Since aerosol are one of the most important factors in the atmospheric correction of remotely sensed data, many methods have been developed in retrieving the aerosol optical depth (AOD). The dark dense vegetation (DDV) and contrast reduction method are two concepts which are mostly used for the estimation of AOD. The main aim of this study is to improve the structure function method, which lies within the contrast reduction approach, in retrieving the AOD over Taiwan. The results compared with in-site observations show significant improvements in the accuracy of the retrieved AOD. On the other hand, the optical depth and size distribution of aerosol are key parameters in the monitoring of the air quality. The two can be constructed in this study as the results indicate a high practicality in applying satellite data for the monitoring of the atmospheric environment.

Building geometric models are elementary but indispensable information in many applications. After years of researches, it is generally recognized that human intervention in the extraction procedure for build­ing geometric models from aerial images is necessary. Therefore, this paper will present a practical strategy that effectively integrates the capability of human interpretation with the computed powers of computers to extract the building models from urban stereo image pairs. In this proposed strategy, we suggest the rel­evant geometric features are extracted and the possible roof boundaries are detected by automatic approaches. Subsequently, the geometrical models of roof patches are reconstructed by interactive operations. The operator only plays the assistant role but very important intelligent interpretation task in this proposed strategy. Experiments show our proposed practical strategy is feasible and applicable. The accuracy of reconstructed roof patches is good enough for some practical applications.

This paper presents a simulated error analysis of the GPS (Global Positioning System) occultation technique, which is used by the joint Taiwan-U.S. COSMIC (Constellation Observing System for Meteorology, Ionosphere, and Climate) project to study the Earth's atmosphere and monitor temperature and pressure profiles. A qualitative description of the GPS occultation technique of atmosphere limb sounding is reviewed, and retrieval algorithms for temperature and pressure profiles determination in the absence of water vapor are summarized in this paper. A simulation study was conducted to investigate the impact of various error sources, including integral truncation error, GPS and COSMIC orbit errors, and measurement noises, on the retrieved temperature and pressure profiles. Simulation results show that the induced effects of the above error sources on atmospheric pressure retrieval were less than O.2mb when precise COSMIC orbit ephemeris was used. The corresponding impacts on the retrieved temperature profiles were small (less than 1K) in the troposphere; however, they were substantially greater in the stratosphere (~ 10K). To date, there still remains an outstanding problem in GPS occultation data analysis - the retrieved temperature often displays wave-like fluctuations in the upper atmosphere (> 45km). According to the simulation results contained in this paper, these wave-like motions do not necessarily represent the actual atmospheric temperature behavior; they could also reflect the induced error pattern of the GPS measurement noises.

The elements of concrete structures suffering from chemical reactions trend to be an expansion phenomenon, which lead to cracks. The concrete structures of inferior quality caused by cracks have much influence on the structures durability and strength. The non-contact measurement is employed because of the physical limitations in manual measurements. This paper deals with semi-automatic crack feature extraction from digital close-range images to infer the relationship between concrete expansion and crack width. Only few manual seed points are required on crotches of crack features. Then, the shape and position may be illustrated with automatic feature extraction in two- dimensional image space. In the process of crack searching, a Gaussian filter is applied in advance to smooth the noise along crack profiles to improve the reliability and precision. Subsequently, the edges of cracks are determined by computing the gradient of the crack profiles. The edges of cracks detected by the proposed algorithm will be compared with the one digitized by manual operations.