Home Journals Articles Updates About
CN
image
收藏切换
Comparative analysis of CPUE standardization of Chinese Pacific saury (Cololabis saira) fishery based on GLM and GAM
收藏切换
PDF
Chuanxiang Hua1, 2, Qingcheng Zhu1, 2, *, Yongchuang Shi1, Yu Liu1, 2
Acta Oceanologica Sinica | 2019, 38(10) : 100 - 110
Less
收藏切换
Acta Oceanologica Sinica | 2019, 38(10): 100-110
• Marine Information Science •
Comparative analysis of CPUE standardization of Chinese Pacific saury (Cololabis saira) fishery based on GLM and GAM
Full
Chuanxiang Hua1, 2, Qingcheng Zhu1, 2, *, Yongchuang Shi1, Yu Liu1, 2
Affiliations
  • 1 College of Marine Sciences, Shanghai Ocean University, Shanghai 201306, China
  • 2 National Engineering Research Center for Pelagic Fishery, Shanghai 201306, China
Published: 2019-10-25 doi: 10.1007/s13131-019-1486-3
Outline
收藏切换

Pacific saury is an important high-seas fishery resource in the Northwest Pacific Ocean for the Chinese Mainland. Reliable and accurate catch per unit effort (CPUE) plays a significant rule in Pacific saury stock assessment. Many statistical models have been used in the previous CPUE standardization research. Here, we compare the performance of Generalized Linear Models (GLMs) and Generalized Additive Models (GAMs) using CPUE data collected from Chinese saury fishery in the Northwest Pacific Ocean from 2003 to 2017 (excluding data from Chinese Taipei), and evaluate the influence of spatial, temporal, environmental variables and vessel length on CPUE. Optimal GLM/GAM models were selected using the Bayesian information criterion (BIC). Explained deviance and 5-fold bootstrap cross-validation results were used to compare the performance of the two model types. Fitted GLMs accounted for 21.57% of the total model-explained deviance, while GAMs accounted for 38.95%. Predictive performance metrics and 5-fold cross-validation results showed that the best GAM performed better than the best GLM. Therefore, we recommend GAM as the preferred model for standardizing CPUE of Pacific saury in the Northwest Pacific Ocean.

Cololabis saira  /  CPUE standardization  /  generalized linear model  /  generalized additive model
Chuanxiang Hua, Qingcheng Zhu, Yongchuang Shi, Yu Liu. Comparative analysis of CPUE standardization of Chinese Pacific saury (Cololabis saira) fishery based on GLM and GAM[J]. Acta Oceanologica Sinica, 2019 , 38 (10) : 100 -110 . DOI: 10.1007/s13131-019-1486-3
Full Text
Less
1 Introduction
Less
Pacific saury (Cololabis saira) is a highly migratory fish, widely distributed in the high seas of the Northwest Pacific Ocean (Lin, 2003; Sun et al., 2003). The species is harvested primarily by Japan, Russia, South Korea, Taiwan Province, and Chinese Mainland. China began Pacific saury fishing in the high seas in 2003 and it has become one of the most important fisheries for China since then. Although various studies regarding the fishing gear and methods (Xu et al., 2005; Yu et al., 2006; Xia, 2008), distribution of fishing grounds (Huang et al., 2005; Hua et al., 2010), and basic biology (Zhang et al., 2013) have been conducted, there is no study focused on catch per unit effort (CPUE) standardization for Chinese Pacific saury fishery in the Northwest Pacific Ocean.
CPUE is commonly used as an important relative index of fish abundance and is one of the most important dataset used in fisheries stock assessment (Nishida and Chen, 2004; Chen et al., 2008). Although abundance index should, ideally, be derived from fishery-independent surveys, it is often based on fishery-dependent data, because fishery-independent data are often costly and difficult to collect (Ward et al., 2013). Nominal CPUE is the uncorrected value obtained directly from commercial fishing data (Maunder and Start, 2003). CPUE is assumed to have a linear relationship with abundance in the assessment; however, CPUE derived from fishery-dependent data may not be the case, because it is often influenced greatly by various factors, such as temporal factors (e.g., Year, Month), spatial factors (e.g., Longitude, Latitude), environmental factors (e.g., sea surface temperature (SST), sea surface height (SSH), sea surface temperature gradient (SSTG), and fishing capacity (e.g., Vessellength) (Harley et al., 2001; Erisman et al., 2011). CPUE can be misleading if these confounding factors are not taken into account.
In order to have more reliable and representative CPUE data for stock assessment, nominal CPUE values need to be standardized using statistical models, which is a process aiming to remove the impacts of the confounding factors (Maunder and Punt, 2004). For providing better management and conservation recommendations, high quality data and continued evaluation of statistical model performance should be highly valued (Martínez-Rincón et al., 2012). In recent decades, many efforts have been made to solve the problems associated with CPUE standardization. Generalized linear models (GLM) and generalized additive models (GAM) are commonly used to standardize CPUE due to the availability of well-tested and user-friendly software to perform calculations (Venables and Dichmont, 2004). Indeed, GLMs are the most common method for standardizing CPUE; they differ from ordinary linear models by allowing fitting of categorical variables and they allow to incorporate non-normal distributions of the response variable. GAMs, on the contrary, are extensions of GLMs which have smooth functions. They are often used to deal with nonlinear relationships between response and explanatory variables (Wood, 2006). However, when standardizing CPUE data, GLMs and GAMs always have their own disadvantages in error structure assumptions, dealing with interaction terms and zero data (Yu et al., 2013). In order to select the better CPUE standardization model, comparative research should be conducted between GLMs and GAMs. Further, to date and to the best of our knowledge, there is no study about CPUE standardization for Pacific saury.
