The Use of the GWPCA-MGWR Model for Studying Spatial Relationships between Environmental Variables and Longline Catches of Yellowfin Tunas

Abstract

The yellowfin tuna represents a significant fishery resource in the Pacific Ocean. Its resource endowment status and spatial variation mechanisms are intricately influenced by marine environments, particularly under varying climate events. Consequently, investigating the spatial variation patterns of dominant environmental factors under diverse climate conditions, and understanding the response of yellowfin tuna catch volume based on the spatial heterogeneity among these environmental factors, presents a formidable challenge. This paper utilizes comprehensive 5°×5° yellowfin tuna longline fishing data and environmental data, including seawater temperature and salinity, published by the Western and Central Pacific Fisheries Commission (WCPFC) and the Inter-American Tropical Tuna Commission (IATTC) for the period 2000–2021 in the Pacific Ocean. In conjunction with the Niño index, a multiscale geographically weighted regression model based on geographically weighted principal component analysis (GWPCA-MGWR) and spatial association between zones (SABZ) is employed for this study. The results indicate the following: (1) The spatial distribution of dominant environmental factors affecting the catch of Pacific yellowfin tuna is primarily divided into two types: seawater temperature dominates in the western Pacific Ocean, while salinity dominates in the eastern Pacific Ocean. When El Niño occurs, the area with seawater temperature as the dominant environmental factor in the western Pacific Ocean further extends eastward, and the water layers where the dominant environmental factors are located develop to deeper depths; when La Niña occurs, there is a clear westward expansion in the area with seawater salinity as the dominant factor in the eastern Pacific Ocean. This change in the spatial distribution pattern of dominant factors is closely related to the movement of the position of the warm pool and cold tongue under ENSO events. (2) The areas with a higher catch of Pacific yellowfin tuna are spatially associated with the dominant environmental factor of mid-deep seawater temperature (105–155 m temperature) to a greater extent than other factors, the highest correlation exceeds 70%, and remain relatively stable under different ENSO events. The formation of this spatial association pattern is related to the vertical movement of yellowfin tuna as affected by subsurface seawater temperature. (3) The GWPCA-MGWR model can fully capture the differences in environmental variability among subregions in the Pacific Ocean under different climatic backgrounds, intuitively reflect the changing areas and influencing boundaries from a macro perspective, and has a relatively accurate prediction on the trend of yellowfin tuna catch in the Pacific Ocean.

1. Introduction

The yellowfin tuna (Thunnus albacares) is a social species predominantly found in tropical and subtropical meso-open waters. It is also a significant catch for global longline fisheries, garnering considerable attention from nations worldwide [1]. The Pacific Ocean serves as the primary fishing ground for yellowfin tuna, contributing to over 60% of global production [2]. However, there is debate regarding whether the Pacific’s yellowfin tuna population has been overfished based on existing resource assessments [3,4,5]. Factors such as the marine environment significantly influence both the catch and the spatial distribution of yellowfin tuna [1,6,7,8]. Notably, climate events like ENSO can alter various oceanic environmental variables, leading to noticeable changes in the tuna’s distribution [9,10,11,12]. Consequently, understanding how multiple environmental variables respond to different climate patterns in relation to the catch and spatial distribution of yellowfin tuna in the Pacific Ocean is crucial.

Currently, numerous studies have been conducted on the relationship between yellowfin tuna catch and the marine environment. These primarily utilize methods such as mark–recapture [1], the acoustic method [6], wavelet analysis [9], time series analysis [10], a machine learning model [13], generalized additive models (GAMs) [8,11,14,15,16], an artificial neural network model (ANNS) [5], and an XGBRF integrated model [17]. These methods effectively reflect the response laws between the amount of yellowfin tuna resources and their spatial distribution, as well as changes in environmental factors, thereby constructing corresponding relationships or predictive models. However, it has been observed that due to the existence of multiple distinct hotspot areas for yellowfin tuna in the Pacific Ocean, there are significant environmental differences within these hotspots. Additionally, there is multicollinearity among marine environmental factors, which often leads these methods to overlook the spatial heterogeneity between the catch situations of yellowfin tuna in different ocean areas and changes in environmental factors or to omit some important influencing factors. Existing research frequently avoids the impact of limitations on model construction by limiting the scope of the research area or using limited fishing data from a specific commercial fleet [5,11,12,15,17,18,19,20,21,22,23,24,25,26]. This artificially disconnects the spatial connection of yellowfin tuna resources, which is not conducive to the macro-analysis research of yellowfin tuna in ocean areas. These methodological limitations cause the lack of research on the spatial distribution of yellowfin tuna catch under different climatic backgrounds in ocean areas and the changing relationship between multiple environmental variables.

