Overview
The
Dunnart Model (see
Figure 1) provides an example of how expert knowledge and limited empirical data can be combined within a Bayesian network, and linked to GIS data, to assist in recovery planning of endangered fauna populations. The model was developed in collaboration with a wildlife ecologist, supported by field data, to determine the probabilistic influence of grazing pressure, density of the invasive shrub prickly acacia (Acacia nilotica), land tenure, soil variability and seasonal variability on dunnart habitat suitability. The model was then applied in a GIS to map the likelihood of suitable dunnart habitat.
Click on the following reference to access a copy of a paper published on the Dunnart Model:
Smith, C. S., Howes, A. L., Price, B. and McAlpine, C. A. (2007). Using a Bayesian belief network to predict suitable habitat of an endangered mammal - the Julia Creek Dunnart (Sminthopsis douglasi). Biological Conservation. 139 (3-4):333-347
Figure 1: Dunnart Model
Using the Model
Open the Dunnart Model by clicking
here. You will notice that the nodes in the model are colour coded. The nodes coloured
green are the
input nodes, from which scenarios can be selected. The
red node is the
output node, which in this case is
habitat suitability.
The model can be used to
a) predict the
habitat suitability of a site for dunnarts, and
b) identify the
most important factors influencing
habitat suitability. Use the scenario selector (left-hand frame) in the DBLi network viewer to select a state for
Distance to Water,
Mapped Land Tenure,
Mapped Prickly Acacia Density,
Mapped Soil Type and
Rainfall. Then click on
Update/Refresh Network. The probabilities in the
Suitability node will update to show the probability of your site being of
High,
Medium or
Low suitability.
In order to identify the
most important factors influencing habitat suitability we can use
diagnostic analysis. This means
selecting a outcome and running the model backwards to observe a change in the probability of inputs. Click on the
Remove Scenario link in the top frame of the network viewer, then select the
High state for
Suitability and click
Update/Refresh Network. Now the model is running backwards, showing the
most likely state of inputs (such as
Distance to Water,
Land Tenure,
Prickly Acacia Density,
Rainfall and
Dominant Soil Type) under a
High suitability scenario.
Figures 2 and
3 show the Dunnart Model with the
High and
Low state for habitat suitability selected. Recording the probability of particular states for model inputs under both the High and Low habitat suitability produces
Table 1, which indicates that the inputs with greatest change in probability are
Rainfall,
Dominant Soil Type, and
Prickly Acacia Density. This means that these factors will most influence habitat suitability because a change in habitat suitability most influences them.
Table 1: Change in the probability of Dunnart Model model inputs when Habitat Suitability is changed from High to Low
| Node and State |
Probability when High
Suitability selected
|
Probability when Low
Suitability selected
|
Difference
|
Rank
|
| Rainfall (Above Average) |
94.3
|
47.6
|
46.7
|
1
|
| Distance to Water (Greater than 5) |
74.9
|
70.8
|
4.1
|
5
|
| Land Tenure (Reserve) |
10.7
|
1.4
|
9.3
|
4
|
| Prickly Acacia Density (Low) |
95.8
|
75.6
|
20.2
|
3
|
| Dominant Soil Type (Clay) |
99.9
|
60.1
|
39.8
|
2
|
Figure 2: Dunnart Model when High habitat suitability is selected
Figure 3: Dunnart Model when Low habitat suitability is selected
