基于随机森林的印度洋长鳍金枪鱼渔场预报

为了提高远洋渔场预报水平和满足渔业生产的需要,提出了一种基于随机森林建立印度洋长鳍金枪鱼(Thunnus alalunga)渔场预报模型的方法。选取2002-2009年各个月份印度洋5°×5°格点渔业环境和时空数据(包括海表温度、叶绿素a浓度、表温距平、叶绿素a浓度距平、海表温度梯度强度和海面高度异常等数据)作为预测变量,利用长鳍金枪鱼的CPUE(Catch per unit effort,单位:尾/千钩数)的三分位点将渔区划分为高CPUE、中等CPUE和低CPUE三种类型,作为响应变量,对数据进行训练。结果表明,当随机森林中决策树达到100以上时,袋外数据OOB(out-of-bag)的分类误差率趋于平稳。将训练得到的随机森林用于2010年印度洋长鳍金枪鱼分月渔场的预测,其概率等值面图与实际生产的渔场分布进行叠加比较,显示高CPUE渔场概率分布与实际渔场的位置及范围变化情况符合。通过ROC(Relative Operating Characteristic)分析,高CPUE、中等CPUE和低CPUE的AUC(Area Under ROC Curve)分别达到0.847、0.743和0.803,表明预测精度较高。最后对中等CPUE渔区预测精度相对较低的原因进行了分析。

Fishing ground forecasting of Thunnus alalung in Indian Ocean based on random forest

To improve the forecasting accuracy of pelagic fishing ground and meet the needs of the practical fishery production, a method of albacore fishing ground forecast model in the Indian Ocean based on random forest was proposed.The Indian Ocean fishery environmental and spatio-temporal data by 5°×5° grid of every month from 2002 to 2009 were taken as predictor variables and the CPUE(catch per unit effort,unit∶inds per 1 000 hooks)of albacore,classified into high CPUE, moderate CPUE and low CPUE by tertile, was selected as the response variable in the random forest training. The training result indicated that the OOB(out-of-bag)misclassification rate tends to be steady when the number of decision trees reached 100. The random forest obtained through the training was applied to the albacore fishing ground monthly forecast in 2010.The isosurface chart on forecasted probabilities was overlapped on the practical fishery production and a comparison was made between them.The result demonstrated that the high CPUE fishing ground probability distribution and the CPUE of actual fishery ground tally well. Through the ROC analysis, AUCs (Area Under ROC Curve)of high CPUE, moderate CPUE and low CPUE were 0.847,0.743 and 0.803 respectively, indicating that the forecast precision was high. Finally the reason of the relatively low precision of medium CPUE probability in the forecast was analyzed.