Overview
The
Pasture Dynamics Model is an example of how a Bayesian Network can be used to integrate knowledge about ecosystem dynamics to build a predictive model. In the past, state and transitional models (STMs) have been used by researchers to describe vegetation dynamics and the possible responses of vegetation to management actions and environmental events. STMs describe vegetation dynamics using diagrams that position vegetation states along several axes representing environmental or management gradients. Possible transitions between these vegetation states are represented using arrows and a table, called a catalogue of transitions, is used to describe the environmental or management conditions under which each transition can occur.
Figure 1 is an STM of a pasture ecosystem in South East Queensland, Australia. This particular STM has five vegetation states (
Palatable Tall Tussock Grasses (PTG),
Unpalatable Tall Tussock Grasses (UPTG),
Short Sward and
Sparse Tall Grasses,
Short Sward, and
Lawn), positioned along three gradients (
Palatability,
Grazing Intensity and
Soil Nutrient Status). The arrows indicates how each vegetation state can move to another vegetation state, for example, Palatable Tall Grasses can move to Lawn and Lawn can move back to Palatable Tall Grasses.
Figure 1: State and Transition Model (STM) for a South East Queensland pasture ecosystem. PTG = Palatable Tall Grasses; UPTG = Unpalatable Tall Grasses. The possible transition is the point at which the pasture is unlikely to return to a better state without extreme management intervention.
The limitation of STMs is that they cannot be used for predictive purposes, because they are static diagrams. To overcome this limitation, an STM can be converted into a Bayesian network, which can subsequently be used for scenario analysis. The Pasture Dynamics Model shown in
Figure 2 is a Bayesian network created from the STM in
Figure 1. The model has been colour coded to show the current state of the pasture and time frame of interest (
blue nodes), possible transitions away from each pasture state (
red nodes), the main factors driving transitions (
yellow nodes), and the sub-factors influencing the main factors (
green nodes).
Figure 2: Pasture dynamics model created from the STM in Figure 1
The model in
Figure 2 has a scenario inserted. The scenario shown is where the current state of pasture is
palatable tall grasses (hence only transitions from palatable tall grasses are possible), the time frame is
5–10-years, spell post-fire is
no, type of grazer is
cattle, summer spelling in time period is
none, drought is
no, supplements in dry season is
yes, stocking rate is
high, distance from camp site is
near, good seasons in time period is
infrequent and fertiliser application is
none. Under this scenario the model is predicting that the most likely transition is to
short sward (
39.6% chance).
Click on the following reference to access a copy of a paper published on the Pasture Dynamics Model:
Bashari, H., Smith, C., Bosch, O.J.K. (2009). Developing decision support tools for rangeland management by combining state and transition models and Bayesian belief networks. Agricultural Systems. 99: 23-34.
Using the Model
The pasture dynamics model can be used for two main decision-making purposes:
(a) prediction and
(b) diagnosis. Predictive analysis can be used to answer
what if questions by selecting particular states for input nodes (a scenario) and using the model to predict the probability of transitions, as shown in
Figure 3. In the example, the model predicts that the chance of a transition away from
palatable tall grass to
lawn is
relatively high (
64%) within a
5 to 10 year time frame (note that the state 5 to 10 years is selected in the time frame node). The model also indicates the probable causes for this transition: high grazing pressure (80% chance) and above-average soil nutrition (95% chance).
Figure 3: Pasture dynamics model being used for prediction (shaded nodes represented the selected scenario)
Diagnostic analysis can be used to answer ‘how’ questions by selecting a desired outcome and using the model to identify the scenario that is most likely to lead to that outcome, as shown in
Figure 5. In this example, the model is used to identify how a land manager might shift
pasture from an
unpalatable tall grass state to a
palatable tall grass state within a
5 to 10 year time frame (note that the state 5 to 10 years is selected in the time frame node). The model shows that this transition is most likely if
grazing pressure and
selective grazing are
absent (see the
selective grazing and
grazing pressure nodes), and this is most likely where
destocking is applied (see the
stocking rate node). The model also shows that
frequent fires are important for achieving
low to no selective grazing (see the
fires in time period node), and in turn,
good seasons are important for achieving
frequent fires (see the
good seasons in time period node).
Figure 4: Pasture dynamics model being used for diagnosis (shaded nodes represented the selected scenario)
