The National Institute of Fisheries Science in Korea is developing marine mammal bycatch reduction devices (BRDs) for high-risk fishing gear, such as trawls. We experimented with two BRD types (guiding nets) attached in front of codend at 30° and 45° angles, and catch losses and mesh size selectivity were investigated. Experimental fishing operations were conducted along the East Coast of South Korea where whales and dolphins are commonly sighted. The catch was classified according to fishing location, BRD type, codend, and covernet, with measurements recorded for body length, maximum girth, and weight. The average selectivity for each haul was analyzed using the ‘selfisher’ package. The catch loss rates with the BRD attached at tilt angles of 30° and 45° were 11% and 29% for common flying squid, 6% and 28% for sailfin sandfish, and 5% and 8% for pearlside. While the mesh selectivity rates for common flying squid and pearlside remained at 0.2–0.5 across all lengths and tilt angles, the mesh selectivity curve for sailfin sandfish was estimated. There were significant differences in catch loss between 30° and 45° angles, with the 30° angle being more effective in catch loss. We observed a masking effect in the codend.

The National Institute of Fisheries Science in Korea is developing marine mammal bycatch reduction devices (BRDs) for midwater trawl gear. In this study, we tested two BRD-type guide nets (inclined net panel) with 30° and 45° tilt angles to prevent marine mammals from reaching the codend and facilitating their escape from the net. Fishing operations were conducted along the east and south coasts of South Korea, and cameras were installed in front of the BRDs to monitor their performance. The catch loss of herring with the 30° guide net was 13% and 11% in number and weight. The catch loss of hairtail was 53% and 51% in number and weight with the 30° guide net. Mackerel showed a 97% catch loss in number and weight with the 45° guide net. The 30° guide net resulted in lower catch loss for rudderfish and jack mackerel compared to the 45° guide net. The jellyfish discard rate of the BRD was 5% and 7% in number and weight with the 30° guide net and 12% and 11% with the 45° guide net, indicating that the 30° guide net was more effective at discarding jellyfish. Mesh selectivity was not strongly related to target species body length.

Marine mammal Bycatch Reduction Devices (BRDs) can result in the escape of target fish species. In the present study, a modified single-layer codend configuration developed from previous studies was evaluated using BRDs with guide net angles of 30° and 45°. Catch was collected separately from the codend and the BRD cover to estimate catch loss rates and characteristics. For caught species, catch loss rates based on the number and weight of individuals were 9% and 13% for herring with the 30° guide net and 16% and 18% with the 45° guide net, respectively. For Korean pomfret, catch loss rates were 24% and 23% at 30° and 4% at 45°. These results suggest that further modification of BRD is necessary. In contrast, catch loss rates of mackerel with the 30° guide net reached 56%, whereas hairtail showed loss rates of 60% and 65% with the 45° guide net. Underwater observations indicated that catch loss was influenced by swimming behavior and entanglement with the guide net. Despite the modification of the codend to a single layer, the present study still observed high catch loss rates. However, no bycatch of marine mammals occurred during the experimental operations.

Guide-net marine mammal bycatch reduction devices (BRDs) are intended to reduce protected species interactions in trawl fisheries, but their use may also cause catch loss of commercial fish and cephalopods. We restructured June and September 2024 sea trial records into a final six-taxon BRD/COD modeling dataset containing 5641 individual records from 24 unique haul clusters. Logistic GLM, random forest, and XGBoost models were evaluated using haul-grouped cross-validation, and random forest SHAP and permutation analyses were used as exploratory variable contribution diagnostics. Model discrimination was modest and split-sensitive, but Brier score and accuracy were interpreted cautiously because the catch loss prevalence was 0.163 and trivial baselines were strong. Body length and headline height were the leading non-species contributors, while nominal guide-net angle remained a relevant design variable but ranked below several biological, tow condition, and realized gear geometry variables. Additional haul-clustered GEE sensitivity analyses treated nominal guide-net angle as a two-level categorical factor (30°/45°). Across four GEE specifications, the 45° angle term was consistently positive, but confidence intervals were wide, p-values were not significant, and the full species-adjusted model showed suspected quasi-separation in the hairtail term. These results indicate that nominal angle contribution cannot be cleanly isolated from species composition and realized gear geometry in this field dataset. We therefore interpret guide-net angle as an exploratory predictive contributor rather than as a causal factor of catch loss.