Can we really use design-free hydroacoustic data from fishing vessels for assessing abundance and distribution of marine species? A proof of concept analysis on Antarctic krill

• Antarctic krill is assessed by design-based acoustic surveys with low spatiotemporal coverage. • Design-free data from fishing vessels is collected extensively but rarely used in management. • Spatiotemporal models fitted to design-free data overestimate krill abundance. • Parallel or orthogonal transects performed by fishing vessels would increase accuracy. Euphausia superba is a well-known Antarctic crustacean of great economic and ecological importance, whose management requires accurate and precise abundance and distribution estimates. Such estimates are difficult to achieve given the remoteness, extension, and large spatio-temporal variability of its geographic distribution. Acoustic data collected on board krill fishing vessels during normal fishing operation has a great potential to enhance such abundance and distribution estimates. In the present work we test the hypothesis that design-free hydroacoustic data collected during regular fishing operations can be used to produce abundance and distribution estimates with similar accuracy and precision than design-based scientific surveys. Thus, we produced and compared distribution and abundance estimates produced using either design-free hydroacoustic data collected during regular fishing operations or design-based data from scientific surveys conducted off the South Orkney Islands during summer 2017 and 2019. Following a Bayesian geostatistical approach that considered and fitted simultaneously the spatial and temporal correlation of the data, we tested different auto-correlation structures and selected the most informative models. The comparison included the means and coefficients of variation (CV) of the probability of presence (p), conditional density (d) and relative abundance index (RAI) estimates. In addition, we also simulated scenarios of parallel and orthogonal transects and obtained RAI estimates from each scenario to compare with design-based and design-free estimates for each year. In 2017, the mean RAI estimated using design-free data (94 421 m 2 ; CV: 14 %) was ∼ 50 % higher than the one estimated with design-based data (60 232 m 2 ; CV: 42 %), both within the fishing area. In 2019, the mean RAI estimated using design-free data (509 413 m 2 CV: 6 %) was ∼ 5-fold higher than the one obtained using design-based data (113 654 m 2 ; CV: 33 %) in the same area. Design-free RAI estimates were highly sensitive to extrapolating the inference area from fishing to the high-density sub-area. On the other hand, changing from an hourly-resolved spatio-temporal model to a purely spatial model resulted in neglectable changes. Despite observed differences in mean estimates, both methods identified similar areas of high presence and density of Antarctic krill north and north-west of the South Orkney Islands. The 2017 estimate from design-free data was probably affected by a larger dispersion of krill, and a less observed effective area during regular fishing operations. Our results show that despite using state-of-the-art methods for processing and analyzing design-free, acoustic data collected by the fishing fleet, it still yielded unreliable RAI estimates. The bias and uncertainty related to design-free data were reduced when parallel or orthogonal transects were applied although orthogonal transects yielded results with increased accuracy as they were only 21 % lower and 0.02 % higher than the true value in 2017 and 2019, respectively. Other possible approach to minimize bias would be integrating hydroacoustic information from multiple vessels.