Postprocess Wetland Predictions

Description

Apply simple, geometry based post-processing to wetland predictions.

Usage

Pixel-based classifications, like Random Trees, can produce sparse predictions since they consider the relationship between target classes and characteristics on a cell-by-cell basis. In the case of wetlands, these isolated cells are unlikely to represent true wetlands, which exist as geomorphic objects. Further, smaller collections of sparse predictions may represent wetlands smaller than a user's minimum mapping wetland unit. This tool applies a simple, geometry based post-processing to raster wetland predictions to return a cleaned up set of predicted wetland polygons. The post-processing workflow performs the following:

1. Majority Filter on the input binary predictions raster (using 8 neighbors and a replacement threshold of half)
2. Converts the filtered raster to polygons, deleting the nonwetland polygons (cell value = 1)
3. Deletes any remaining polygons smaller than the input minimum wetland size
4. Uses Eliminate Polygon Part to fill holes smaller than the input minimum wetland size within remaining wetlands

Parameters

Parameter Name	Type	Direction	Data Type	Dialog Reference
Input Pixel-Based Wetland Predictions Raster	Required	Input	Raster Layer	Input raster containing binary wetland predictions. Values of 0 are interpreted as predicted wetlands. Values of 1 are interpreted as predicted nonwetlands.
Input Minimum Wetland Size (sq. m)	Required	Input	Double	Minimum wetland size (sq. meters) for the output post-processed wetlands.
Output Post Processed Wetland Polygons	Required	Output	Feature Layer	Output post-processed wetlands feature class.