application to forestry have been actively researched through three decades, this technology has been little operationalized in the Central European forest sector. That means specific recommendations are needed in order to apply ABA for forest biomass modelling in this region. The present study was directed to filling such gaps while examining the effect of input ABA parameters on AGB model quality in conditions of mixed mountainous forests in Central Europe. Specific objectives were to assess whether the strength of the AGB model can be impacted by 1) canopy conditions (leaf-on and leaf-off), 2) airborne LiDAR point density (2.5, 5.0, 7.5, 10.0 points/m²), 3) field methods to estimate AGB (with regeneration components or without), and 4) machine learning methods (AdaBoost, Random decision forest, multilayer neural network, and Bayesian ridge regression). The results show that canopy conditions and airborne LiDAR point densities did not affect the strength of the AGB model, but that model's strength was affected by the vegetation regeneration component in the field biomass reference and by the machine learning method tested for modelling. AdaBoost and random decision forest were the most successful methods. To evaluate the quality of an AGB model it is recommended to combine several individual evaluation functions into the model score. The study highlights several recommendations to follow when estimating AGB from ALS using an ABA in Central European forests.
Background
Forest inventories have always been a primary information source concerning the forest ecosystem state. Various applied survey approaches arise from the numerous important factors during sampling scheme planning. Paramount aspects include the survey goal and scale, target population inherent variation and patterns, and available resources. The last factor commonly inhibits the goal, and compromises have to be made. Airborne laser scanning (ALS) has been intensively tested as a cost-effective option for forest inventories. Despite existing foundations, research has provided disparate results. Environmental conditions are one of the factors greatly influencing inventory performance. Therefore, a need for site-related sampling optimization is well founded. Moreover, as stands are the basic operational unit of managed forest holdings, few related studies have presented stand-level results. As such, herein, we tested the sampling intensity influence on the performance of the ALS-enhanced stand-level inventory.
Results
Distributions of possible errors were plotted by comparing ALS model estimates, with reference values derived from field surveys of 3300 sample plots and more than 300 control stands located in 5 forest districts. No improvement in results was observed due to the scanning density. The variance in obtained errors stabilized in the interval of 200–300 sample plots, maintaining the bias within +/− 5% and the precision above 80%. The sample plot area affected scores mostly when transitioning from 100 to 200 m ² . Only a slight gain was observed when bigger plots were used.
Conclusions
ALS-enhanced inventories effectively address the demand for comprehensive and detailed information on the structure of single stands over vast areas. Knowledge of the relation between the sampling intensity and accuracy of ALS estimates allows the determination of certain sampling intensity thresholds. This should be useful when matching the required sample size and accuracy with available resources. Site optimization may be necessary, as certain errors may occur due to the sampling scheme, estimator type or forest site, making these factors worth further consideration.
Traditional field-based forest inventories tend to be expensive, time-consuming, and cover only a limited area of a forested region. Remote sensing (RS), especially airborne laser scanning (ALS) has opened new possibilities for operational forest inventories, particularly at the single-tree level, and in the prediction of single-tree characteristics. Throughout the world, forests have varying characteristics that necessitate the development of modern, effective, and versatile tools for ALS data processing. To address this need, we aimed to develop a tool for individual tree detection (ITD) utilising a self-calibrating algorithm procedure and to verify its accuracy using the complicated forest structure of near natural forests in the temperate zone.
This study was carried out in the Polish part of the Białowieża Forest (BF). The airborne laser scanner (ALS) and color-infrared (CIR) datasets were acquired for more than 60 000 ha. Field-based measurements were performed to provide reference data at the single tree level. We introduced a novel ITD method that is self-calibrated and uses a hierarchical analyses of the canopy height model.
There were more than 20 000 000 of trees in first layer in BF above 7 m height. Trees visible from above were divided into coniferous, deciduous and mixed trees that were then matched with an accuracy of 85 %, 85 % and 75 %, respectively. Compared to existing methods, the proposed method is more flexible and achieves better results, especially for deciduous species. Before application of the presented method to other regions, the calibration based on the developed optimisation procedure is needed.
Many volume functions use diameter and height as input variables. Measurement and prediction errors in heights propagate into the error of the wood volume estimate for a tree or a plot. We quantified these errors (and bias) with census data of species, heights, and stem diameters in a 1.44 ha hardwood stand in Lower Saxony (GER) dominated by beech (Fagus sylvatica L.) and with an understorey of hardwood species. We simulated simple random sampling with fixed area and variable radius plots and distorted the actual height measurements with three magnitudes of random Gaussian errors. The inventory protocol for height measurement results in a selection of three to four trees per plot, while heights for the remaining trees were predicted via two alternative models. Tree and plot level RMSE% of wood volume with a mix of measured and predicted tree heights were—across the three magnitudes of measurement errors, 18–81% greater than root-mean-squared errors obtained when all trees were measured for heights with errors. The results from this study contribute the statistical side of recommendations on the number of height measurements per forest inventory plot, where inferences on optimal numbers of height measurements will also depend on the costs.
Random forest and similar Machine Learning techniques are already used to generate spatial predictions, but spatial location of points (geography) is often ignored in the modeling process. Spatial auto-correlation, especially if still existent in the cross-validation residuals, indicates that the predictions are maybe biased, and this is suboptimal. This paper presents a random forest for spatial predictions framework (RFsp) where buffer distances from observation points are used as explanatory variables, thus incorporating geographical proximity effects into the prediction process. The RFsp framework is illustrated with examples that use textbook datasets and apply spatial and spatio-temporal prediction to numeric, binary, categorical, multivariate and spatiotemporal variables. Performance of the RFsp framework is compared with the state-of-the-art kriging techniques using fivefold cross-validation with refitting. The results show that RFsp can obtain equally accurate and unbiased predictions as different versions of kriging. Advantages of using RFsp over kriging are that it needs no rigid statistical assumptions about the distribution and stationarity of the target variable, it is more flexible towards incorporating, combining and extending covariates of different types, and it possibly yields more informative maps characterizing the prediction error. RFsp appears to be especially attractive for building multivariate spatial prediction models that can be used as “knowledge engines” in various geoscience fields. Some disadvantages of RFsp are the exponentially growing computational intensity with increase of calibration data and covariates and the high sensitivity of predictions to input data quality. The key to the success of the RFsp framework might be the training data quality—especially quality of spatial sampling (to minimize extrapolation problems and any type of bias in data), and quality of model validation (to ensure that accuracy is not effected by overfitting). For many data sets, especially those with lower number of points and covariates and close-to-linear relationships, model-based geostatistics can still lead to more accurate predictions than RFsp.
