南美沙丁鱼资源量影响因子的选择及预报模型比较

Selection and Forecasting Model of Marine Environmental Factors Affecting Sardinops Sagax Abundance

根据1970—2012年智利捕捞南美沙丁鱼的年产量数据,以及厄尔尼诺指数、海平面气压、南方涛动指数等海洋环境和气候因子共9个数据,采用相关性分析确定影响南美沙丁鱼资源量的主要海洋环境和气候因子;利用多种神经网络模型对南美沙丁鱼资源量与相关性分析选取的主要海洋环境和气候因子进行建模拟合,预测南美沙丁鱼资源量。通过对17种不同神经网络模型的研究,以拟合残差、偏差解释率、预报标准差3个因素综合分析,确定南美沙丁鱼资源量的最优预报模型。研究表明,结构为10-8-1的神经网络模型拟合残差仅为0.003 8,偏差解释率高达98%,预报标准差为21%,可作为南美沙丁鱼资源量的预报模型,该研究结果可为预测南美沙丁鱼资源动向提供依据。

Sardinops sagax is an important oceanic species, which is widely distributed in the southeast Pacific Ocean. According to the annual production data of South American sardines during 1970-2012 from Chile, and nine kinds of marine environmental and climate data such as EI-Nino, Sea level pressure, Southern oscillation index, the correlation analysis is used to determine the main factor affecting sardines resources in South America. And a variety of neural network structure are used to build the forecasting model for sardines resources. Based on the results of simulation residual, deviation rate, forecast standard deviation are explained from 17 different structure of neural network, and the optimal prediction model of sardines resources in South America will be determined. The results indicate that the simulation residual obtained from the neural network structure for the 10-8-1 model is only 0.0038, and the explained deviation rate is as high as 98%, and the standard deviation is 21%, which means that it can be used as a prediction model to estimate sardines resources in South America. This model will provide a basis for predicting sardines resources in South America and for sustainable development and utilization of this resources.