Unsupervised Classification
- Assigning land cover classes to satellite image pixels
- Supervised classification: We tell the software which region of image corresponds to which class
- Unsupervised classification: Software automaticallly classes based on spectral similarity
Composite Band
- Search and open "Composite Band"
- Add all seven (1 to 7; not otherway around)
- Alternatively, Windows > Image Analysis -> select 7 bands then composite
- For the produced layer(s), select Std-dev instead of percent clip on Image Analysis
- Follow false coloring as per source guidelines
- Natural color, Landsat 8/9 OLI -> RGB -> 4 3 2
- Left click on Red/Green/Blue
- Select Dhaka on bgd_admbnda_...
- Extract by Mask
- Input = Composite
- Feature mask data = bgd_...
- Change RGB bands on extracted dhaka
Classification
- Windows > Image Analysis: Select extracted dhaka (not just check) then Std-dev
- Customize > Toolbars > Image Classification
- Classification > Iso Cluster Unsupervised...:
- Input raster bands - extracted dhaka
- Number of classes: 6
- LULC: Land Use Land Cover; Land Use is how I use the land; Land Cover is how it is actually
- Add basemap > Imagery
- Common classes:
- Waterbody
- Agriculture
- Vegetation
- Buildup/Settlement
- Baseland
- Group same classes that are distinced
- LULC/IsoClassified layer > Properties > Symbology > Select similar symbols > Right click > Group values
- Changes are volatile; would reset when opened again on another computer
- To make changes permenant: Reclassify
- Input = IsoClassified
- To group, set new value same
Clac Area
- Raster to Polygon
- Input = reclasses LULC
- Field = Value
- Polygon under which class is represented by "gridcode" (open attribute table of output polygon)
- Add new field "Area" then calc area
- Geoprocessing > Dissolve: by gridcode of polygon
- Open attribute table of dissolved polygon to get total area by LULC; optionally add another field for friendly name