Raster Meets Vector Data - Open Source Geospatial Tools: Applications in Earth Observation - page 129

In addition, we create a second vector dataset where all isolated forest objects
smaller than 15 pixels are removed. We apply the sieving filter from Sect. 9.1 to the
raster dataset fmap_ie.tif with the connectivity defined in eight directions. This
corresponds to a minimum mapping unit of 0

938 ha (15 pixels of 25 m 2 ).

.

gdal_sieve.py -st 15 -8 fmap_ie.tif fmap_ie_15_8.tif
gdal_merge.py -n 2 -init 0 -pct -o forest_ie_15_8.tif
→ fmap_ie_15_8.tif
gdal_polygonize.py -mask forest_ie_15_8.tif fmap_ie_15_8.tif -f
→

SQLite fmap_ie_15_8.sqlite FTYP

The results between the polygonizedmaps with and without prior sieving is shown
in Fig. 9.1 . Filtering polygons based on the area can also be performed on vectors

(a) (b)

forest map
forest
no forest

0 2.5 5 km

(c)

Fig. 9.1 Raster to vector conversion using gdal_polygonize.py . a Raster forest cover map.
b Vector forest cover map retaining all polygons. c Vector forest cover map after sieving filter