In this study, we used GLM and GAM based on Chinese fishery data (2003–2017) to conduct a comparative study on CPUE standardization of the Chinese Pacific saury fishery in the Northwest Pacific Ocean. Firstly, we selected the optimal GLM/GAM model using the Bayesian information criterion (BIC). We then used the explained Deviance and results from 5-fold bootstrap cross-validations to compare the predictive accuracy of the two models (Rodríguez-Marín et al., 2003; Ortiz and Arocha, 2004). The goal of this study was to identify the best method to the standardization of Pacific saury CPUE data and improve the quality of future stock assessment for Pacific saury.
2 Materials and methods
Less
2.1 Study area
As shown in Fig. 1, the study area is outside the Exclusive Economic Zone (EEZ) of Japan and Russia. This area is at the junction of the Kuroshio warm current and the Oyashio cold current, which together provide an adequate foundation for marine life, and the area is in fact one of the high-yield sea areas of the world (Watanabe et al., 2006).
2.2 Data sources
2.2.1 Fishery data
Fishery data were obtained from the Technical Group for Saury Fishery, Distant-water Fishery Society of China. These data included the date (with a time resolution of days), longitude, latitude, yield, vessel name, and vessel length, among others. Nominal CPUE was defined as the fishing yield of a vessel per day, in units of tons per vessel per day.
2.2.2 Environment data
SST was obtained from the National Oceanic and Atmospheric Administration (NOAA, ftp.nodc.noaa.gov). The spatial-temporal resolution of SST data is 0.1°×0.1° grid per day. SSH was obtained from Archiving Validation and Interpolation of Satellite Oceanographic data (AVISO, www.aviso.altimetry.fr). The spatial-temporal resolution of the data is SSH daily at 0.25°×0.25° grid.
SSTG was estimated from Gradient Magnitude (GM) (Ortiz and Arocha, 2004; Howell and Kobayashi, 2006), and expressed as
$SST{G_{i,j}} = \sqrt {{{\left({\frac{{SS{T_{i + 1,\,j}} - SS{T_{i - 1,j}}}}{{\Delta x}}} \right)}^2} + {{\left({\frac{{SS{T_{i,j + 1}} - SS{T_{i,j - 1}}}}{{\Delta y}}} \right)}^2}},$
where SSTi+1, j, SSTi–1, j, SSTi, j+1 and SSTi, j–1 were SST values of four adjacent grids respectively, i and j are row and column number, respectively, ∆x is the longitudinal distance between (j–1)th and (j+1)th columns (km), ∆y is the latitudinal distance between (i–1)th and (i+1)th rows (km), SSTGi, j is SSTG value of the current grid (°C/km).
In order to match fishery data and environmental data, the present study used the environmental data of the nearest grid corresponding to the grid where the fishery data existed on the same date.
2.3 Methods for analysis
2.3.1 Selection of variables
GLM and GAM both require response variables and explanatory variables that are independent from each other. Since there were no zero-catch data, the natural logarithm of CPUE (i.e., ln(CPUE)) was used as the response variable (Campbell, 2004).
The selection of explanatory variables takes into account the following facts: (1) Pacific saury is a highly migratory fish, and the distribution of its fishing grounds shows large variation during the fishing period (June–November) each year (Tian et al., 2003; Shen et al., 2004); therefore, the explanatory variables include temporal variables (Year and Month), spatial variables (Longitude and Latitude), and temporal-spatial interaction terms (Year×Longitude, Year×Latitude, Month×Longitude, and Month×Latitude); (2) the formation of Pacific saury fishing grounds is tightly associated with the marine environment (Zhu et al., 2006a, b; Zou and Zhu, 2006; Yan et al., 2012; Zhang et al., 2015). Thus, explanatory variables include SST, SSTG and SSH; (3) in a real fishing process, vessel performance may affect fishing efficiency; thus, explanatory variables, such as Vessellength, were taken into account in this study.
Mutual independence of explanatory variables was checked by the variance inflation factor (VIF) and Spearman’s correlation coefficient (Table 1). In this table, the data under the dotted line are Spearman’s correlation coefficient among explanatory variables, data above the line are the corresponding P values. The maximum VIF<10, indicated there was no serious multi-collinearity (Tien et al., 2011; Menard, 1995).
