基于提升回归树的东、黄海鲐鱼渔场预报模型研究

发布时间：2018-05-18 00:42

本文选题：鲐鱼 + 提升回归树模型　；参考：《上海海洋大学》2016年博士论文

【摘要】：鲐鱼(Scomber japonicus)是沿海性中上层鱼类,广泛分布于西北大西洋沿岸。中国东、黄海拥有丰富的鲐鱼资源。到90年代后期,鲐鱼的年产量超过30万吨,成为我国近海主要的经济鱼种之一。鲐鱼是一种洄游性鱼类,其洄游路径鲐鱼是一种季节洄游性鱼类,其作业渔场位置与洄游路线密切相关,同时也受到海洋环境条件变动的影响,呈现出较大的年际变化,准确的渔场预报能够指导灯光围网渔业企业合理安排船组的生产位置,缩短寻找渔场的时间,减少成本并提高渔获产量,这对我国大型灯光围网鲐鱼渔业具有重要的作用。为此,论文针对鲐鱼渔场预报模型种类不多、模型研究不系统的状况,在总结现有的渔情分析和渔场预报模型的基础上,引入了提升回归树模型,系统介绍了提升回归树模型的构建、求解和算法实现以及模型参数的选择等。论文以东、黄海鲐鱼为案例,构建了基于提升回归树的渔场预报模型,一方面拓展了现有的渔情分析和渔场预报模型的理论和方法,可为东、黄海鲐鱼的渔场预报和渔业资源管理提供理论支持;另一方面,论文的研究涵盖了数据模型、渔场学基础和预报模型三个方面,也为我国在柔鱼、金枪鱼等远洋渔业渔场预报模型的构建提供了参考。主要研究成果如下:(1)东、黄海鲐鱼渔场的时空分布特征。研究发现,2003~2011年大型鲐鱼灯光围网渔船在每年的7~9月份主要在东海作业,10~12月份主要在黄海作业。各个月份的变异系数均大于0.3,表明同一月份下的渔获量的年间波动较大。在东海,8月份的渔获量多年月平均值最高,且变异系数最小;而7月份的渔获量平均值最低,且变异系数超过了0.5,波动较为剧烈。在黄海,11月份的多年月平均产量最高,且变异系数最小;而10月份的渔获量多年月平均值较高,但变异系数超过了0.6,波动最为剧烈;12月份的渔获量多年月平均值最小,且波动也较大。表明8月份和11月份分别是大型灯光围网渔业在东海和黄海海域较为稳定的生产期。在东海,2003年的渔获量最高、2011年的渔获量最低;在黄海,2008年的渔获量最高、2006年的渔获量最低。总体而言,东海的多年平均产量百分比约占60.81%,而黄海的多年平均产量百分比约占39.19%。东海的多年平均渔获量高于黄海,方差和变异系数则比黄海要低,这说明东海是2003~2011年间大型灯光围网渔业的主要产区,其年产量较高,且产量相对稳定。在东海,2003~2011年鲐鱼大型灯光围网渔业的渔获量百分比在经度上的分布的年间差异较大。2003~2011年各年的渔获量主要分布在123°e~127°e海域,除2009年以外,该区域的渔获量均占到了全年渔获量的95%以上,其余区域则渔获量极低。虽然该区域在2009年的渔获量所占比例较其它年份稍低,但除约40%的渔获量分布于122°e~123°e海域外,2009年的大部分渔获量依然分布于123°e~127°e范围内。而在123°e~127°e范围内,各年份的分布差异也较为明显,其中2003、2004、2007和2011年渔获量偏向近海,全年渔获量的75%~90%位于125°e以西,尤其是2003和2007年,渔获量的75%~90%均分布于123°e~124°e范围内;2005、2008和2010年渔获量则偏向东海外海,全年渔获量的60%~90%位于125°e以东海域;2006年渔获量则主要分布于124°e~126°e海域。在纬度方向上,2003~2011年各年的渔获量主要分布在26°n~30°n海域,除2003、2005和2007年以外,该区域的渔获量均占到了全年渔获量的80%~95%以上。在大部分年份,30°n以北的海域的鲐鱼渔获量相对较低,但在2003年30°n~31°n海域的渔获量占到了全年的40%以上,2005年则超过了全年渔获量的50%,2007年占全年渔获量的70%以上。分析发现,东海海域各年经度渔获量比重的年间相关系数中,2003年和2007年的渔获量比重相关系数较高,2006、2008、2010、2011年的相关系数较高,2004年与2011年具有一定的相似性,但和其余年份差异较大。2005年和2009年两个年份与其它所有年份在经度上的空间分布均存在较大的差异。东海海域各年纬度渔获量比重的年间相关系数中,2003~2011年渔获量在纬度上的空间分布明显分为3个类型:首先,2004、2006、2008、2009、2010和2011年这几年的相关系数均较高,表示这些年份中大型鲐鱼灯光围网渔业的渔获量百分比在纬度上的分布较为相似;其次,2005年和2007年的相关系数也达到了0.9,说明其渔获量百分比在纬度上的分布具有相似性;最后,2003年与其它所有年份的相关系数均小于0.4,与2005年和2007年的相关系数小于0,说明2003年的渔获量百分比在纬度上的分布与其它所有年份均有较大差异。在黄海海域,2003~2011年鲐鱼大型灯光围网渔业各年间的渔获量百分比在经度上的分布趋势基本相同,均主要分布在123°e~126°e海域,该区域的渔获量均占到了全年渔获量的90%以上,其余区域则渔获量极低。除2006年外,在127°e以东海域均没有渔获量分布。在所有年份中,122°e以西海域的渔获量百分比也均不超过1%。纬度方向上,32°n~36°n的所有纬度范围均有最高渔获量比重分布,如2004年和2009年分布在32°n~33°n海域,2008、2010和2011年分布在33°n~34°n海域,2005年在34°n~35°n海域,2003、2006和2007年则在35°n~36°n海域。分析发现,黄海海域各年经度渔获量比重的年间相关系数中,除2009年之外,2003~2011年多数年份间的渔获量比重相关系数均较高,说明在黄海海区各年份间渔获量比重在经度上的空间分布差异不大。其中,2009~2011年的相关系数较高,空间分布差异较小,但2009年与其余各年份的相关系数均较小;2003~2004、2006~2008年的相关系数也均大于0.95,说明这些年份渔获量比重在经度上的空间分布差异极小;2004年和2005年的相关系数为0.89,说明其渔获量比重的空间分布也有一定的相似性,且2005年2006年、2008年的相关系数也大于0.9。黄海海域各年纬度渔获量比重的年间相关系数分析中,2003~2011年渔获量百分比的相关系数仅个别年间较高,如2003、2006和2007年,2009年和2011年,2011年和2010年等,但这些年份与其它年份之间的相关系数均较低,因此没有形成明显的类别。这表明,2003~2011年黄海海区的大型鲐鱼灯光围网的渔获量百分比在纬度上的空间分布具有明显的差异性,仅有少数年份的空间分布具有相似,而多数年份的分布情况并不相同。此外,研究还发现,2003~2011年东、黄海鲐鱼渔场重心的纬度在所有的年份随时间上的变动大体上是一致的,即每年的7~8月份位于32°n以南的东海渔场,9月或10月份开始向北渔场转移,11~12月份渔场重心由黄海北部向南移动,到12月份,渔场重心一般会南移至32°n左右,有些年份可能会移动到东海海域。从细节上看,东、黄海鲐鱼渔场重心纬度的变化情况又有如下一定的差异。首先是渔场由东海转向黄海的时间,9月或10月是东、黄海鲐鱼作业渔场转换的时间,但每年的转场时间并不一致。其中2003、2004、2009、2011年的转场时间较早,一般是在9月份的第一个星期或第二个星期开始向黄海移动,在10月份之前完成作业渔场位置的转换。而2005、2006、2007、2009和2010年的转场时间较晚,一般在当年10月份的第一或第二周才开始向黄海移动,在11月份之前完全移动到黄海渔场。