In the process of producing digital elevation model with airborne LiDAR, different procedures are required for products of different specified level. This thesis was written from April to June, 2007. Reference to a draft for specifications and standard operation procedures for airborne LiDAR survey, MOI, DEM products were classified to three levels. When producing products higher than level two, it requires manual editing to assure the correctness. In this study, several issues of manual editing are investigated. Ten map-sheets covering five major land-cover types are selected for the comparison. The area of each map-sheet is about 2.5 km by 2.5 km. Preliminary results show that for the same map-sheets, manual editing by different operators may produce different results and the time required is significantly different. In general, the height difference of the DEMs from two individuals is 8 cm and it gives a standard deviation of 25 cm.

The main purpose of this study was to assess the effect of ecosystem classification systems at various scales on environmental parameters using remote sensing techniques. The processes included applying hybrid classification to generate a land-use map of the north Taiwan using Landsat-5 TM image in 1995; using the DTM and the SEBAL model to calculate 16 environmental parameters and compare the differences among different land-use types; and assessing the effects of 2 ecosystem classification systems (i.e., geographic climate method and watershed division method) at various scales on environmental parameters using stepwise discriminant analysis. The results indicated that the study area was classified into 7 land-use types. They were forest-land, building, farm-land, baring farm-land, water body, cloud, and shadow. The Comparison of 16 environmental parameters among 5 land-use types (excluding cloud and shadow) showed that forestland had higher value with cosine of solar incidence angle, twenty-four hour extraterrestrial radiation, net radiation, normalized difference vegetation index, emissivity, estimating friction velocity, surface roughness for momentum transport, sensible heat flux, soil heat flux, evapotranspiration, and had lower value with transmittance, air density, surface albedo, surface albedo at the top of atmosphere, aerodynamic resistance to heat transport, surface temperature. As for assessing the effect of 2 ecosystem classification systems at various scales on environmental parameters, the result pointed out that ecosystem classification systems at various scales indeed caused the variation of environmental parameters according to the selected parameters and the number of parameters for discriminating 5 land-use types. However, among environmental parameters, normalized difference vegetation index and emissivity were the most important factors regardless of ecosystem classification systems at various scales.

Estimation of plant growth over a large paddy field provides the needed information for site-specific management of crop including rice. With the estimated leaf area index (LAI) as input, growth status of rice crop may be evaluated and yield production at harvest may be assessed through functioning algorithms. This study measured the near-ground hyperspectral reflectance of rice canopy periodically on the dates of plant samplings during crop development, and then established the relationship between LAImeasured (i.e. the measured LAI) and NDVINB (i.e. normalized difference vegetation index calculated from narrowbands of hyperspectral reflectance) from the collected data. Rice plants of different planting densities, in the range of 0.28-2.78  105 hills ha-1, were grown in the field to produce varied values (m2 m-2) of LAI along plant growth for such purpose in the first and the second cropping seasons of 2006. A total of thirty-six multi-spectral images of the study area taken by Formosat-2 satellite on days of ground samplings were also acquired to calculate the broadband values of NDVI (NDVIBB) and input to the LAImeasured─NDVINB relationship to obtain the estimated LAI (LAIBB). Results indicate that the high-temporal and high-spatial-resolution images of Formosat-2 satellite are good source for monitoring plant growth of rice crop by providing reasonable estimated values of LAI. Such a capability of Formosat-2 spectral images enables their applicability in areas of precision farming.

High spatial resolution images applied in this study include digital orthophotos of 0.5m grid and pan-sharpened SPOT images of 2.5m grid. A land-cove/land-use classification scheme upto level 3 is adopted, with randomly-sampled field checks for accuracy verification. Manual interpretation with the assistance of on-screen tool-kit is adopted for the discrimination of various land-cover/land-use types on orthophotos. Whereas automatic classifications both supervised and unsupervised approaches are applied for satellite images. Preliminary results show that aerial photographs give an accuracy of 97~98% for agriculture, forestry, hydraulic and telecommunication land units. Satisfied results of level 3 of building up areas can never been achieved solely by photo-interpretation. For satellite classification, an accuracy of 60~75% can be achieved for level 1 and level 2 for the forestry and hydraulic land units. If the classes of level 3 are to be achieved, more ancillary information from GIS data-base should be incorporated. Level 3 can not be attained using satellite images. Multi-temporal images with complementary GIS polygons of field parcels can give rice parcels an overall accuracy of 96.47%. Whereas, the average accuracy (85.76%) is higher than the produce’s accuracy (73.38%) and user’s accuracy (74.23%). It is concluded that temporal plant histogram information is helpful and critical for the classification of agriculture lands.

For the land cover classification with remote sensing images, the spectral differences among classes provide the vital information. Besides the spectral information, the spatial relation of pixels also carries useful information for classification. Along with the increase of spatial resolution for satellite images, the spatial information is expected to be more abundant than before. Object based classification utilizes the spatial segmentation procedure prior to the classification. This study investigates the performance of ECHO, Definiens, and compared with the maximum likelihood, back-propagation, and support vector machines, classification schemes. Three SPOT-5 images with different complexity are selected for the experiments. The spatial resolution of these two images is 2.5 m, which were produced through the fusion process. From the experiments, it is shown that object based classification schemes provide stable result, while the SVMs showed dependency of data set.

Oil spill on the sea surface produced by human activities is disastrous to the ecological environment. How to detect, monitor and track the oil spill are always very important tasks. Due to oil spill often occur in open sea, remotely sensed image provides an effective technology to monitor the sea area. But the oil spill only occur in a very small are in the image, therefore, to detect the oil spill on the sea surface is a challenge problem. This study focuses on the detection of oil slick on sea surface using both spectral and spatial information of the remotely sensed images. Since the oil spill area usually only occupies a few pixels in the image scene, it can be considered as anomaly. By applying anomaly detection algorithm with only spectral information, the oil spill area can be extracted along with other marine phenomena as interference. In order to eliminate the interferences, spatial features are introduced. The experimental result shows the proposed method combining the spatial features of oil spill with spectral information improves the oil spill detection performance.