Generating Tree-Lists by Fusing Individual Tree Detection and Nearest Neighbor Imputation Using Airborne LiDAR Data

Active remotely sensed data can be used to perform a variety of forestry tasks including stand characterization, inventory, and management of forest and fire behavior modeling. The present work investigates the potential of Airborne Laser Scanning (ALS) derived methods applied in the deciduous forest by processing an individual tree detection (ITD) based on canopy Height model (CHM) and tree segmentation of larger-area point clouds. Different algorithms are tested and their performances are evaluated to show which of them can provide the most adequate number of trees compared with the ground truth. Tree scale information is used in order to determine stand age. The forest height, structure, and density are specified by applying individual tree Detection (ITD) to calculate some forest attributes such as stem volume, forest uniformity, and biomass estimation. The major aim of this post is to examine the state of the forest to monitor it in real-time. We assume that utilizing the LM algorithm, which was originally built for ITD from LiDAR data, trees should be automatically distinguished from the ALS-derived CHM with reasonable accuracy. As a result, the present research work studies the fixed treetop window size (FWS), fixed smoothing window size (SWS), and variable window (VW) effect on ITD performance (RMSE=3.4% and R=0.88). It is obvious, from the obtained results that smaller window sizes result in more trees. In fact, the smallest trees obscured by the largest trees containing the highest points in the neighborhood are often ignored by large windows. Crown delineation is also explored to extract the height of the trees, radius crown and, 3D coordinates and to compare them to those detected by a Low Bluetooth sensor “iBeacon.”.

We examined the problem of selecting predictor variables for Nearest Neighbor (NN) imputation in remote sensing based forest inventory. Eighty-three variables were calculated from Airborne Laser Scanning data and aerial images, with responses being either dominant height or a set of five common stand attributes. Three different approaches were compared with select predictor variables. Analyses were repeated with three different NN imputation methods using a varying number of predictor variables. Results indicated that variable selection is justified, but it must be done properly. The most accurate method to select predictors was to minimize error using Simulated Annealing. For a single response, the most accurate imputation method was Random Forest proximity matrix-based imputation, whereas Most Similar Neighbor was the most accurate for five responses. An optimization-based distance metric also worked well. We also examined the degree to which different imputation methods are prone to overfitting as well as how to properly do cross-validation in NN imputation.

As oil palm is cultivated in large-scale plantations, prior and post-planting operations are laborious and time expensive, and also manual implementation of some of these operations often results in incorrect measurements and information. Successful oil palm management in prior and post-planting operations requires effective techniques to collect precise information. Unmanned aerial vehicle (UAV) imagery is a low-cost alternative to field-based assessment but requires the development of methods to easily and accurately extract the required information. The individual oil palm tree detection and height assessment are important and labor-intensive tasks for large-scale plantations where trees taller than 15 m are being replanted due to the high cost of harvesting and low yield. Therefore, in this chapter we demonstrate a general work flow for individual oil palm detection and height assessment using UAV imagery. We explain local maximum (LM) and template matching (TM) techniques as two commonly used approaches in individual tree detection. The accuracy of each method was evaluated on 20 randomly selected plots on UAV image with recall, precision, and F-score method. The F-score for LM method is higher than TM methods which, respectively, are 0.83 and 0.60. From 17,252 oil palm trees in the 20 plots, LM algorithm could detect 1395 (almost 80%) and TM algorithm, 967 (almost 55%). Three hundred fifty-seven and 785 trees have been missed, respectively, in LM and TM approaches. In both cases, background vegetation incorrectly has been labeled as oil palm tree, where 141 and 322 objects have been falsely detected as oil palm tree in LM and TM approaches. LM worked better in almost all of the plots; however the performance decreased in highly dense area. Contradictory to TM approach, in LM approach, shadows did not affect the performance as it was reflected in precision value. In densely cultivated plots due to leaves overlapping of neighboring trees, algorithm failed, and also in sparsely cultivated plots, shadows caused some commission errors. Inherited distortion of UAV image also caused some omission errors in TM approach. Individual detected tree with LM algorithm was used in the next part for oil palm height estimation overlaying with canopy height model.

