Classification of non-vegetated areas using Formosat-2 high spatiotemporal imagery: the case of Tseng-Wen Reservoir catchment area (Taiwan)

The occurrence of landslides in the catchment area is a potential threat to the water quality and the lifespan of a reservoir. Due to the limitations of spatial coverage in ground surveys and of temporal resolution in aerial photos, it is difficult to monitor such events in the entire catchment area at short intervals. Formosat-2 is the first commercial satellite dedicated to site surveillance with a high-spatial-resolution sensor placed in a daily revisit orbit (2 m in panchromatic and 8 m in multi-spectral). In this research, a new approach is proposed to identify the non-vegetated areas in the multi-temporal and multi-spectral images taken by Formosat-2 by integrating the Getis statistic, the spectral index and the unsupervised K-means classification. With this new approach, we analyse a total of 16 pairs of Formosat-2 images, taken in the catchment area of Tseng-Wen Reservoir from February to December 2006 at an interval of three to four weeks. The results show that newly developed non-vegetated areas are closely related to earthquakes and rainfall. Once the slump material is generated by an earthquake, a comparatively low amount of rainfall will trigger its flushing. However, once the slump material has gone, there are no significant changes in the non-vegetated areas, even with severe weather events such as typhoons or storms. This suggests that the most critical time for protecting the reservoir is right after an earthquake and before the next rain. If the slump material is not managed or removed during this crucial period of time, eventually it will fall into the reservoir. Since the catchment area of Tseng-Wen Reservoir is protected and restricted from access, most of the non-vegetated areas should be closely related to landslides caused by natural processes (such as rainfall or earthquake) rather than man-made processes (such as tree cutting or degradation of vegetation). This research demonstrates the potential of Formosat-2 imagery in monitoring the spatial and temporal variations of landslides in the catchment areas of reservoirs.