| Chub mackerel(Scomber japonicus) is a coastal pelagic fish, widely distributed in the northwest coast of the north Pacific. By the late 1990’s, the annual catch of chub mackerel riches more than 300 thousand tons, and chub mackerel become one of the main economic species in China coastal waters. Mackerel is a migratory fish, the migration path of chub mackerel is seasonal, location of fishing grounds and migration routes is closely related, but also by the impact of changes in the marine environment, showing a greater inter-annual variability. Accurate forecasts of the fishing grounds can guide the lighting purse seine fishery enterprises arranging the production, reducing fishing time and the fishing cost then enhancing the fishing yield, which is very important to the large type lighting purse seine mackerel fishery.So this paper aim at the mackerel fishing ground forecasting model, based on the summary of existing fishing condition analysis and fishing ground forecasting model the regression tree model is introduced. The paper systematically introduced the boosting regression tree’s model construction, algorithm realization, and the choice of model parameters etc. The paper take chub mackerel in the East China Sea and the Yellow Sea as an example to construct the fishing ground forecasting model based on the boosted regression trees. On the one hand extends the theory and method of the existing fishing condition analysis and fishing ground forecasting model, and provide theoretical support for mackerel fishing ground forecasting and fishery resources management. On the other hand, the research covers three aspects of the data model, fisheries science basis and prediction model, and can provide reference to the squid and tuna fisheries. The main research results are as follows:The study found that the large mackerel purse seine fishing boats mainly working in the East China Sea from July to September, and in the Yellow Sea from October to December every year in 2003~2011. The coefficient of variation of each month was greater than 0.3, indicating the variants of the total catch of the same month. In the East China Sea, average catches was the highest in August, and minimum coefficient of variation; and the lowest in July with coefficient of variation exceeded 0.5. In the Yellow Sea, the average yield in November was the highest with minimum coefficient of variation; and October’s catch was higher but the coefficient of variation greater than 0.6. Average catch of December was the smallest. This indicate that August and November were the stable production period of large light purse seine fishery in the East China Sea and the Yellow SeaIn the East China Sea, the catch of 2003 was the highest, 2011 had the minimum catches. In the Yellow Sea, the catch of 2008 was the highest, and 2006 the lowest. Overall, the average annual yield of the East China Sea were about 60.81% of the total catch, while the average annual yield of the Yellow Sea accounted for about 39.19%. The average catch of the East China Sea is higher than that in the Yellow Sea, and the variance and the coefficient of variation is lower than the Yellow Sea, indicating that the East China Sea is the main producing areas during 2003-2011, with high annual catch and stable production. In the East China Sea, the distribution of total catch on longitude is very different. Most of total catch was distributed in 123°E-127°E area. The total catch of this region are accounted for more than 95% of the year, except for 2009. The catch from the rest of the region is extremely low. Although the total catch from this region in 2009 is slightly lower, most of the catch still came from this region with addition to about 40% of the catch distribution at 122°E-123°E area. And in 123 °E-127°E area, the differences in the distribution of each year is also obvious, 75%- 90% of the catch of the 2003, 2004, 2007 and 2011 are distributed in the west of 125°E, especially in 2003 and 2007, 75%- 90% of catch are distributed in 123°E-124°E area. The catch of 2005, 2008 and 2010 tend to off the coast of the East China Sea, the annual catch of 60% to 90% are in east of 125°E area. The catch of 2006 is mainly distribution in 124°E-126°E. In the latitudinal direction, the main catch are mainly distributed in 26°N-30°N waters, except for the 2003, 2005 and 2007, and catches of the region are accounted for 80%- 95% of the total catch. In most years, the catch from the north of 30°N is relatively low. But in 2003, the catch from 30°N-31°N accounted for more than 40% of the total catch, in 2005 more than 50% and in 2007 accounted for more than 70%.Analysis showed that in the East China Sea, among the correlations between the percentage on longitude of each year, the correlation coefficient of 2003 and 2007 is high, correlation coefficients between year 2006, 2008 and 2010 and 2011 are also high, and year 2004 and 2011 has certain similarity, but not with the rest of the years. There was a large difference in the spatial distribution in longitude between 2005 and 2009 and other year. In the East China Sea, the correlation coefficients between each year of the percentage of catch on latitude during 2003-2011 can be divided into 3 types: first, the correlation coefficients of 2004, 2006, 2008, 2009, 2010 and 2011 were higher, indicating the distribution of the percentage catch in latitude in these years is similar. Secondly, the correlation coefficient of 2005 and 2007 reached 0.9, indicated that the distribution of percentage catch in latitude is similar in these two years. Finally, the correlation coefficients of 2003 and all other years were less than 0.4 and the correlation coefficient between 2005 and 2007 is less than 0, indicating the distribution of percent catch in latitude in 2003 is quite different with other years.In the Yellow Sea, the distribution of percentage catch in longitude during 2003-2011 is basically the same and distributed in 123°E-126 °E, and catch from this region were accounted for more than 90% of the total catch. The catch of the rest of the region was extremely low. There was no catch in the east of the 127°E except for 2006. In all years, the catch from the west of 122°E were not more than 1%. On latitude direction, 32°N-36°N has the highest percentage catch, as distributed in 2004 and 2009 in 