| Summary: | In the last two decades, laser scanning systems made the transition from scientific research to the commercial market. Laser scanning has a large variety of applications such as digital elevation models, forest inventory and man-made object reconstruction, and became the most required input data for flood plain and hydraulic models. This system is generally called as a discrete laser scanning system. A discrete laser scanning system sends a pulse to the ground surface and records the return signal resulting from the illumination of the surface. The area of the illuminated surface is defined as the footprint size of the laser shot. The two-way travel time of the laser pulse allows to determine the distance of the laser system to the surface. Traditional systems are unable to record the complete return signal, but typically store only one to four distances to objects in the laser footprint. A new system developed to overcome the above limitations is the so-called full waveform scanning system. The system sends out a pulse of a certain width and amplitude. After reflection of the pulse on the objects surface, the system records the complete returning pulse signal. This complete signal is the so-called full waveform. Compared to traditional scanning systems, a full waveform system retrieves more information that should still be extracted though from the waveform shape. The shape of the full waveform contains information on the characteristics of the illuminated footprint, like object information (tree and building height), forest structure and ground surface characteristics (e.g. forest species surface roughness and slope) as well as a land cover type (water, bare earth, or urban areas). Harding (1998), and Blair et al. (1999) showed that using observations from a full waveform laser system it is possible to achieve accurate forest structure and biomass estimates. However, the system considered in this case was just operated from an airplane flying at a low altitude of a few kilometers above the ground surface with a medium footprint size of about 10-20 m. Moreover, data acquisition could only be performed in a small area. Some typical systems used in these days were the Scanning Lidar Imager of Canopies by Echo Recovery system (SLICER), the Laser Vegetation Imaging Sensor (LVIS), and the commercial airborne full waveform scanning system from RIEGL the LMS-Q560 (2003). In 2003, moreover, NASA launched the first satellite full waveform system, the so-called Ice Cloud and land Elevation Satellite system (ICESat), carrying the Geoscience Laser Altimetry System (GLAS) instrument. The purpose of ICESat was to collect, among others, measurements concerning the Antarctica and Greenland ice sheets and their mass balance, concerning land vegetation and concerning the atmosphere. This space borne system acquired data between 2003 and 2009 over the entire earth from 600 km altitude, with a footprint size of about 70 m and a distance between consecutive footprints of approximately 175 m. However, due to for example the high operational altitude, this system was also affected by many error sources, like instrumental and operational problems, atmospheric effects and surface conditions. Processing such data is a challenging task; and development of a validation method together with the development of new applications of large footprint full waveform data are the main targets of my thesis work. For this purpose, the contents of this thesis is organized in six chapters In chapter 1, the background and scope of the research is introduced. In chapter 2, full waveform sensor and instrument development is presented. The main content starts in chapter 3. This chapter focuses on processing and parameterization of large footprint ICESat full waveform data. Data handling is described in detail. It is presented how a pulse interacts to objects on the earth surface and what information a return pulse contains. Moreover, it is shown how waveform parameters can be obtained by fitting a number of Gaussian numbers to a waveform using least squares. waveform in a sense of least square estimation. Two alternative methods to obtain information on waveforms, i.e. waveform deconvolution and waveform simulation were also implemented and are discussed. At the end of this chapter, an overview is given of waveform parameters and their possible physical interpretation and application. This chapter aims to provide a tutorial to wider audiences/readers who would like to use satellite laser full waveform data for their own purpose. In order to evaluate the accuracy and the precision of elevation and height estimates that can be obtained using the ICESat full waveform system, the topic of validation is studied in chapter 4. In this chapter, first background and related work are discussed. Two validating cases discussed in detail : (i) Comparison between the bare land elevations derived from ICESat full waveform data and airborne laser scanning data over The Netherlands; and (ii) Identification and investigation of error sources of the ICESat full waveform system by comparing waveform pairs that have overlapping footprints. For the first validation case, filtering constraints have been investigated and further developed to avoid influence of data anomalies. Examples of factors having a negative influence on a comparison include on one hand waveforms affected by saturation or cloudy conditions and on the other hand footprints that are not covered well by the available airborne data points The comparison between ICESat and accurate airborne laser data over The Netherlands confirms that the ICESat full waveform accuracy and precision strongly depends on land cover type. With respect to different land cover type, the accuracy was about -21 cm with a standard deviation of about 20 cm over bare earth. The accuracy was -24 cm with a standard deviation of 28 cm over urban areas and -9 cm with a largest standard deviation of about 45 cm over forest. As expected, the accuracy decreases when the complexity of the surface increases (e.g. from bare land to urban)). The accuracy of ICESat derived elevations over water could not be assessed because of few effects: changes in water level, lack of airborne data points over water surfaces, etc., It is concluded that if a) proper filtering is applied, and b) the terrain relief is small, the last mode of an ICESat waveform