Tree species classification in subtropical forests using small-footprint full-waveform LiDAR data

The accurate classification of tree species is critical for the management of forest ecosystems, particularly subtropical forests, which are highly diverse and complex ecosystems. While airborne Light Detection and Ranging (LiDAR) technology offers significant potential to estimate forest structural attributes, the capacity of this new tool to classify species is less well known. In this research, full-waveform metrics were extracted by a voxel-based composite waveform approach and examined with a Random Forests classifier to discriminate six subtropical tree species (i.e., Masson pine (Pinus massoniana Lamb.)), Chinese fir (Cunninghamia lanceolata (Lamb.) Hook.), Slash pines (Pinus elliottii Engelm.), Sawtooth oak (Quercus acutissima Carruth.) and Chinese holly (Ilex chinensis Sims.) at three levels of discrimination. As part of the analysis, the optimal voxel size for modelling the composite waveforms was investigated, the most important predictor metrics for species classification assessed and the effect of scan angle on species discrimination examined. Results demonstrate that all tree species were classified with relatively high accuracy (68.6% for six classes, 75.8% for four main species and 86.2% for conifers and broadleaved trees). Full-waveform metrics (based on height of median energy, waveform distance and number of waveform peaks) demonstrated high classification importance and were stable among various voxel sizes. The results also suggest that the voxel based approach can alleviate some of the issues associated with large scan angles. In summary, the results indicate that full-waveform LIDAR data have significant potential for tree species classification in the subtropical forests.     

Tree species identification in mixed coniferous forest using airborne laser scanning

This study tests the capacity of relatively low density (<1 return/m²) airborne laser scanner data for discriminating between Douglas-fir, western larch, ponderosa pine, and lodgepole pine in a western North American montane forest and it evaluates the relative importance of intensity, height, and return type metrics for classifying tree species. Collectively, Exploratory Data Analysis, Pearson Correlation, ANOVA, and Linear Discriminant Analysis show that structural and intensity characteristics generated from LIDAR data are useful for classifying species at dominant and individual tree levels in multi-aged, mixed conifer forests. Proportions of return types and mean intensities are significantly different between species ($p$-value < 0.001) for plot-level dominant species and individual trees. Classification accuracies based on single variables range from 49%–61% at the dominant species level and 37%–52% for individual trees. The accuracy can be improved to 95% and 68% respectively by using multiple variables. The inclusion of proportion of return type greatly improves the classification accuracy at the dominant species level, but not for individual trees, while canopy height improves the accuracy at both levels. Overall differences in intensity and return type between species largely reflect variations in the physical structure of trees and stands. These results are consistent with the findings of others and point to airborne laser scanning as a useful source of data for species classification. However, there are still many knowledge gaps that prevent accurate mapping of species using ALS data alone, particularly with relatively sparse datasets like the one used in this study. Further investigations using other datasets in different forest types will likely result in improvements to species identification and mapping for some time to come.     

Backscatter coefficient as an attribute for the classification of full-waveform airborne laser scanning data in urban areas

Airborne laser scanning (ALS) data are increasingly being used for land cover classification. The amplitudes of echoes from targets, available from full-waveform ALS data, have been found to be useful in the classification of land cover. However, the amplitude of an echo is dependent on various factors such as the range and incidence angle, which makes it difficult to develop a classification method which can be applied to full-waveform ALS data from different sites, scanning geometries and sensors. Additional information available from full-waveform ALS data, such as range and echo width, can be used for radiometric calibration, and to derive backscatter cross section. The backscatter cross section of a target is the physical cross sectional area of an idealised isotropic target, which has the same intensity as the selected target. The backscatter coefficient is the backscatter cross section per unit area. In this study, the amplitude, backscatter cross section and backscatter coefficient of echoes from ALS point cloud data collected from two different sites are analysed based on urban land cover classes. The application of decision tree classifiers developed using data from the first study area on the second demonstrates the advantage of using the backscatter coefficient in classification methods, along with spatial attributes. It is shown that the accuracy of classification of the second study area using the backscatter coefficient (kappa coefficient 0.89) is higher than those using the amplitude (kappa coefficient 0.67) or backscatter cross section (kappa coefficient 0.68). This attribute is especially useful for separating road and grass.     

