Detection and stratification approaches for aerial surveys of deer in prairie–parklands

Abstract

Context For management decisions that require accurate and precise estimates of large mammal population numbers, aerial surveys are considered reliable despite their cost. However, aerial surveys may still suffer from underestimation because of undetected animals and low precision as a result of inefficient sampling designs. Aims We assess detection of two species of deer in prairie-parkland communities of western Canada and evaluate a suite of survey design features for improving the accuracy and precision of population estimates from aerial surveys. Methods Modelling detection of deer was based on 100 sightability trials involving 54 radio-collared white-tailed and 46 mule deer. We used empirical survey data to simulate surveys under three alternative stratification approaches, schemes for grouping strata, and allocations of survey effort and compared the precision and accuracy of the resulting population estimates. Key results We observed deer in 83 of the 100 trials, with detection decreasing with small group size, reduced deer activity, low snow cover, high forest cover and observer fatigue. Survey precision and accuracy were highest when stratification was based on natural breaks, calculated via Jenks optimisation, in the values of resource-selection function (RSF), although improvement was less pronounced for estimates of mule deer abundance. Optimal or equal allocation of sampling effort among strata outperformed proportional allocation of sampling effort. Use of RSF for stratification reduced the coefficient of variation (CV) in estimates of deer numbers from 38% to 23% for white-tailed deer and from 33% to 27% for mule deer compared with past approaches. Conclusions Stratification based on RSF values improved precision of deer surveys the most; however, using even simple measures related to habitat selection can improve population estimates. Where deer are highly aggregated, reliably recording all variables needed to implement sightability models can prove problematic; however, survey detection adjustments are nevertheless important to account for the relatively small, but still significant, proportion of missed animals in open prairie–parkland environments. Implications Field experiments to assess aerial survey design are impractical because of cost. We illustrate how simulated surveys using empirical data can be useful to evaluate alternative survey designs to improve population estimates in a region when high accuracy or precision are needed to address management questions or to calibrate more cost-effective approaches.