We organized and performed airborne LIDAR mapping in the Metropolitan Taipei area in order to produce high-resolution and high-precision digital elevation models for geological research. Applying the newly acquired LIDAR DEMs from the MOEA Central Geological Survey, we analyzed geologic and geomorphic features of the Keelung River drainage area. The studied features include topographic scarps and lineaments, river terraces, drainage basins, and landslide scarps. Based on our analysis of the acquired LIDAR DEMs, we arrived at the following findings and conclusions: 1) The river basin analysis based on the LIDAR DTM indicates that the development of the Keelung River was influenced by regional structural patterns and topographic slopes. Tectonic influences appear to be relatively small for the development of the Keelung River. 2) Along the Keelung River, the LIDAR DTM indicates anomalous terrace morphology and river drainage at the middle section of the river. We interpreted this area to be caused by paleo-landslides, which form a dammed lake in the river and subsequently developed the main terrace of the same level for large area.

Full-waveform LiDAR is an emerging active data acquisition tool in many applications. In addition to accurate positions of point clouds, full-waveform LiDAR provides complete wave signals of returned laser ray, which can be subsequently used to derive important characteristics of the targets. The waveform parameter extraction and analysis are two important operations for full-waveform LiDAR applications. In this research, the Gaussian modeling function with second derivative method was utilized for waveform fitting. Features extracted from the waveform, including width, amplitude, and backscatter cross-section, in conjunction with traditional LiDAR features, such as normalized height and intensity, and greenness index from image were used as primary attributes for land cover classification. Two classifiers were used and compared in this study, including Naïve Bayes and Random Forests. Experimental results demonstrate that using the second derivative method can produce higher fitting success rate and better classification results. The land cover classification results indicate that full-waveform features are helpful for distinguishing different vegetation targets. In addition, the decision-tree-based Random Forests classifier is more suitable for land cover classification of LiDAR data used in this study.

Terrestrial LiDAR can be used to acquire high accuracy 3D spatial data in rapid way. However, point clouds delivered by a terrestrial LiDAR are referenced to a local coordinate system defined by the laser scanner. With the limitation of viewing coverage of a single scanning station, combining point clouds acquired from multiple scanning stations is usually needed to completely cover the target landscape. For most applications, raw point clouds should be transformed into a global coordinate system (geo-referencing). The geo-referencing of terrestrial LiDAR data is currently relied on known control points. But using control points for geo-referencing is often troublesome and laborious. In this research, we perform network adjustment with GPS observations, and we focus on the problem that whether the role of control points in network adjustment can be substituted by GPS observation of scanning station or not. There are two cases in our experiments. The accuracy of the results are under 10 centimeters in E, N and h directions. By the validation of our experiments, we can know that it can perform network adjustment with GPS observations without control points. The accuracy are good enough for application.

Aerial images are faithful records of landscape at that time. Historical aerial images directly witness landscape phenomena at that time of the covered region, including land-use, natural source, roads and residential area, etc. The study of historical aerial images has proven an effective way to reveal some historical phenomena related to cultural, economic, and natural changes, which are linked to the spatial information in the past. Abundant historical aerial images have been acquired in Taiwan for a variety of missions. The earliest aerial images were taken by U.S. military during and after World War II (from 1943 to 1952). Later than that many images were acquired routinely by Taiwan Air Force (1954 to 1976), Aerial Survey Office (since 1976). For the purpose of retrieving spatial information from the historical images, image registration and rectification corresponding to the ground coordinates should be done in advance. In this study, we take Tai-jiang Inner Sea as an example to register and rectify temporal aerial images into nowadays national coordinate system (TWD97). Images acquired by Aerial Survey Office are standard aerial images which can be registered through a computation process of bundle adjustment. However, the main purposes of images taken by U.S. military and Taiwan Air Force were mainly applied for military reconnaissance. These images are lack of camera parameters and the overlapped of them are not standard. So they do not fit the process of bundle adjustment. In this study, we attempt to obtain the information of registration and rectification by solving affine transformation parameters from all overlapped images at the same period. Although this method are not as strict as aerial triangulation, it can work without camera parameters. The experimental results showed that the accuracy of images taken by U.S. military can up to 6.5 m by applying affine transformation. The accuracy was poor than images taken by Aerial Survey Office using aerial triangulation (2m). But we believe this can meet the analysis of temporal geographic information and promote the application of historical aerial images.

Paddy-rice is an important food source in Taiwan, and government devotes a lot of resource to estimate annual food production. In the past, the paddy-rice field area estimation is obtained by field survey, which is both time and man-power consuming and requires a much higher budget. Land cover classification through remote sensing imagery and data mining technique is a well-accepted and popular approach for crop production estimation, thanks to the advances of remote sensing technology in recent decades. It is thus an important research topic that how to develop a methodology or algorithm which can improve image classification accuracy rate. In the present study an aerial photo over the Tanzi county of Taichung city was analyzed. Two different optimization algorithms, Artificial Bee Colony and Fuzzy C-Means Particle Swarm Optimization, were incorporated to the clustering analysis. The result indicates Artificial Bee Colony based clustering approach renders better classification accuracy rate than that based on Fuzzy C-Means Particle Swarm Optimization, which are 89.5% and 80.75% respectively. This study shows bionics computation is a feasible approach and alternative for remote sensing image classification.