In an effort to develop more holistic ecosystem approaches to resource asse
ssment and management, landscapes need to be stratified into homogeneous ge
ographic regions. These regions can then be used in a monitoring framework
to develop reliable estimates of ecosystem productivity. A regional charact
erization of the woodland biome has been developed for South Africa, deline
ated by satellite imagery and using environmental data and a rigorous stati
stical methodology. Distribution maps of key environmental variables are an
alyzed by factor analysis, an iterative clustering technique and maximum li
kelihood classification to quantify and identify homogeneous physio-climati
c units.
A spatial clustering technique was used to identify regions, which are stat
istically different with regard to five physiographic, climatic and edaphic
variables deemed important within southern African savanna woodlands. The
woodland biome of South Africa at 1km resolution was successively divided.
Thirty year mean monthly temperature, total plant-available water balance o
f soil, elevation, landscape topographic position, and landscape soil ferti
lity were used as input classification variables.
The map data were submitted to a factor analysis and varimax axis rotation.
The factor analysis removes correlations from the input variables, reduces
the dimensionality, and normalizes the axis measurements. A cluster analys
is was performed on the three principal factor scores using a modified iter
ative optimization clustering procedure to determine the finest level of cl
asses statistically permitable. Twenty-seven identified unimodal cluster si
gnatures were then submitted to a maximum likelihood classification where t
he statistical probability of the GIS cell assignment is carried out to det
ermine class membership. The final map of custom physio-climatic regions is
described, and these custom regions are compared with a vegetation potenti
al map of the woodland types identified in the South African summer rainfal
l zone.