The geographically weighted principal component analysis (GWPCA) model effectively avoids the impact of multicollinearity among marine environmental factors. It fully accounts for variations in dominant environmental factors across different subregions and the synergistic effects of these variables within specific regions of the Pacific Ocean. At the same time, the multiscale geographically weighted regression (MGWR) model assigns each marine environmental factor a unique spatial smoothing scale. This approach not only reduces model bias but also yields a more accurate geospatial representation. It facilitates the exploration of yellowfin tuna catch responses to environmental changes across multiple spatial scales, enabling more precise analyses and evaluations. Both models have garnered attention in fields such as soil science [27,28], economics [29,30], sociology [31,32], agriculture [33,34], and others. Their efficacy in examining heterogeneity across spatial scales and analyzing response patterns is well established. This research amalgamates the strengths of both models to introduce a multiscale geographically weighted regression model grounded in GWPCA: GWPCA-MGWR. This integration capitalizes on the distinct advantages of both models, fostering advancements in related studies on Pacific yellowfin tuna. The SABZ is a suitable method for quantitatively comparing the degree of association in the spatial distribution of regionalized variables. It predominantly quantifies the degree of association among variables within a singular region by determining the overlapping area between them. This approach enables the seamless integration of variables from discrete domains back into continuous ones, thereby enhancing the precision of correlation calculations among variables [35]. It compensates for the shortcomings of traditional counting methods to represent the magnitude of association [35] and can better reflect the spatial relationship between yellowfin tuna catch in the Pacific Ocean and environmental factor changes.

This study incorporates the GWPCA-MGWR model into yellowfin tuna fishery resource research, addressing the existing gap in studies on the response of yellowfin tuna catch to environmental changes in the Pacific Ocean. This research utilizes fishery data from the Western and Central Pacific Fisheries Commission (WCPFC) and Inter-American Tropical Tuna Commission (IATTC), which includes longline fishing data for yellowfin tuna published between 2000 and 2021 in the Pacific Ocean. By employing the GWPCA-MGWR method in conjunction with the Niño index, this paper examines the variation and spatial differences in the leading environmental factors impacting yellowfin tuna catch under varying ENSO events in the Pacific Ocean. Furthermore, it integrates spatial association between zone methods to delve into the relationship between the spatial distribution status of yellowfin tuna catch under different ENSO events and the pattern of marine environmental change. This leads to the establishment of relevant prediction models and verifies their accuracy. The findings provide a more robust scientific foundation and reference for studying the rational development and utilization of yellowfin tuna fishery resources in the Pacific Ocean and for constructing related models.

2. Data and Method

2.1. Data Sources

2.1.1. Fishery Data

The fishery data utilized in this study were sourced from the official websites of the WCPFC and IATTC. These sites provided fishery production data with a spatial resolution of 5° × 5° and a temporal scale spanning months. The collected data included details such as operational duration, the longitude and latitude of operations, and the quantity and weight of the species captured. The chosen time frame spanned from 2000 to 2021. Figure 1 depicts the partitioning of the WCPFC and the IATTC utilized for conducting a yellowfin tuna fishery resource assessment. This figure also delineates the geographical area under investigation in this study, which spans from 130° E to 80° W and from 40° S to 45° N.

In the figure, the numbers represent WCPFC partition labels, while the letters represent IATTC partition labels.

2.1.2. Environmental Data

The biological characteristics of yellowfin tuna suggest that many vital life activities, such as spawning and foraging, are intrinsically linked to the marine environment. Consequently, changes in this environment can significantly impact the resource status and spatial distribution of yellowfin tuna. Studies have shown that temperature [36,37,38] and seawater flow velocity [13,39] at varying depths influence the body temperature status and spawning behavior of yellowfin tuna. Furthermore, salinity [40,41] at different depths affects the quantity and distribution of nutrients required by these tunas [42], which subsequently impacts their foraging, migration, and other behaviors. Additionally, dissolved oxygen concentration [19,43] directly influences the physiological state, organ morphological structure, and movement of yellowfin tuna. This study incorporates public data to select 21 environmental factors related to the spatial distribution of Pacific yellowfin tuna, as detailed in

Table 1. Environmental factors and their sources.

Variables	Unit	Source
Dissolved oxygen concentration (DO)	mg/L	Commentary on the Coupled Model Intercomparison Project Phase 6 https://esgf-node.llnl.gov/search/cmip6/ accessed on 1 March 2023
Mixed layer depth (MLD)	m	Oregon State University Data Center http://www.science.oregonstate.edu/ accessed on 1 March 2023
Net primary productivity (NPP)	Mg/m2/day
Sea level anomaly (SLA)	m	Copernicus Marine Environment Monitoring Service http://marine.copernicus.eu/ accessed on 1 March 2023
Temperature of Sea at Various Depths (T5, T55, T105, T155, T205, T255)	°C	NOAA/NWS Space Weather Prediction Center https://cfs.ncep.noaa.gov/ accessed on 1 March 2023
Salinity of Sea at Various Depths (S5, S55, S105, S155, S205, S255)	PSU
Sea water velocity (U5, U55, U105, V5, V55, V105)	m/s

In the table, T5 represents the temperature of the sea at a depth of five meters; S5 represents the salinity of the sea at a depth of five meters; U5 represents the east–west sea water velocity; U5 represents the south–north sea water velocity, etc.

