User menu
Three distant-water tuna longline countries, Japan, Korea and Taiwan, have started a collaborative study since December 2019 for producing the joint abundance indices using integrated fishery data of these fleets to contribute to the upcoming stock assessments of yellowfin tuna in the Indian Ocean. The intention is to produce reliable indices by increasing the spatial and temporal coverage of fishery data. In this paper, some preliminary results using data up to 2019 fisheries were provided to update the WPTT on the progress of this activity.
As an underlying analysis, a clustering approach was utilized to account for the inter-annual changes of the target in each fishery in each region. Due to high dimensionality of fishery data with species composition, a two-step procedure was employed. A “K-means clustering” method with a pre-specified large number of initial clusters was firstly applied to fine scale fishery data in order to reduce the dimension of data, and then the aggregated data based on the first step were used in the subsequent “hierarchical clustering”. The whole process was repeated through a certain number of iterations with different random initial clusters to seek a set of the smallest sum of within-cluster variation. The outputs of the finalized cluster were then used to assign the cluster label on fishery target to each catch-effort data.
For standardizing the catch-per-unit-effort data, the conventional linear models and delta-lognormal linear models were employed for data of monthly and 1° grid resolution in each region. In addition to the implicit target species through the clustering, geographical and temporal covariates were used in the regression structures. The models were diagnosed by the standard residual plots and influence analysis. Although the results shown in this paper were still preliminary because of a delay and difficulty in the data-sharing process, a final set of results based on the updated data including 2020 fishery outcomes will be submitted before the upcoming yellowfin tuna stock assessment meeting scheduled in October 2021 for use as inputs for the update of its stock assessment.
Besides these conventional regression methods, analyses using an advanced spatio-temporal model, vector-autoregressive spatio-temporal model (VAST), were attempted for developing abundance indices with additional consideration of spatio-temporal correlations and targets as well as the life stage of yellowfin tuna. In the VAST analysis, the convergence was not achieved enough when aggregating the three fisheries data yet, but the codes were developed well and ready to use for the finalization of results.
As other future works, the regional scaling will be applied for the conventional regression models so that a constant catchability can be assumed across the regions in the stock assessment models. The regional trends in the standardized CPUE are then compared to those from the VAST analysis, where catchability is constant by default and the regional scaling is not required.
