Using the gridded fishery data to estimate the habitat preferences of bigeye tuna (Thunnus obesus) in the Indian Ocean is challenging, as it is still not clear what type of model is appropriate to make reliable habitat predictions. In this study, we tested two classes of habitat models: Generalized Additive Models (including Gaussian distribution GAM, Poisson distribution GAM, Negative Binomial distribution GAM, Tweedie class distribution GAM, and Zero-inflated distribution GAM) and Maximum Entropy Model (MaxEnt) for forecasting the habitat of bigeye tuna in the Indian Ocean, using the 5°×5° monthly gridded fishery data from 2008 to 2015 provided by Indian Ocean Tuna Commission (IOTC) and the environmental factors (i.e., the vertical temperature, salinity, dissolved oxygen concentration, thermocline depth, vertical shear of ocean current, concentration of sea surface chlorophyll-a (chl-a), and the eddy kinetic energy (EKE)). We compared the models’ fitting ability, predictive performance, predicted results, and the ecological interpretation within the models. The results showed (1) GAMs provided better model fit and predictive performance than MaxEnt; (2) GA-GAM showed the best model fit performance by using the log transformation of the standardized CPUE as response variable; (3) TW-GAM was suitable for predicting the habitat of adult bigeye tuna with the over-dispersed fishery datasets. (4) Results suggested that the vertical temperature, dissolved oxygen concentration, and thermocline depth could be used as reliable predictors in forecasting the spatial-temporal distribution of the adult bigeye tuna in the Indian Ocean.

中文翻译：

---

基于网格化金枪鱼延绳钓渔业数据的印度洋成年大眼金枪鱼栖息地模型比较研究

利用网格化渔业数据估计大眼金枪鱼（Thunnus obesus）的栖息地偏好) 在印度洋具有挑战性，因为目前尚不清楚哪种类型的模型适合进行可靠的栖息地预测。在本研究中，我们测试了两类栖息地模型：广义加性模型（包括高斯分布 GAM、泊松分布 GAM、负二项分布 GAM、特威迪类分布 GAM 和零膨胀分布 GAM）和最大熵模型 (MaxEnt)利用印度洋金枪鱼委员会（IOTC）提供的 2008 年至 2015 年 5°×5° 月度网格渔业数据和环境因素（即垂直温度、盐度、溶解氧）预测印度洋大眼金枪鱼的栖息地浓度，温跃层深度，海流的垂直切变，海面的浓度受叶绿素一个（chl-一个) 和涡动能 (EKE))。我们比较了模型的拟合能力、预测性能、预测结果以及模型内的生态解释。结果表明 (1) GAMs 提供了比 MaxEnt 更好的模型拟合和预测性能；(2) GA-GAM通过使用标准化CPUE的对数变换作为响应变量表现出最佳的模型拟合性能；(3) TW-GAM适用于过度分散的渔业数据集预测成年大眼金枪鱼的栖息地。(4) 结果表明垂直温度、溶解氧浓度和温跃层深度可作为预测印度洋大眼金枪鱼成鱼时空分布的可靠预测指标。