Towards developing thresholds for waterbirds that take into account turnover

To attain international importance and thus protection as a Ramsar site or as a Special Protection Area (SPA) a wetland site must either “regularly” support at least 20,000 waterbirds or seabirds, or 1% of the individuals of a population of a species or subspecies of waterbird. In most cases, sites have been designated by using the maxima of individual counts. These counts will underestimate volume (i.e. total number) of birds passing through the site if turnover of birds occurs. Using count data, observations of individually marked birds and survival and recruitment mark-recapture models, we present three different methods (V1, V2 & V3) implemented in the StopOver Duration Analysis or SODA program (Choquet & Pradel 2007) for estimating the total volume of birds passing through a site. We use simulated data to determine their performance using both biased and unbiased data. Specifically, we tested whether the estimates of volume were biased where the following parameters varied: proportion of birds marked, daily resighting rate, timing of arrival, proportion of transients in the population, heterogeneity in the resighting rates (i.e. some individuals with a high or low resighting rate), variation in arrival and stopover time and count error. With a relatively simple dataset (single arrival, no biases), the proportion of individuals marked had little effect on the reliability of the resulting volume estimates for both V1 and V3. Estimates of volume from V2 were always overestimated. The major factor that caused a small positive bias in V1 and V3 was the resighting probability. Lower resighting probabilities caused a small positive bias in the volume estimates. Resighting heterogeneity (i.e. some birds more likely to be seen than others) caused a substantial positive bias for all estimators. Transience (i.e. some birds stopping over for shorter time than others) caused no bias in V1 and V3, but a strong negative bias in V2.Transience seemed to reduce the positive bias due to heterogeneity in V1 and V3 when both were present. The use of trap-dependent models (i.e. those that allow individuals to have differential recapture rates) showed some promise for V3 as little bias in the volume estimate was observed when there was a moderate amount of variation in individuals’ resighting rates. V1 & V3 performed well under scenarios of varying arrival and stopover duration as well as where error in the counts was introduced. V2 was consistently biased (see Table 4.1) The V3 method performed well and consistently had the highest precision; it is the method we recommend to use to estimate volume. It is important that goodness of fit tests are used to determine biases in the data and appropriate models are used in Program SODA. Although some biases in the data have little effect on the resulting volume estimates, care must be taken when setting up a study to reduce bias. We present eight different ways of ensuring that bias is reduced during the collection of data. Practical ways to deal with biases are discussed. Recommendations (see section 4.2 for further details) are to: (i) Count at the same time as reading colour rings; (ii) Count at approximately one-third of the length of stay interval, e.g. if the species is thought to stay ten days on a site during passage then count every 5 days; (iii) aim to resight > 30 individuals during every count period, although preferably more; obtain as far as is possible representative samples of the population being studied; (iv) the timing of marking of the study species, the number of sites included, and the timing of counts is discussed.