Airborne laser scanning based forest inventories employ two major methods: individual tree detection (ITD) and the area-based statistical approach (ABSA). ITD is based on the assumption that trees are of a certain form and can be delineated using airborne laser scanning techniques, whereas ABSA is an empirical method based on the relations between area-level forest attributes and laser echo height distributions. These two methods are compared here within the same test area in terms of their usefulness for estimating mean forest stand characteristics and tree size distributions. All evaluations were performed using leave-one-out cross validation. The average errors in volume and basal area did not differ significantly between the methods. ABSA resulted in overall better accuracies when estimating the diameter and height of the basal area median tree and the number of stems, whereas ITD produced significantly biased estimates for the number of stems and the mean tree size. Tree size distributions were estimated with slightly better accuracy using ABSA. More comprehensive investigations revealed that both methods were not able to estimate forest structure (tree size distribution and spatial distribution of tree locations), which in turn, affected the estimation accuracies.

Uncovering forest dynamics using historical forest inventory data and Landsat time series

Fast-growing Eucalyptus spp. forest plantations and their resultant wood products are economically important and may provide a low-cost means to sequester carbon for greenhouse gas reduction. The development of advanced and optimized frameworks for estimating forest plantation attributes from lidar remote sensing data combined with statistical modeling approaches is a step towards forest inventory operationalization and might improve industry efficiency in monitoring and managing forest resources. In this study, we first developed and tested a framework for modeling individual tree attributes in fast-growing Eucalyptus forest plantation using airborne lidar data and linear mixed-effect models (LME) and assessed the gain in accuracy compared to a conventional linear fixed-effects model (LFE). Second, we evaluated the potential of using the tree-level estimates for determining tree attribute uniformity across different stand ages. In the field, tree measurements, such as tree geolocation, species, genotype, age, height (Ht), and diameter at breast height (dbh) were collected through conventional forest inventory practices, and tree-level aboveground carbon (AGC) was estimated using allometric equations. Individual trees were detected and delineated from lidar-derived canopy height models (CHM), and crown-level metrics (e.g., crown volume and crown projected area) were computed from the lidar 3-D point cloud. Field and lidar-derived crown metrics were combined for ht, dbh, and AGC modeling using an LME. We fitted a varying intercept and slope model, setting species, genotype, and stand (alone and nested) as random effects. For comparison, we also modeled the same attributes using a conventional LFE model. The tree attribute estimates derived from the best LME model were used for assessing forest uniformity at the tree level using the Lorenz curves and Gini coefficient (GC). We successfully detected 96.6% of the trees from the lidar-derived CHM. The best LME model for estimating the tree attributes was composed of the stand as a random effect variable, and canopy height, crown volume, and crown projected area as fixed effects. The %RMSE values for tree-level height, dbh, and AGC were 8.9%, 12.1%, and 23.7% for the LFE model and improved to 7.3%, 7.1%, and 13.6%, respectively, for the LME model. Tree attributes uniformity was assessed with the Lorenz curves and tree-level estimations, especially for the older stands. All stands showed a high level of tree uniformity with GC values approximately 0.2. This study demonstrates that accurate detection of individual trees and their associated crown metrics can be used to estimate Ht, dbh, and AGC stocks as well as forest uniformity in fast-growing Eucalyptus plantations forests using lidar data as inputs to LME models. This further underscores the high potential of our proposed approach to monitor standing stock and growth in Eucalyptus—and similar forest plantations for carbon dynamics and forest product planning.

Nearest neighbor imputation of species-level, plot-scale forest structure attributes from LiDAR data

A graph-based progressive morphological filtering (GPMF) method for generating canopy height models using ALS data

Detection and delineation of individual trees is important for the estimation of forest information using Light Detection and Ranging (LiDAR) point clouds. However, over-segmentation of canopy height models (CHMs) is frequently caused by subcrowns and branches, whereas under-segmentation is caused by adjacent crowns, especially in deciduous forests. To eliminate such problems, an approach for delineating individual trees based on curvature attributes derived from LiDAR data is proposed. To detect accurate treetops for reducing over-segmentation, the dominant sizes of tree crowns are first determined and used to fit the dome-shaped tree crowns in CHMs, and then the minimum curvature values of fitted CHMs at different scales are calculated for treetop detection. For delineation of individual trees, tree features such as height variation and curvature are chosen to form a feature vector as a pixel-level logistic regression classifier. Then an ensemble classifier combining multiple logistic regression classifiers is trained to analyze the similarity between neighboring points to reduce under-segmentation. Finally, segment maps at different scales are integrated to generate the final segment map. Our experimental results on 2 test areas demonstrate that the proposed method has promising applications for eliminating over-segmentation and under-segmentation in deciduous forests.