. The environmental data used in this study cover the period from 2000 to 2021, with all data points matched with fishery data on a monthly time scale and a 5° × 5° spatial scale.

2.1.3. Determination of ENSO Events

The El Niño–Southern Oscillation (ENSO) is a potent sea–air interaction phenomenon capable of triggering global climate chain reactions [44]. At its core, the ENSO phenomenon is characterized by anomalous fluctuations in subsurface Pacific Ocean temperatures. Specifically, abnormal warming events are referred to as El Niño events, while their cold counterparts are termed La Niña events [45,46]. Current research indicates that the occurrence of ENSO events leads to significant alterations in various environmental factors associated with yellowfin tuna fishing grounds. These include changes in Pacific Ocean currents, thermoclines, nutrient levels, and seawater temperature salinity. Such changes can profoundly influence the distribution of yellowfin tuna resources [11,44].

The judgment of El Niño and La Niña events in this study is based on the Niño index from the NOAA Climate Forecast Center (NOAA Climate Forecast Center), and the judgment standard is according to the ‘Method for Judging El Niño/La Niña Events’ issued by the China Meteorological Administration in 2017. When the Niño index exceeds +0.5 for 5 consecutive months, it is judged as an El Niño event, and when the Niño index falls below −0.5 for 5 consecutive months, it is judged as a La Niña event [47,48].

Table 2. Schedule of different ENSO events.

ENSO Event Type	Start and End Time (Year.Month)
Moderate El Niño	2009.10–2010.03
Moderate La Niña	2010.06–2010.12
Weak La Niña	2011.09–2011.12
Normal year	2014.03–2014.09
Super-strong El Niño	2015.10–2016.03
Weak El Niño	2018.10–2019.03

displays the six different climate backgrounds identified during the study period.

2.2. Models and Methods

2.2.1. Selection of Environmental Factors

Given the intricate relationship between environmental factors and model construction feasibility, selecting an excessive number of factors not only fails to accurately represent this relationship but also compromises model performance, resulting in inaccurate outcomes. Consequently, a comprehensive correlation analysis was performed on all environmental factors and yellowfin tuna catch data from 2000 to 2021, as presented in

Table 3. Correlation of environmental variables with yellowfin tuna catch.

Environment Variables	Correlation	p-Value
T5	0.50	0.00
T55	0.53	0.00
T105	0.57	0.00
T155	0.53	0.00
T205	0.38	0.00
T255	0.18	0.00
S5	0.15	0.00
S55	0.17	0.00
S105	0.31	0.00
S155	0.42	0.00
S205	0.41	0.00
S255	0.31	0.00
NPP	−0.22	0.00
MLD	0.07	0.15
SLA	−0.06	0.24
DO	−0.24	0.00
V5	−0.13	0.00
V55	0.14	0.00
V105	0.17	0.00
U5	−0.20	0.00
U55	−0.14	0.00
U105	−0.03	0.49

. Based on the correlation findings, twelve environmental variables—T5, T55, T105, T155, T205, S105, S155, S205, S255, NPP, DO, and U5—were identified with an absolute correlation index value exceeding 0.2 and thus selected for further study.

2.2.2. Principal Component Analysis

Principal component analysis (PCA), a multivariate statistical method extensively utilized across various research domains, is frequently employed to investigate trends in multivariable changes [49]. PCA optimizes the correlation of high-dimensional multivariate data, subsequently transforming it into a reduced number of low-dimensional independent variables that significantly influence the data. These transformed variables are referred to as principal components [27]. The first principal component (PC1) accounts for the majority of the data content, while each subsequent principal component elucidates the maximum degree of variability not accounted for by its preceding component. This process effectively eliminates redundant information, integrates pertinent data, and mitigates the impact of collinearity among factors [50].