Current forest growing stock inventory methods used in Poland are based on statistical methods using field measurements of trees on circular sample plots. Such measurements are carried out with traditional equipment, i.e. callipers and range finders. Nowadays, remote sensing based inventory techniques are becoming more popular and have already been applied in North America and some Scandinavian countries. Remote sensing based forest inventories require a certain amount of ground sample plots, which serve either as reference data used for model calibration and/or as a validation dataset for the assessment of the accuracy of modelled variables.
Using a set of 900 ground sample plots and Airborne Laser Scanner (ALS) from the Milicz forest district, a statistical model for the estimation of plot growing stock volume was developed. Next, the developed model was once again fitted to different variants of sample plot size and number of sample plots. Each variant was selected from a full 900 sample plot set. The selection started from 800, 700, 600, ..., down to 25 plots, respectively, and was carried out in proportion to the dominant tree age range. To account for the area effect, each plot number variant was similarly tested with various sample plot areas, i.e. 500, 400, ..., 100 m2. Sampling in each variant was repeated in order to take into account the effect of a single selection. The results showed a strong relationship between obtained modelling errors and the size and number of used sample plots. It has been
demonstrated that the number of sample plots has no influence on the accuracy of
GSV estimation above about 300-400 sample plots (about 500 sample plots
for bias), whereas sample plot size has a visible impact on estimation accuracy,
which reduces with decreasing sample plot size, regardless of the number of
sample plots. If it is about precision, results showed that the influence of a single
selection to be relevant only below 300-400 plots (about 500 for bias) and the
same trend can be observed in each sample plot size variant. The results showed it is possible to strongly reduce the number of ground sample plots (minimum 300-400), while still maintaining decent accuracy and precision levels, at least in similarly investigated forest conditions.
Airborne Laser Scanning data—also known as Light Detection and Ranging (LiDAR)—enables the accurate three-dimensional characterization of vertical forest structure. Airborne Laser Scanning data have proven to be an information-rich asset for forest managers, enabling the generation of highly detailed digital elevation models and the estimation of a range of forest inventory attributes (e.g., height, basal area, and volume). Good practice guidance synthesizes current knowledge from the scientific literature and practical experience to provide non-experts more detailed information about complex topics. With this guide, our goal is to inform and enable readers interested in using Airborne Laser Scanning data to characterize, in an operational forest inventory context, large forest areas in a cost-effective manner.
The aim of the research was to assess the accuracy of the stratified sampling method used to estimate the standing volume of a forest district and to compare it with the accuracy of simple random sampling methods. The paper presents the variability of the variables affecting the accuracy of the stratified sampling method. We attempted to find the ways to increase this accuracy. The research was based on the empirical material collected on approximately 42,000 sample plots with a size of 50−500 m², and with an average of 737 plots per forest district. The standard deviation of the merchantable volume of trees on sample plots ranged from 87 to 213 m³/ha, with an average of 128 m³/ha. The coefficient of variation ranged from 5.3 to 28.5% (the average 40.8%). Using a simple random sampling method, the standard error of the volume ranged from 3.3 to 10.0 m³/ha (the average 4.8 m³/ha) and the relative error – from 1.01 to 3.41% (the average 1.55%). The absolute error of the stratified sampling method under which strata are formed on the basis of the main tree species and its age ranged from 2.9 to 7.4 m³/ha, the average 4.2 m³/ha, and the relative error ranged from 0.65 to 1.95%, 1.02% on average (tab.). The accuracy of the stratified sampling method was by 15% higher than that of the simple random sampling method. We found that the relationship between the volume of a sample plot and the main tree species and its age measured by the correlation coefficient was 0.453 on average. For the relationship between volume and age of stands this coefficient was on average 0.422, while between volume and main tree species – only 0.118. Stand age – as an auxiliary variable in formation of strata – proved to be of moderate usefulness resulting from a small difference in the standing volume of stands in older age classes. Main tree species turned out to be of slight usefulness in formation of strata, therefore it seems reasonable to find some other auxiliary variables to replace it.
Recent research has shown that image-derived point clouds (IPCs) are a highly competitive alternative to airborne laser scanning (ALS) data in the context of selected forest inventory activities. However, there is still a need for investigating different kinds of aerial images used for point cloud generation. This study compares the effectiveness of IPCs derived from true colour (RGB) and colour infrared (CIR) aerial images with ALS data for growing stock volume estimation of single canopy layer Scots pine stands. A multiple linear regression method was used to create predictive models. All models predicted growing stock volume with low root mean square errors – ALS: 15.2%, IPC-CIR: 17.0% and IPC-RGB: 17.5%. The following variables for each data type were found to be the most robust: ALS – mean height of points, percentage of all returns above mean height of points, interquartile range of point heights; IPC-CIR – mean height of points, percentage of all returns above mode height of points, canopy relief ratio; IPC-RGB – mean height of points and canopy relief ratio. Our results show that for single canopy layer Scots pine dominated stands it is possible to predict growing stock volume using IPCs with a comparable accuracy as using ALS data. The comparable performance of IPC-RGB and IPC-CIR based models suggests that a mixed usage of RGB and CIR data in retrospective studies could be possible.
Airborne laser scanning (ALS) based stand level forest inventory has been used in Finland and other Nordic countries for several years. In the Russian Federation, ALS is not extensively used for forest inventory purposes, despite a long history of research into the use of lasers for forest measurement that dates back to the 1970s. Furthermore, there is also no generally accepted ALS-based methodology that meets the official inventory requirements of the Russian Federation. In this paper, a method developed for Finnish forest conditions is applied to ALS-based forest inventory in the Perm region of Russia. Sparse Bayesian regression is used with ALS data, SPOT satellite images and field reference data to estimate five forest parameters for three species groups (pine, spruce, deciduous): total mean volume, basal area, mean tree diameter, mean tree height, and number of stems per hectare. Parameter estimates are validated at both the plot level and stand level, and the validation results are compared to results published for three Finnish test areas. Overall, relative root mean square errors (RMSE) were higher for forest parameters in the Perm region than for the Finnish sites at both the plot and stand level. At the stand level, relative RMSE generally decreased with increasing stand size and was lower when considered overall than for individual species groups.