Integration of full-waveform LiDAR and hyperspectral data to enhance tea and areca classification

Hyperspectral image and full-waveform light detection and ranging (LiDAR) data provide useful spectral and geometric information for classifying land cover. Hyperspectral images contain a large number of bands, thus providing land-cover discrimination. Waveform LiDAR systems record the entire time-varying intensity of a return signal and supply detailed information on geometric distribution of land cover. This study developed an efficient multi-sensor data fusion approach that integrates hyperspectral data and full-waveform LiDAR information on the basis of minimum noise fraction and principal component analysis. Then, support vector machine was used to classify land cover in mountainous areas. Results showed that using multi-sensor fused data achieved better accuracy than using a hyperspectral image alone, with overall accuracy increasing from 83% to 91% using population error matrices, for the test site. The classification accuracies of forest and tea farms exhibited significant improvement when fused data were used. For example, classification results were more complete and compact in tea farms based on fused data. Fused data considered spectral and geometric land-cover information, and increased the discriminability of vegetation classes that provided similar spectral signatures.     

Support vector machines for tree species identification using LiDAR-derived structure and intensity variables

Tree species identification and forest type classification are critical for sustainable forest management and native forest conservation. Recent success in forest classification and tree species identification using LiDAR (light detection and ranging)-derived variables has been reported in many studies. However, there is still considerable scope for further improvement in classification accuracy. It has driven research into more efficient classifiers such as support vector machines (SVMs) to take maximum advantage of the information extracted from LiDAR data for potential increases in the accuracy of tree species classification. This study demonstrated the success of the SVMs for the identification of the Myrtle Beech (the dominant species of the Australian cool temperate rainforest in the study area) and adjacent tree species – notably, the Silver Wattle at individual tree level using LiDAR-derived structure and intensity variables. An overall accuracy of 92.8% was achieved from the SVM approach, showing significant advantages of the SVMs over the traditional classification methods such as linear discriminant analysis in terms of classification accuracy.     

Mapping multiple tree species classes using a hierarchical procedure with optimized node variables and thresholds based on high spatial resolution satellite data

ABSTRACT

Tree species distribution mapping using remotely sensed data has long been an important research area. However, previous studies have rarely established a comprehensive and efficient classification procedure to obtain an accurate result. This study proposes a hierarchical classification procedure with optimized node variables and thresholds to classify tree species based on high spatial resolution satellite imagery. A classification tree structure consisting of parent and leaf nodes was designed based on user experience and visual interpretation. Spectral, textural, and topographic variables were extracted based on pre-segmented images. The random forest algorithm was used to select variables by ranking the impact of all variables. An iterating approach was used to optimize variables and thresholds in each loop by comprehensively considering the test accuracy and selected variables. The threshold range for each selected variable was determined by a statistical method considering the mean and standard deviation for two subnode types at each parent node. Classification of tree species was implemented using the optimized variables and thresholds. The results show that (1) the proposed procedure can accurately map the tree species distribution, with an overall accuracy of over 86% for both training and test stages; (2) critical variables for each class can be identified using this proposed procedure, and optimal variables of most tree plantation nodes are spectra related; (3) the overall forest classification accuracy using the proposed method is more accurate than that using the random forest (RF) and classification and regression tree (CART). The proposed approach provides results with 3.21% and 7.56% higher overall land cover classification accuracy and 4.68% and 10.28% higher overall forest classification accuracy than RF and CART, respectively.     

Analysis of full-waveform LiDAR data for classification of an orange orchard scene

Full-waveform laser scanning data acquired with a Riegl LMS-Q560 instrument were used to classify an orange orchard into orange trees, grass and ground using waveform parameters alone. Gaussian decomposition was performed on this data capture from the National Airborne Field Experiment in November 2006 using a custom peak-detection procedure and a trust-region-reflective algorithm for fitting Gauss functions. Calibration was carried out using waveforms returned from a road surface, and the backscattering coefficient γ was derived for every waveform peak. The processed data were then analysed according to the number of returns detected within each waveform and classified into three classes based on pulse width and γ. For single-peak waveforms the scatterplot of γ versus pulse width was used to distinguish between ground, grass and orange trees. In the case of multiple returns, the relationship between first (or first plus middle) and last return γ values was used to separate ground from other targets. Refinement of this classification, and further sub-classification into grass and orange trees was performed using the γ versus pulse width scatterplots of last returns. In all cases the separation was carried out using a decision tree with empirical relationships between the waveform parameters. Ground points were successfully separated from orange tree points. The most difficult class to separate and verify was grass, but those points in general corresponded well with the grass areas identified in the aerial photography. The overall accuracy reached 91%, using photography and relative elevation as ground truth. The overall accuracy for two classes, orange tree and combined class of grass and ground, yielded 95%. Finally, the backscattering coefficient γ of single-peak waveforms was also used to derive reflectance values of the three classes. The reflectance of the orange tree class (0.31) and ground class (0.60) are consistent with published values at the wavelength of the Riegl scanner (1550 nm). The grass class reflectance (0.46) falls in between the other two classes as might be expected, as this class has a mixture of the contributions of both vegetation and ground reflectance properties.     