2.2.3. GWPCA-MGWR

The multiscale geographically weighted regression method based on geographically weighted principal component analysis (GWPCA-MGWR), amalgamates the benefits of both geographically weighted principal component analysis and multiscale geographically weighted regression. Prior to executing the multiscale geographically weighted regression analysis, the independent variables undergo a process of geographically weighted principal component analysis. Subsequently, several uncorrelated principal component score values, extracted as independent variables for MGWR, are subjected to multiscale geographically weighted regression. This process aims to analyze the impact of environmental factors on the spatial distribution of yellowfin tuna catch. The specific steps involved in this procedure are as follows:

(1)

Geographically Weighted Principal Component Analysis (GWPCA)

The GWPCA model, an advancement of the traditional PCA [51], integrates spatial information of variables during the computation of principal components. This integration addresses the spatial heterogeneity issue often overlooked by conventional PCA [49,52]. The GWPCA model thoroughly examines the spatial effects between factors and uncovers previously obscured details that may influence the spatial distribution pattern of Pacific yellowfin tuna [30,33]. In the GWPCA model, the principal components are derived from the variance-covariance matrix. This matrix is weighted based on spatial distance, which is computed as follows:

$$\sum (u_i,v_i)=X^{T}W(u_i,v_i)X$$

(1)

In this study, $(u_i,v_i)$ represents the spatial location coordinate of i within the designated research area. X denotes the original variable matrix, while $W(u_i,v_i)$ signifies the spatial location weight matrix, which is computed using a kernel function selection. Specifically, this study employs the Gaussian kernel to ascertain the spatial location weight [31,32].

A further decomposition of the derived variance-covariance matrix yields eigenvalues and eigenvectors. These can be computed as follows:

$$L\left(u_i,v_i\right)V\left(u_i,v_i\right)L{\left(u_i,v_i\right)}^T=\sum (u_i,v_i)$$

(2)

In the given formula, $L\left(u_i,v_i\right)$ represents the matrix of eigenvectors, while $V\left(u_i,v_i\right)$ denotes the diagonal matrix of eigenvalues [31,32,49]. Subsequently, the score value matrix $T\left(u_i,v_i\right)$ for the data can be derived. The calculation formula is as follows:

$$T\left(u_i,v_i\right)=XL\left(u_i,v_i\right)$$

(3)

The eigenvectors delineate the directions of the original data within the new Principal Component (PC) space. The transformation of data from its original feature space to the principal component space is facilitated by multiplying the original data matrix with its corresponding eigenvector matrix. This procedure effectively converts the data’s projection in the original coordinate system into one in the principal component space, thereby enhancing our understanding of the data’s structure and variations within this new coordinate framework.

Furthermore, GWPCA offers the capability to compute geographically weighted correlations between variables. This function is instrumental in characterizing the interactions of variables across diverse spatial locations. The computation formula for this purpose is as follows:

$$r_i=\frac{cov(y_{1i},y_{2i})}{s\left(y_{1i}\right)s\left(y_{2i}\right)}$$

(4)

In the given formula, $cov(y_{1i},y_{2i})$ denotes the GW covariance between $y_{1i}$ and $y_{2i}$. Similarly, $s\left(y_{1i}\right)$ and $s\left(y_{2i}\right)$ represent the respective GW standard deviations of these variables.

(2)

Multiscale Geographically Weighted Regression (MGWR)

The MGWR model is an advanced extension of the geographically weighted regression (GWR) model [53]. This enhancement addresses the inherent limitation of GWR, which presumes that each independent variable’s impact on the dependent variable is uniform across all scales or uses a single bandwidth. The MGWR model calculates the optimal bandwidth for local parameters in each iteration, introduces multiple bandwidths, and enhances the model’s reliability through the utilization of the backward-fitting algorithm [53,54,55]. This has been successfully applied to fishery forecasting with promising results [56]. The calculation formula for this model is as follows:

$$y_i={\displaystyle {\sum }_{j=0}^m}{\beta }_{bwj}\left(u_i,v_i\right)x_{ij}+{\epsilon }_i$$

(5)

In the given formula, $y_i$ represents the predicted yield of point i within the study area. $x_{ij}$ denotes the score associated with input GWPCA, while bwj signifies the bandwidth of the regression coefficient for the jth variable.

In summary, the output content of GWPCA offers critical insights that are often overlooked in traditional PCA [49]. The MGWR model more accurately captures the interrelationships between variables and has been utilized in fishery resource research. By employing the GWPCA-MGWR approach, we can comprehensively capture the shifts in factor roles across different subregions within the study area, simultaneously discerning the impact of dependent variables on independent ones. This method facilitates regression function fitting and yields more precise predictive outcomes [56].

2.2.4. SABZ

The tool for spatial association between zones (SABZ) is utilized to ascertain the degree of spatial correlation between two regionalized variables within the same region. This degree of spatial association is expressed by calculating the overlapping area between these regions [35]. The strength of the association between regions is determined by the overlap between each regionalized region. If two regions correspond spatially closely, then the association degree between the two spatially varying variables is at its highest. This method conceptually aligns with the traditional geographic detector model but measures the degree of spatial association through the calculation of overlapping areas. This approach compensates for the lack of a spatial concept in traditional counting methods and addresses the weakness of the geographic detector model in quantifying the correlation between numerical and categorical spatial variables [35]. This study employs the SABZ in the spatial modeling toolset of ArcGIS 10.8 software to determine the spatial relationship between yellowfin tuna catch and dominant environmental factors.

