Departmental Bulletin Paper An Evaluation of Initial Weighting Factors for Length Composition Data Used in Integrated Stock Assessment Models
トウゴウ シゲン ヒョウカ モデル ニ モチイル タイチョウ ソセイ データ ノ オモミヅケ ショキチ ヒョウカ

芝野, あゆみ  ,  金岩, 稔  ,  石原, 幸雄  ,  氏, 良介  ,  志村, 健  ,  竹内, 幸夫  ,  余川, 浩太郎  ,  Ayumi, Shibano  ,  Minoru, Kanaiwa  ,  Yukio, Ishihara  ,  Ryosuke, Uji  ,  Tsuyoshi, Shimura  ,  Yukio, Takeuchi  ,  Kotaro, Yokawa  ,  東京農業大学生物産業学研究科生物産業学専攻  ,  東京農業大学生物産業学部アクアバイオ学科  ,  鳥取県水産試験場  ,  鳥取県水産課  ,  鳥取県水産課  ,  水産総合研究センター国際水産資源研究所  ,  水産総合研究センター国際水産資源研究所  ,  Department of Bioindustry, Graduate School of Bioindustry, Tokyo University of Agriculture  ,  Department of Aquatic Bioscience, Faculty of Bioindustry, Tokyo University of Agriculture  ,  Tottori Prefectural Fisheries Experimental Station  ,  Tottori Prefectural Fisheries Division  ,  Tottori Prefectural Fisheries Division  ,  National Research Institute of Far Seas Fisheries  ,  National Research Institute of Far Seas Fisheries

61 ( 1 )  , pp.17 - 30 , 2016-06-20
In stock analysis using integrated age-structured models, weighting factors of each likelihood component (e.g., for abundance indices and size composition data) directly affect the estimation of the model. In a past stock assessment of Pacific bluefin tuna (PBF) conducted in 2012, conflicts were recognized between the length composition dataset of a purse seine fishery operating in the Sea of Japan (PS-SoJ) and the abundance indices derived from other fisheries. After careful consideration, the initial weighting factor used in iterative reweighting method for the length composition dataset of PS-SoJ was reduced. In this paper, we provide a further analysis of procedure for determining the initial weighting factor of the likelihood component for the length composition dataset of PS-SoJ used in the stock assessment model of PBF. We tested five scenarios involving alternative initial weighting factors. Firstly, we estimated an effective sample size as an initial weighting factor considering inherent accuracy and precision of the length composition dataset in the context of cluster sampling. Next, we illustrated the partial likelihood of estimated key parameter (i.e. R0) as indicator of model fit for the tested scenarios to suggest optimal methods. In the results, the model fit was improved in a scenario where the point estimate of the effective sample size except for highly uncertain years was used. The differences between the statistical point estimates of stock dynamics in each scenario would have significant effect on management considering PBF’s situation. In conclusion, we discussed appropriate methods of setting initial weighting factors for a variety of data conditions.

Number of accesses :  

Other information