Full-waveform Laser Scanning Data Research Articles

Processing and analysis of airborne full-waveform laser scanning data for the characterization of forest structure and fuel properties

<p class="Bodytext">This PhD thesis addresses the development of full-waveform airborne laser scanning (ALS<sub>FW</sub>) processing and analysis methods to characterize the vertical forest structure, in particular the understory vegetation. In this sense, the influence of several factors such as pulse density, voxel parameters (voxel size and assignation value), scan angle at acquisition, radiometric correction and regression methods is analyzed on the extraction of ALS<sub>FW</sub> metric values and on the estimate of forest attributes. Additionally, a new software tool to process ALS<sub>FW</sub> data is presented, which includes new metrics related to understory vegetation. On the other hand, occlusion caused by vegetation in the ALS<sub>FW</sub>, discrete airborne laser scanning (ALS<sub>D</sub>) and terrestrial laser scanning (TLS) signal is characterized along the vertical structure. Finally, understory vegetation density is detected and determined by ALS<sub>FW</sub> data, as well as characterized by using the new proposed metrics.</p>

VOXEL BASED REPRESENTATION OF FULL-WAVEFORM AIRBORNE LASER SCANNER DATA FOR FORESTRY APPLICATIONS

Abstract. The advantages of using airborne full-waveform laser scanner data in forest applications, e.g. for the description of the vertical vegetation structure or accurate biomass estimation, have been emphasized in many publications. To exploit the full potential offered by airborne full-waveform laser scanning data, the development of voxel based methods for data analysis is essential. In contrast to existing approaches based on the extraction of discrete 3D points by a Gaussian decomposition, it is very promising to derive the voxel attributes from the digitised waveform directly. For this purpose, the waveform data have to be transferred into a 3D voxel representation. This requires a series of radiometric and geometric transformations of the raw full-waveform laser scanner data. Thus, the paper deals with the geometric aspects and describes a processing chain from the raw waveform data to an attenuationcorrected volumetric forest stand reconstruction. The integration of attenuation-corrected waveform data into the voxel space is realised with an efficient parametric voxel traversal method operating on an octree data structure. The voxel attributes are derived from the amplitudes of the attenuation-corrected waveforms. Additionally, a new 3D filtering approach is presented to eliminate non-object voxel. Applying these methods to real full-waveform laser scanning data, a voxel based representation of a spruce was generated combining three flight strips from different viewing directions.

VOXEL BASED REPRESENTATION OF FULL-WAVEFORM AIRBORNE LASER SCANNER DATA FOR FORESTRY APPLICATIONS

The advantages of using airborne full-waveform laser scanner data in forest applications, e.g. for the description of the vertical vegetation structure or accurate biomass estimation, have been emphasized in many publications. To exploit the full potential offered by airborne full-waveform laser scanning data, the development of voxel based methods for data analysis is essential. In contrast to existing approaches based on the extraction of discrete 3D points by a Gaussian decomposition, it is very promising to derive the voxel attributes from the digitised waveform directly. For this purpose, the waveform data have to be transferred into a 3D voxel representation. This requires a series of radiometric and geometric transformations of the raw full-waveform laser scanner data. Thus, the paper deals with the geometric aspects and describes a processing chain from the raw waveform data to an attenuationcorrected volumetric forest stand reconstruction. <br><br> The integration of attenuation-corrected waveform data into the voxel space is realised with an efficient parametric voxel traversal method operating on an octree data structure. The voxel attributes are derived from the amplitudes of the attenuation-corrected waveforms. Additionally, a new 3D filtering approach is presented to eliminate non-object voxel. Applying these methods to real full-waveform laser scanning data, a voxel based representation of a spruce was generated combining three flight strips from different viewing directions.

FOREST AREA DERIVATION FROM SENTINEL-1 DATA

The recently launched Sentinel-1A provides the high resolution Synthetic Aperture Radar (SAR) data with very high temporal coverage over large parts of European continent. Short revisit time and dual polarization availability supports its usability for forestry applications. The following study presents an analysis of the potential of the multi-temporal dual-polarization Sentinel-1A data for the forest area derivation using the standard methods based on Otsu thresholding and K-means clustering. Sentinel-1 data collected in winter season 2014-2015 over a test area in eastern Austria were used to derive forest area mask with spatial resolution of 10m and minimum mapping unit of 500 m<sup>2</sup>. The validation with reference forest mask derived from airborne full-waveform laser scanning data revealed overall accuracy of 92 % and kappa statistics of 0.81. Even better results can be achieved when using external mask for urban areas, which might be misclassified as forests when using the introduced approach based on SAR data only. The Sentinel-1 data and the described methods are well suited for forest change detection between consecutive years.