其次是黄海渔场重心位置,每年的10~11月,大型灯光围网渔船主要在黄海海域生产,每年黄海渔场的渔场重心位置也有差异。2005和2009年的渔场重心明显偏北,已经达到37°n~38°n左右,而其余年份的渔场重心则主要在36°n~37°n左右。最后是12月份的渔场重心位置差异,每年的12月,大型灯光围网渔业的作业位置相对10~11月份主要变化是向南移动,但12月份本身的渔场重心也有一定的差异。2006年之前,12月份的渔场重心一般在34°n左右,且12月内的4个星期的渔场重心变化不大。而2006年之后,每年12月份的渔场重心均有较大差异,如2007和2009年,渔场重心从第一个星期开始向南移动,到第四周时已经移动到28°n左右。而2011年,12月的渔场重心主要在33°n~34°n范围内小幅移动。(2)建模方法对东、黄海鲐鱼渔场预报模型精度的影响。研究发现,根据渔业数据划分高产区和非高产区进行建模的三种模型不能正确地预测中心渔场和非中心渔场,从kappa系数auc值上看,这三种模型还达不到可用的标准。对于2011年实际作业记录的统计来看,实际作业记录、下网次数和渔获量均分布在模型预报渔场概率小于0.5的区域。根据实际作业记录与预报渔场概率叠加来看,这三种模型对于中心渔场的预报无法与实际作业位置吻合,渔场的移动情况也与实际情况不同。这说明对于东、黄海鲐鱼渔场预报而言,不同的建模方法对渔场预报的精度具有决定性的影响,使用高产区/低产区划分方法建立的渔场预报模型的预报精度无法达到实用的要求,而以渔场/背景方法建立的预报模型则能够满足实际渔场预报的精度要求。从模型比较的结果来看,虽然高产区/低产区划分方法在柔鱼和金枪鱼等鱼种的渔场预报中比较常用,但在东、黄海鲐鱼渔场预报模型中并不适用,而以渔场/背景方法建立的鲐鱼预报模型则能够满足实际鲐鱼渔场预报的精度要求,因此该方法是适用的。(3)基于提升回归树的东、黄海鲐鱼渔场预报模型。研究发现,基于提升回归树的东黄海鲐鱼渔场预报模型所预报的2011年7~9月的渔场主要位于东海中南部26.5°n~31°n、122.5°e~127°e区域以及29°n~31°n、124°e以西的舟山渔场,其中9月份东海中南部渔场向东北方向稍有移动,并且在36°n附近的黄海海域也有渔场分布,但预报的渔场概率不高。10~12月的预报渔场主要位于黄海海域,随时间推移预报渔场有向黄海南部移动的趋势,12月份的主要预报渔场已南移至33.5°n,最南达到东海北部海域,同时东海中南部也有小范围的渔场分布。总体上看,除了9月份黄海海域和10月份东海北部海域实际作业渔场概率预测值偏低之外,预报渔场的位置与实际作业位置基本吻合,其随时间的位置移动也基本与实际情况相符,这说明模型的预报渔场在空间分布上是合理的。从基于验证数据集的评价结果来看,基于提升回归树的渔场预报模型预报准确率较高(AUC值0.935),且能有效地区分低产和高产区域。从空间位置上来看,预报渔场的范围也基本与实际作业位置吻合。研究表明,采用基于提升回归树的渔场预报模型来预报东、黄海鲐鱼渔场是可行的。同时,从鲐鱼渔业资源管理和保护的角度来说,中心渔场或高产渔区同时也是鲐鱼资源保护的关键区域。因此,准确的预测这些区域的位置,对于鲐鱼资源的管理和保护也有重要的意义。
[Abstract]:Scomber japonicus (mackerel mackerel) is a coastal middle upper fish, widely distributed in the coastal the Atlantic. China East, the Yellow Sea has rich mackerel resources. By the late 90s, the total output of mackerel was more than 300 thousand tons, and mackerel was a migratory fish, the migration route of mackerel was a season. The location of migratory fish is closely related to the migratory route, and it is also influenced by the changes in the marine environmental conditions, showing a large interannual change. The accurate prediction of fishing grounds can guide the lighting seine fishery enterprises to arrange the production position of the ship group reasonably, shorten the time for finding the fishing ground, reduce the cost and increase the catch production. This has an important role in the large light Seine mackerel fishery in China. To this end, the paper has introduced the lifting regression tree model on the basis of the existing fishing situation analysis and the fishing field forecasting model, and systematically introduced the construction of the lifting regression tree model. In this paper, a fishing field prediction model based on the lifting regression tree is constructed in the case of the mackerel in the East and the the Yellow Sea. On the one hand, the theory and method of the existing fishing situation analysis and the fishing field prediction model are extended, and the theory and the theory support for the fishing ground prediction and fishery resources management of the mackerel in East, the Yellow Sea, and the mackerel are provided. The study covered three aspects of the data model, the foundation of the fishing field and the forecast model, and also provided a reference for the construction of the prediction model for the ocean fishing grounds of the