Chapter 6: Defining the detection unit
The detection unit is the building block for the SCR models. The choice of detection unit determines what type of density model we estimate. We have learned in the previous chapters that if individual animals are the detection unit, we estimate animal density with the SCR model (number of animals per hectare). If groups are the detection unit, we estimate group density (number of groups per hectare). If calls are the detection unit, we estimate call density (number of calls per hectare per time unit).
Gibbons generally occur in small territorial groups which engage in duetted singing, generally by the dominant breeding pair of the groups from within the boundaries of their territories. This enables us to survey gibbons acoustically rather than detecting them visually. It also enables us to attach calls to individual groups (or territories) and track groups across occasions. As a result, we have various options on how to define the detection unit. The main choice is between groups or calls; however, if we choose groups, we further have to choose between groups during a single sampling occasion or during multiple occasions.
Additonal difficulties arise due to that not all groups sing every day and that each group differs in the amount of singing they engage in both between groups and within a group between days. This leads us to having to estimate additional parameters besides the SCR model parameters if the interest is to estimate absolute density of groups. What sort of parameters depends on the choice in detection unit we make.
The following decision tree summarises the options. The four topics depicted in green are covered in the four lectures in this chapter, one of these (attaching calls to the groups) relates to localising calling animals and the other three relate to the additional parameters that need to be estimated.
Decision tree for defining the detection unit
Groups as the detection unit
6.2 Lecture: Attach calls to groups
Surveying gibbons generally relies on detecting them acoustically rather than visually. When choosing groups of gibbons as a detection unit, the individual calls that were detected during the sampling occasion need to be attached to individual groups. This is generally done via triangulation of individual calls from the different groups.
We briefly explain how triangulation of a single call that was detected by three detectors works using the angle recorded along with each detection.
We use a short example of 21 call detections made by three detectors to illustrate how these are matched between detectors and how to triangulate the location of the call (or the location of the animal when it made the call). Using prior knowledge about territory size, boundaries are then drawn around groups of located animals based on proximity to each other. The individual group territories located through this process are the detection units and form the basis for the capture history.
We demonstrate how this process can potentially create biased results (e.g. wrong number of territories, territories in the wrong location) by small errors in the angles of the call detections recorded in the field, highlighting the importance of accurate angle measurements.
Accounting for the silent groups
Not all groups sing every day. If we survey groups during a single occasion, we effectively only survey the groups that sing during that occasion and our density estimates from the SCR model will be negatively biased. We can alleviate this issue in two ways:
When using groups during a single occasion as the detection unit, we need to estimate the proportion of occasions on which a group calls. But we do not need to track groups across occasions. See chapter 6.3 for more details.
When using groups during multiple occasions as the detection unit, SCR estimates the proportion of occasions on which a group calls. It is part of the g0 estimate. However, we need to be able to link groups across occasions reliably. See chapter 6.4 for more details.
Groups during a single occasion
6.3 Lecture: Estimate the proportion of occasions that groups call
When choosing groups during a single occasion as the detection unit the SCR model fitted to the data will only represent an estimate of the groups that called during that occasion. There is no information in the survey data that we could use to account for this potential bias in absolute density. Hence, we need to collect additional data that allows us to estimate the proportion of occasions during which groups call. This allows us to adjust our density estimate from the SCR model so that we obtain an estimate of absolute density. In this lecture we present one option for how these data may be collected.
The example we present includes focal follows of the groups detected during the sampling occasions for several additional occasions. Here, focal follows refer to following the groups during the same time frame as the SCR sampling occasion on several additional days and recording whether the groups call. Additional options exist - please refer to the published literature for these.
Groups during multiple occasions
6.4 Lecture: Track groups across multiple occasions
When using groups during multiple occasions as the detection unit, we need to track groups across occasions. This method relies heavily on mapping territories correctly for each occasion which we described in a previous lecture. It further entails matching territories between sampling occasions. For example, if group 1 was detected during the first occasion, we need to determine if this group was detected again for each additional occasion. Difficulties can arise, for example, if territories were mapped inaccurately for the individual occasions resulting in poor overlapping of mapped territories between occasions.
Using this method, a g0 estimate of <1 from the SCR model may be the result of either or both reasons that groups may have been missed near the detector: 1. not all groups were calling during the occasion and 2. not all calling groups near the detector were detected.
Calls as the detection unit
6.5 Lecture: Calls
When using calls as the detection unit, we do not need to attach calls to individual animals or to groups. We also do not need to identify different groups or territories via triangulation. Here, the individual calls form the basis of the capture history and the SCR analysis.
We use the same 21 call detections as in the previous lecture and build the capture history from it, now using calls as the detection unit. This exercise illustrates the point that we do not need to know which animal made which call. However, we do need to be able to match calls between detectors.
Now, the density estimate from the SCR model is call density. If we are interested in obtaining an estimate of group density, we need to estimate the call rate, i.e. how many calls does a group make on average during the sampling occasion.
We learned in a previous lecture that gibbons make a variety of different calls and engage in singing bouts. Hence, we need to define the detection unit carefully, e.g. choosing only great calls of the female (if applicable to your study species), and calling rate needs to be estimated appropriately (e.g. how many great calls does a group produce on average during the sampling occasion).