To effectively provide scientific-based data of forest cover changes on public and private landuse
types, in this study, forestland in Shitan District of Miaoli County was selected as the research
experimental area, and 8 cloud-free FORMOSAT-2 (FS2) images from 2005 to 2014 in the area
were collected. To improve the reliability of the analyses, a digital elevation model (DEM) was
used to calculate the common shadow areas of the sun incident geometry of each date’s FS2 image,
and after filtering out common shadow areas from each image, the normalized difference vegetation
index (NDVI) of each FS2 image was established. Image segmentation with an object-based
mean shift algorithm was applied to extract non-vegetation-covered patches (NVCPs). The overall
accuracy of the 8-date NVCP extractions exceeded 81%. Combined with analysis of geoprocessing
and landscape metrics, the spatial distribution pattern of NVCPs in each land-use type was evaluated.
To measure the degree of forest landscape fragmentation, 2 levels of landscape metrics (class
and landscape) were selected to access landscape structural changes during these years. Comparing
the ratio of the area of NVCPs, the average slope which was more than 16° and occurred with
main land-use types in individual images, results showed that the ratios of the area of NVCPs were
larger in the order of land-use types of fruit trees, bare land, and dry land, and the ratios of mixed
forests, broadleaf forests, bamboo forests, and betel nut (areca) plantations slowly increased year
to year. In addition, utilizing a local spatial statistical indicator to analyze the spatial distribution
of NVCPs during the 10 years, 2 hotspots of NVCPs were detected. The methods and outcomes of
this study can provide spatial decision-making information for managing forest land-use in shallow
mountain areas.