soft fish and tuna fish in China. The main research results are as follows: (1) the spatial and temporal distribution characteristics of the mackerel fishing grounds in the East and the Yellow Sea. The study found that the large mackerel light Seine fishing in 2003~2011 was found. The ship is mainly in the East China Sea in the 7~9 month of the year, and the month of 10~12 is mainly in the Yellow Sea. The variation coefficient of each month is more than 0.3, which indicates that the year in the year of January has a great fluctuation. In the East China Sea, the average value of the catch was the highest in August, and the coefficient of variation was the smallest; and the average of the catch in July was the lowest and variation. In the Yellow Sea, the average annual monthly output in November is the highest and the variation coefficient is the smallest in the Yellow Sea. In October, the annual monthly mean value of the catch is higher, but the coefficient of variation exceeds 0.6 and the fluctuation is the most intense. In December, the annual mean value of the catch is the smallest and the fluctuation is larger. In August and November, respectively, respectively. It is a relatively stable production period of large light seine fishery in the East China Sea and the Yellow Sea sea. In the East China Sea, the catch was highest in the East China Sea in 2003 and the lowest catch in 2011. In the Yellow Sea, the catch was highest in 2008 and the lowest catch in 2006. In general, the average annual output of the East China Sea was about 60.81% per hundred percent, and the average annual output of the Yellow Sea was 100 percent. The average annual catches in the East China Sea of 39.19%. are higher than that in the Yellow Sea, and the variance and coefficient of variation are lower than that in the Yellow Sea. This indicates that the East China Sea is the main production area of the large light seine fishery in the period of 2003~2011, and its annual output is higher and the yield is relatively stable. In the East China Sea, the percentage of the catch percentage of the mackerel large type light seine fishery in the East China Sea is on the longitude. The annual variation of the distribution in the.2003~2011 year was mainly distributed in the 123 degree e~127 / e sea area. Except for 2009, the catch accounted for more than 95% of the annual catch, and the rest area was very low. Although the proportion of the area in 2009 was slightly lower than that of the other years, except about 40% of the catch. In the area of 122 degree e~123 e, most of the catches in 2009 are still distributed in the range of 123 e~127 e. In the range of 123 degrees e~127 e, the distribution differences in each year are also obvious, and the catch of 200320042007 and 2011 is near to the sea, and the annual catch 75%~90% is located in the west of 125 degree e, especially in 2003 and 2007. The 75%~90% is distributed in the range of 123 degree e~124 e, and the catch in 20052008 and 2010 is biased toward the sea of the East China Sea, and the annual catch 60%~90% is located in the East Sea area of 125 degree e. In 2006, the catch is mainly distributed in the 124 degree e~126 degree e sea area. In the latitude direction, the