Forest inventory and management requirements are changing rapidly in the context of an increasingly complex set of economic, environmental, and social policy objectives. Advanced remote sensing technologies provide data to assist in addressing these escalating information needs and to support the subsequent development and parameterization of models for an even broader range of information needs. This special issue contains papers that use a variety of remote sensing technologies to derive forest inventory or inventory-related information. Herein, we review the potential of 4 advanced remote sensing technologies, which we posit as having the greatest potential to influence forest inventories designed to characterize forest resource information for strategic, tactical, and operational planning: airborne laser scanning (ALS), terrestrial laser scanning (TLS), digital aerial photogrammetry (DAP), and high spatial resolution (HSR)/very high spatial resolution (VHSR) satellite optical imagery. ALS, in particular, has proven to be a transformative technology, offering forest inventories the required spatial detail and accuracy across large areas and a diverse range of forest types. The coupling of DAP with ALS technologies will likely have the greatest impact on forest inventory practices in the next decade, providing capacity for a broader suite of attributes, as well as for monitoring growth over time.
Maps of standing timber volume provide valuable decision support for forest managers and have therefore been the subject of recent studies. For map production, field observations are commonly combined with area-wide remote sensing data in order to formulate prediction models, which are then applied over the entire inventory area. The accuracy of such maps has frequently been described by parameters such as the root mean square error of the prediction model. The aim of this study was to additionally address the accuracy of timber volume classes, which are used to better represent the map predictions. However, the use of constant class intervals neglects the possibility that the precision of the underlying prediction model may not be constant across the entire volume range, resulting in pronounced gradients between class accuracies. This study proposes an optimization technique that automatically identifies a classification scheme which accounts for the properties of the underlying model and the implied properties of the remote sensing support information. We demonstrate the approach in a mountainous study site in Eastern Switzerland covering a forest area of 2000 hectares using a multiple linear regression model approach. A LiDAR-based canopy height model (CHM) provided the auxiliary information; timber volume observations from the latest forest inventory were used for model calibration and map validation. The coefficient of determination (R2 = 0.64) and the cross-validated root mean square error (RMSECV = 123.79 m3 ha-1) were only slightly smaller than those of studies in less steep and heterogeneous landscapes. For a large set of pre-defined number ofclasses, the optimization model successfully identified those classification schemes that achieved the highest possible accuracies for each class.
Small-area estimation has received considerable attention in recent years because of a growing demand for reliable small-area statistics. The direct-survey estimators, based only on the data from a given small area (or small domain), are likely to yield unacceptably large standard errors because of small sample size in the domain. Therefore, alternative estimators that borrow strength from other related small areas have been proposed in the literature to improve the efficiency. These estimators use models, either implicitly or explicitly, that connect the small areas through supplementary (e.g., census and administrative) data. For example, simple synthetic estimators are based on implicit modeling. In this article, three small-area models, of Battese, Harter, and Fuller (1988), Dempster, Rubin, and Tsutakawa (1981), and Fay and Herriot (1979), are investigated. These models are all special cases of a general mixed linear model involving fixed and random effects, and a small-area mean can be expressed as a linear combination of fixed effects and realized values of random effects. Using the general theory of Henderson (1975) for a mixed linear model, a two-stage estimator (or predictor) of a small-area mean under each model is obtained, by first deriving the best linear unbiased estimator (or predictor) assuming that the variance components that determine the variance-covariance matrix are known, and then replacing the variance components in the estimator with their estimators. Second-order approximation to the mean squared error (MSE) of the two-stage estimator and the estimator of MSE approximation are obtained under normality. Finally, the results of a Monte Carlo study on the efficiency of two-stage estimators and the accuracy of the proposed approximation to MSE and its estimator are summarized. The MSE approximation provides a reliable measure of uncertainty associated with the two-stage estimator. It can also provide asymptotically valid confidence intervals on a small-area mean, as the number of small areas tends to ∞.
Due to the availability of good and reasonably priced auxiliary data, the use of model-based regression-synthetic estimators for small area estimation is popular in operational settings. Examples are forest management inventories, where a linking model is used in combination with airborne laser scanning data to estimate stand-level forest parameters where no or too few observations are collected within the stand. This paper focuses on different approaches to estimating the variances of those estimates. We compared a variance estimator which is based on the estimation of superpopulation parameters with variance estimators which are based on predictions of finite population values. One of the latter variance estimators considered the spatial autocorrelation of the residuals whereas the other one did not. The estimators were applied using timber volume on stand level as the variable of interest and photogrammetric image matching data as auxiliary information. Norwegian National Forest Inventory (NFI) data were used for model calibration and independent data clustered within stands were used for validation. The empirical coverage proportion (ECP) of confidence intervals (CIs) of the variance estimators which are based on predictions of finite population values was considerably higher than the ECP of the CI of the variance estimator which is based on the estimation of superpopulation parameters. The ECP further increased when considering the spatial autocorrelation of the residuals. The study also explores the link between confidence intervals that are based on variance estimates as well as the well-known confidence and prediction intervals of regression models.
In this study, eight airborne laser scanning (ALS)-based single tree detection methods are benchmarked and investigated. The methods were applied to a unique dataset originating from different regions of the Alpine Space covering different study areas, forest types, and structures. This is the first benchmark ever performed for different forests within the Alps. The evaluation of the detection results was carried out in a reproducible way by automatically matching them to precise in situ forest inventory data using a restricted nearest neighbor detection approach. Quantitative statistical parameters such as percentages of correctly matched trees and omission and commission errors are presented. The proposed automated matching procedure presented herein shows an overall accuracy of 97%. Method based analysis, investigations per forest type, and an overall benchmark performance are presented. The best matching rate was obtained for single-layered coniferous forests. Dominated trees were challenging for all methods. The overall performance shows a matching rate of 47%, which is comparable to results of other benchmarks performed in the past. The study provides new insight regarding the potential and limits of tree detection with ALS and underlines some key aspects regarding the choice of method when performing single tree detection for the various forest types encountered in alpine regions.
