produced at a 250 m spatial resolution, covering the Taihu watershed area from
2000 to 2013. The interannual temporal dynamics (Fig. 11) for five land cover
macro-classes were chosen as classification targets (urban dense, urban sparse,
agricultural low intensity, agricultural high intensity, natural vegetation). For
each year, two synthetic multi-temporal features were extracted: NDVI annual
mean value and NDVI annual standard deviation of the monthly mean. Annual
NDVI features (NDVI mean and NDVI stdev ) were clustered into the five categories
using ISODATA algorithm and were subsequently labeled according to this
classification scheme.
2000
2005
2009
2013
Fig. 12 Land cover maps produced starting from MODIS composites for specific years in the
range 2000–2013
100
P. Villa et al.
2000 to 2013. The interannual temporal dynamics (Fig. 11) for five land cover
macro-classes were chosen as classification targets (urban dense, urban sparse,
agricultural low intensity, agricultural high intensity, natural vegetation). For
each year, two synthetic multi-temporal features were extracted: NDVI annual
mean value and NDVI annual standard deviation of the monthly mean. Annual
NDVI features (NDVI mean and NDVI stdev ) were clustered into the five categories
using ISODATA algorithm and were subsequently labeled according to this
classification scheme.
2000
2005
2009
2013
Fig. 12 Land cover maps produced starting from MODIS composites for specific years in the
range 2000–2013
100
P. Villa et al.