catch of each year in 2003~2011 year is mainly distributed in the 26 degrees n~30 degree n sea area, except 20032005 and 200. In 7 years, the catches in the region accounted for more than 80%~95% of the annual catch. In most years, the catch of mackerel in the sea area north of 30 N was relatively low, but the catch in the 30 degree n~31 n sea area in 2003 accounted for more than 40% in the whole year. In 2005, it exceeded 50% of the annual catch. In 2007, it accounted for more than 70% of the annual catch. It is found that the correlation coefficient of the specific gravity of the annual catches in the East China Sea is higher in 2003 and 2007, and the correlation coefficient of the 2006200820102011 years is higher. In 2004 and 2011, it has a certain similarity, but the difference is larger in.2005 and 2009 in two years and in all the other years in the rest of the year in the other years. There are great differences in the spatial distribution of the longitude. In the annual correlation coefficient of the annual latitudinal catches in the East China Sea, the spatial distribution of 2003~2011 catch at latitude is obviously divided into 3 types: first, the correlation coefficient of 20042006200820092010 and 2011 is higher, indicating the large mackerel light in these years. The distribution of catch percentage in the seine fishery is similar in latitude; secondly, the correlation coefficient of 2005 and 2007 also reaches 0.9, indicating that the distribution of its catch percentage is similar in latitude; finally, the correlation coefficient of all the years in 2003 is less than 0.4, and the correlation coefficient of 2005 and 2007 is less than 0, indicating 2. The distribution of catch percentage in latitude 003 years has a great difference from all the other years. In the the Yellow Sea sea area, the distribution trend of the catch percentage in the longitude of mackerel mackerel large light seine fishery in 2003~2011 is basically the same in the longitude, which is mainly distributed in the 123 e~126 e sea area. More than 90% of the quantity and the rest of the area are very low. Except for 2006, there is no catch distribution in the East Sea area of 127 degree e. In all years, the percentage of catch in the West Sea area of 122 degree e is not more than the direction of 1%. latitude. All latitude of 32 degree n~36 degree n has the highest catch proportion distribution, such as 2004 and 2009 distribution at 32 degree n. ~33 degree n sea area, 20082010 and 2011 distribution in 33 degree n~34 n sea area, 2005 in 34 degree n~35 n sea area, 20032006 and 2007 in 35 degree n sea area. There is little difference in the spatial distribution of catches in the longitude of each year in the Huanghai Sea area. Among them, the correlation coefficient of 2009~2011 year is higher and the spatial distribution difference is small, but the correlation coefficient between 2009 and the rest of the years is smaller, and the correlation coefficient of 2003~20042006~2008 year is also larger than 0.95, indicating that the proportion of the catch of these years is longitude. The spatial distribution difference is very small; the correlation coefficient of 2004 and 2005 is 0.89, indicating that