Estimating forest stand density and structure using Bayesian individual tree detection, stochastic geometry, and distribution matching

ABSTRACTWe examine the nearest neighbor (NN) imputation of species-specific logwood volumes using airborne laser scanning (ALS) data and aerial images. We compare different remote sensing (RS) data combinations as predictor variables in an area-based prediction of logwood volumes using separate training and validation data. We include multispectral leaf-on ALS data, bi-temporal leaf-off ALS data and aerial images in the analyses. Two response configurations are used in the NN imputations: (1) simultaneous imputation in which species-specific logwood volumes are response variables, and (2) separate imputation by tree species in which the attributes of one tree species at a time are response variables. Although an unrealistic alternative in practical implementation, the combination of leaf-on and leaf-off ALS metrics as predictors proved to be the most successful RS data combination, according to the RMSE values associated with the predicted species-specific and dominant logwood volumes. The results showed that older leaf-off ALS data perform well in combination with leaf-on ALS data. In general, predictive performance was better with simultaneous imputation than with separate imputation by tree species. Our finding promotes an awareness of how best to utilize various RS data in future forest inventories.

BackgroundDetermining the spatial distribution of tree heights at the regional area scale is significant when performing forest above-ground biomass estimates in forest resource management research. The geometric-optical mutual shadowing (GOMS) model can be used to invert the forest canopy structural parameters at the regional scale. However, this method can obtain only the ratios among the horizontal canopy diameter (CD), tree height, clear height, and vertical CD. In this paper, we used a semi-variance model to calculate the CD using high spatial resolution images and expanded this method to the regional scale. We then combined the CD results with the forest canopy structural parameter inversion results from the GOMS model to calculate tree heights at the regional scale.ResultsThe semi-variance model can be used to calculate the CD at the regional scale that closely matches (mainly with in a range from − 1 to 1 m) the CD derived from the canopy height model (CHM) data. The difference between tree heights calculated by the GOMS model and the tree heights derived from the CHM data was small, with a root mean square error (RMSE) of 1.96 for a 500-m area with high fractional vegetation cover (FVC) (i.e., forest area coverage index values greater than 0.8). Both the inaccuracy of the tree height derived from the CHM data and the unmatched spatial resolution of different datasets will influence the accuracy of the inverted tree height. And the error caused by the unmatched spatial resolution is small in dense forest.ConclusionsThe semi-variance model can be used to calculate the CD at the regional scale, together with the canopy structure parameters inverted by the GOMS model, the mean tree height at the regional scale can be obtained. Our study provides a new approach for calculating tree height and provides further directions for the application of the GOMS model.

