Definitions and assumptions

Distance sampling in the field

Collecting field data for distance sampling involves:

  1. Surveying multiple line transects or points across a landscape
  2. At each sighting of your target species, record:
    1. Number of animal or sign seen (group size)
    2. Compass bearing to the sighting
    3. Estimated distance from you to the sighting location

Surveys can occur:

  • Within a single season, or
  • Across multiple seasons, if you’re interested in changes in density through time, or variations in density at different times of year

Surveying for sign

As well as live animals, you can survey for animal signs such as:

  • 🐾 Tracks
  • 🐣 Nests
  • 💩 Dung

Translate signs to density

If you’re estimating the density of animals signs, you need to combine that with information on sign creation and decay rates to estimate population density

Encounter rate

Encounter rate is the number of sightings per unit of distance you walk along your transects

For example:

  • You walk three 2km transects (a total of 6km)
  • You observe 24 animals during that survey
  • Encounter rate would be 24 sightings divided by 6 km = 4 animals per kilometre walked

Encounter rates will obviously vary between species and sites, and can also be affected by many other variables

Encounter rates are useful because they provide us with information on the efficiency of our survey technique.  We learn more about encounter rates in the Survey design course, as we use them to help us determine the distance that we need to survey in order to collect sufficient sightings for analysis

Replicates versus repeats

When analysing line transect or point count data, it’s important to distinguish between replicate and repeat surveys

Replicates are transects in different locations, and therefore sampling a different part of the population

The number of replicate transects is your total sample size

Repeats are return visits to the same transect

By repeating a survey of the same transects you collect more sightings to help construct your detection function, but you are not increasing your sample size or the precision of your density estimates

We learn more about replicates versus repeats in the Analysis and Survey design courses

Assumptions: Random placement & certain detection

Distance sampling relies on four assumptions for trustworthy results

Random placement: Transects are randomly placed with respect to the distribution of animals

The animals themselves do not need to be distributed at random, as long as the transects are randomly placed in the environment, and do not follow features such as roads that may themselves influence the animals’ locations

Detection is certain: You will always detect animals (or sign) that are directly on the transect line

In other words, detection probability on the transect line itself is 1

Assumptions: No movement & accurate measurements

No movement: The animal is detected before it moves in response to the observer

Your record of the animal must match its true location during the survey, so that you can accurately model the pattern of detectability at different distances from the transect.  Movements in response to the observer before detection will make your density estimate less reliable

Accurate measurements: Distances and angles to the animal are measured accurately

Alternatively, animals are assigned to the correct distance interval, for example 0-5m, 5-10m etc

Environmental covariates

Covariates are environmental variables that influence density and/or detection of your species

Covariates can be:

  • 🌐 Spatial variables that affect:
    • Density, e.g. land-cover class, prey density, or
    • Detection of the species, e.g. vegetation, terrain, or
  • Temporal variables that affect:
    • Detection, e.g. weather conditions, observer experience

To meet the assumptions of distance sampling, probability of detection and density must either be:

  • Constant across all of your transects, OR
  • Variable in a way that can be explained by the covariates you include in your model