the spatial distribution of its catch proportion is also similar, and the correlation coefficient of 2006 2005 in 2008 is greater than that of the year correlation series analysis of the annual latitudinal catch of 0.9. the Yellow Sea sea area, and the correlation of the percentage of fishing catch in 2003~2011 years. The coefficient is higher only in a few years, such as 20032006 and 2007, 2009 and 2011, 2011 and 2010, but the correlation coefficient between the years and the other years is low, so there is no obvious category. This shows that the spatial distribution of the catch percentage of the large mackerel light enclosure in the the Yellow Sea sea area of the the Yellow Sea sea area is in the latitude. There are obvious differences, only a few years of spatial distribution are similar, and the distribution of most years is different. In addition, the study also found that the latitude of the center of gravity of the mackerel fishing ground in the Yellow Sea, in 2003~2011, was generally consistent in all the years, that is, the annual 7~8 month is located in the East China Sea fishing ground south of 32 degrees N, 9 The center of center of gravity of the fishing ground moved southward from northern the Yellow Sea in 11~12 month or October. By December, the center of gravity of the fishing ground generally moved southward to about 32 n, and some years may move to the sea area. In detail, the variation of the latitude of the center of gravity of the mackerel in East and the Yellow Sea has a certain difference. First, the fishing ground is from the fishing ground. The time for the East China Sea to turn to the Yellow Sea, in September or October, is the time for the conversion of the mackerel in East and the Yellow Sea, but the time of the transfer is not consistent each year. The 2003200420092011 years of the transfer time is earlier, generally moving to the Yellow Sea in the first or second weeks of September, and completing the position of the fishing ground before October. In 2005200620072009 and 2010, the transfer time was late, generally in the first or second weeks of the year in October, it began to move to the Yellow Sea, and moved to the the Yellow Sea fishing ground before November. Next, the center of the the Yellow Sea fishing ground, each year of the 10~11 month, the large light Seine fishing boats were mainly produced in the the Yellow Sea sea, and each year the Yellow Sea fishing grounds The center of gravity of the fishing ground is also different.2005 and the center of the fishing ground in 2009 is obviously north, which has reached about 37 n~38 n, and the rest of the fishing ground is mainly about 36 n~37 degrees N. Finally, the difference of the center of gravity of the fishing ground in December, and in December, the main change of the operation position of the large light seine fishery is relative to the month of 10~11. South movement, but the center of gravity of its own fishing ground in December has a certain difference.2006 years ago, the center of the fishing ground in December was generally around 34 n, and the center of gravity of the fishing ground changed little in the 4 weeks in December, and after 2006, the center of gravity of the fishing