The Preliminary Study Of Abundance Distribution For Dosidicus Gigas In The High Sea Off Peru Based On Geostatistics

Dosidicus gigas is a pelagic species belonging to one of the biggest cephalopods in the world, distributed in the area of California(37°N)to Chile(40°S). From the 1990 s, the D.gigas off Peru was explored massively. The oceanic area off Peru became one of the most important pelagic fishing grounds for China. D.gigas has a shorten-life and could spawn all of the year, easily affected by the change of oceanic environment. It is necessary for fishery sustainable exploration and management to learn about the spatiotemporal distributing regulation of D.gigas. Thus, further study for the spatial distribution, difference and change mechanism of D.gigas resources, understanding the distributed regulation of high abundance and forecasting the resources will be important for studying the spatial spread of resources and supplying theoretical references for sustainable exploration and scientific management.In this paper, geostatistical methods including semi-variogram and kriging were used for studying the distribution of D.gigas resource and fishing ground combined with oceanic remote sensing, geographical information system and statistics based on the historical fishery data. The main conclusions were as follows:(1) The spatial heterogeneity analysis for D.gigas based on CPUE. This study took the statistical data of squid catches for example during June to September from 2003 to 2012, using the global and local autocorrelation and semi-variogram to explore the spatial heterogeneity and find the key environmental factors. The results as follows: 1) the z scores of Morans’ s I and Getis-Ord Gi* showed that D.gigas off Peru distributed for patchy aggregation and had a strong spatial dependence. 2) The optimal models were Exponential, Spherical, Gaussian and Spherical model for June to September, respectively. The nugget coefficient were 9.53%、20.40%、33.61% and 24.59 %, respectively, which reflected the strong spatial dependence. These results was consistent with the global and local analysis results. 3) The correlation test result suggested that chlorophyll-a(Chl-a) was the key factor for preoviposition period(in June), and sea surface temperature(SST) was important for spawning period(in July to August).(2) Comparison between cokriging based on an auxiliary variable and ordinary kriging. Analyzing the spatial distribution of D.gigas off Peru using geographical information system and geostatistical methods. The residual sum of squares(RSS) and the coefficient of determination(R2) were used to select the optimal semi-variogram models from different environmental factors( SST, Chl-a, sea surface salinity: SSS, sea surface height: SSH) as auxiliary variables. The cross-validation and relative error(RI) were used to measure the predicting precision for ordinary kriging(OK) and cokriging(COK). The results showed that SSS, SSH and Chl-a were significantly related to CPUE in June. SSS, SSH and SST were significantly related to CPUE in July. SSS and SST were significantly related to CPUE in August. And only SST was related to CPUE in September. The optimal models were all Gaussian model based on Chl-a, SSS, SSS and SST from June to September, respectively. The spatial distribution of D.gigas in June had the strong spatial dependence, and had the medium spatial dependence in July to September. The cross-validation results showed cokriging increased prediction precision by 1.51% in June, 3.00% in July, 1.72% in August and 1.18% in September, compared to ordinary kriging separately.(3) The study for D.gigasabundance off Peru by cokring based on the comprehensive environmental factor. The main variable CPUE and the environmental variable were standardized to values from 0 to 1. Principal component analysis(PCA) was used to combine the standardized environmental variables as a comprehensive variable. The results showed the comprehensive variables were correlated with CPUE during June to September. The fishing ground predicted by COK was consistent with the actual CPUE distribution. The prediction had good accuracy(mean error: ME<0.01) for June to September. Though the precision(mean square error: RMSE=0.4128) in June was the best and in August was the worst(RMSE=0.5169), the value were similar in four months. The predicted standard deviation in June and July(RMSSE-1<0.1) had better effectiveness than those in August and September(RMSSE-1>0.1).From the ME, RMSE and RMSSE points of view, it is a certain reliable for this cokriging method for predicting during June to September.(4) Comparison between cokring and general linear model to study the abundance of Dosidicus gigas off Peru. This study took the statistical data of squid catches, and SSS, SST, SSH and Chl-a for example during June to September from 2003 to 2012.75% samples were taken from total samples to build general linear model(GLM) and semivariogram. Optimal GLM was selected by Akaike Information Criterion(AIC) and best fitted semi-variogramwas selected by Residual Sum of Squares(RSS) and coefficient of determination(R2). Another 25% data were used to validate the accuracy of models(external validation). At the same time, the 75% were used self-validation for models.According to the result of external validation, self validation and the ration of external and self validation to compare the accuracy of GLM and cokriging methods.The result showed that the effective environmental factors in GLM during June to September were different, which suggested that environment had different influence on D.gigas ontogenesis. And the environmental factors in GLM were different from those in semi-variogram. In self-validation, the mean error(ME) was small in both GLM and cokriging, but its scale was larger in cokriging than in GLM. This incicated that a good accuracy was shown in GLM. In external validation, ME was close in GLM and cokriging which were both little. The absolute ratio of external validation and self validation was smaller in cokriging than in GLM, which indicated that cokriging had better robustness than GLM.Just given the predicted error, cokriging had the similar effectiveness as GLM.(5) The spatial distribution of D. gigas predicted by ordinary kriging based on fishing effort and CPUE. Standardize fishing effort and CPUE from June to September in 2003 to 2012 to the value of 0 to 1 in a grid of 0.5°×0.5°. Semi-variogram and ordinary kriging methods were used to study the spatial distribution of D.gigas off Peru. The results were as following:1) except in September, the nugget coefficient from fishing effort were smaller than those from CPUE during June to August, which suggested the random error was smaller in fishing effort, and the spatial structure was stronger. As a whole, the spatial dependence and aggregation of fishing effort were both better than CPUE. 2) Except in June, the errors from fishing effort were larger than those from CPUE during June to September, which indicated the accuracy of ordinary kriging based on fishing effort was not better than that based on CPUE. However, the precision of fishing effort was higher than CPUE in June to August, and it was adverse result in September, which indicated that the prediction precision was better from fishing effort than from CPUE. 3) After accounting the average of fishing effort and CPUE, the result of ordinary kriging had a good job in accuracy and precision which was improved by 14.04%,1 8.40%, 22.34%, 39.65% during June to Sptember respectively. Thus, it is reliable to predicting fishing ground based on fishing effort and CPUE.