There are a variety of ways to supplement the process of knowledge synthesis that leads to the definition of state and transition models. Both classification and ordination of temporal vegetation data and/or experimental vegetation manipulation data can help identify vegetation states and transitions. Ideally, one would access long term monitoring data, although this kind of data is scarce. This is especially true for oak woodlands and other plant communities where several decades may still be insufficient to observe change.
One solution to this problem is to substitute an understanding of vegetation variation in space for the knowledge of variations in time. It is likely that, at the regional level, vegetation states present at one point in time are representative of the variety of states that could be found over time. This type of information can be used to delineate groups of vegetation states potentially linked by transitions.
A factorial model of vegetation is useful for the design of state and transition models (Major, 1951):
These factors operate on different time scales:
(1) parent material and topography, are not time dependent in a practical sense; and (2)
regional climate and organisms (flora, fauna, humans). This distinction overlaps nicely
with the amount of control that management can exert on these factors. At one end of the
spectrum, parent material and topography cannot be modified, but since they do not vary
over time they can be seen as constants which set the stage for interaction of other
factors. As these two factors vary over a landscape, their patterns partly account for the
vegetation mosaic. Climate is more variable over time, especially in arid and semi arid
regions. At the other end of the spectrum, the presence of organisms and their impact are
the most variable factors and, at the same time, the most amenable to management
influence. The contribution of any state factor to changes in vegetation in a chosen
landscape becomes negligibly small if the factor is almost constant in the area. Spatial
analyses of the correlations of vegetation data with climate, parent material and
topographic position (the abiotic factors) can help delimit regions where these factors
are sufficiently constant in their influence on vegetation. Within these regions, the
observed variations in vegetation should be the result of a combination of natural
disturbance and management actions in the context of short term variations in climate. In
California's Mediterranean climate, site water balance is an important determinant of
potential vegetation states that results from the interplay of the abiotic factors.
The identification of domains where the abiotic factors are relatively
constant in their influence, will facilitate the linkage of vegetation states by
transitions.
A subset of 1,177 plots situated north and east of the Central Valley were grouped into seven "vegetation domains" using abiotic variables. The analysis was done using supervised clustering. The physiognomic groups in each domain provide a basis for the identification of relevant vegetation states in a particular bioclimatic zone of California.
The most important abiotic variables in the analysis were: yearly average precipitation, January mean minimum temperature, July maximum temperature and distance from the coast. However, on a smaller scale (i.e. within one of the domains), other variables such as available soil-water capacity and exposure (slope and aspect) probably play an important mitigating role. Therefore, it is important to develop zonations that are specific to the particular question being asked.
A high level of agreement was achieved between the types of vegetation in the precipitation zones delineated by an analysis of the VTM plots and the maps of hardwood rangelands (see map above) established from remotely sensed data (Vayssières and Plant, 1998). Evidence from other sources confirm that changes in vegetation response to management occurs at one of the thresholds in precipitation. Standiford et al (1991) found that many more (321.2) blue oak saplings per hectare at the high end of their precipitation gradient (560 to 660 mm/year) than in the zones below (e.g. 39.2 saplings/ha for 430-560 mm/year). This indicates that the level of precipitation that separates dry and mesic domains (at 600 mm) corresponds to a major shift in regeneration potential, which has important consequences for management. McClaran and Bartolome (1985) identified a relationship between rainfall, blue oak canopy cover and effects on grass forage production. Further work is needed to determine if some other threshold values used to delineate abiotic domains correspond to marked changes in plant response. Potential changes in response that are of interest to management include tree resprouting rates, potential for shrub cover and oak growth rates.
By limiting the geographic zone of application and identifying potential vegetation states, the physiognomic groups/abiotic domains approach provided a group of hardwood experts with a basis for the formulation of a state and transition model for one of the domains.
Supervised clustering reclassified 2000 VTM plots. The reclassification resulted in 26 physiognomic groups, based on the relative proportion of overstory tree basal area, and understory tree, shrub, herb, grass and ground cover (see decision tree).
This new classification was tested using data from a seperate 1000 plots dominated by oak trees. To validate the new classification, life forms and species of each of the new groups were checked for consistency and homogeneity. A total of 3000 plots are used to characterize oak woodland vegetation surrouding the Central Valley.
Examination of the relationships between physiognomic groups and cover types showed that the former are more useful to describe vegetation dynamics (Vayssieres and Plant, 1998). The physiognomic groups approach delineates combinations of life form abundance that have important management implications. For instance, the amount of grass cover in the understory influences value for grazing livestock, while the amount of overstory tree cover affects the productivity of the grass-forb understory. Also, shrubs need to reach a certain level of cover and height before they constitute a ladder-fuel that may allow ground fires to reach into the tree crown.
Conventional clustering and ordination methods failed to produce "natural" groups from life form abundance data to group vegetation plots into physiognomic groups. A new method of cluster analysis was designed to group objects based on attributes but "under the supervision" of one or more other attributes (Vayssieres, 1998). Given the multi-layered nature of variability in vegetation, there is a need for an array of ecologically sound classifications of vegetation. Supervised clustering is a particularly interesting tool in this context because it allows constructing classifications centered on one layer of variability under the supervision of a classification derived from another level of variability.
Return to State and Transition Models for California's Hardwood Rangelands