3. Results and Analysis

3.1. Environmental Factors Information Redundancy and Heterogeneity Test

3.1.1. Environmental Factor Information Redundancy Test and Traditional PCA

To enhance the analysis of the influence of environmental factors on the spatial distribution of Pacific yellowfin tuna catch, a traditional principal component analysis (PCA) is initially conducted on these factors. Before applying PCA, a data redundancy check is performed on the environmental data to ensure the validity of the analysis. Subsequently, both the environmental and fishery data spanning 21 years from 2000 to 2021 are integrated for an in-depth examination of the inherent relationships between these factors. Figure 2 illustrates the correlation diagram between these environmental factors.

The degree of correlation among various environmental factors can effectively indicate data redundancy and information crossover between these factors. Upon examining the overall correlation of environmental factors, it is evident that there is a significant relationship among different environmental factors. Apart from the notable positive correlation between temperature and salinity across different water depth layers, a significant positive correlation also exists between salinity and temperature in the water layer below 100 m, with correlation coefficients exceeding 0.5. However, DO exhibit a certain negative correlation with surface temperatures T5 and T55, with correlation coefficients of −0.63 and −0.57, respectively. NPP also shows a negative correlation with temperature across different water depth layers, with correlation coefficients of −0.55, −0.66, −0.67, −0.65, and −0.58, respectively. The significant correlation among these environmental factors suggests they may contain identical information. Furthermore, the Kaiser-Meyer-Olkin (KMO) test results for the data indicate a KMO value of 0.69, which exceeds the boundary value of 0.6, fully supporting the effectiveness of using PCA in this study.

Traditional principal component analysis (PCA) can effectively identify the primary statistical characteristics of environmental factors, reflecting their overall average level and unveiling complex intrinsic interactions among selected variables.

Table 4. The proportion of principal components for environmental factors.

PC	Total	Percent Variance	Cumulative Variance Percentage
1	7.323	61.027	61.027
2	1.922	16.014	77.041
3	1.223	10.193	87.234
4	0.631	5.257	92.491
5	0.395	3.289	95.780

illustrates the traditional PCA results for these environmental factors. The eigenvalues exceeding 1 represent the first three principal components. Notably, the first principal component (PC1) accounts for 61%, while the second (PC2) and third principal components (PC3) account for 16% and 10%, respectively. Collectively, these first three principal components explain 87.2% of the total data variation. Consequently, they are instrumental in elucidating the most significant changes in data structure [57].

The loading sizes of environmental factors on the three principal components, indicative of their individual contributions to each component, also mirror the roles these factors play in the overarching changes in the environment. The loading values of these factors in the first three principal components are presented in

Table 5. The first three principal component loadings of each factor.

Variables	PC1	PC2	PC3
T5	0.755	−0.547	0.024
T55	0.857	−0.460	0.116
T105	0.900	−0.257	0.244
T155	0.928	−0.001	0.266
T205	0.892	0.239	0.207
S105	0.859	0.388	−0.195
S155	0.897	0.378	−0.165
S205	0.869	0.434	−0.194
S255	0.741	0.420	−0.338
NPP	−0.637	0.197	−0.535
DO	−0.399	0.619	0.493
U5	−0.332	0.454	0.538

. In the first principal component, temperature and salinity loading values for various water layers are predominantly positive. Notably, the T155 loading value is the highest at 0.928, suggesting that PC1 encapsulates significant temperature and salinity data. This also implies that the primary driver of environmental changes remains temperature and salinity. The second principal component’s largest loading value is for DO at 0.619, indicating that fluctuations in dissolved oxygen concentration in seawater can influence marine environments in a correlated direction. In the third principal component, U5 and NPP emerge as key factors, with loading values of −0.535 and 0.538, respectively. The data suggest that the overarching alterations in the marine environment can be partially attributed to changes in U5 and NPP. However, given that the value of U5 load is negative, it implies that the modifications in the marine environment triggered by U5 are inversely proportional to the direction of change in U5 itself.

Traditional principal component analysis (PCA) outputs are derived from comprehensive domain statistics, which fail to capture the local characteristics of data [58]. This limitation can lead to misinterpretations of the results due to potential spatial heterogeneity among environmental factors. For instance, spatial heterogeneity may influence the identification of dominant variables exerting the most significant influence in each principal component, among other issues [49].