Utility of image point cloud data towards generating enhanced multitemporal multisensor land cover maps

Multitemporal land cover classification over urban areas is challenging, especially when using heterogeneous data sources with variable quality attributes. A prominent challenge is that classes with similar spectral signatures (such as trees and grass) tend to be confused with one another. In this paper, we evaluate the efficacy of image point cloud (IPC) data combined with suitable Bayesian analysis based time-series rectification techniques to improve the classification accuracy in a multitemporal context. The proposed method uses hidden Markov models (HMMs) to rectify land covers that are initially classified by a random forest (RF) algorithm. This land cover classification method is tested using time series of remote sensing data from a heterogeneous and rapidly changing urban landscape (Kuopio city, Finland) observed from 2006 to 2014. The data consisted of aerial images (5 years), Landsat data (all 9 years) and airborne laser scanning data (1 year). The results of the study demonstrate that the addition of three-dimensional image point cloud data derived from aerial stereo images as predictor variables improved overall classification accuracy, around three percentage points. Additionally, HMM-based post processing reduces significantly the number of spurious year-to-year changes. Using a set of 240 validation points, we estimated that this step improved overall classification accuracy by around 3.0 percentage points, and up to 6 to 10 percentage points for some classes. The overall accuracy of the final product was 91% (kappa = 0.88). Our analysis shows that around 1.9% of the area around Kuopio city, representing a total area of approximately 0.61 km², experienced changes in land cover over the nine years considered.     

Individual tree crown approach for predicting site index in boreal forests using airborne laser scanning and hyperspectral data

Site productivity is essential information for sustainable forest management and site index (SI) is the most common quantitative measure of it. The SI is usually determined for individual tree species based on tree height and the age of the 100 largest trees per hectare according to stem diameter. The present study aimed to demonstrate and validate a methodology for the determination of SI using remotely sensed data, in particular fused airborne laser scanning (ALS) and airborne hyperspectral data in a forest site in Norway. The applied approach was based on individual tree crown (ITC) delineation: tree species, tree height, diameter at breast height (DBH), and age were modelled and predicted at ITC level using 10-fold cross validation. Four dominant ITCs per 400 m² plot were selected as input to predict SI at plot level for Norway spruce (Picea abies (L.) Karst.) and Scots pine (Pinus sylvestris L.). We applied an experimental setup with different subsets of dominant ITCs with different combinations of attributes (predicted or field-derived) for SI predictions. The results revealed that the selection of the dominant ITCs based on the largest DBH independent of tree species, predicted the SI with similar accuracy as ITCs matched with field-derived dominant trees (RMSE: 27.6% vs 23.3%). The SI accuracies were at the same level when dominant species were determined from the remotely sensed or field data (RMSE: 27.6% vs 27.8%). However, when the predicted tree age was used the SI accuracy decreased compared to field-derived age (RMSE: 27.6% vs 7.6%). In general, SI was overpredicted for both tree species in the mature forest, while there was an underprediction in the young forest. In conclusion, the proposed approach for SI determination based on ITC delineation and a combination of ALS and hyperspectral data is an efficient and stable procedure, which has the potential to predict SI in forest areas at various spatial scales and additionally to improve existing SI maps in Norway.     

大光斑激光雷达数据的森林冠层高度反演

针对传统的LM波形分解算法在GLAS大光斑波形数据处理中容易陷于局部最优解,限制了GLAS大光斑激光雷达数据在森林结构参数反演方面应用的问题,该文结合GLAS大光斑数据特征,引进优化后的EM算法对大光斑全波形数据进行分解,获取波形参数最优值。结合波形前缘长度和波形后缘长度,建立树高反演模型,并与LM分解算法建立的模型进行对比分析。研究结果表明,通过改进的EM算法对GLAS大光斑激光雷达数据进行处理,波形特征参数的获取更为精确,达到了较高的树高反演精度。     