2.3.2 GLM
GLMs are the most common models for standardizing CPUE data. The key assumption of a GLM is that the relationship between some function of the expected value of the response variable and the explanatory variables, is linear:
${{g}}\left({{\mu _i}} \right) = X_i^{\rm T}\beta,$
where g is the differentiable and monotonic link function, μi=E(Yi), Xi is the explanatory variable for the ith response variable, β is a vector of the parameters, and Yi is the ith random variable. GLM assumes a normal error distribution, and the full GLM is expressed as
$\begin{split}\ln \left({CPUE} \right) =& Year + Month + Longitude + Latitude + SST +\\& SSTG + SSH + Vessellength + Interaction + \varepsilon,\end{split}$
where CPUE is the fishing catch of a vessel per day, and Interactions is an interaction term representing the interactive effect of spatial and temporal factors for Pacific saury. Full model includes all the possible combination of Year, Month, Longitude, and Latitude as interaction terms; ε is the residual, which is assumed to have a normal distribution; Year is a categorical variable of 15 years (2003–2017). Month is a categorical variable including the eight calendar months from May to December. We attempted two cases (categorical and splined variable) for Longitude and Latitude, which divided at intervals of 1°. We also investigated two cases (categorical or splined variable) for each explanatory variable of environment. Vessellength is a categorical or continuous variable of 60–75 m vessels, which will affect the catchability (Table 2).
2.3.3 GAM
GAMs are extensions of the generalized linear models, which can be used to describe nonlinear relationships between response variables and explanatory variables (Tseng et al., 2013), as shown below:
$g\left({{\mu _i}} \right) = \alpha + \mathop \sum \limits_{i = 1} {f_i}\left({{X_i}} \right) + {\varepsilon _i},$
where fi is a smoothing function. The full GAM in this study is expressed as
$\begin{split}{\rm{ln}}\left({CPUE} \right) =& Year + Month + s\left({Longitude} \right) + s\left({Latitude} \right) + \\ & s\left({SST} \right) + s\left({SSTG} \right) +s\left({SSH} \right) + s\left({Vessellength} \right) + \\ & s\left({Interactions} \right) + \varepsilon,\end{split}$
where s( ) denotes the smoother functions. The explanatory variables used in GAM are the same as GLM (Table 2). Temporal-spatial interaction terms, including Year×Longitude, Year×Latitude, Month×Longitude, Month×Latitude, and all possible combinations were considered in this study.
2.3.4 Model evaluation
Explanatory variables were added to the GLM/GAM in turn and GLM models/GAM models with different number of explanatory variables were obtained (Shono, 2005). We used the Bayesian information criterion (BIC) to select the best model with minimum BIC in each of the GLM and GAM analyses (Quinn and Keough, 2002; Watanabe et al., 2006). The BIC was calculated as follows:
$BIC = m{\rm {ln}}\left(n \right) + n{\rm {ln}}\left({RSS/n} \right),$
where m is the number of parameters in the model, n is the number of observed values (data points), and RSS is the sum of squared residuals.
For model diagnostics, percent explained deviance was calculated in addition to q-q plot and residual plots. The most common method to evaluate the performance of different models is k-fold cross-validation (Arlot and Celisse, 2010). When conducting k-fold cross-validation, k sets of subsamples of roughly equal size are produced from the original samples. One set of subsamples is saved as the validation data for testing the model, and the rest sets are used as training data. Then, the cross-validation is repeated k times, each of the k sets used exactly once as the validation data. The estimation can be obtained as the average of k results from folds. The 5-fold cross-validation procedure has shown good performance in model selection (Kohavi, 2001). Thus, here we used this procedure to evaluate the performance of each model. We conducted a 5-fold cross validation test for the final model selection between the best models derived from GLM and GAM and the process in cross validation are repeated for 1 000 times. In this test, Spearman’s correlation between the predicted and observed CPUEs, and mean of squared errors between two CPUEs were calculated to evaluate prediction performance.
2.3.5 Calculation of nominal CPUE and standardized CPUE
The yearly nominal CPUE and standardized CPUE can be calculated by the following formula:
$CPU{E_i} = \frac{1}{{{n_i}}} \times \sum\limits_{k = 1}^{{n_i}} {CPU{E_k}}.$
Nominal CPUE values are calculated as the means of original CPUE, standardized CPUE values are calculated as the means of fitted CPUE from the best model, where CPUEi (nominal/standardized CPUE) is the CPUE index in the ith year, ni is the observation number in the ith year, and CPUEk is the kth CPUE data (fitted/original CPUE) in the ith year.
The bootstrapped 95% confidence intervals of Standardized CPUE of the optimal GLM and GAM were calculated. All statistical analyses were conducted using Matlab2016b.
3 Results
Less
3.1 Statistical distribution test of response variable (ln(CPUE))
The scatter points of ln(CPUE) in the normal q-q plot appeared aligned on a straight line (Fig. 2a). Frequency distribution (Fig. 2b) indicated that ln(CPUE) showed approximate normal distribution and was acceptable for using as a response variable in the GLM and GAM.
3.2 GLM analysis
The best GLM model selected by BIC is shown in Table 3. A summary of the best fitting model is shown in Table 4. All explanatory variables, including the interaction terms, were statistically significant (P<0.05). In the 5-fold cross validation test, high correlation was observed for the best GLM model (Table 5). The q-q plot and the residual plots for the explanatory variables for evaluating the distribution assumption are shown in Fig. 3. Residuals showed an approximately normal distribution around 0, which indicated that the model assumptions were satisfied.