The aim of this paper was to examine different plot selection strategies of field training plots in forest inventory using
airborne laser scanner (ALS) data. The applied plot selection strategies were random selection, random selection within pre-stratification
according to forest type, selection of plots according to geographical location and selection of plots based on properties
of the ALS data given as a priori information. The study was conducted by means of simulation utilizing existing and independent training and validation plot
data and the performance was evaluated by assessing bias and the root mean square error (RMSE). The accuracy of simultaneously
derived biophysical stand properties, i.e. volume, number of stems and Lorey’s mean height, was examined using non-parametric
modelling. The use of ALS data as a priori information provided the most accurate results in the case of stand volume and number of stems the RMSE being less than 15
and 30 per cent, respectively. For the mean height, also the other selection strategies were as good but the most accurate
alternative varied according to number of training plots used. In most cases, the RMSE values for the mean height were between
8 and 9 per cent. The bias of the different strategies followed the same patterns as the corresponding RMSE values.
The aim of this study was to evaluate the automatically determined parameters of tree crowns, which are then used in two-phase inventory method of growing stock. Research was performed in forest stands of different age, located in the Stolowe Mountains National Park (south-western Poland) where the dominant species was Norway spruce (Picea abies L.). On the test area of approximately 500 hectares, 35 sample plots were measured. On LIDAR-based Crow Height Model (CHM), in the places corresponding to the position of 500 m2 circular ground samples, automatic segmentation was carried out. The extent of the crown was associated with the height of the tree. Two variants of the assigning separate crowns in the sample were used: (1) according to the centroid position, (2) according to the location of any fragment of the crown inside the sample plot boundary. In each of the variants five series of measurements with different relative height of the range 0.65-0.8 (with a gap of 0.05) were carried out. Relationship between the volume of living trees measured on the ground and LIDAR parameters (average height of trees and tree canopy projection area) automatically measured based on CHM was determined. Multiple correlation coefficient differed depending on the location to extract coverage crowns and ranged from 0.687 to 0.788. The variant of counting of all trees with crowns or pieces inside the sample appeared to perform better. The relationship between the above-indicated characteristics was stronger after elimination of dead trees (about 0.1 for each of cases). This means that for measurements of the growing stock in forests under protection, it is necessary to improve the process of automatic segmentation of the crown, by identification and elimination of dead trees - usually present at sample plot.
Forest harvest planning to maximize economic benefits also has to consider additional criteria such as the biodiversity functioning of the managed forest. The biodiversity requirements are determined by the size, shape, and distribution of harvest units and forest stands. A multiple criteria approach is presented where the harvesting volume is maximized while the environmental aspects are also considered. Multiple criteria programming and integer programming techniques are used to find an optimal program of forest harvesting with respect to both economic and environmental requirements. The practicality of the model is shown in a case study for one particular forest management unit. Different optimal solutions are calculated depending on changes made to the criteria weights. This model includes strict spatial constraints, multiple objective functions with three objectives, and alternative solutions according to the real manager's priority. The results show that the spatial pattern and other spatial demands affect the harvest possibilities. It was confirmed that a compromise solution from both forest management and nature conservation could be achieved using the presented harvest scheduling approach.
The United States Forest Service Forest Inventory and Analysis (FIA) Program provides a diverse selection of data used to assess the status of the nation's forests using sample locations dispersed throughout the country. Airborne laser scanning (ALS) systems are capable of producing accurate measurements of individual tree dimensions and also possess the ability to characterize forest structure in three dimensions. This study investigates the potential of discrete return ALS data for modeling forest aboveground biomass (AGBM) and gross volume (gV) at FIA plot locations in the Malheur National Forest, eastern Oregon utilizing three analysis levels: (1) individual subplot (r = 7.32 m); (2) plot, comprising four clustered subplots; and (3) hectare plot (r = 56.42 m). A methodology for the creation of three point cloud-based airborne LiDAR metric sets is presented. Models for estimating AGBM and gV based on LiDAR-derived height metrics were built and validated utilizing FIA estimates of AGBM and gV derived using regional allometric equations. Simple linear regression models based on the plot-level analysis out OPEN ACCESS Remote Sens. 2015, 7 230 performed subplot-level and hectare-level models, producing R 2 values of 0.83 and 0.81 for AGBM and gV, utilizing mean height and the 90th height percentile as predictors, respectively. Similar results were found for multiple regression models, where plot-level analysis produced models with R 2 values of 0.87 and 0.88 for AGBM and gV, utilizing multiple height percentile metrics as predictor variables. Results suggest that the current FIA plot design can be used with dense airborne LiDAR data to produce area-based estimates of AGBM and gV, and that the increased spatial scale of hectare plots may be inappropriate for modeling AGBM of gV unless exhaustive tree tallies are available. Overall, this study demonstrates that ALS data can be used to create models that describe the AGBM and gV of Pacific Northwest FIA plots and highlights the potential of estimates derived from ALS data to augment current FIA data collection procedures by providing a temporary intermediate estimation of AGBM and gV for plots with outdated field measurements.
Forest compartments are usually delineated according to artificial or natural boundaries and usually include portions of different strata. While volume estimation of each stratum can be performed from field plots located within each stratum, volume estimation in portions of the stratum may be problematic owing to the small number (or even the absence) of plots falling in those portions. If upper canopy heights from airborne laser scanning are available at the pixel level for the whole survey area, these data are used as auxiliary information. A ratio model presuming a proportional relationship between transformed heights (e.g., power of heights) and volumes at the pixel level is adopted to guide estimation. From this model, the volume within any portion of the survey area is estimated as the proportionality factor estimate multiplied by the total of transformed heights in that portion. This estimator is considered from the model-based, design-based, and hybrid perspectives. Variances and their estimators are derived under the three approaches together with the corresponding confidence intervals. The volume estimator and the variance estimators are checked from the design-based point of view by a simulation study performed on a real forest in northwestern Italy. An application to a public forest estate in the same zone is performed.
