Humans often look at wild places and guess animals are either abundant because they see large numbers of animals or animals are limited because they observe low numbers or little sign of activity. In reality, our estimates of animal numbers may be limited because of our inability to accurately detect animals and predict habitat occupancy or persistence over different seasons. Scientists and land managers are tasked with gaining a better understanding of species by predicting population size based upon a limited number of observations or counts recorded over a set period of time. Since there are limited funds and time to make complete estimates of animals, successful conservation efforts often require population estimation using computer models that take into account sampling effort and the probability of detecting the desired species.

Improving Population Estimates by Reducing Sample Bias:

In this study, scientists used computer simulations to evaluate the population dynamics of migratory songbirds in the Upper Mississippi River National Fish and Wildlife Refuge in the Midwest. These so-called “dynamic occupancy models” estimated populations of birds in various rates of decline. Four population-level parameters were used over multiple seasons and multiple site visits within a given study area. The four parameters included: the probability of detection during a given survey given the species of interest is present at a site in a particular season (detection probability); the probability that a site is occupied by the species of interest in a particular season (occupancy probability); the probability that an occupied site from the previous season was still occupied in the present season (persistence probability); and the probability that an unoccupied site in the given season becomes occupied in the subsequent season (colonization probability).
Model results indicate bias in estimates of species occurrence decreases with increasing sampling effort, i.e., increase in number of sampling sites per season and visits to those sampling sites. The model demonstrated a roughly linear association between bias and probability of site occupancy, with this probability being overestimated at low population levels and underestimated at high population levels. Colonization and extinction biases were dependent upon probabilities of detection and occupancy.

The most influential factor in estimation of all four parameters was low detection probability. This is good news, because this is the single parameter that may be modified by researchers. As examples, detection probabilities may often be increased by 1) surveying for a greater period of time (e.g., 10 min instead of 5 min per visit), 2) surveying during ideal weather, 3) choosing an optimal time of day, or 4) using experienced observers, relative bias of population estimation will decrease (Fig. 1).

Figure 1: The probability of at least one positive detection as a function of the detection probability p and number of visits per site. For 5 visits, when p=0.5, the probability of at least one detection is 0.97. When p=0.1, the detection probability drops to0.41.

For more details on model results follow link.

Science Impact:

Researchers or resource managers intending to study the occupancy dynamics of a species need to be aware of how much data these models require to achieve accurate and/or reasonably precise occupancy estimates. A rare or hard to detect species will require much more field work to obtain enough data to obtain accurate estimates than will a common or easily detectable species. A simplified recipe for study design would be to estimate the detection probability, and then determine the number of visits necessary to give a site probability near 90%. Then, the current study suggests that bias in occupancy probability will be largely eliminated with 60 sites; increasing the number of sites to 120 will yield improved estimation of probabilities of colonization and extinction.

For more information, see
McKann, P., B. R. Gray, and W. E. Thogmartin. 2013. Small sample bias in dynamic occupancy models. Journal of Wildlife Management 77:172–180.