Raster classification and mapping of ecological units of Southern California

For a series of studies on the ecosystem service values of chaparral in
Southern California, we developed a raster data layer providing an
ecological unit classification of the Southern California landscape. This
raster dataset is at a 30 meter pixel resolution and partitions the
landscape into 37 different ecological unit types. This dataset was
derived through a GIS-based cluster analysis of 10 different physiographic
variables, namely soil suborder type, terrain geomorphon type, flow
accumulation, slope, solar irradiation, annual precipitation, annual
minimum temperature, actual evapotranspiration, and climatic water
deficit. This partitioning was based on physiographic variables rather
than vegetation types because of the wish to have the ecological units
reflect biophysical characteristics rather than the historical land use
patterns that may influence vegetation. The cluster analysis was performed
across a set of 10,000 points randomly placed on a GIS layer stack for the
10 variables. These random points were grouped into 37 discrete clusters
using an algorithm called partitioning around medoids. This assignment of
points to clusters was then used to train a random forest classifier,
which in turn was run across the GIS stack to produce the output raster
layer. This dataset is described in the following book chapter
publication: Underwood, Emma C., Allan D. Hollander, Patrick R. Huber, and
Charlie Schrader-Patton. 2018. “Mapping the Value of National Forest
Landscapes for Ecosystem Service Provision.” In Valuing Chaparral, 245–70.
Springer Series on Environmental Management. Springer, Cham.
https://doi.org/10.1007/978-3-319-68303-4_9.