Post-Processing

Reference > Post-Processing

The 3-Post-Process stage cleans the binary presence/absence rasters produced by training and prediction. It uses two Datasheets, Filtering and Rule-Based Restrictions, and produces four versions of each binary map: unfiltered, filtered, restricted, and restricted and filtered.

Filtering

Filtering removes isolated presence pixels (speckle) and fills small gaps, based on how many of a pixel’s 8 neighbours are presence.

Field Description
Apply filtering When checked (default), applies the filter.
Min neighbors to keep A presence pixel stays presence only if at least this many of its 8 neighbours are also presence; otherwise it becomes absence. Value from 0 to 8. Defaults to 8 when blank. Lower values remove less.
Min neighbors to fill After the step above, an absence pixel becomes presence if at least this many of its 8 neighbours are presence. Value from 0 to 8. Defaults to 8 when blank. Lower values fill more.

Rule-Based Restrictions

Restrictions use other rasters to force pixels to presence or absence where the classifier is known to be wrong, for example removing predicted snow below a given elevation or water on steep slopes. Add one row per rule. Rules are applied in order, and later rules override earlier ones.

Field Description
Condition Raster A single-band raster on the same grid (extent, resolution and coordinate reference system) as the predictions. Rules whose raster is missing or doesn’t match the grid are skipped with a warning.
Minimum Value Lower bound of the condition range (inclusive).
Maximum Value Upper bound of the condition range (inclusive). To match a single category, set the minimum and maximum to the same value. If the minimum is larger than the maximum, the two are swapped.
Reclassify Value The class given to pixels where the condition raster falls within the range: 0 (absence) or 1 (presence). Pixels outside the range, or where the condition raster has no data, are unchanged.