Estimates of forest elephant abundance: projecting the relationship between precision and effort.
A key element to African elephant [Loxodonta africana] conservation is monitoring how different forms of human activity influence elephant distribution and abundance. Information on human-elephant interaction is critical for guiding conservation action at all levels, from very local management interventions, through national protected area planning, to international policy initiatives such as the ban on ivory trade. However, methods for monitoring elephant numbers often produce rather poor estimate precision. Therefore, a major challenge for elephant monitoring programmes at all levels is to allocate survey effort efficiently: the greatest possible statistical power to detect the effects of particular forms of human activity must be obtained from limited survey effort. This study examined an approach to survey design based on the assumption that the precision of an abundance estimate is directly proportional to sampling effort. Field data from three sites in Central Africa (collected between 1997 and 1999) were used to evaluate the potential of this approach for projecting how much effort will be necessary to survey and monitor forest elephants. Variance in elephant dung pile encounter rate was proportional to sampling effort, but variance was also influenced by mean encounter rate. The relationship between variance and mean encounter rate followed a power law more closely than the linear model suggested by many previous authors. The linear model tended to substantially underestimate the effort level required for areas with high encounter rates. A bootstrapping exercise suggested that modelling the precision-effort relationship across strata should produce more accurate projections than estimating a distinct precision-effort relationship for each stratum. Results suggest that the survey design framework examined could substantially improve the statistical power of forest elephant monitoring programmes, although the use of a power law (rather than linear) model seems preferable. A similar approach could be appropriate for designing savanna elephant monitoring programmes and for addressing other issues, including rates of hunting and human elephant conflict. Furthermore, the parameter estimates reported in this study could be useful in designing elephant survey and monitoring programmes at other forested sites in Central Africa.