Spatial Distance Sampling for Wildlife Surveys That Aren't Random

Dall's sheep in Alaska. For them, spending less survey effort in poor habitat is a practical way to cut costs. NPS.

The problem

Distance sampling estimates animal numbers from how far detected animals are from a survey transect. The conventional method assumes animals are spread uniformly relative to the transects — which requires random placement. But random designs are often impractical, especially in mountains, where transects run along elevation contours or follow roads and ridgelines whose habitat may not represent the whole area.

The approach

Full-likelihood spatial distance-sampling models model detection and occurrence together, using habitat information tied to location. Through simulations and a real Dall's sheep survey in Alaska, NPS researchers tested whether the approach stays unbiased when transects aren't random.

Findings

• Unbiased: the approach was generally unbiased, even in extreme cases where habitat quality ran opposite to distance from the transect.
• Cheaper surveys: for Dall's sheep, designs with less effort in low-quality habitat are a practical way to cut logistical costs — if the data are analyzed with a spatial model.
• Broader value: the results confirm earlier work that spatial distance sampling is a useful answer when non-random designs are unavoidable. Since surveys are expensive, valid alternatives to random designs mean more information for more species.

Sources

Based on "Spatial Distance Sampling is Useful with Non-Random Sampling Designs," National Park Service, summarizing Joshua H. Schmidt and W. W. Deacy (2021), "Using spatial distance sampling models to optimize survey effort and address violations of the design assumption," Ecological Solutions and Evidence 2(3); a work of the United States government in the public domain.