FOREST AREA DERIVATION FROM SENTINEL-1 DATA

Abstract. The recently launched Sentinel-1A provides the high resolution Synthetic Aperture Radar (SAR) data with very high temporal coverage over large parts of European continent. Short revisit time and dual polarization availability supports its usability for forestry applications. The following study presents an analysis of the potential of the multi-temporal dual-polarization Sentinel-1A data for the forest area derivation using the standard methods based on Otsu thresholding and K-means clustering. Sentinel-1 data collected in winter season 2014-2015 over a test area in eastern Austria were used to derive forest area mask with spatial resolution of 10m and minimum mapping unit of 500 m2. The validation with reference forest mask derived from airborne full-waveform laser scanning data revealed overall accuracy of 92 % and kappa statistics of 0.81. Even better results can be achieved when using external mask for urban areas, which might be misclassified as forests when using the introduced approach based on SAR data only. The Sentinel-1 data and the described methods are well suited for forest change detection between consecutive years.

Radiometric Correction of Terrestrial LiDAR Data for Mapping of Harvest Residues Density

Abstract. In precision agriculture detailed geoinformation on plant and soil properties plays an important role. Laser scanning already has been used to describe in-field variations of plant growth in 3D and over time and can serve as valuable complementary topographic data set for remote sensing, such as deriving soil properties from hyperspectral sensors. In this study full-waveform laser scanning data acquired with a Riegl VZ-400 instrument is used to classify 3D point clouds into post-harvest straw residues and bare soil. A workflow for point cloud based classification is presented using radiometric and geometric point features. A radiometric correction is performed by using a range-correction function f(r), which is derived from lab experiments with a reference target of known reflectance. Thereafter, the corrected signal amplitude and local height features are explored with respect to the target classes. The following procedure includes feature calculation, decision tree analysis, point cloud classification and finally result validation using detailed classified reference RGB images. The classification tree separates the classes of harvest residues and bare soil with an accuracy of 96% by using geometric and radiometric features. The LiDAR-derived harvest residue coverage value of 75% lies in accordance with the image-based reference (coverage of 68%). The results indicate the high potential of radiometric features for natural surface classification, particularly in combination with geometric features.

Automated reconstruction of tree and canopy structure for modeling the internal canopy radiation regime

Understanding canopy radiation regimes is critical to successfully modeling vegetation growth and function. For instance, the vertical distribution of photosynthetically active radiation (PAR) affects vegetation growth, informative upon carbon and energy cycling. Availing upon advances in information capture and computing power, geometrically explicit modeling of forest structure becomes increasingly possible. A primary challenge however is acquiring the forest mensuration data required to parameterize these models and the related automation of modeling forest structure. In this research, to address these issues we employ a novel and automated approach that capitalizes upon the rich information afforded by ground-based laser scanning technology. The method is implemented in two steps: in the first step, geometric explicit models of canopy structure are created from the ground-based laser scanning data. These geometric explicit models are used to simulate the vertical range to first hit. In the second step, we derive canopy gap probability from full waveform laser scanning data which have been used in a number of studies for characterization of radiation transmission (Jupp et al., 2009; Yang et al., 2010) and do not require any geometric explicit modeling. The radiative consistency of the geometric explicit models from step 1 is validated against the gap probabilities of step 2. The results show a strong relationship between the radiative transmission properties of the geometric models and canopy gap probabilities at plot level (R=0.91 to 0.97), while the geometric models suggest the additional benefit to serve as a bridge in scaling between shoot level and canopy level radiation.

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 (1550nm). 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.

DETECTION OF WATER SURFACES IN FULL-WAVEFORM LASER SCANNING DATA

Abstract. Airborne laser scanning has become a standard method for recording topographic data. A new generation of laser scanners digitises the complete waveform of the backscattered signal and thus offers the possibility of analysing the signal shape. As a product of the laser scanning, a digital surface model (DSM) or a digital terrain model (DTM) can be derived. In water regions, data acquisition by laser scanning is limited to the water surface because the near-infrared laser pulses hardly penetrate water. Therefore, a height model generated from laser scanner point clouds over water regions does not represent the actual terrain. The generation of a DTM thus requires the detection of water surfaces. In this study, a method for the detection and classification of water surfaces in airborne laser scanning data is proposed. The method works with both geometrical features (e.g. height or height variation) and characteristics of the pulses derived from the full waveform of the returned signal (e.g. intensity or pulse width). In our strategy, based on fuzzy logic, all classification parameters are derived automatically from training areas. According to their statistical distributions, the features are considered with individual weights. The aim of this paper is to analyse crucial features for classification and to investigate the potential of full waveform laser scanning data for this application. We present results from different areas with lakes and rivers, analysing the contribution of the individual groups of features for the detection of water surfaces.

