Hyper-spectral and LIDAR images provide a series of image data for geoscience applications. However, few applications of this image material have been proposed to establish a land creep prediction system. The spatial resolution of a LIDAR image is approximately 50 cm, while the typical creep displacement is about 20 cm. It is thus a difficult task to predict the land creep displacement by the use of LIDAR images. Nonetheless, the present study addresses the above issues by using Regional-based growing Object Classification (ROC). The ROC technique is a computational object model that is transformed into a ROI module by the connecting neighboring pixels of similar intensity. The creep displacement is predicted by analyzing the DEM model, which is generated by LIDAR images. The study region is the retaining wall and the slide of a hillslope along Formosa Freeway. The creep displacement is so small that a series of hyperspectral and LIDAR images taken at the same geological location but at different time periods are used for investigation. The hyperspectral data is used to generate the ROC model, and LIDAR data is used to create the DEM model for measuring heaving/settlement. This study proposed two different approaches to investigate the results: the average of difference and baseline correction methods. The average of displacement obtained by the average of difference method is 24 mm, and that obtained by the baseline-correction method is 19 mm. The result indicates the baseline-correction method is a better prediction model, and its regression formula can be used to estimate the amount of heaving/settlement in the early stage of analysis.
Keywords: Hyper-spectral, LIDAR, Creep and Creep monitoring.