Biodiversity in Taiwan is interfered by an invasive plant: Leucaena leucocephala (L. leucocephala), which has been considered one of the most important contemporary ecological problems. This study identified spatial extents of L. leucocephala in the Hengchun Peninsula in 1988 and 2007 using multi-temporal data from SPOT images with maximum likelihood classifier. To understand the spatial dynamics of L. leucocephala through time, we investigated the relationships between the species expansion and topography and land use. For classified maps of L. leucocephala and each image date, overall accuracies above 85% were obtained. Results of the regional analysis showed that the area was an increase in L. leucocephala cover, and a preferential invasion was found at lower elevations, gradual slopes, proximity to the roads and farmlands. Further, the results of statistic analysis indicated that the correlations of the elevation, slope, road and farmland with L. leucocephala invasion were considerably. Therefore, enhancement was required for the L. leucocephala management on human disturbance zones.

Conventional approach of landslide mapping is either solely on basis of spectral or spatial characteristics of the 2D images. Landslides are usually confusing with houses, roads, and other bare lands because these ground features have similar spectral patterns. In this study, 3D airborne LiDAR data are integrated with SPOT images for landslide classification for improving classification accuracy. A study area is selected in a sub-basin of Shimen Reservoir. SPOT images and LiDAR data are taken in the same month after the typhoon of Longwang in November of 2005. The LiDAR-derived data include DEM slope gradient, DSM slope gradient, and roughness data such as Fractal dimension, diversity, dominance and relative richness. These derivatives are then combined with spectral bands for classification algorithms including maximum likelihood and Mahalanobis. Comparisons are made for various approaches such as with or without LiDAR derivatives and various classification algorithms. It is concluded that with the inclusion of LiDAR derivatives and proper arrangement of classification procedures, an improvement of more than 11% of user’s accuracy and 27% of producer’s accuracy in diversity by maximum Likelihood classification algorithms can be achieved. Secondary, the diversity provides 11% of user’s accuracy and 21% producer’s accuracy by maximum likelihood classification algorithms.

WorldView-1 is the first commercial satellite that provides 50cm spatial resolution. It contains both the on-board data and sensor-oriented rational polynomial coefficients (RPC). The objective of this investigation is to build up the geometric models and perform the geometric analysis for WorldView-1. The geometric correction models include rigorous sensor model (RSM) and rational function model (RFM). In the RSM, we use on-board data to build the relationship between object and image space. The correction models comprise the compensation of orientation parameters and correction of image coordinates. The RFM uses sensor-oriented RPC to establish the transformation of object and image space. Then, a small number of ground control points are applied to refine the model. Experimental results indicate that the results of RSM and RFM are similar. Both of the methods may reach sub-pixel accuracy when 9 control points are applied.

Considerable attention has been given for monitoring and simulating the impacts of global environment change on forest watershed hydrology. Therefore, the purpose of this study is to integrate the remote sensing techniques, the general circulation models(GCMs), and the Markov model to investigate the effects of land use change and climate change on the Dan-Shui watershed, moreover, to compare the difference of stream flow simulations under various GCMs scenarios(i.e., CGCM1 and HADCM3).The processes were as follows: Land-use maps of the Dan-Shui watershed were firstly generated using hybrid classification and Landsat-5 images in 1995 and 2002. The study area was classified into 5 land-use types. They are forest-land, building, farm-land, baring farm-land, and water body. Secondly, the Markov model built based on those land-use data was applied to predict future land-use changes, furthermore, the generated results were used to estimate the cover coefficients (CV) of evapotranspiration based on future conditions and the GWLF water balance model. Results from the Markov prediction indicated that the area of building increased from 17.71% in 1995 and 20.17% in 2002 to 40.24% in 2030, 48.91% in 2058, and 58.16% in 2086. In addition, the estimations of the CV showed that the value decreased gradually from 0.726 and 0.704 to 0.558, 0.478 and 0.408. Finally, the GWLF model was integrated with the Markov predictions to analyze the influence of future land-use changes and climate changes on stream flow simulation under the CGCM1 and HADCM3 scenarios, and also to assess the impacts on the watershed. The results indicated that, due to the increase of building area and the decrease of the CV value, monthly stream flow, monthly average stream flow, and annual total stream flow all changed under two climate change scenarios, regardless of the short-, middle- or long-term simulation,. Furthermore, land-use changes and climate changes would lead to increase the occurrences of floods.

Efficiently obtaining information of forest regions and tree canopy volume is important for forestry management. Airborne Light Detection and Ranging (LiDAR) data is able to provide high resolution three dimensional coordinates of surface features, but does not contain spectrum information. In contrast, remote sensing imagery offers copious spectrum information that can be used to locate forest regions. Therefore, integrating the complementary LiDAR data and remote sensing imagery is an effective strategy for the estimation of forest area and canopy volume. Two data sets are required to estimate forest canopy volume: digital elevation models (DEMs) and canopy height models (CHMs). In this study, ground point clouds are first extracted from airborne LiDAR data to generate a DEM. Subsequently, the DEM is utilized as the basic datum for calculating the forest canopy volume. LiDAR data and remote sensing imagery are then combined to generate the canopy LiDAR data. The sub-grid volume accumulated between the canopy grid surface and the DEM provides the forest canopy volume. The study sites include the campus of National Cheng Kung University (NCKU) and Nan-Wha Reservoir forest area. The tree height, crown diameter, crown projected area, and canopy volume are determined for both study areas. Furthermore, the forest area and canopy volume are estimated over the study areas using the proposed techniques. The results show forest information can be acquired effectively using airborne LiDAR data and remote sensing imagery.

Traditionally, because the lack of observations of each surveyed position in GPS kinematic positioning assignments, the differential GPS (DGPS) technique has been frequently used to reduce the most of the GPS positioning errors and then to improve that of accuracy of the kinematic positioning results. However, in the DGPS strategy, the longer the observed-baselines were measured, the more difficult the observations were resolved. Recently, Precise Point Positioning (PPP) has been applied in the various estimations of GPS kinematic positioning. In this paper, the kinematic positioning accuracy of PPP was analyzed by comparing with those of the estimated results from DGPS and PPP, respectively. The airborne GPS observations from 5-day aerial triangulation in the south area of Taipei and the 5 GPS permanent-site observations with 1 second sample rate were estimated by DGPS and PPP in this paper, respectively. This paper not only showed that comparisons of kinematic positioning results from both of DGPS and PPP but also compared the estimations of principle point position by GPS and traditional aerial triangulation methods. From the analyzed results, under the state of using 43 ground- control-points and all of the baseline lengths were not more than 50 km, the average standard deviations of difference of the kinematic positioning position from PPP and DGPS were ± 0.012 m in N direction, ± 0.018 m in E direction, and ± 0.045 m in H direction, respectively, showing that the PPP results were consonant with that of DGPS. Furthermore, the posterior accuracy were reached to ± 0.068 m, ± 0.064 m, and ± 0.059 m in N, E, and H direction in 97 % adjusted observations, respectively.