Simulations of IDCR and SOWER Antarctic minke whale abundance surveys

Abstract

The International Whaling Commission (IWC) conducted abundance line‐transect surveys of Antarctic minke whales (Balaenoptera bonaerensis) during the International Decade of Cetacean Research (IDCR) between 1978/79 and 2003/04. These surveys continued throughout 2004/05–2009/10 under the Southern Ocean Whale and Ecosystem Research (SOWER) programme. Several methods used to analyse these data resulted in very different abundance estimates. In order to evaluate the accuracy of these methods, the IWC Scientific Committee recommended using simulated line‐transect data. The software package TRANSIM was therefore modified to generate simulated line‐transect data that emulated an IDRC/SOWER survey in a study area populated with whales that followed the distribution, density and behaviour of Antarctic minke whales. Simulations incorporated factors that could potentially bias resulting abundance estimates, and in order to investigate the effects of these factors on the resulting abundance estimates, simulated datasets were generated for different scenarios, each of which incorporated combinations of factors that were chosen using the mirror‐image fold‐over partial factorial method that is designed to identify factors related to biased estimates. This paper describes the programme used to generate these datasets and reports on the robustness of the statistical methods applied to analyse them. In addition, a brief discussion is included on how this information was used when the actual line‐transect data were analysed. The statistical methods investigated were the hazard probability method (OK), integrated hazard probability method (IM), spatial point independence method (SPL) and standard method. After developers analysed the simulated data, they used this information to improve their analytical methods. The results reported here focus on the performance of the most advanced versions of the OK and SPL methods compared with the standard method as of 2009. In general, SPL estimates of group density were negatively biased, OK estimates were positively biased, standard estimates were the most negatively biased and IM estimates were the least biased. Density estimates from SPL were less biased than estimates from OK in the complex scenarios that contained spatial gradients and correlations between density, location and weather conditions. The OK method was less robust when: (1) closing‐mode data collection procedures were simulated; (2) whale groups were diving in a non‐synchronous fashion; and (3) there were complex spatial patterns. In almost every case, the density estimates from the OK method were greater than the SPL methods, on average by 10–15%. These observations assisted the development of more robust analysis methods and highlighted which factors should be addressed when analysing the actual data.