Accurately estimating wildlife abundance based on field surveys in remote areas has been a big challenge to ecologists. The Tibetan wild ass (Equus kiang), an endemic species on the Qinghai-Tibet Plateau, is a dominant species living at the montane grasslands ecosystem. Although numerous surveys had been conducted, its total abundance is still unknown. We carried out field surveys in summers from 2014 to 2018 following the distance sampling protocol, covering an area of 1.2 million km2 on the Qinghai-Tibet Plateau. The total length of the survey routes is 25,000 km. We observed 9233 individuals of Tibetan wild ass. We developed a novel method to accurately estimate the abundance of the Tibetan wild ass. First, we used species distribution models (e.g. Poisson regression, random forest, Maxent) to quantify the relationship between species occurrences and 22 environmental variables (including elevation, land use, human footprint index, and 19 climate variables), and predicted the population density in the whole study area. Secondly, we compared the model prediction and field survey results, and calculated the ratio between them. Thirdly, we used the ratio to adjust the predicted abundance.
Results/Conclusions
The estimated total abundance of Tibetan wild ass is about 300,000. To evaluate the potential bias of the estimation, we quantified three types of uncertainty: 1. Survey uncertainty. The detection function of distance sampling was used, and the integration of 1- detect-rate at the range of 0-1000 meters was calculated to represent survey uncertainty. 2. Model uncertainty. The index 1- R2 of the best species distribution model represented how likely the wild ass was wrongly predicted in space. 3. Adjustment uncertainty. The predicted abundance in a quadrat and the observed number of individuals within that quadrat were different, and such difference varied in space. We used the standard deviation of ratio between predicted and observed abundance at different surveys segments as the adjustment uncertainty. The estimation uncertainty for the total abundance is the product of above three uncertainties, and the 95% confident was 180,000 to 520,000. Our new method for estimating species abundance is suitable for species which distribution is well constrained by environmental variables; meanwhile, distance sampling is needed for adjusting model predictions.