A Thorough Accuracy Estimation of DTM Produced From Airborne Full-Waveform Laser Scanning Data of Unmanaged Eucalypt Plantations

A digital terrain model (DTM) is needed to extract forest variables. Its accuracy affects that of these variables, and thus it has to be known. Airborne laser scanning (ALS) is being increasingly used to produce DTM. The potential of full-waveform ALS (FWALS), however, for DTM production in forest areas has not been thoroughly evaluated, particularly under adverse conditions of heterogeneous forest stands, such as the typically extensively managed eucalypt plantations of north-central Portugal studied here. In the study area of 900 ha, an exhaustive assessment of FWALS-derived DTM accuracy was carried out based on the field survey of 43 plots. In each of these circular plots of roughly 400 m <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> , the coordinates were measured for all trees and all prominent terrain points using Global Navigation Satellite Systems receivers and total station. In addition, the quality of the FWALS data was assessed by surveying 1 m × 1 m grids of points on three bare areas. The accuracy assessment measures are computed by assuming normal and nonnormal distributions of the differences between the h values of FWALS terrain points and those of control points at the same planimetric locations. While the accuracy of the FWALS data is as expected, around 3 cm, that of the DTM of the forested area exceeds expectations. In fact, the DTM accuracy of 12 cm is higher than that obtained in prior ALS studies carried out in less adverse conditions than those studied here. Thus, the current FWALS-based methodology to derive DTM appears highly suitable to extracting forest variables.

Formula omitted]-spline deconvolution for differential target cross-section determination in full-waveform laser scanning data

In full-waveform laser scanning, short laser pulses are emitted and travel towards Earth and object surfaces. The sensor samples the waveform of the emitted pulse and its complete backscattered echo as a function of time. This technique allows for the three-dimensional reconstruction of the terrain, natural and man-made objects, and for the derivation of (geo-)physical quantities such as the differential target cross-section. The retrieval of the differential target cross-section requires deconvolution which is an ill-posed problem. In this study, we present a novel technique for the computation of the differential target cross-section using B -splines. This class of mathematical functions enables a well-posed linear approach for deconvolution. Furthermore, it is not dependent on the symmetry of the temporal profiles of the emitted laser waveform and the received echoes, as approaches previously suggested. In this paper, the algorithm for deconvolution is presented in detail and validated for both synthetic and real-world full-waveform laser scanner data.

Rigorous pulse detection from full-waveform airborne laser scanning data

Current discrete airborne laser scanning (ALS) systems suffer from a limited number of multiple returns and the loss of information about range estimation. To improve range resolution and accuracy, latest generation small-footprint, full-waveform laser scanning data were investigated and an approach developed to determine the number of targets within complex waveforms. The approach, termed rigorous Gaussian detection (RGD), is based on the popular Gaussian decomposition method but is implemented with rigorous estimates of initial values and a sophisticated iteration procedure. Second derivatives of waveforms and the number of samples on the edges of visible peaks were analysed to find overlapping targets. Weak pulses were detected where the residuals of Gaussian fitting were high and corresponding pulse widths were close to that of the transmitted pulse. The performance was compared with two other pulse detection techniques from commercial software, namely the centre of gravity (COG) method and the Gaussian pulse fitting (GPF) method with the Levenberg–Marquardt iterative algorithm. The results proved that the developed pulse detection method resolved multi-targets well. For example, in a sample of 201 dual-target waveforms, the RGD method correctly detected two targets in all the samples, considerably outperforming the COG and GPF methods, which achieved 57% and 67% success rates, respectively. When measuring the target heights using the different methods, the RGD method, with a root mean square error (RMSE) of 0.06 m, proved twice as accurate as the next best method, COG. The developed approach therefore compares favourably against the two commercial pulse detection methods considered, where their failure to detect hidden targets within complex waveforms and weak pulses resulted in a decrease of multi-target resolution, unreliable three-dimensional (3D) points, and also the loss of information from weak pulses.