3.3 GAM analysis
The best GAM model selected by BIC is shown in Table 6. A summary of fitting to the best model is shown in Table 7. All explanatory variables, including the interaction terms, were statistically significant. In the 5-fold cross validation test, a high correlation was observed for the best GAM model (Table 8). The q-q plot and the residual plots for the explanatory variables for evaluating the distribution assumption are shown in Fig. 4. Residuals were plotted around 0 and displayed a normal distribution despite a few observed biases.
Figure 5 displays the residuals and explanatory variables fitted by the best GAM. The mean residuals of the GAM model over the years ranged from –1.15 to 0.94. The year 2011 showed the largest residual values for the GAM model and the year 2008 showed the largest negative residual values. The mean residuals of the GAM model over the months ranged from –0.98 to 0.32. September showed the largest positive residual values for the GAM model and December had the largest negative residual values compared with that from other month. For the spatial explanatory variables, Longitude and Latitude showed relatively smaller residuals, which were close to zero. As environmental explanatory variables, SST, SSTG and SSH showed similar distributions of residuals, which were also close to zero. The mean residuals of the GAM model over Vessellength ranged from –0.21 to 0.28.
Effects of temporal, spatial, environmental, and fisheries operational variables on Pacific saury CPUE are shown in Fig. 6. CPUE gradually declined in 2004–2008 and reached its lowest point in 2008. CPUE increased steadily from 2008 to 2014, and then it decreased slightly in 2014 (Fig. 6a). During the main fishing season (May–December), CPUE showed a gradual increase with the Month (Fig. 6b). The spatial factor Longitude had a great influence on the CPUE in the range of 145°–155°E (Fig. 6c). The effect of Latitude on CPUE decreased gradually from the South to the North (Fig. 6d). As for environmental factors, SST had different effects on CPUE within different temperature ranges. When the temperature was between 10°C and 18°C, the effect of SST on CPUE was relatively stable. Between 18°C and 23°C, the effect showed a decreasing trend first, which then reversed (Fig. 6e). The effect on CPUE is a gradual increase with increasing SSTG (Fig. 6f). The effect of SSH also showed some fluctuations (Fig. 6g). Vessellength was one of the most important factors affecting CPUE; as Figure. 6h shows, Vessellength had a positive correlation with CPUE.
3.4 Yearly trend of standardized CPUE
We used the best GLM and GAM models to estimate the yearly trend of standardized CPUE values with a 95% confidence interval (Fig. 7a). The estimates of standardized and nominal CPUE values from 2003 to 2017 are shown in Table 9. Besides 2004–2005 and 2008–2009, standardized CPUE values by the GLM model were significantly lower than the corresponding nominal CPUE values. As for the GAM model, the standardized CPUE values were significantly lower than the corresponding nominal CPUE values. However, there was little difference between CPUEs values standardized by GLM and GAM; this may be related to the assumption of relationships between CPUEs and explanatory variables. So linear relations maybe weak between CPUEs with explanatory variables, deviance explained is 21.57% (Table 3), however, more nonlinearity relations maybe available for them, deviance explained is 38.95% (Table 6). CPUE decreased from May to August (Fig. 7b), whereas it increased and then decreased from September to December, with the highest CPUE value occurring in October.
4 Discussion
Less
4.1 Effect of temporal-spatial factors on CPUE
The GLM and GAM analyses indicated that temporal (i.e., Year, Month) and spatial factors (i.e., Longitude and Latitude), all had a significant effect on CPUE values (p<0.05). GAM analysis suggested that the two most important factors, which accounted for the largest percentage of the CPUE value, were Longitude and Year. Fluctuations of resources, as well as changes in marine environmental conditions, climate, and fishing effort over time, eventually led to yearly and seasonal fluctuations in CPUE. During 2005–2008, the annual mean CPUE value showed obvious decreases and dropped to the lowest level over the past 12 years in 2008 (Fig. 7a). In 2009, CPUE showed a significant increase, but then gradually decreased. Changes in fishing effort may be the main cause of yearly fluctuations and declines in CPUE value. Further, CPUE differed significantly among months (p<0.01), decreasing from May to August (Fig. 7b), while increasing and then decreasing from September to December, peaking in October, a result that was consistent with those reported by Wu et al. (2015) and Yan (2012). The seasonal migration of saury and the production time of the saury fishery are the main reasons for the seasonal fluctuation in CPUE value; thus, the time variable month showed an impact on CPUE value.
The results of this study indicated that, in the range from 38.5° to 44°N and from 145° to 155°E, overall, CPUE values increased with increasing Longitude and Latitude but fluctuated, peaking at a marine site near 44°N, 155°E; further, this fluctuation may be associated with the migratory route of Pacific saury. Before August, Pacific saury migrates from the South to the North for feeding, and they start to reverse the route from near 46°N in September for feeding and over-winter migration (Tian et al., 2003; Shen et al., 2004). During its northward migration, Pacific saury is widely distributed, and the fishing grounds are relatively dispersed; whereas, during its southward migration, plankton growth is promoted by the increasing strength of the Oyashio Current and the richness in nutrients of seawater, which causes the school of Pacific saury to gradually concentrate near the coastal waters (He et al., 1999), thereby increasing the fishery catch continually until the end of the fishing season in early December.