Olsson, H. 2012. Prediction of stem attributes by combining airborne laser scanning and measurements from harvesters. Silva Fennica 46(2): 227–239. In this study, a new method was validated for the first time that predicts stem attributes for a forest area without any manual measurements of tree stems by combining harvester measure-ments and Airborne Laser Scanning (ALS) data. A new algorithm for automatic segmentation of tree crowns from ALS data based on tree crown models was developed. The test site was located in boreal forest (64º06'N, 19º10'E) dominated by Norway spruce (Picea abies) and Scots Pine (Pinus sylvestris).The trees were harvested on field plots, and each harvested tree was linked to the nearest tree crown segment derived from ALS data. In this way, a reference database was created with both stem data from the harvester and ALS derived features for linked tree crowns. To estimate stem attributes for a tree crown segment in parts of the forest where trees not yet have been harvested, tree stems are imputed from the most similar crown segment in the reference database according to features extracted from ALS data. The imputa-tion of harvester data was validated on a sub-stand-level, i.e. 2–4 aggregated 10 m radius plots, and the obtained RMSE of stem volume, mean tree height, mean stem diameter, and stem density (stems per ha) estimates were 11%, 8%, 12%, and 19%, respectively. The imputation of stem data collected by harvesters could in the future be used for bucking simulations of not yet harvested forest stands in order to predict wood assortments.
We consider the small-area estimation problem for forest inventories with two-phase sampling schemes. We propose an improvement to the synthetic estimator, when the true mean of the auxiliary variables over the small area is unknown and must be estimated, and likewise to the residual corrected small-area estimator. We derive the asymptotic design-based variances of these new estimators, the pseudo-synthetic and pseudo-small-area estimators, by also incorporating the design-based variance of the regression coefficients. We then propose a very simple mathematical device that transforms pseudo-small-area estimators into pseudo-synthetic estimators, which is very convenient for deriving asymptotic variances. The results are extended to cluster and two-stage sampling at the plot level. A case study and a simulation illustrate the theory.
There is increasing interest in the use of airborne laser scanning (ALS; also referred to as airborne Light Detection and Ranging or LiDAR) for forest inventory purposes in Canada. Timber volume is one of the key inventory attributes that is commonly estimated using ALS data, and estimates of volume can be validated using post-harvest measures. ALS data and the area-based approach were used to develop an enhanced forest inventory for the Hinton Forest Management Area (FMA) in central Alberta. Weight scale measures of coniferous merchantable volume from 272 stands harvested between 2008 and 2010 were used as validation data for both conventional and ALS-based estimates. Overall, conventional estimates of coniferous merchantable volume derived from cover type adjusted volume tables were found to underestimate weight scale volumes by 19.8%. Conversely, estimates generated from the ALS data overestimated weight scale volumes by 0.6%. ALS-based estimates provide wall-to-wall, spatially explicit estimates of merchantable volume, enable within-stand variability in merchantable volume to be characterized, and are beneficial for strengthening linkages between strategic and operational forest planning.
Over the last decade, Oregon State University (OSU) College Forests has successfully implemented a plan for its 4600-ha forest in the Willamette Valley of Oregon that called for significant timber harvest under a range of silvicultural systems, including clear-cutting, while providing teaching, research and extension opportunities. Despite discovery of endangered species, increasing recreational use and demands from OSU faculty for research and field teaching uses, the college has been able to implement the plan and produce the expected volume and revenue. Central to this success was the forest inventory installed in the mid-1080s, combined with a growth and yield model calibrated to the forest. Rather than using class averages like most public forests, the College Forests use projected stand estimates, which enable accurate aggregation and disaggregation of the forest stands for both strategic planning and detailed harvest scheduling. This approach both enables accurate initial estimates of the harvest schedule and facilitates the inevitable shifts in the timings and location of the harvest as unforeseen events occur. One such event was the arrival of northern spotted owls (Strix occidentalis), a federally listed threatened species. Utilizing this inventory and projection system, the college identified stands that functioned as spotted owl habitat, adjusted the location of harvests to conserve that habitat during the plan period (10 years) and still met its harvest and revenue targets. In the recent plan revision, the college has taken the next step to dynamically schedule investment and harvest in ways that will maintain desired amounts of the spotted owl habitat in the long run.
This study compares methods to estimate stem volume, stem number and basal area from Airborne Laser Scanning (ALS) data for 68 field plots in a hemi-boreal, spruce dominated forest (Lat. 58 degrees N, Long. 13 degrees E). The stem volume was estimated with five different regression models: one model based on height and density metrics from the ALS data derived from the whole field plot, two models based on similar combinations derived from 0.5 m raster cells, and two models based on canopy volumes from the ALS data. The best result was achieved with a model based on height and density metrics derived from 0.5 m raster cells (Root Mean Square Error or RMSE 37.3%) and the worst with a model based on height and density metrics derived from the whole field plot (RMSE 41.9%). The stem number and the basal area were estimated with: (i) area-based regression models using height and density metrics from the ALS data; and (ii) single tree-based information derived from local maxima in a normalized digital surface model (nDSM) mean filtered with different conditions. The estimates from the regression model were more accurate (RMSE 52.7% for stem number and 21.5% for basal area) than those derived from the nDSM (RMSE 63.4%-91.9% and 57.0%-175.5%, respectively). The accuracy of the estimates from the nDSM varied depending on the filter size and the conditions of the applied filter. This suggests that conditional filtering is useful but sensitive to the conditions.
Tree crown attributes are important parameters during the assessment and monitoring of forest ecosystems. Canopy height models (CHMs) derived from airborne laser scanning (ALS) data have proved to be a reliable source for extracting different biophysical characteristics of single trees and at stand level. However, ALS-derived tree measurements (e.g., mean crown diameter) can be negatively affected by pits that appear in the CHMs. Thus, we propose a novel method for generating pit-free CHMs from ALS point clouds for estimating crown attributes (i.e., area and mean diameter) at the species level. The method automatically calculates a threshold for a pixel based on the range of height values within neighbouring pixels; if the pixel falls below the threshold then it is recognized as a pitted pixel. The pit is then filled with the median of the values of the neighbouring pixels. Manually delineated individual tree crowns (ITC) of four deciduous and two coniferous species on Colour Infrared (CIR) stereo images were used as a reference in the analysis. In addition, a variety of different algorithms for constructing CHMs were compared to investigate the performance of different CHMs in similar forest conditions. Comparisons between the estimated and observed crown area (R² = 0.95, RMSE% = 19.12% for all individuals) and mean diameter (R² = 0.92, RMSE% = 12.16% for all individuals) revealed that ITC attributes were correctly estimated by segmentation of the pit-free CHM proposed in this study. The goodness of matching and geometry revealed that the delineated crowns correctly matched up to the reference data and had identical geometry in approximately 70% of cases. The results showed that the proposed method produced a CHM that estimates crown attributes more accurately than the other investigated CHMs. Furthermore, the findings suggest that the proposed algorithm used to fill pits with the median of height observed in surrounding pixels significantly improve the accuracy of the results the species level due to a higher correlation between the estimated and observed crown attributes. Based on these results, we concluded that the proposed pit filling method is capable of providing an automatic and objective solution for constructing pit-free CHMs for assessing individual crown attributes of mixed forest stands.
