Resource Selection by African Buffalo (Syncerus caffer) in the Caprivi Strip, Namibia
I performed 64 logistic regressions in total; two for each buffalo. One regression had a day interaction in order to assess the significance to the covariates at day vs. night time. The results of the logistic regressions can be seen in tables 1 and 2. The main results are summarized below.
Distance to Features
In general, buffalo showed avoidance of barriers, including fences and roads. Rivers were significantly positive (avoidance) in two cases and negative (preference) in two cases. Distance to pan was equally positively and negatively significant. Distance to agricultural fields was largely negatively significant; indicating preference. This is likely due to design flaws, as in many of the cases agricultural fields were located just opposite a large river that the buffalo use as a water source. The river poses a large enough barrier to restrict buffalo from crossing. With a day interaction added, most distances to features were not significant predictors of buffalo presence.
Vegetation variables (EVI, fraction tree cover, vegetation type)
The majority (43/50) of vegetation structure coefficients that were significant were negative, indicating a preference for wetland over drier vegetation types. Out of the vegetation types, grassland had the most (4) significant positive co-efficients, indicating preference over wetland habitats. With a day interaction, the vegetation types were not very significant, although in a one case, buffalo showed an avoidance for woodland and grassland during the day, and one animal showed a preference for shrubland during the day.
Fraction tree cover was significant for only 3 animals without the day interaction; two of these were negative (avoidance) and one was positive (preference). Adding in the day interaction reversed this pattern; two buffalo showed preference for higher tree cover during the day and one showed avoidance. Preference for tree cover during the day could be explained by predation avoidance behaviour.
EVI was significant for 28 out of the 32 buffalo. The co-efficients were a mix of negative and positive; this is likely due to the spatial heterogeneity in the Caprivi Strip.
I generated two probability surfaces in order to display my results geographically. I chose collar 77265, which represents a buffalo whose home range is in the Mudumu area of the Caprivi (figure 5). The probability surfaces represent two different 16-day periods (the temporal resolution is restricted by the EVI data). The first probability surface (figure 6) represents January 1-16, 2011. I chose this date range to represent the height of the wet season in the Caprivi. Figure 7 displays the other variables used for this logistic regression. The second probability surface (figure 8) represents June 20 - July 4, 2011. This EVI tile was chosen to represent the dry season. Figure 9 displays the variables used for this layer, with EVI in the background. Note the difference in EVI between January and July.