4.2 Effect of environmental factors on CPUE
The distribution of Pacific saury is tightly associated with marine environmental factors (Takasuka et al., 2014; He et al., 1999; Wang et al., 2012), as reported in many studies based on environmental factor data obtained via remote sensing (Stephens and MacCall, 2004). The SST data in this study were obtained via on-board monitoring. Our results indicated that CPUE was significantly correlated with SST (p<0.01), and slowly increased with increasing SST within a temperature range from 11 to 15°C, which was consistent with the findings reported by Zhu et al. (2006b). CPUE was significantly correlated with SSH (p<0.01) and SSTG (p<0.01). For the SSH, CPUE increased with increasing SSH within a range from 0.2 to 0.6 m and have some fluctuation between a range from 0.6 to 1.2 m. CPUE increased with increasing SSTG within a range from 0.05 to 0.2°C/km. A similar result was obtained by Tian (2004) while studying Pacific saury fishery data in Japan. It can be seen that environmental factors have a significant impact on CPUE and should be taken into account in future studies.
4.3 Effect of vessel performance on CPUE
Pacific saury fishing is conducted using a stick-held net under light induction (Yu et al., 2006), where fish-gathering lamps are used to induce a school of fish into a fishing net on one side of the vessel, and the net is then hauled to catch the fish (Yang et al., 2005). It is generally accepted that the Vessellength significantly affects Pacific saury fishing. Indeed, the results from the GLM and GAM models indicated that the Vessellength significantly affected CPUE (p<0.01); the reason for this may be that vessel length affected the working space in Pacific saury fishing vessels, the size of the fishing gear, the convenience of the fishing process, and the capacity for processing and freezing caught fish, all of which, together, resulted in a large impact on CPUE of Pacific saury fishing vessels.
4.4 Comparison between GLM and GAM
GLMs and GAMs are commonly used to standardize CPUE values; however, both show advantages and limitations. GLMs assume that there is a linear relationship between response variables and the explanatory variables. However, nonlinear relationships are common between fish densities and environmental factors (Walsh and Kleiber, 2001; Denis et al., 2002). For example, the three most important predictors in the main effects models were SST, Latitude, and Longitude, all of which may have non-linear relationships with Pacific saury CPUE. Contrarily, GAMs are extensions of GLMs in which the explanatory variables have been replaced with smooth functions to deal with nonlinear relationships between the response variable and explanatory variables. However, the smoothening functions of GAM models cannot infer predictions outside of the range of the training data that were used to build the model effectively (Frescino et al., 2001). Any value in the test sets out of the training data range would be assigned to the closest maximum or minimum values of the training data set. Nonetheless, both GLM and GAM showed good performance for estimating relative abundance index in this study. Our experience confirms that different models are suitable for different situations and data sources.
In our study, GLM analysis indicated that the Year, Month, Longitude, Latitude, SST, SSTG, SSH, Vessellength and the interaction terms Year×Month, Month×Longitude, and Month×Latitude, Longitude×Latitude were all significant variables with highly significant effects on CPUE (p<0.05), whereas GAM analysis indicated that interaction terms include Year×Month, Year×Longitude, Month×Longitude, Month×Latitude, Longitude×Latitude. The GAM model effectively explained a higher average percentage of Deviance than the GLM model (Tables 3 and 6). We can see that linear relations may be weak between CPUE values and an explanatory variable (deviance explained was only 21.57% in this case); however, more nonlinear relations were available for CPUE values, whereby deviance explained reached 38.95%. Comparing the results of cross-validation tests in GLM and GAM analyses (Tables 5 and 8), higher Spearman's correlation and lower mean squared error (MSE) between observed and predicted values of test data were observed by GAM. Therefore, GAM is likely to be more suitable than GLM for CPUE standardization of Pacific saury fishing in the Northwest Pacific Ocean.
In summary, the application of statistical models such as GAMs, which perform better than GLMs with the nonlinearity of predictors and spatial autocorrelation, should be given priority for the future CPUE standardization. Furthermore, in most cases, fishery data are commercial fishery data, which do not correspond to a design. The predicted relative resource abundance index will be unreliable and remain considerable uncertainty if the survey area is not enough due to the lack of sampling locations or biased designs. In this situation, GAMs should be preferred (Yu et al., 2013). Therefore, GAMs are appropriate for standardization of Pacific saury CPUE. This study contributes to Pacific saury fishery research by recommending potential statistical approaches for standardizing CPUE, which can provide a solid support to fishery stock assessment.