机载LiDAR数据估算样地和单木尺度森林地上生物量

利用机载激光雷达点云数据,结合大量实测单木结构信息,分别从样地和单木尺度估算了森林地上生物量AGB。首先,利用局部最大值单木提取算法提取了每个样地内的单木结构参数,并针对样地和单木尺度分别计算了一组激光雷达变量。然后,利用激光雷达变量和地上生物量及其两者的对数形式,从样地和单木尺度分别构建了估算模型。最后,针对两种尺度估算过程中存在的不确定性进行了详细讨论。结果表明:(1)样地和单木尺度模型估算的森林地上生物量与地面实测值都具有明显的相关性,且对数模型估算效果要优于非对数模型;(2)样地尺度模型估算效果(R2=0.84,rRMSE=0.23)明显优于单木尺度模型(R2=0.61,rRMSE=0.46);(3)按树木类型分别进行估算可以提高单木地上生物量的估算精度;(4)不论是样地还是单木尺度地上生物量估算都存在一定的不确定性,与样地尺度相比,单木尺度估算过程的不确定性更大,这种不确定性主要来自单木识别过程。     

Tree survey and allometric models for tiger bush in northern Senegal and comparison with tree parameters derived from high resolution satellite data

A tree survey and an analysis of high resolution satellite data were performed to characterise the woody vegetation within a 10 × 10 km² area around a site located close to the town of Dahra in the semi-arid northern part of Senegal. The surveyed parameters were tree species, height, tree crown radius, and diameter at breast height (DBH), for which allometric models were determined. An object-based classification method was used to determine tree crown cover (TCC) from Quickbird data. The average TCC from the tree survey and the respective TCC from remote sensing were both about 3.0%. For areas beyond the surveyed areas TCC varied between 3.0% and 4.5%. Furthermore, an empirical correction factor for tree clumping was obtained, which considerably improved the estimated number of trees and the estimated average tree crown area and radius. An allometric model linking TCC to tree stem crosssectional area (CSA) was developed, which allows to estimate tree biomass from remote sensing. The allometric models for the three main tree species found performed well and had r²-values of about 0.7–0.8.     

利用国产GF-1卫星数据实现山区细小河流河宽的自动提取

以国产GF-1卫星影像为数据源,选取皇甫川流域内山区细小河流密集的上游1421 km²作为研究区域,针对因山区河流河道狭窄、形态复杂等导致的河流边界提取难度大、精度差、河宽无法自动提取的难题,首先利用改进的变异系数法筛选水体指数,再采用改进的决策树法结合DEM河网精确获取河流边界,最后通过自动化河宽提取算法实现对山区细小河流及其河宽的自动提取。结果表明,本文方法对山区河流判别的总体精度为89.5%,有效地排除了山体阴影等地物的干扰。对河宽为0~10 m的极细河流,本文方法提取河宽的误差为18.54%;10~30 m的细小河流,提取误差为12.07%。     

Assessing integration of intensity,polarimetric scattering,interferometric coherence and spatial texture metrics in PALSAR-derived land cover classification

Synthetic aperture radar (SAR) is an important alternative to optical remote sensing due to its ability to acquire data regardless of weather conditions and day/night cycle. The Phased Array type L-band SAR (PALSAR) onboard the Advanced Land Observing Satellite (ALOS) provided new opportunities for vegetation and land cover mapping. Most previous studies employing PALSAR investigated the use of one or two feature types (e.g. intensity, coherence); however, little effort has been devoted to assessing the simultaneous integration of multiple types of features. In this study, we bridged this gap by evaluating the potential of using numerous metrics expressing four feature types: intensity, polarimetric scattering, interferometric coherence and spatial texture. Our case study was conducted in Central New York State, USA using multitemporal PALSAR imagery from 2010. The land cover classification implemented an ensemble learning algorithm, namely random forest. Accuracies of each classified map produced from different combinations of features were assessed on a pixel-by-pixel basis using validation data obtained from a stratified random sample. Among the different combinations of feature types evaluated, intensity was the most indispensable because intensity was included in all of the highest accuracy scenarios. However, relative to using only intensity metrics, combining all four feature types increased overall accuracy by 7%. Producer’s and user’s accuracies of the four vegetation classes improved considerably for the best performing combination of features when compared to classifications using only a single feature type.     

结合面向对象和深度特征的高分影像树种分类

针对传统手工提取特征方法需要专业领域知识,提取高质量特征困难的问题,将深度迁移学习技术引入到高分影像树种分类中,提出一种结合面向对象和深度特征的高分影像树种分类方法。为了获取树种的精确边界,该方法首先利用多尺度分割技术分割整幅遥感影像,并选择训练样本作为深度卷积神经网络的输入。为了避免样本数量少导致过拟合问题,采用迁移学习方法,使用ImageNet上训练的VGG16模型参数初始化深度卷积神经网络,并利用全局平局池化压缩参数,在网络最后添加1024个节点的全连接层和7个节点的Softmax分类器,利用反向传播和Adam优化算法训练网络。最后分类整幅遥感影像,生成树种专题地图。以安徽省滁州市的皇甫山国家森林公园为研究区,QuickBird高分影像作为数据源,采用本文方法进行树种分类。试验结果表明,本文方法树种分类总体精度和Kappa系数分别为78.98%和0.685 0,在保证树种精度的同时实现了端到端的树种分类。     

