关键词: 短蛸, 海州湾, 随机森林模型, 人工神经网络模型, 广义提升回归模型

Abstract: In recent years, a variety of important fishery resources in China’s coastal waters have declined. Octopus ocellatus has the characteristics of short life cycle and rapid growth, with great contributions to fisheries of China’s coastal waters. However, we know little about the habitat distribution characteristics of O. ocellatus and its relationship with environmental factors, which is not conducive to better protection and utilization of its resources. Here, we analyzed the distribution characteristics of O. ocellatus and its relationship with environmental factors using three machine learning methods, i.e., random forest model, artificial neural network model, and generalized boosted regression models, based on the survey data of fishery resources and habitat in Haizhou Bay during spring of 2011 and 2013-2017. Among the three models, random forest model had great advantages in the fitting effect and prediction ability. The model analysis results showed that sea bottom temperature, seawater depth and sea bottom salinity had significant effects on the habitat distribution of O. ocellatus. The relative resource density of O. octopus increased first and then decreased with the increases of sea bottom temperature, seawater depth, and sea bottom salinity. Based on environmental data simulated by the FVCOM model, we predicted the habitat distribution of O. ocellatus in Haizhou Bay using random forest model and found that O. ocellatus was mainly distributed in the area between 34.5°-35.8° N and 119.7°-121° E.

Key words: Octopus ocellatus, Haizhou Bay, random forest model, artificial neural network model, generalized boosted regression models