5 Conclusions
Less
Fishery management and conservation depend heavily on accurate fish stock assessments which always need reliable relative abundance index. In order to get a more reliable fishery CPUE, a well-performed CPUE standardization model is needed. In this study we evaluated the performance of statistical methods including GLMs and GAMs, using CPUE data collected from the Chinese saury fishery in the Northwest Pacific Ocean from 2003 to 2017. Further, we evaluated the impact of spatial, temporal, environmental variables and vessel length on CPUE. The significant variables were used individually in the GLM/GAM to select an optimal GLM/GAM model based on BIC. Subsequently, we used deviance explained and results from 5-fold bootstrap cross-validations to compare the performance of the two types of model. The standardized indices of abundance from the two models studied here suggested a gradual fluctuation trend in Pacific saury CPUE for the Northwest Pacific Ocean. The performance of GLM and GAM were evaluated to determine the most robust model for standardizing CPUE as an index of abundance for Pacific saury. GAM models were more suitable than GLM models for fitting the fishery data. The reason was that GAMs could fit the nonlinear relationships that exist between the response variables and the explanatory variables. These methods could be reproduced efficiently and used to examine the spatial/temporal dynamics of fishing activities.
Although we have obtained some valuable research results, we believe there are still some important shortcomings: (1) The relationship between fish distribution and the environmental factors over time is not static and fishing space is important for discovering the real fish abundance variation. (2) Fish species are not randomly distributed in the ocean, but tend to cluster in certain habitats and may be rare or totally absent in others. (3) AIC is the most popular criterion for model selection when GLMs/GAMs are used to estimate CPUE data. However, AIC may overestimate the effect of the number of parameters in the case of small samples, which may cause unreliable results. In future studies, we will emphasize on model improvement in order to provide better recommendations for management and conservation.
6 Data availability statement
Less
Conflict of interest: The authors declare that they have no conflict of interest.
Data availability: The datasets created during and/or analyzed during the current study are available from the corresponding author upon reasonable request.
Human and animal rights: All applicable international, national, and/or institutional guidelines for the care and use of animals were followed.
Acknowledgements
Less
We thank Siquan Tian, Bai Li, Jie Cao and Luoliang Xu for their valuable comments and advice. Thanks are also given to other laboratory colleagues for field and laboratory work assistance.
Funding
Less
  • The National Sci-Tech Support Plan “Fishing Technology and New Resources in Oceanic Fisheries” under contract No. 2013BAD13B05.
References
Less
Arlot S, Celisse A. 2010. A survey of cross-validation procedures for model selection. Statistics Surveys, 4: 40–79, doi: 10.1214/09-SS054
Campbell R A. 2004. CPUE standardisation and the construction of indices of stock abundance in a spatially varying fishery using general linear models. Fisheries Research, 70(2–3): 209–227, doi: 10.1016/j.fishres.2004.08.026
Chen Xinjun, Liu Bilin, Chen Yong. 2008. A review of the development of Chinese distant-water squid jigging fisheries. Fisheries Research, 89(3): 211–221, doi: 10.1016/j.fishres.2007.10.012
Denis V, Lejeune J, Robin J P. 2002. Spatio-temporal analysis of commercial trawler data using general additive models: patterns of Loliginid squid abundance in the north-east Atlantic. ICES Journal of Marine Science, 59(3): 633–648, doi: 10.1006/jmsc.2001.1178
Erisman B E, Allen L G, Claisse J T, et al. 2011. The illusion of plenty: hyperstability masks collapses in two recreational fisheries that target fish spawning aggregations. Canadian Journal of Fisheries and Aquatic Sciences, 68(10): 1705–1716, doi: 10.1139/f2011-090
Frescino T S, Edwards T C Jr, Moisen G G. 2001. Modeling spatially explicit forest structural attributes using generalized additive models. Journal of Vegetation Science, 12(1): 15–26, doi: 10.1111/j.1654-1103.2001.tb02613.x
Harley S J, Myers R A, Dunn A. 2001. Is catch-per-unit-effort proportional to abundance?. Canadian Journal of Fisheries and Aquatic Sciences, 58(9): 1760–1772, doi: 10.1139/f01-112
He Min, Song Wenling, Chen Xingfang. 1999. Typhoon activity in the Northwest Pacific in relation to El Niño/La nina events. Journal of Tropical Meteorology (in Chinese), 15(1): 17–25
Howell E A, Kobayashi D R. 2006. El Niño effects in the Palmyra Atoll region: oceanographic changes and bigeye tuna (Thunnus obesus) catch rate variability. Fisheries Oceanography, 15(2): 477–489
