Abstract
Growing threats to primates in tropical forests make robust and long-term population abundance assessments increasingly important for conservation. Concomitantly, monitoring becomes particularly relevant in countries with primate habitat. Yet monitoring schemes in these countries often suffer from logistic constraints and/or poor rigor in data collection, and a lack of consideration of sources of bias in analysis. To address the need for feasible monitoring schemes and flexible analytical tools for robust trend estimates, we analyzed data collected by local technicians on abundance of three species of arboreal monkey in the Udzungwa Mountains of Tanzania (two Colobus species and one Cercopithecus), an area of international importance for primate endemism and conservation. We counted primate social groups along eight line transects in two forest blocks in the area, one protected and one unprotected, over a span of 11 years. We applied a recently proposed open metapopulation model to estimate abundance trends while controlling for confounding effects of observer, site, and season. Primate populations were stable in the protected forest, while the colobines, including the endemic Udzungwa red colobus, declined severely in the unprotected forest. Targeted hunting pressure at this second site is the most plausible explanation for the trend observed. The unexplained variability in detection probability among transects was greater than the variability due to observers, indicating consistency in data collection among observers. There were no significant differences in both primate abundance and detectability between wet and dry seasons, supporting the choice of sampling during the dry season only based on minimizing practical constraints. Results show that simple monitoring routines implemented by trained local technicians can effectively detect changes in primate populations in tropical countries. The hierarchical Bayesian model formulation adopted provides a flexible tool to determine temporal trends with full account for any imbalance in the data set and for imperfect detection.