32°N-33°N, and 2008, 2010 and 2011 in 33°N-34°N, 2005 in 34°N-35°N, 2003, 2006 and 2007 in 35°N-36 °N.The study found that in the Yellow Sea the correlation coefficients of the percentage catch on the longitude are mostly high during 2003-2011, except for 2009. This indicated that the distribution of percentage catch in longitude are almost the same. Among them, from 2009 to 2011, the correlation coefficients are higher, but year 2009 has lower correlation coefficients with other years. From 2003 to 2004 and from 2007 to 2008, the correlation coefficients were greater than 0.95, indicating that the difference of the distribution in these years is very small. In 2004 and 2005, the correlation coefficient was 0.89, indicating that the spatial distribution also has a certain similarity, and in 2005 and 2006, 2008, the correlation coefficients are also greater than 0.9. Among the correlation coefficients of percentage catch on latitude in the Yellow Sea, it is only high in a few year, such as 2003, 2006 and 2007, 2009 and 2011, 2011 and 2010, but they have lower correlation coefficients with each other. This suggests that there was obvious difference in the distribution of percentage catch in latitude in the Yellow Sea, only a few years is similar, most of the year is different.The study found that variation of the center of gravity of the fishing ground of chub mackerel is generally consistent, namely at 32°N in South East China Sea in July and August, then began to transfer to the north in September or October, then move to the south in the north of the Yellow Sea, in November. In December the center of gravity of the fishing ground will generally southward to about 32°N, some years may move to the East China Sea. From the details view, the movement of the center of gravity follows certain differences. Firstly, the time for the fishing ground to move from the East China Sea to the Yellow Sea. It’s early in year 2003, 2004, 2009 and 2011 and later in year 2005, 2006, 2007, 2009 and 2010. Secondly, the position of the center of gravity of the Yellow Sea fishing ground. The large lighting purse seine fishing vessels mainly product in the Yellow Sea, and the center of gravity of every year is different. In 2005 and 2009 the center of gravity of the fishing ground was in the north, has reached about 37°- 38°N, and the rest of the year the center of gravity of the fishing ground were in about 36°N-37°N. Thirdly, the center of gravity of the fishing ground in December. The working position of the large lighting purse seine fishery will move south in December. Before 2006, the fishing ground in December were generally at about 34 °N, and did not change much within the 4 weeks of December. And after 2006, the fishing ground in December are very different, for example, in 2007 and 2009, the fishing ground began to move south in the first week, and has been moved to about 28°N in the fourth week of December. But in 2011 the fishing ground were mainly in the range of 33°N-34°N.The study found that three models based on the division of high yielding region and low yielding region cannot correctly predict the central fishing ground. According to the Kappa index and the AUC values, these models couldn’t reach the standards of "available". According to the actual operation fishing records of year 2011, the total catch and the fishing effort are mostly distributed in the region which has the predicted probability of fishing ground less than 0.5. According to the overlay map of the actual fishing records and the forecasted probability of the fishing ground, the three models couldn’t predict the center fishing ground, and the movement of the predicted fishing ground is different from the actual situation. This indicate that for the fishing ground forecasting application of chub mackerel in the East China Sea and the Yellow Sea the modeling methods of fishing ground forecasting model a decisive influence to the performance of the forecasting models. The method based on the high / low zone division couldn’t meet the practical requirements, and the method based on the fishing ground/background division can meet the requirements of actual fishing ground forecasting accuracy. From the results of the model comparison, it shows that the method based on the high / low zone division does not apply in the chub mackerel fishing ground forecasting, although it’s the commonly used in squid and tuna fishing ground forecasting. The method based on the fishing ground/background division can meet the requirements of actual chub mackerel fishing ground forecasting accuracy, so the method is applicable here.The study found that the predicted fishing grounds of July to September in 2011 mainly located in the south of the East China Sea at 26.5°N-31°N, 12.25°E-127°E region and 29°N-31°N, 124°E to the west of Zhoushan fishing ground. In September the fishing ground in the middle and southern East China Sea move slightly to the northeast, and also distributed in 36°N in the Yellow Sea. From October to December the forecasted fishing grounds are mainly located in the Yellow Sea, with the trend to move south with time. In December the main forecasted fishing grounds have been shifted southward to 33.5°N, to the north of the East China Sea, and there are also small fishing ground in the southern East China Sea. Mostly, the forecasted fishing ground forecast and the actual position is consistent, and its movement with time also match the actual situation, indicating spatial distribution of the forecasted fishing ground is reasonable. Based on the evaluation of the testing dataset, one can see that the fishing ground forecasting model based on the boosted regression tree(AUC value 0.935) can predict the center fishing ground correctly and the forecasting fishing ground is basically consistent with the actual operation position.Research shows that the fishing ground forecasting model based on the boosted regression tree is feasible. At the same time, from the perspective of fishery resource management and protection, the central fishing ground or high fishing areas are also the key region of chub mackerel resource protection. Therefore, the accurate prediction of the location of these areas, is vital significant for the management and protection of the chub mackerel resources. |
PDF Full Text Request