3.1.2. Spatial Heterogeneity Test of Environmental Factors

The premise of this study is the existence of spatial heterogeneity among factors, addressed through the use of the geographically weighted principal component analysis (GWPCA) model. To ascertain whether such heterogeneity exists within the environmental factors of the study area, we utilize geographically weighted correlations (GW correlations). This method allows us to examine the local relationships between these variables. Although GW correlations analysis is not the primary focus of this research, it serves to illustrate the underlying spatial dynamics. For this analysis, we choose T105 and DO as examples. T105 is an environmental factor with a significant load value in the first principal component of traditional PCA, while DO has the highest load value in the second principal component. These examples are pivotal as they further underscore the presence of spatial heterogeneity among environmental factors within the study area. The results, depicted in Figure 3, show that the correlation between temperature and dissolved oxygen exhibits spatial heterogeneity, suggesting a localized heterogeneous relationship among environmental variables in the study area. This observation implies that traditional principal component analysis may not fully capture the nuanced variations in changes in environmental factors. Therefore, it is crucial to consider local heterogeneity when conducting principal component analysis.

3.2. Spatial Distribution of PC Score Values under Different ENSO Events

Previous research has demonstrated varying degrees of multicollinearity issues among marine environmental factors. This complexity has limited the scope of studies investigating the relationship between yellowfin tuna catch and these environmental factors. However, the score values derived from the GWPCA-MGWR model’s output can accurately represent changes in data within a new principal component space. In other words, these score value results provide an approximate reconstruction of the original data based on this principal component space. Consequently, this facilitates an easier analysis and comprehension of the data in this new space [57]. Therefore, utilizing score values effectively mitigates the impact of multicollinearity without necessitating the removal of environmental factors. As per Table 2, the fishery data under varying ENSO event scenarios were independently amalgamated with environmental data and subjected to GWPCA analysis. Traditional PCA results indicate that the initial three components can account for 87.2% of the data structure variation. The proportion of variance explained by the first three principal components of GWPCA ranges from 84.6% to 98.5%. This suggests that, when considering variance proportion alone, the GWPCA model outperforms the traditional PCA model.

A further examination of the spatial distribution of the first principal component score value across different ENSO events is depicted in Figure 4. Overall, the score value’s spatial distribution exhibits pronounced latitudinal variations, with high values primarily concentrated within the sea areas of 20°S–0°, 150°E–115°W. These high-value areas display a north–south symmetrical distribution pattern, with minimal changes in score value in the South Pacific Ocean and western Pacific Ocean. In terms of ENSO event intensity, as the event’s occurrence strength diminishes, a specific range of low-value areas emerges on either side of the study area, while the range of high-value areas contracts. As the ENSO event intensity escalates to a medium level, both sides of the low-value area range expand further, leading to a notable decrease in the central East Pacific Ocean’s score value area and a further reduction in high-value area distribution. When the ENSO event intensifies to an extremely strong level, both sides of the low-value area range continue to expand, while median and high-value areas exhibit minimal changes, predominantly occurring in the Central Pacific Ocean where the score value decreases to low-value changes. The spatial variation in score values was predominantly influenced by the intensity of ENSO events, with minimal differences observed between various types of ENSO events. The distinction between La Niña and El Niño events was primarily evident when considering the intensity; a more pronounced decrease in score value was observed during El Niño events compared to La Niña events.

3.3. Response of Yellowfin Tuna Catch Distribution to Environmental Factors under Different ENSO Events

Prior research indicates that the PCA scores of marine environmental factors influencing the spatial distribution of yellowfin tuna catch in the Pacific Ocean exhibit significant variability under varying ENSO climate conditions. To delve deeper into the response of this catch to alterations in environmental factors across different ENSO events, the GWPCA-MGWR model was employed.

The variables identified as the winning factors in the GWPCA-MGWR model’s output results encapsulate the dominant environmental change factors within the study area. These variables provide a comprehensive representation of the variations in marine environmental changes under varying ENSO events. Traditional principal component analysis has indicated that the primary environmental factors influencing the spatial distribution of yellowfin tuna in the Pacific Ocean are predominantly reflected in the first principal component. Consequently, this research focuses on the winning variable of the first principal component for analysis, as illustrated in Figure 5. The predominant influence of environmental factors on the distribution of yellowfin tuna catch in the Pacific Ocean exhibits distinct spatial alterations across different ENSO events.

As illustrated in Figure 5, the variations in winning variables reveal significant spatial disparities in the shifts of dominant environmental factors within the Pacific Ocean under different El Niño–Southern Oscillation (ENSO) events. From a spatial standpoint, the North Pacific Ocean predominantly exhibits minimal environmental fluctuations, whereas the Central Pacific Ocean demonstrates the opposite trend. The alterations in dominant environmental factors across subregions during varying ENSO events are multifaceted and dynamic. While the dominant environmental factor in the South Pacific region undergoes significant changes during ENSO events, the outcomes and trajectories of these changes exhibit less variability across events. This suggests that the alterations are more consistent. From the perspective of different ENSO events, when moderate-intensity El Niño and weak La Niña events occur, the dominant environmental factor changes in the Pacific Ocean are basically the same; when the El Niño intensity increases to a super-strong level, it will cause the western ocean area’s dominant environmental factor to change to the mesopelagic temperature, while the eastern ocean area near the equator will further change to 105 m deep seawater salinity; meanwhile, when a moderate La Niña event occurs, the dominant factor in the western Pacific region mainly changes to the mid-depth layer seawater temperature, while the eastern Pacific Ocean basically all becomes 155 m deep seawater salinity.

