Fuel Classifications - Wildland Fuel Fundamentals and Applications - page 113

Instead, it uses a set of ecological descriptions mostly based on vegetation and stand
history to aid in fuelbed identification (Ottmar et al. 2007 ). As a result, there is
often redundancy across many fuel classification categories; the properties of one
fuelbed may be quite similar to other fuelbeds sampled in another part of the coun-
try or for another vegetation type, especially for the fine woody debris components.
Linking opportunistic classification categories to spatial data layer attributes is also
problematic because it is difficult to consistently validate an assigned class in the
field because there is no fuel classification key. Another problem is that since the
variation across fuelbeds is not incorporated into the classification design, there can
be an infinite number of possible categories (fuelbeds), and conversely, there can
be many locally relevant fuelbeds that are missing in the final classification. Keane
et al. ( 2006 ), for example, mapped FCCS categories across central Utah but found
that over 30 % of the land area had vegetation attributes that did not match sampled
FCCS classes. This issue makes opportunistic classifications somewhat difficult to
learn because it is always changing and new classes are always being added.

7.2.3 Classification

Classification, as previously mentioned, is the process of systematically and com-
prehensively clustering items (fuelbeds) into unique groups based on selected attri-
butes—mainly loading by fuel components. Usually, this involves numerical clus-
tering and complex statistical techniques that attempt to directly identify unique
groups based on the variation of the attributes selected to develop the classification
(Gauch and Whittaker 1981 ; Orloci 1967 ). Once unique groups are identified, a
comprehensive key based on the analysis variables (e.g., loading) can be devised to
objectively identify the classification category for a field-assessed observation. This
approach partitions the variation in the field data to reduce redundancy and produce
a singular classification.
Few existing fuel classifications were built using this direct, top-down classifi-
cation approach. In perhaps the first effort at directly classifying fuels, Fahnestock
( 1970 ) developed two keys that evaluated various fuel attributes, including particle
size, compactness, vertical position, and horizontal continuity, to key to unique spread
rate and crowning potential classes. Dimitrakopoulos ( 2001 ) created a fuels classi-
fication for Greece by clustering flammability variables, such as heat content, ash
content, and particle density, into unique groups using hierarchical cluster analysis
and canonical discriminant analysis for Mediterranean shrublands. The fuel loading
models (FLMs) of Lutes et al. ( 2009 ) is distinctive in that field-collected fuel loading
data were used to simulate smoke emissions and soil heating, and these simulation
results, along with loading, were used to create unique classes using advanced clus-
tering and then a unique key was created using regression tree analyses. As a result,
this classification effectively integrated the resolution of the fire models for which
the FLMs would eventually be used into the classification design (Fig. 7.3 ).
An advantage of the direct classification approach is that resultant classifica-
tions are fully supported by the data that were used to create them, and therefore,