With the maturation of methods for estimating aboveground forest biomass by remote sensing, researchers increasingly need test data, particularly ground reference data, that are large enough to fine-tune existing approaches and test their robustness under diverse conditions. In this context, realistic synthetic datasets present an interesting alternative to costly and limited field data. Here, we present a new approach to simulate realistic canopy height and cover type data by combining an individual-tree forest simulator with real LiDAR point clouds of individual trees. We demonstrate the utility of our approach by reexamining the influence of field plot size on the predictive power of remote-sensing models for biomass estimation. Our approach with a complete (wall-to-wall) field reference dataset and matching synthetic remote sensing data allowed us to not only perform internal cross-validations with field plots that were used to fit the model (as in studies with real data), but to also consider the quality of model predictions to a standardized spatial grid or to the entire region. Our results confirm earlier reports of smaller predictive errors with increased field plot sizes under internal model validation (RMSE of 125 t/ha at 10 m field plot size and RMSE of 40 t/ha for 50 m field plots). However, we show that this is mainly an artifact of comparing the models with the same data they were fit, thus with validation data of different scales. When validating on a grid with standardized scale, smaller field plots performed almost equally well as larger field plots (small RMSE decrease between 4 t/ha and 7 t/ha when going from 10 m to 50 m plots), and even outperformed them if we assumed that increasing the plot size means that fewer field plots can be obtained (small RMSE increase between 9 t/ha and 12 t/ha when going from 10 m to 50 m plots). We conclude that synthetic remote sensing datasets are a useful tool for method testing. The suggested approach may be used to reexamine our current methodological understanding, which is often based on tests with real data of very limited sizes, as well as to optimize workflows, model choices and data collection. A wider use of synthetic data could be instrumental in improving remote sensing methodology.
The book presents the state-of-the-art of forest resources assessments and monitoring and provides links to practical applications of forest and natural resource assessment programs. It gives an overview of current forest inventory systems and discusses forest mensuration, sampling techniques, remote sensing applications, geographic and forest information systems, and multi-resource forest inventory. In addition to the assessment of the productive functions of forests, particular attention is given to the quantification of non-wood goods and services and the relationship of forests to other landscape elements. All methodology is presented in the framework of sustainable management of the multiple functions that forests provide to the natural environment and to society.
The book was developed as a reference text for (forest) biometricians, practitioners involved in forest and natural resources assessment and monitoring programs, and graduate students with a strong interest in becoming forest inventory specialists.
This article provides an overview of methods for estimating small-area means or totals when the area-specific sample sizes are small. Methods studied include indirect methods based on implicit models and methods based on explicit models.
Model-based predictions of a domain (stand) mean of a study variable Y is synthetic when there are no direct observations of Y within the domain of interest. In the absence of estimators of domain effects in a model used for domain-level predictions, the use of model-based estimators of variance can lead to overly narrow confidence intervals with poor coverage and a serious underestimation of uncertainty. This study proposes a new mean squared error estimator (MSE 4) based on an estimator of the among-domain variance in Y derived with fitted values of ŷ. In simulated sampling, with at most one sample taken from a domain, 95% confidence intervals based on MSE 4 provided better, yet unsatisfactory, coverage than possible with available alternatives. A shift to sampling designs affording estimators of domain effects is recommended.
Keywords: forest inventory; sampling design; small-area estimation; spatial autocorrelation
DOI: https://doi.org/10.5849/forsci.16-056
Affiliations: Canadian Forest Service, Canadian Forest Service, Victoria, BC, Canada.
Appeared or available online: November 17, 2016
(document).ready(function() { var shortdescription = (".originaldescription").text().replace(/\\&/g, '&').replace(/\\, '<').replace(/\\>/g, '>').replace(/\\t/g, ' ').replace(/\\n/g, '
'); if (shortdescription.length > 350) { shortdescription = " " + shortdescription.substring(0,250) + "... more"; } (".shortdescription a").click(function(e) { e.preventDefault(); (".originaldescription").slideDown(); }); });
Forest surveys, in the form of both stand management and strategic inventories, have a long history of using remotely sensed data to support and enhance their design and estimation processes. By the use of airborne laser scanning data this capacity has emerged as one of its most important and prominent applications. The chapter includes a brief overview of forest inventory uses of remotely sensed data, a section on aspects of ground sampling that can be managed to optimize estimation of relationships between ground and airborne laser scanning (ALS) data, and a section on stand management inventories. The latter section reviews underlying and motivating factors crucial to the primary focus of the chapter, formal statistical inference for ALS-assisted forest inventories. Inferential methods are described for two primary cases, full and partial ALS coverage. Within each case, estimators for both design-based and model-based inference are presented.
The aim of this chapter is to give an overview of the development of ALS as an operational tool for forest management inventories in Norway. The chapter will shed light on some of the technical and institutional challenges that were faced. Interaction between the scientific community and private sector was seen as a critical factor for successful adoption of the new technology for practical purposes and it will briefly be described. A description of local adoptions of the methods and of research conducted to improve the technical and economic performance will be given. Finally, some future needs and directions will be discussed. It is believed that the lessons learned in Norway may be found useful for similar efforts in other countries.
