Resource Selection by African Buffalo (Syncerus caffer) in the Caprivi Strip, Namibia
The resouce selection function used in this analysis allowed me to identify variables, both environmental and anthropogenic, that are important in determining buffalo space and resource use patterns. Spatially mapping probability distributions of valuable species such as buffalo can inform land use and resource planning. Identifying which features on the landscape, as well as vegetation types, are used by animals, can help conservation planners to create parks that address the need for migratory species to have intact corridors. This is especially important in the Caprivi Strip due to its narrow geography and the KAZA TFCA that is being planned in the region.
The 250m spatial resolution of the EVI tiles restricts the predictive ability of the probability surface. A 250m square area could contain a large amount of spatial variability, in terms of resources important to buffalo, including water availability and vegetation. The resolution could be increased by excluding EVI from the analysis, although this is undesirable as greenness is expected to be a powerful predictor of buffalo space use (and this is confirmed by the results).
EVI was the only temporally variable predictor included in this analysis, and has a resolution of 16 days. Most of the other covariates included are likely to change over time, as the human population expands in the Caprivi and land use decisions are made. The analysis would benefit from the addition of other temporally variable predictors, such as fire, and TRMM, which weren't included in this analysis due to time constraints. Despite these limitations, the results of the RSF can nevertheless provide insight into buffalo preference/avoidance behaviours with respect to landscape variables and environmental factors, which can contribute to future planning.
Choice of Predictor Variables
Although the dataset was split by home range, the predictor variables were aggregate for the whole of the Caprivi. This poses a potential problem, especially with the 'Distance to Field' variable, where correlation is high but the variable is likely not to influence movement at all. For example, many buffalo populations reside on one side of a major river, with agricultural fields on the opposite side. The minimum distance to field variable counted these fields as the closest, although the river is an un-crossable barrier for buffalo. In many instances, distance to field was highly significant and the co-efficient was negative, indicating that buffalo preferred to spend time closer to agricultural fields. This is likely false, and masking the true pattern which is that buffalo are restricted by water use and spend time closer to rivers. The analysis would benefit from splitting layers such as fields and pans into different regions, and only using those that are accessible to the buffalo population in the analysis.
The analysis used a 'day' variable in order to assess the relative significance of resource selection at during the day compared to night time. This is a coarse resolution variable, and more insight would likely be gained from splitting the observations further into morning, afternoon, evening, and night. Additionally, some measure of seasonality would be beneficial for the analysis, as the Caprivi has large differences in seasonality from wet to dry season.
This analysis is an example of the modifiable areal unit problem. A preliminary logistic regression was performed on all 32 buffalo together, and the results were quite different from the analysis performed on each animal split down. The Caprivi is spatially heterogeneous and the results of an RSF will be highly dependent on the scale at which the analysis is performed. This analylsis was performed using home ranges including 100% of data points per individual buffalo to create each RSF 'study area'. This is a generous analysis as it includes data points that may be outside the normal range of each animal. For example, one buffalo has recently begun moving quite far South and East of the Caprivi, down into Botswana and toward the Okavango delta. It's impossible to tell at this point in time whether this pattern is part of a regular (or semi-regular) migratory route, or a new movement by the invididual. Without high resolution spatial data, it's also impossible to tell whether this is an individual buffalo, or an entire herd.
Extrapolating from Individual to Population
The data used in this analysis represents only 32 individual buffalo. These individuals were chosen to represent different populations living in the Caprivi, but they are in no way comprehensive, and it is very difficult to predict whether the individual buffalo represents the herd as a whole. Buffalo are highly social animals, and tend to herd, with the exception of old males. Within each herd there is a wide variety of age and health. These are important factors affecting behaviour and movement patterns. Although an RSF analysis can be very insightful and highlight interesting movement patterns, it is important to recognize the limitations of assessing individuals, as well as the resolution of the data, both temporally and spatially.