Hua Chuanxiang, Zhu Qingcheng, Xu Wei. 2010. Fishing ground distribution of cololabis saira in the Northwestern Pacific. Shandong Fisheries (in Chinese), 27(10): 10–13
Huang Hongliang, Zhang Xun, Xu Baosheng, et al. 2005. Preliminary analysis on the fishing grounds of Cololabis saira in the North Pacific Ocean. Marine Fisheries, 27(3): 206–212
Kohavi R. 2001. A study of cross-validation and bootstrap for accuracy estimation and model selection. International Joint Conference on Artificial Intelligence. Stanford, CA: Morgan Kaufmann Publishers Inc
Lin Longshan. 2003. Fishery survey of Stick-held Net for Cololabis saira in Taiwan. Marine Fisheries (in Chinese), (4): 200–203
Martínez-Rincón R O, Ortega-García S, Vaca-Rodríguez J G. 2012. Comparative performance of generalized additive models and boosted regression trees for statistical modeling of incidental catch of wahoo (Acanthocybium solandri) in the Mexican tuna purse-seine fishery. Ecological Modelling, 233: 20–25, doi: 10.1016/j.ecolmodel.2012.03.006
Maunder M, Punt A E. 2004. Standardizing catch and effort data: a review of recent approaches. Fisheries Research, 70(2–3): 141–159, doi: 10.1016/j.fishres.2004.08.002
Maunder M N, Start P J. 2003. Fitting fisheries models to standardised CPUE abundance indices. Fisheries Research, 63(2): 43–50
Menard SW. 1995. Applied Logistic Regression Analysis. Thousand Oaks, CA: SAGE
Nishida T, Chen Dinggeng. 2004. Incorporating spatial autocorrelation into the general linear model with an application to the yellowfin tuna (Thunnus albacares) longline CPUE data. Fisheries Research, 70(2–3): 265–274, doi: 10.1016/j.fishres.2004.08.008
Ortiz M, Arocha F. 2004. Alternative error distribution models for standardization of catch rates of non-target species from a pelagic longline fishery: billfish species in the Venezuelan tuna longline fishery. Fisheries Research, 70(2–3): 275–297, doi: 10.1016/j.fishres.2004.08.028
Quinn G P, Keough M J. 2002. Experimental Design and Data Analysis for Biologists. Cambridge: Cambridge University Press
Rodríguez-Marín E, Arrizabalaga H, Ortiz M, et al. 2003. Standardization of bluefin tuna, (Thunnus thynnus) catch per unit effort in the baitboat fishery of the Bay of Biscay (Eastern Atlantic). ICES Journal of Marine Science, 60(1): 1216–1231
Shen Jianhua, Han Shixin, Fan Wei, et al. 2004. Saury Resource and Fishing Grounds in the Northwest Pacific. Marine Fisheries (in Chinese), 26(1): 61–65
Shono H. 2005. Is model selection using Akaike’s information criterion appropriate for catch per unit effort standardization in large samples?. Fisheries Science, 71(5): 978–986, doi: 10.1111/j.1444-2906.2005.01054.x
Stephens A, Maccall A. 2004. A multispecies approach to subsetting logbook data for purposes of estimating CPUE. Fisheries Research, 70(2–3): 299–310, doi: 10.1016/j.fishres.2004.08.009
Sun Manchang, Ye Xuchang, Zhang Jian, et al. 2003. Probe into Pacific saury fisheries in the northwest Pacific Ocean. Marine Fisheries (in Chinese), 25(3): 112–115
Takasuka A, Kuroda H, Takeshi O, et al. 2014. Occurrence and density of Pacific saury Cololabis saira larvae and juveniles in relation to environmental factors during the winter spawning season in the Kuroshio Current system. Fisheries Oceanography, 23(4): 304–321, doi: 10.1111/fog.12065
Tian Y J, Akamine T, Suda M. 2003. Variations in the abundance of Pacific saury (Cololabis saira) from the Northwestern Pacific in relation to oceanic-climate changes. Fisheries Research, 60(2–3): 439–454, doi: 10.1016/S0165-7836(02)00143-1
Tian Y J, Ueno Y, Suda M, et al. 2004. Decadal variability in the abundance of Pacific saury and its response to climatic/oceanic regime shifts in the northwestern subtropical Pacific during the last half century. Journal of Marine Systems, 52: 235–257, doi: 10.1016/j.jmarsys.2004.04.004
Tien B D, Lofman O, Revhaug I, et al. 2011. Landslide susceptibility analysis in the Hoa Binh province of Vietnam using statistical index and logistic regression. Natural Hazards, 59(3): 1413–1444, doi: 10.1007/s11069-011-9844-2
Tseng C T, Su N J, Sun C L, et al. 2013. Spatial and temporal variability of the Pacific saury (Cololabis saira) distribution in the northwestern Pacific Ocean. ICES Journal of Marine Science, 70(5): 991–999, doi: 10.1093/icesjms/fss205
Venables W N, Dichmont C M. 2004. GLMs, GAMs and GLMMs: an overview of theory for applications in fisheries research. Fisheries Research, 70(2–3): 319–337, doi: 10.1016/j.fishres.2004.08.011
Walsh W A, Kleiber P. 2001. Generalized additive model and regression tree analyses of blue shark (Prionace glauca) catch rates by the Hawaii-based commercial longline fishery. Fisheries Research, 53(2): 115–131, doi: 10.1016/S0165-7836(00)00306-4
Wang Zhizu, Zuo Juncheng, Chen Meixiang, et al. 2012. Relationship between El Niño and sea surface temperature variation in coastal region of Yellow Sea and East China Sea. Journal of Hohai University (Natural Sciences) (in Chinese), 40(4): 461–468