ground in December was quite different, such as 2007 and 2009, the center of gravity of the fishing ground began to south from the first week. The movement has moved to about 28 n in the fourth week. In 2011, the center of gravity of the fishing ground in December was mainly moved in the range of 33 degrees n~34 n. (2) the modeling method influenced the precision of the model of the mackerel fishing ground in the Yellow Sea. The study found that the three models of modeling high yield and non high yield areas according to the fishery data were not correctly predicted. At the center fishing ground and non central fishing ground, the three models can not reach the standard of availability from the value of the kappa coefficient AUC. For the statistics of actual operation records in 2011, the actual operation records, the number of down nets and the catch are distributed in the area where the probability of the model forecast fishing ground is less than 0.5. According to the actual operation record and the prediction of the fishing ground probability superposition In view of the three models, the prediction of the central fishing ground can not coincide with the actual operation position, and the movement of the fishing ground is also different from the actual situation. This shows that for the East and the Yellow Sea mackerel fishing ground forecast, the different modeling methods have a decisive influence on the precision of the fishing ground forecast, and the fishing ground is established by the method of high yield / low yield division. The prediction accuracy of the field prediction model can not meet the practical requirements, and the prediction model established by the fishing ground / background method can meet the precision requirements of the actual fishing ground prediction. From the model comparison, although the high yield / low yield division method is more commonly used in the fishing ground prediction of the fish species such as the soft fish and the tuna, but in the East, the Yellow Sea The mackerel prediction model of mackerel fishing ground is not applicable, and the mackerel prediction model based on the fishing ground / background method can meet the precision requirements of the actual mackerel fishing ground prediction. Therefore, this method is applicable. (3) the model of mackerel fishing ground prediction based on the east of the lifting regression tree and the the Yellow Sea mackerel fishing ground prediction model. The fishing ground of 2011 7~9 month predicted by the model is mainly located in the 26.5 degree n~31 degree n, 122.5 degree e~127 degree E region and 29 [n~31] n and 124 degree e to the Zhoushan fishing ground. In September, the fishing ground in the East China Sea of the East China Sea is slightly moved north-east, and there are fishing grounds in the Huang Haihai domain near 36 degrees, but the probability of the forecast is not high.1. The forecast fishing grounds for 0~12 months are mainly located in the the Yellow Sea sea area. As time goes on, the fishing ground is predicted to move towards the Yellow Hainan section. The main forecast for December is fishing.
【学位授予单位】：上海海洋大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S934

【相似文献】