The SABZ was utilized to analyze the relationship between yellowfin tuna catch distribution and changes in dominant environmental factors under various ENSO conditions. This was achieved by integrating actual yield data. The study area was categorized into one to five levels based on the yield of yellowfin tuna, from the highest to lowest. A further aggregation of environmental factors was conducted into surface sea temperature (T5, T55), middle layer sea temperature (T105, T155), deep layer sea temperature (T205), middle layer salinity (S105, S155), deep layer salinity (S205, S255), and additional factors such as NPP, U5, and DO. The spatial correlation results are presented in

Table 6. The proportion of spatial association between dominant environmental factors and yield grading under different ENSO conditions. (In the figure, colors represent numerical magnitudes, with warm tones indicating high values and cool tones indicating low values).

Catch Grades	ENSO Event Types	Surface Sea Temperature	Middle Layer Temperature	Deep Layer Temperature	Middle Layer Salinity	Deep Layer Salinity	Other Factors
I (High)	Normal year	0.26	0.03	0.45	0.03	-	0.24
	Weak El Niño	0.12	0.28	0.44	-	0.16	-
	Moderate El Niño	0.08	0.15	0.46	0.08	0.23	-
	Super-strong El Niño	0.10	0.14	0.57	0.19	-	-
	Weak La Niña	0.13	0.08	0.61	0.03	0.16	-
	Moderate La Niña	0.18	0.31	0.22	0.18	0.11	-
II	Normal year	0.57	-	0.25	0.14	-	0.04
	Weak El Niño	0.17	0.31	0.29	0.09	0.11	0.03
	Moderate El Niño	0.41	0.05	0.36	0.14	0.05	-
	Super-strong El Niño	0.05	0.33	0.43	0.19	-	-
	Weak La Niña	0.13	0.30	0.30	0.13	0.13	-
	Moderate La Niña	0.42	0.19	0.12	0.19	0.08	-
III	Normal year	0.56	0.17	0.17	0.11	-	-
	Weak El Niño	0.37	0.24	0.24	0.10	0.05	-
	Moderate El Niño	0.21	0.21	0.54	0.04	-	-
	Super-strong El Niño	0.40	0.36	0.12	0.12	-	-
	Weak La Niña	0.30	0.30	0.15	0.21	0.03	-
	Moderate La Niña	0.50	0.08	0.12	0.27	0.04	-
IV	Normal year	0.50	0.29	0.04	0.14	-	0.04
	Weak El Niño	0.36	0.19	0.02	0.23	-	0.19
	Moderate El Niño	0.50	0.03	0.32	0.12	-	0.03
	Super-strong El Niño	0.31	0.31	0.14	0.25	-	-
	Weak La Niña	0.50	0.28	0.08	0.14	-	-
	Moderate La Niña	0.16	0.16	0.12	0.52	0.04	-
V (Low)	Normal year	0.55	0.08	0.02	0.05	0.07	0.22
	Weak El Niño	0.40	0.09	0.13	0.29	0.01	0.08
	Moderate El Niño	0.44	0.12	0.12	0.14	-	0.17
	Super-strong El Niño	0.31	0.37	0.02	0.31	-	-
	Weak La Niña	0.47	0.12	0.09	0.22	0.01	0.10
	Moderate La Niña	0.22	0.14	0.21	0.41	0.01	-

.

As illustrated in Table 6, during instances of El Niño or La Niña events, the spatial association between yellowfin tuna zones at all levels was observed to be significantly high. This association was particularly pronounced with areas characterized by seawater temperature and salinity dominance, with a marked preference for those dominated by seawater temperature. The catch grade I of Pacific Ocean yellowfin tuna displayed a clear spatial association with areas dominated by mesopelagic seawater temperature, under conditions other than moderate La Niña events. Conversely, areas with higher fishing catches during moderate La Niña events were primarily characterized by intermediate sea temperature. For areas classified as level II, they exhibited a high degree of spatial association with surface seawater temperature under normal annual conditions. However, when different ENSO events occurred, these areas transitioned from being dominated by surface seawater temperature to those dominated by meso-depth layer seawater temperature. Notably, during extreme El Niño events, over 75% of the area classified as level II was dominated by meso-depth layer seawater temperature. Areas graded as level III followed a similar trend to level II but with an additional shift towards deep seawater temperature during moderate El Niño events. Importantly, regardless of the intensity of La Niña events, there was a significant increase in the degree of spatial association between level III areas and mid-layer salinity. Finally, areas graded as level IV and V were both dominated by surface seawater temperature under varying conditions. However, under moderate La Niña events, they were dominated by mid-layer seawater salinity areas and also exhibited a shift towards mid-layer salinity during other ENSO events.