Function estimation/approximation is viewed from the perspective of numerical optimization iti function space, rather than parameter space. A connection is made between stagewise additive expansions and steepest-descent minimization. A general gradient descent "boosting" paradigm is developed for additive expansions based on any fitting criterion. Specific algorithms are presented for least-squares, least absolute deviation, and Huber-M loss functions for regression, and multiclass logistic likelihood for classification. Special enhancements are derived for the particular case where the individual additive components are regression trees, and tools for interpreting such "TreeBoost" models are presented. Gradient boosting of regression trees produces competitives highly robust, interpretable procedures for both regression and classification, especially appropriate for mining less than clean data. Connections between this approach and the boosting methods of Freund and Shapire and Friedman, Hastie and Tibshirani are discussed.
Random forests are a combination of tree predictors such that each tree depends on the values of a random vector sampled independently and with the same distribution for all trees in the forest. The generalization error for forests converges a.s. to a limit as the number of trees in the forest becomes large. The generalization error of a forest of tree classifiers depends on the strength of the individual trees in the forest and the correlation between them. Using a random selection of features to split each node yields error rates that compare favorably to Adaboost (Y. Freund & R. Schapire, Machine Learning: Proceedings of the Thirteenth International conference, ∗∗∗, 148–156), but are more robust with respect to noise. Internal estimates monitor error, strength, and correlation and these are used to show the response to increasing the number of features used in the splitting. Internal estimates are also used to measure variable importance. These ideas are also applicable to regression.
This study proposed modifying the conceptual approach that is commonly used to model development of stand attribute estimates using airborne LiDAR data. New models were developed using an area-based approach to predict wood volume, stem volume, aboveground biomass, and basal-area across a wide range of canopy structures, sites and LiDAR characteristics. This new modeling approach does not adopt standard approaches of stepwise regression using a series of height metrics derived from airborne LiDAR. Rather, it used four metrics describing complementary 3D structural aspects of the stand canopy. The first three metrics were related to mean canopy height, height heterogeneity, and horizontal canopy distribution. A fourth metric was calculated as the coefficient of variation of the leaf area density profile. This fourth metric provided information on understory vegetation. The models that were developed with the four structural metrics provided higher estimation accuracy on stand attributes than models using height metrics alone, while also avoiding data over-fitting. Overall, the models provided prediction error levels ranging from 12.4% to 24.2%, depending upon forest type and stand attribute. The more homogeneous coniferous stand provided the highest estimation accuracy. Estimation errors were significantly reduced in mixed forest when separate models were developed for individual stand types (coniferous, mixed and deciduous stands) instead of a general model for all stand types. Model robustness was also evaluated in leaf-off and leaf-on conditions where both conditions provided similar estimation errors.
Scanning lidar remote sensing systems have recently become available for use in ecological applications. Unlike conventional microwave and optical sensors, lidar sensors directly measure the distribution of vegetation material along the vertical axis and can be used to provide three-dimensional, or volumetric, characterizations of vegetation structure. Ecological applications of scanning lidar have hitherto used one-dimensional indices to characterize canopy height. A novel three-dimensional analysis of lidar waveforms was developed to characterize the total volume and spatial organization of vegetation material and empty space within the forest canopy. These aspects of the physical structure of canopies have been infrequently measured, from either field or remote methods. We applied this analysis to 22 plots in Douglas-fir/western hemlock stands on the west slope of the Cascades Range in Oregon. Each plot had coincident lidar data and field measurements of stand structure. We compared results from the novel analysis to two earlier methods of canopy description. Using the indices of canopy structure from all three methods of description as independent variables in a stepwise multiple regression, we were able to make nonasymptotic predictions of biomass and leaf area index (LAI) over a wide range, up to 1200 Mg ha−1 of biomass and an LAI of 12, with 90% and 75% of variance explained, respectively. Furthermore, we were able to make accurate estimates of other stand structure attributes, including the mean and standard deviation of diameter at breast height, the number of stems greater than 100 cm in diameter, and independent estimates of the basal area of Douglas-fir and western hemlock. These measurements can be directly related to indices of forest stand structural complexity, such as those developed for old-growth forest characterization. Indices of canopy structure developed using the novel, three-dimensional analysis accounted for most of the variables used in predictive equations generated by the stepwise multiple regression.
A key requirement of sustainable forest management is accurate, timely, and comprehensive information on forest resources, which is provided through forest inventories. In Canada, forest inventories are conventionally undertaken through the delineation and interpretation of forest stands using aerial photography, supported by data from permanent and temporary sample plots. In recent years, Airborne Laser Scanning (ALS) data have been shown to provide accurate estimates of a range of forest structural attributes. As a result, ALS has emerged as an increasingly common data source for enhanced forest inventory programs. Capture of species compositional information with ALS, based upon the nature of the data, is less reliable than structural variables, with species information typically derived from spectral or textural interpretation of aerial photography or very high spatial resolution digital imagery. Utilizing national allometric equations for the major species found in British Columbia, Canada, and a series of individual tree-level simulations, we analyzed (i) how incorrect species identification can influence individual tree volume prediction; (ii) which of the possible species substitutions result in higher volume errors; (iii) how the error in height that is typical for photogrammetry-based and ALS-based forest inventories impacts individual tree volume estimates; and (iv) the impact of combined errors in both species composition and height on overall individual tree volume estimates. Our results indicate that species information is important for volume calculations, and that the use of generic (i.e. all species) or cover-type allometric equations can lead to large errors in volume estimates. We also found that, even with a 50% error in species composition (whereby incorrect species-specific equations are substituted), volume estimates derived from species-specific allometric equations were more accurate than estimates derived from generic or cover-type equations. Our findings indicate that errors in species composition have less impact on individual tree volume estimates than errors in height measurement. The implications of these results are that, with very accurate estimates of height provided by ALS and knowledge of what dominant species is expected in a stand, accurate estimates of volume can be generated in the absence of more detailed species composition information.
The random forest (RF) algorithm by Leo Breiman has become a standard data analysis tool in bioinformatics. It has shown excellent performance in settings where the number of variables is much larger than the number of observations, can cope with complex interaction structures as well as highly correlated variables and return measures of variable importance. This paper synthesizes 10 years of RF development with emphasis on applications to bioinformatics and computational biology. Special attention is paid to practical aspects such as the selection of parameters, available RF implementations, and important pitfalls and biases of RF and its variable importance measures (VIMs). The paper surveys recent developments of the methodology relevant to bioinformatics as well as some representative examples of RF applications in this context and possible directions for future research. © 2012 Wiley Periodicals, Inc.
