Probabilistic Representation of Geographic Characteristics

The first step in building any species viability influence diagram is to identify regions of the study area (here, Kenya) that are large enough to have the potential of sustaining the species under favorable conditions. These regions should be as homogeneous in terms of land use and climate as possible. Strict homogeneity however, is not required because the influence diagram uses chance nodes to represent within-region heterogeneity.

To represent regions of relatively homogeneous cheetah habitat, Gros (1998) divides Kenya into seven regions. Several of the data sources used here however, index wildlife observations by public administration district (see Tartter 1983, pp. 194-195). Hence, regions will be defined as collections of these districts. The Gross (1998) regions (Marsabit, Eastern, Samburu, Tsavo surroundings, Masailand, Laikipia, and Nakuru) are used here with the addition of regions for the densely settled western districts (Western), the central farmlands (Central), Turkana (Turkana), and coastal areas (Coastal) (see Figure 3).

Values of C cover the most important climate regimes in Kenya and are very arid, arid, semi-arid, and non-arid.

The variable U takes on the values given in Gros (1998): nomadic-camel, nomadic-cattle, and ranching along with the additional value of farming. Any change in unprotected land use relative to 1971 (Ominde 1971) is represented as the result of management decisions and hence U is influenced only by the decision nodes q and m. This approach is a simplification because in reality, Kenya's land use distribution changed from 1887 to 1971. Future land use distributions however, will be influenced at least in-part by governmental policy (node m). The discrete variable R_t takes on the two values 0-.5 and .5-.99 fraction of the region's area that is protected. The majority of the national parks and game reserves used in this study have been created since 1945 (Kaplan et al. 1976, p. 65). Therefore, this node is influenced by the nodes t and q, i.e., $P(R_t = \mbox{0-.5}\vert t,region) = \beta_{R_t}^{(e)}$ if t < 1945 and $P(R_t = \mbox{0-.5}\vert t,region) = \beta_{R_t}^{(l)}$ if t > 1945 (superscript ``e'' for early and ``l'' for late).