Combining QuickBird,LiDAR, and GIS topography indices to identify a single native tree species in a complex landscape using an object-based classification approach

There are now a wide range of techniques that can be combined for image analysis. These include the use of object-based classifications rather than pixel-based classifiers, the use of LiDAR to determine vegetation height and vertical structure, as well terrain variables such as topographic wetness index and slope that can be calculated using GIS. This research investigates the benefits of combining these techniques to identify individual tree species. A QuickBird image and low point density LiDAR data for a coastal region in New Zealand was used to examine the possibility of mapping Pohutukawa trees which are regarded as an iconic tree in New Zealand. The study area included a mix of buildings and vegetation types. After image and LiDAR preparation, single tree objects were identified using a range of techniques including: a threshold of above ground height to eliminate ground based objects; Normalised Difference Vegetation Index and elevation difference between the first and last return of LiDAR data to distinguish vegetation from buildings; geometric information to separate clusters of trees from single trees, and treetop identification and region growing techniques to separate tree clusters into single tree crowns. Important feature variables were identified using Random Forest, and the Support Vector Machine provided the classification. The combined techniques using LiDAR and spectral data produced an overall accuracy of 85.4% (Kappa 80.6%). Classification using just the spectral data produced an overall accuracy of 75.8% (Kappa 67.8%). The research findings demonstrate how the combining of LiDAR and spectral data improves classification for Pohutukawa trees.     

基于决策树分类的地表覆盖遥感信息提取

地处西南的渝北地区地表覆盖类型复杂、土地利用多元化,仅依赖于光谱特征的传统遥感信息提取方法难以获得较高的分类精度。利用决策树分类技术对渝北地区的TM遥感影像进行分类,除光谱信息外还结合地质、NDVI、PCI等多源数据进行实验。结果表明,总精度和Kappa系数分别为88.42%和0.854 7,较传统的监督分类和仅依赖于光谱特征的决策树分类方法有较大提高,这也表明基于多源数据的决策树分类技术对地表覆盖复杂地区的遥感影像分类比较适用,是遥感信息提取的一种有效手段。     

全波形激光雷达回波分解方法

全波形激光雷达的回波中携带了被测目标的距离与特征信息,为了获取这些信息,本文提出了一种回波分解方法。本方法将原始的全波形回波分解为几个独立的高斯脉冲,并得到其函数表达式,从而提取出被测目标的距离等信息。分解过程中,首先,采用可变阈值的经验模态分解滤波法(EMD-soft)对原始波形进行滤波和噪声水平评估;其次,采用一套应对多种波形组成的初始参数估计方法,获取后续拟合所需的初始参数;最后,采用LM(Levenberg-Marquardt)优化算法对回波进行拟合优化,从而获取全波形回波中包含的独立高斯脉冲及其函数表达式。仿真波形的分解实验表明,分解误差在0.1 ns量级,换算成距离误差为15 mm,通过实验室自制的全波形激光雷达实验系统获取的回波的分解实验表明,分解的距离误差小于0.1 m。对比另外两种高斯分解方法对于相同仿真与实验数据的分解结果可以看出,本方法在分解成功率与精度上都有较大的提高。回波分解后的独立高斯脉冲中,除距离外还含有被测目标的反射率、粗糙度、面型等丰富的信息,回波分解方法作为回波分析的基础,将在遥感、测绘等生产与科研领域中发挥非常重要的作用。     

机器学习算法在森林地上生物量估算中的应用

森林地上生物量是森林生产力的重要评价指标,对其进行高效监测对维持全球碳平衡和保护生态系统具有重要意义。本文首先基于冠层高度模型数据,通过分水岭分割算法得到单木冠幅边界;然后在单木冠幅范围内提取23个LiDAR变量,结合佩诺布斯科特试验森林的87组实测数据,利用随机森林和支持向量机建立森林地上生物量估算模型;最后对样地模型估算的结果进行了比较,讨论了预测结果及其精度。结果表明:本文选用的随机森林模型和支持向量机模型在估算森林地上生物量的应用中获得了较高的精度;并且,随机森林模型在基于机载雷达数据估测森林地上生物量中的估算精度更高,模型泛化能力更强,制图精度也更好,具有更好的适用性。