This article is categorized under: Algorithmic Development > Hierarchies and Trees
Algorithmic Development > Statistics
Application Areas > Health Care
In this paper two sampling and estimation strategies for regional forest inventory were investigated in detail and results were presented for various geographical scales. Airborne laser scanner (ALS) data were acquired to augment data from a systematic sample of National Forest Inventory (NFI) ground plots in Hedmark County, Norway (27,390 km2). Approximately 50% of the NFI field plots were covered by the systematic ALS sample of 53 parallel flight lines spaced 6 km apart. The area was stratified into eight cover classes and independent log-transformed regression models were developed for each class to predict total above-ground dry biomass (AGB). The two laser-ground estimation strategies tested were a model-dependent (MD), two-phase approach that rests on the assumption that the predictive models are correctly specified, and a model-assisted (MA) approach with a two-stage probability sampling design which utilizes design-unbiased estimators. ALS AGB estimates were reported by land cover class and compared to the NFI ground estimates. The ALS-based MA and MD mean estimates differed from the NFI AGB estimates by about 2% and 8%, respectively, for the entire County. At the county level the smallest estimated standard error (SE) for the estimates was obtained using the field data alone. However, the SEs calculated from field and ALS data were based on unequal numbers of ground plots. When considering only the NFI plots in the ALS strips, the smallest SEs were obtained using the MD framework. However, we also illustrated the sensitivity of the estimates of applying different plausible models. All the applied estimators assumed simple random sampling while the selection of flight lines as well as ground plots followed a systematic design. Thus, the estimates of SE were most likely conservative. Simulated sampling undertaken in a parallel research effort suggests that the overestimation of the SEs was probably much larger for the ALS-based estimates compared to the NFI estimates. ALS-based estimates were also derived for sub-county political units and thereby demonstrated how limited sample sizes affect the standard error of the biomass estimates.
Estimates of growing stock volume are reported by the national forest inventories (NFI) of most countries and may serve as the basis for aboveground biomass and carbon estimates as required by an increasing number of international agreements. The probability-based (design-based) statistical estimators traditionally used by NFIs to calculate estimates are generally unbiased and entail only limited computational complexity. However, these estimators often do not produce sufficiently precise estimates for areas with small sample sizes. Model-based estimators may overcome this disadvantage, but they also may be biased and estimation of variances may be computationally intensive. For a minor region within Hedmark County, Norway, the study objective was to compare estimates of mean forest growing stock volume per unit area obtained using probability- and model-based estimators. Three of the estimators rely to varying degrees on maps that were constructed using a nonlinear logistic regression model, forest inventory data, and lidar data. For model-based estimators, methods for evaluating quality of fit of the models and reducing the computational intensity were also investigated. Three conclusions were drawn: the logistic regression model exhibited no serious lack of fit to the data; estimators enhanced using maps produced greater precision than estimates based on only the plot observations; and third, model-based synthetic estimators benefit from sample sizes for larger areas when applied to smaller subsets of the larger areas.
Tree height is an important variable in forest inventory programs but is typically time-consuming and costly to measure in the field using conventional techniques. Airborne light detection and ranging (lidar) provides individual tree height measurements that are highly correlated with field-derived measurements, but the imprecision of conventional field techniques does not allow for definitive assessments regarding the absolute accuracy of lidar tree height measurements and the relative influence of beam divergence setting (i.e., laser footprint size), species type, and digital terrain model (DTM) error on the accuracy of height measurements. In this study, we developed a methodology for acquiring accurate individual tree height measurements (
The two main approaches to deriving forest variables from laser-scanning data are the statistical area-based approach (ABA) and individual tree detection (ITD). With ITD it is feasible to acquire single tree information, as in field measurements. Here, ITD was used for measuring training data for the ABA. In addition to automatic ITD (ITDauto), we tested a combination of ITDauto and visual interpretation (ITDvisual). ITDvisual had two stages: in the first, ITDauto was carried out and in the second, the results of the ITDauto were visually corrected by interpreting three-dimensional laser point clouds. The field data comprised 509 circular plots (r=10m) that were divided equally for testing and training. ITD-derived forest variables were used for training the ABA and the accuracies of the k-most similar neighbor (k-MSN) imputations were evaluated and compared with the ABA trained with traditional measurements. The root-mean-squared error (RMSE) in the mean volume was 24.8%, 25.9%, and 27.2% with the ABA trained with field measurements, ITDauto, and ITDvisual, respectively. When ITD methods were applied in acquiring training data, the mean volume, basal area, and basal area-weighted mean diameter were underestimated in the ABA by 2.7−9.2%. This project constituted a pilot study for using ITD measurements as training data for the ABA. Further studies are needed to reduce the bias and to determine the accuracy obtained in imputation of species-specific variables. The method could be applied in areas with sparse road networks or when the costs of fieldwork must be minimized.
Mean tree height, dominant height, mean diameter, stem number, basal area and timber volume of 116 georeferenced field sample plots were estimated from various canopy height and canopy density metrics derived by means of a small-footprint laser scanner over young and mature forest stands using regression analysis. The sample plots were distributed systematically throughout a 6500 ha study area, and the size of each plot was 232.9 m 2 . Regressions for coniferous forest explained 60-97% of the variability in ground reference values of the six studied characteristics. A proposed practical two-phase procedure for prediction of corresponding characteristics of entire forest stands was tested. Fifty-seven test plots within the study area with a size of approximately 3740 m 2 each were divided into 232.9 m 2 regular grid cells. The six examined characteristics were predicted for each grid cell from the corresponding laser data using the estimated regression equations. Average values for each test plot were computed and compared with ground-based estimates measured over the entire plot. The bias and standard deviations of the differences between predicted and ground reference values (in parentheses) of mean height, dominant height, mean diameter, stem number, basal area and volume were −0.58 to −0.85 m (0.64-1.01 m), −0.60 to −0.99 m (0.67-0.84 m), 0.15-0.74 cm (1.33-2.42 cm), 34-108 ha −1 (97-466 ha −1 ), 0.43-2.51 m 2 ha −1 (1.83-3.94 m 2 ha −1 ) and 5.9-16.1 m 3 ha −1 (15.1-35.1 m 3 ha −1 ), respectively.