Poaching is contributing to rapid declines in elephant populations across Africa. Following high-profile changes in the political environment, the overall number of illegally killed elephants in Africa seems to be falling, but to evaluate potential conservation interventions we must understand the processes driving poaching rates at local and global scales. Here we show that annual poaching rates in 53 sites strongly correlate with proxies of ivory demand in the main Chinese markets, whereas between-country and between-site variation is strongly associated with indicators of corruption and poverty. Our analysis reveals a recent decline in annual poaching mortality rate from an estimated peak of over 10% in 2011 to <4% in 2017. Based on these findings, we suggest that continued investment in law enforcement could further reduce poaching, but is unlikely to succeed without action that simultaneously reduces ivory demand and tackles corruption and poverty.
Dispersal is crucial for population viability and thus a popular target for conservation measures. However, the ability of individuals to move between habitat patches is notoriously difficult to estimate. One solution is to quantify functional connectivity via realistic individual‐based movement models. Such simulation models, however, are difficult to build and even more difficult to parameterize. Here, we use the example of natal little owl (Athene noctua) dispersal to develop a new analysis chain for the calibration of individual‐based dispersal models using a hybrid of statistical parameter estimation and Approximate Bayesian Computation (ABC). Specifically, we use locations of 126 radio‐tracked juveniles to first estimate habitat utilization by generalized additive models (GAMs) and the biased random bridges (BRB) method. We then include the estimated parameters in a spatially explicit individual‐based model (IBM) of little owl dispersal and calibrate further movement parameters using ABC. To derive efficient summary statistics, we use a new dimension reduction method based on random forest (RF) regression. Finally, we use the calibrated IBM to predict the dispersal potential of little owls from local populations in southwestern Germany to suitable habitat patches in northern Switzerland. We show that pre‐calibrating habitat preference parameters while inferring movement behavioral parameters via ABC is a computationally efficient solution to obtain a plausible IBM parameterization. We also find that dimension reduction via RF regression outperforms the widely used least squares regression, which we applied as a benchmark approach. Estimated movement parameters for the individuals reveal plausible inter‐individual and inter‐sexual differences in movement behavior during natal dispersal. In agreement with a sex‐biased dispersal distance in little owls, females show longer individual flights and higher directional persistence. Simulations from the fitted model indicate that a (re)colonization of northern Switzerland is generally possible, albeit restricted. We conclude that the presented analysis chain is a sensible work‐flow to assess dispersal connectivity across species and ecosystems. It embraces species‐ and individual‐specific behavioral responses to the landscape and allows likelihood‐based calibration, despite an irregular sampling design. Our study highlights existing, yet narrow dispersal corridors, which may require enhancements to facilitate a recolonization of little owl habitat patches in northern Switzerland.
Generalized degrees of freedom (GDF), as defined by Ye ( 1998 Ye, J. (1998). On measuring and correcting the effects of data mining and model selection. Journal of the American Statistical Association 93(441):120–131. [Google Scholar] JASA 93:120–131), represent the sensitivity of model fits to perturbations of the data. Such GDF can be computed for any statistical model, making it possible, in principle, to derive the effective number of parameters in machine-learning approaches and thus compute information-theoretical measures of fit. We compare GDF with cross-validation and find that the latter provides a less computer-intensive and more robust alternative. For Bernoulli-distributed data, GDF estimates were unstable and inconsistently sensitive to the number of data points perturbed simultaneously. Cross-validation, in contrast, performs well also for binary data, and for very different machine-learning approaches.