Trees play a vital role in the urban ecosystem, providing benefits for society, ecology and economy. In particular in urban areas, trees mitigate the urban heat island effect, filter air pollution, regulate microclimate and hydrology, bond carbon dioxide, and provide spaces for recreation and leisure, among others. Despite these diverse positive effects, detailed information on the number, location, dimensions and other characteristics of urban trees remains scarce. For this reason, most cities in Germany currently aim to establish a tree information system for efficient and targeted management of their tree inventory. However, traditional terrestrial surveying is time-consuming and costly and therefore only suitable to a limited extent. In addition, the municipal tree cadastre usually only includes urban trees on public propertyand thus does not cover the complete stock. Against this background, remote sensing acquisitions with very-high spatial resolution (VHR) of less than one meter offer promising capabilities for area-wide detection, delineation, and characterization of urban trees. In this study, we use VHR aerial imagery as well as a derived canopy height model (CHM) for detection and delineation of urban trees. Different methods for individual tree detection using local maximum (LM) filtering andLaplacian of Gaussian (LoG) blob detectionare compared and evaluated. For tree crown delineation, marker-controlled watershed segmentation (MCWS), clustering using Voronoi tessellation, and region growing are implemented as segmentationtechniques. The detection of individual trees and delieation of tree crowns are validated against about1,000 reference trees from visual interpretation via stereophotogrammetry.In addition, we relate our results to street tree location data of Munich, which was derived from mobile terrestrial laser scanning (TLS).The characterization of urban trees is realized based on the 3-dimensional shape of individual tree segments as well as auxillary data sets of land use and building density. According to our analyses, there are 1.54 million trees in Munich.Compared to available reference trees, tree detection was evaluated with highest values of F-score, precision, and recall of 0.95, 0.99, and 0.94, respectively. Results of tree crown segmentation revealed an overall accuracy of 88.1 % compared to crowns of reference trees. Based on auxillary land use information, urban trees were categorized into street trees, (public) park trees, as well as trees in (private) residential gardens.In Munich, 9.1 % are characterized as street trees, 38.4 % are allocated in residential gardens and 33.1 % stand in public parks. The remaining 19.4 % oftree segments were found onother land use such as agricultural areas, parking lots, or along railroad tracks. According to these categories, the height and crown area of urban trees are analyzed and related to the distance to the city center. In a more general manner, this analysis was performed in relation to the building density in Munich. As expected, relatively few trees were found close to the city center and generally on areas with high building density. However, these areas are particularly associated with the greatest challenges in the context of sustainable and climate change-adapted urban development.In this study, we demonstrate that information derived from remote sensing contributes new spatial and quantitative knowledge on urban trees, providing the basis for sustainable management and informed decision-making in cities.

Classification of species at the individual tree level would be beneficial to many applications including forest landscape visualization, forest management, and biodiversity monitoring. This article develops a patch-based classification algorithm of individual tree species based on convolutional neural network. The individual trees are first detected using the local maximum method from the canopy height model, as derived from light detection and ranging (LiDAR) data. The detected individual trees are then cropped into patches for classification based on the tree apexes, and three spatial scale image patches are chosen for analysis and discussion. A modified ResNet50 deep network is further employed for the cropped individual tree patches classification. The patch-based method accounts for the contexture information of a tree and does not require the feature selection or the feature reduction processes. About 1388 training samples including Ficus microcarpa Linn. f., Delonix regia, Chorisia speciosa A.St.-Hil., Dimocarpus longan Lour., Musa nana Lour., Carica papaya, and Others (the other tree species except the above six) were collected from both field work and visual interpretation. Aerial images, LiDAR data, and Worldview images were used for the tree species classification. For 362 test tree samples, the results of patch size 64 achieve the best accuracies, and the proposed method outperforms the traditional machine learning method with the overall accuracy of 89.06% + 0.58% using aerial images only. Transferability Study to the Luhu Park also indicated the feasibility of our method. While challenges in individual tree detection and multisource data fusion remain, the solution shows the potential in characterizing tree species at the individual tree level using remote sensing data.

Forest-related statistics, including forest biomass, carbon sink, and the prevention of forest fires, can be obtained by estimating stand density. In this study, a dataset with the laser pulse density of 225.5 pulses/m2 was obtained using airborne laser scanning in a tropical broadleaf forest. Three digital surface models (DSMs) were generated using first-echo, last-echo, and highest first-echo data. Three canopy height models (CHMs) were obtained by deducting the digital elevation model from the three DSMs. The cell sizes (Csizes) of the CHMs were 1, 0.5, and 0.2 m. In addition, stand density was estimated using CHM data and following the local maximum method. The stand density of 35 sample regions was acquired via in-situ measurement. The results indicated that the root-mean-square error ( R M S E ) ranged between 1.68 and 2.43; the R M S E difference was only 0.78, indicating that stand density was effectively estimated in both cases. Furthermore, regression models were used to correct the error in stand density estimations; the R M S E after correction was called R M S E ′ . A comparison of the R M S E and R M S E ′ showed that the average value decreased from 12.35 to 2.66, meaning that the regression model could effectively reduce the error. Finally, a comparison of the effects of different laser pulse densities on the R M S E value showed that, in order to obtain the minimum R M S E for stand density, the laser pulse density must be greater than 10, 30, and 125 pulses/m2 at Csizes of 1, 0.5, and 0.2 m, respectively.