This report describes design-based and model-based estimates of abundance for a group of seabirds who reside in large aggregations (scoters) and another group of seabirds that are less consistently clumped (‘gulls’) from three types of aerial surveys in Carmarthen Bay. Each type of survey was conducted on four dates during the month of March 2009 using different amounts of effort.
This work explores patterns in estimates derived from different data collection and data analysis methods for both types of seabird groups. The work follows on from an examination of a portion of this data set conducted by Rexstad and Buckland (2009).
No clear pattern emerged from examining the point estimates for the various survey-analysis method combinations. We have no measure of the true sizes of either of these populations against which to measure the point estimates. However we can assess the relative precision of these combinations using the coefficient of variation (ratio of the standard error of the point estimate to the point estimate itself).
The clumped distribution of scoters resulted in estimates with poor precision for most of the survey type-analysis method combinations. The average coefficient of variation across all survey-analysis combinations was slightly smaller for gulls (0.45) than for scoters (0.50). The range in these average coefficients of variation ran from 0.25 for design-based estimates from visual data to 0.85 for model-based estimates of digital still transect data. Among survey-analysis combinations for gulls, design-based visual and model-based digital video produced estimates with the highest average precision.
Coefficients of variation were higher for scoters. The only survey-analysis method that produced an average coefficient of variation smaller than 0.40 was the model-based estimate of digital video data, perhaps partially a consequence of the larger number of transects used for this survey method. However the range in average coefficients of variation was smaller for the scoter data than for the gull data. The similarity in precision of design- and model-based estimates for all survey types was higher for the scoter data than for the gull data, perhaps reflecting movement in and out of the study area during survey days by gulls.
Design-based estimators produced on average lower coefficients of variation for visual and digital still transect data for both gulls and scoters, but model-based estimators produced more precise estimates for digital video data.
Two points emerge from this comparison. First, an approach based on estimating calibration factors to make estimates from different approaches comparable is unlikely to be useful. The Carmarthen Bay surveys were conducted under controlled conditions with close attention to detail such as carrying out the surveys on the same days, at approximately the same time of day. Even under these conditions, estimates for the different methods were very variable, with little indication of any consistent differences between methods. Second, for a relatively small body of water such as Carmarthen Bay, application of a pro forma survey design such as ‘place transects at 2km intervals’ does not produce estimates with sufficient precision to detect change. There is no substitute for bespoke survey designs created with knowledge from pilot surveys so that sufficient effort can be deployed to produce defensible estimates. This insufficient amount of effort was highlighted for the common scoter, whose highly aggregated distribution would have necessitated even higher coverage of the study region.