Validating the use of generalized additive models and at-sea surveys to estimate size and temporal trends of seabird populations

1. The population size of seabirds is often difficult to estimate from surveys at breeding colonies because of factors such as burrow nesting and nocturnal breeding. The reliability of alternative surveys conducted at sea has not previously been validated. Sample-based estimates from at-sea surveys could be biased and have poor precision due to non-random survey design and the uneven distribution of seabirds at sea. 2. We conducted at-sea surveys of western gulls Larus occidentalis and common guillemots Uria aalge of the Farallon Islands, California, USA, and waved albatrosses Phoebastria irrorata of the Galapagos. The observed counts were modelled using generalized additive models (GAMs), with a correction factor for bird movement relative to the ship included in an offset. The models were used to produce estimates of the size of the seabird populations at sea, which were compared with independent colony-based estimates, adjusted to account for the number of non-breeders. 3. Gull and guillemot populations were estimated separately for each of 10 survey years. Temporal trends were estimated by smoothing through the annual values. The albatross data from 7 survey years were pooled to obtain a single estimate of average population size. 4. The coefficients of variation (CVs) of the annual estimates were approximately 10%, 15-20% and 15-45% prior to smoothing for the albatross, gull and guillemot, respectively. The CVs of the smoothed estimates were about 10% for the gull and 15% for the guillemot. These represent substantial improvements in precision over previous sample-based estimates from at-sea surveys. 5. The colony-based estimates usually lay within the 95% confidence intervals (CIs) of both the annual and smoothed sea-based estimates, showing that the sea-based methods worked well. 6. Synthesis and applications. We conclude that GAMs of at-sea seabird survey data, collected under suitable protocols and corrected for bird movement, can accurately estimate population size. Given sufficient demographic information, these methods can provide a valuable tool for the management of populations that are difficult to census at the breeding colony.