The question of leech habitat use versus target species habitat described in ‘Occupancy modelling’ points towards an important aspect: the definition of detection probability in leech-based occupancy studies. The most basic definition of detection probability in occupancy models is the probability that the species is detected, given that it is present . However, in any realistic situation, the probability structure is much more complex. For example, if birds are detected by song, the probability of detection constitutes two pieces i) the probability that a bird sings and ii) the probability that the song is detected by an observer and correctly assigned to a species (Fig. 3a-c illustrates the components that contribute to the detection probability for some common survey methods compared to leeches).|
In the case of leech sampling, the series of probabilities leading to species detection is even more complex (Fig. 3c). On the basic level, i) both the leech and target species have to be present at the sampling site. One is not necessarily conditional on the other, but it seems reasonable to suspect some correlation between the two (leeches should not occur where there are no hosts). Leech habitat preferences, for example, feed into the probability of a leech being present at a sampling site. Conditional on both leech and target species being present, ii) the leech then has to feed on the target species, and here, possible leech host preferences can influence detection. On the next level, iii) a collector has to detect the leech. The probability of detecting a leech could be influenced by habitat, weather, time of day and ability of the collector. Conditional on a leech filled with remnants of blood of the target species being collected in the field, then come iv) the lab related probabilities – that DNA can be extracted, amplified and correctly identified to species level. For example, if a reference sequence of a particular species is missing it is impossible to match the sequences obtained from the sample, and thus the species will be not detected.
In the analysis of the resulting species detection/non-detection data, all these levels of the detection process get balled up into a single “detection probability”. It is important to keep in mind the different processes that feed into this parameter, as variation in any of these levels across sampling sites or times or even leeches, can lead to biased results if not accounted for. Fortunately, methods such as occupancy modeling enable detection probability to be modeled as a function of spatial and temporal covariates. This is regardless of where in the process (e.g. from biology of the leech to sensitivity of genetic assay) these occur. For example, covariates could include measurable variables (both continuous or categorical) related to the genetic analyses, such as amount of blood extracted from a leech, measures of DNA quality, etc. These measures, however, may not be as straightforward to obtain when pools of leeches constitute a sample, rather than a single leech.