represents the terrain/bare land height with both an accuracy and precision at the decimeter level. Also feature height estimation of features like trees or buildings were studied. The difference between ICESat and airborne derived feature heights are acceptable over forested and bare land areas. However, the result over buildings are not satisfactory. The main reason is that ICESat derived feature height estimations are sensitive to feature height variations occurring at spatial distances smaller than the size of the ICESat footprints. For features that are homogeneous at the scale of ICESat footprints, ICESat waveform analysis is a suitable method for estimating feature heights. For urban environments, incorporation of an additionally accurate Digital Elevation Model might still enable the monitoring of feature height changes. For the second validation case a database was constructed, consisting of more than one hundred thousand (>100000) repeated, partly overlapping ICESat waveforms over Europe. The aim is to identify the cause of changes in waveform parameters obtained from (partly) overlaying waveforms Even in such a size of database it turned out almost impossible to identify suited waveform pairs. The first problem was the lack of almost completely overlaying footprints, and second, the lack of waveform pairs situated at stable areas. Unfortunately, the number of almost completely overlapping footprints from repeated ICESat measurements that could be obtained from campaign L1 to L3e in the region between 36N and 71N latitude and 11W and 33E longitude was very limited. Performing a full worldwide search over all campaigns is however expected to result in a quite large data set of suitable almost perfectly overlapping repeated footprints. Such data set could therefore be used to identify and resolve remaining ICESat processing issues. After applying all corrections identified in the calibration/validation procedure, the ICESat product is ready to be used in different applications considering the surface of the Earth. How to develop applications by using large footprint full waveform data is discussed in chapter 5. By considering the full waveform parameters derived from the waveform processing procedures of chapter 3, two applications of ICESat full waveform data were studied in detail. A new contribution described in this chapter compares repeated ICESat waveform pairs to assess forest change. By analyzing corresponding waveforms acquired in different campaigns at nearly coinciding footprints, canopy changes caused by seasonal influences were detected. A general new aspect introduced in this chapter is to analyze waveform parameters in terms of a pair of full waveforms that were acquired at approximately the same location. Comparison of the waveform parameters of such pairs not only allows the detection of seasonal influences on forest type, it may also become possible to further detect and estimate important forest change parameters, like forest growth and deforestation. As a first step seasonal changes over broadleaf, mixed-wood, and needleleaf forests between winter and summer epochs of 2003, along near-coincident ground tracks were quantified. It was found that although the maximum tree height barely changes over 6 months, i.e., less than 2.2% for the three forest types, a suited waveform parameter can detect forest canopy change mostly for broadleaf (a 148% change, winter to summer) and less for conifers (a 36% change). Alternative waveform parameters to describe forest changes are also discussed. An application of the seasonal change in waveform parameters is to use these parameters to classify footprint areas directly into forest type classes. Preliminary results, with a kappa value of 0.57, provide a baseline against which improvements in both data and methodology can be gauged in future. Future work should include a method for correcting slope-induced changes in the waveform results, both for individual waveforms and for waveform pairs. In addition, further improvement is expected by including a neighbourhood analysis, that is by incorporating spatial correlation between close by waveforms and their changes. An individual quality descriptor of the tree parameter values could be obtained by quantifying and propagating errors encountered during the waveform processing. Moreover, tree parameter values should be validated against either data from field measurements or data from other sensors. Finally, the parameter definitions should be refined and validated in order to improve agreement with biophysical characteristics of the forest. In the second application, it is demonstrated how ICESat full waveforms can be used for land cover classification. It was the first time that the possibility of using ICESat data for this purpose was investigated. Over The Netherlands, ICESat footprint locations were classified into four classes: high vegetation (high trees or forest), urban, water, and bare land/low vegetation. The following waveform parameters were used as class attributes: return energy, waveform extent, waveform start and number of Gaussian components. It is concluded that the accuracy of classification equals 73% in comparison to a confusion matrix based on the CORINE land cover database 2000 (CLC2000) covering the same study area. In addition, it is shown that ICESat waveforms could be used to build a feature vector consisting of suited waveform parameters that can be consecutively applied to classify the land cover class of the footprint location. The full waveform parameters derived from chapter 3 together with total return energy able to discriminate between high vegetation, urban areas, bare land/low vegetation and water. This result showed the feasibility of land cover classification with spaceborne lasers. As the ICESat satellite has a near polar orbit, coverage is global, but still a main disadvantage of ICESat data is that only tracks were mapped. With the development of new systems, an area-wise coverage may become possible in future. In that case, automatic classification of large footprint full coverage full waveform laser data may lead to land cover classification results of the same high quality as can be obtained from optical data. At the end of the thesis, significant information is summarized and remaining steps but also potential new applications are described. DEOS Aerospace Engineering
|
|---|