Ward H G M, Askey P J, Post J R. 2013. A mechanistic understanding of hyperstability in catch per unit effort and density-dependent catchability in a multistock recreational fishery. Canadian Journal of Fisheries and Aquatic Sciences, 70(10): 1542–1550, doi: 10.1139/cjfas-2013-0264
Watanabe K, Tanaka E, Yamada S, et al. 2006. Spatial and temporal migration modeling for stock of Pacific saury Cololabis saira (Brevoort), incorporating effect of sea surface temperature. Fisheries Science, 72(6): 1153–1165, doi: 10.1111/j.1444-2906.2006.01272.x
Wood S N. 2006. Generalized Additive Models: An Introduction with R. London: Chapman and Hall/CRC, 410
Wu Yue, Huang Hongliang, Liu Jian, et al. 2015. Spatiotemporal distribution pattern of saury fishing grounds and catch yield per unit effort in the Northern Pacific high sea in 2014. Fishery Modernization (in Chinese), 42(3): 61–64
Xia Hui. 2008. The illumination distribution model of the pacific saury (Cololabis saira) stick-held dip net fishing (in Chinese) [dissertation]. Shanghai: Shanghai Ocean University, 1-54
Xu Wei, Zhu Qingcheng, Zhang xiancun, et al. 2005. Bouke net fishing technology of Pacific saury in the Northweatern Pacific. Shangdong Fisheries (in Chinese), 22(10): 43–46
Yan Lei. 2012. The Relationship between the distribution of saury fishing ground and its environmental factors (in Chinese) [dissertation]. Shanghai: Shanghai Ocean University, 1–51
Yan Lei, Zhu Qingcheng, Zhang Yang, et al. 2012. Fishing ground distribution of saury and its correlation with SST in the Northern Pacific high sea in 2010. Journal of Shanghai Ocean University (in Chinese), 21(4): 609–615
Yang Xiulan, Wang Pengfei, Jiao Yulong, et al. 2005. Study on the culture technique in the middle stage and the growing character of Apostichopus japonicus. Shandong Fishery (in Chinese), 22(10): 43–46
Yu Hao, Jiao Yan, Carstensen L W. 2013. Performance comparison between spatial interpolation and GLM/GAM in estimating relative abundance indices through a simulation study. Fisheries Research, 147: 186–195, doi: 10.1016/j.fishres.2013.06.002
Yu Yuefeng, Zhang Xun, Huang Hongliang, et al. 2006. Study on attracting fish method of stick-held net for Cololabis saira. Journal of Zhejiang Ocean University (Natural Science) (in Chinese), 25(2): 154–156
Zhang Xiaomin, Zhu Qingcheng, Hua Chuanxiang. 2015. Fishing ground distribution of saury and its correlation with marine environment factors in the Northern Pacific high sea in 2013. Journal of Shanghai Ocean University (in Chinese), 24(5): 773–782
Zhang Yang, Zhu Qingcheng, Yan Lei, et al. 2013. Preliminary study on biological characteristics of Cololabis Saira in the Northwest Pacific ocean in Spring. Transactions of Oceanology and Limnology (in Chinese), (1): 53–60
Zhu Qingcheng, Hua Chuanxiang, Xu Wei, et al. 2006a. The fishing ground distribution of Cololabis saira and its relationship with water temperature factors in the Northwestern Pacific from July to September. Marine Fisheries (in Chinese), 28(3): 228–233
Zhu Guoping, Zhu Qingcheng, Chen Jintao, et al. 2006b. Preliminary study on relationship between Cololabis saira fishing ground and temperature factor in the Northern Pacific Ocean. Marine Sciences (in Chinese), 30(7): 91–96
Zou Xiaorong, Zhu Qingcheng. 2006. Preliminary analysis on the relationship between the distribution of fishing ground of pacific saury (Cololabis saira) and SST in northwest pacific. Journal of Zhanjiang Ocean University (in Chinese), 26(6): 26–30
Appendix
Less
Year 2019 volume 38 Issue 10
PDF
36
19
Cite this Article
BibTeX
Article Info
doi: 10.1007/s13131-019-1486-3
  • Receive Date：2018-11-02
  • Online Date：2026-04-01
  • Published：2019-10-25
Article Data
Affiliations
History
  • Received：2018-11-02
  • Accepted：2019-01-22
Funding
The National Sci-Tech Support Plan “Fishing Technology and New Resources in Oceanic Fisheries” under contract No. 2013BAD13B05.
Affiliations
    1 College of Marine Sciences, Shanghai Ocean University, Shanghai 201306, China
    2 National Engineering Research Center for Pelagic Fishery, Shanghai 201306, China

Corresponding:

References
Share
https://castjournals.cast.org.cn/joweb/aos/EN/10.1007/s13131-019-1486-3
Share to
QR

Scan QR to access full text

Cite this article
BibTeX
Citations
表12种不同金属材料的力学参数
科
Family		 属数
Number of
genus		 种数
Number of
species		 占总种数比例
Percentage of
total species (%)		 属
Genus		 种数
Number of
species		 占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏
  • BibTeX
  • EndNote
  • RefWorks
  • TxT
Links
Articles
Latest Articles
Most Read
Collections
Updates
Events
News
Multimedia
About
About Us
Contact
No. 86 Xueyuan South Road, Haidian District, Beijing
100081
010-62199257
qkjq@cast.org.cn

Copyright © 2025 China Association for Science and Technology. All rights reserved. For all open access content, the relevant licensing terms apply.
Sponsored by the Office of the Leading Group for Cybersecurity and Informatization of CAST, and supported by Science and Technology Review Publishing House