In summary, the high fishing catch area of yellowfin tuna exhibits a significant spatial association with areas characterized by meso-deep seawater temperature. Conversely, when the fishing catch diminishes to medium and low categories, the predominant environmental factor in these areas shifts from meso-deep seawater temperature to surface seawater temperature. Notably, during different El Niño–Southern Oscillation (ENSO) events, there is a trend towards regions dominated by meso-deep seawater temperature. Conversely, during La Niña events, the dominant factor shifts towards areas influenced by mid-layer seawater salinity.

The spatial distribution of Pacific yellowfin tuna catch, as predicted by the GWPCA-MGWR model and compared to actual data, is depicted in Figure 6. This figure reveals a significant disparity between the prediction results of models developed based on varying ENSO events. When juxtaposed with the actual distribution, it becomes evident that all but one model—those constructed under conditions of a weak La Niña event—exhibit notable overestimations in the western Central Pacific Ocean. Conversely, there is a marked underestimation in the Central Pacific Ocean. The model predicated on environmental conditions during a weak El Niño event demonstrates effective predictive capabilities in the high-value areas of the Central Pacific Ocean; however, its overall overestimation of Pacific yellowfin tuna catch remains pronounced. In general, the distribution trend of predicted yellowfin tuna catch in the Pacific Ocean, as established by the GWPCA-MGWR model under environmental factor conditions during a weak La Niña event, aligns more closely with the actual 2021 catch distribution than the real distribution. Given that 2021 was a typical year for weak La Niña events in the Pacific Ocean, it suggests that the GWPCA-MGWR model could be more effective in studying the spatial distribution of yellowfin tuna catch under varying ENSO events.

4. Discussion

4.1. Variations in Pacific-Dominated Environmental Factors under Different ENSO Events

This study employed the GWPCA-MGWR model to examine the dominance of environmental factors influencing the spatial distribution of Pacific yellowfin tuna catch across various ENSO events (Figure 5). The findings revealed distinct spatial variations in the environmental factors impacting yellowfin tuna catch in the Pacific Ocean under different ENSO scenarios. Broadly, these dominant environmental factors were categorized into two primary regions: the western Pacific Ocean, where seawater temperature was predominant, and the eastern Pacific Ocean, where salinity and net primary productivity were the most influential. During El Niño events, the region dominated by temperature in the western Pacific exhibited an eastward expansion and a trend towards deeper water layers. In contrast, during La Niña events, the area influenced by salinity in the eastern Pacific expanded further westward, especially when La Niña manifested at an intermediate intensity.

Upon reviewing existing research, it is evident that the shifts in dominant environmental factors during different ENSO events align with the spatial variations in the Pacific warm pool and cold tongue. These changes are primarily influenced by the upper air pressure belt over the Pacific Ocean, which continuously transports surface warm water to the western Pacific. This results in a persistent heat accumulation in this region, leading to a high-sea temperature “warm pool.” Concurrently, colder deeper seawater from the eastern Pacific surfaces up, along with inorganic salts from depth, forming a low-sea temperature, high-salinity, and high-net primary productivity “cold tongue” [44,59,60,61,62]. There is a spatial consistency between the existence of these two phenomena and the overall distribution of dominant environmental factors. Furthermore, when an ENSO event occurs, observable spatial changes in both the Pacific warm pool and cold tongue are noted. During an El Niño event, the western Pacific warm pool expands eastward, leading to a decrease in heat and a thinning of the warm water layer [45,46,59,63,64]. This results in a shallow thermocline. In this study, the eastward movement trend of areas dominated by temperature aligns with an El Niño event. Conversely, during a La Niña event, nutrient-rich bottom cold seawater from the eastern Pacific flows westward into the western Pacific, causing abnormal seawater salinity levels at both surface and subsurface levels [45,64,65,66]. This study suggests that the westward migration phenomenon dominated by salinity during a La Niña event may be related to such seawater salinity anomalies. In summary, the spatial range of the Pacific warm pool and cold tongue exhibits consistent variation under different ENSO events, corresponding to dominant environmental factors in this study. Consequently, it can be inferred that the alteration in the dominance of environmental factors influencing the spatial distribution of yellowfin tuna catch volume during various ENSO events correlates with the spatial changes in the Pacific warm pool area and cold tongue area. These changes are influenced by flow velocity and other alterations, as well as their temperature–salinity properties during El Niño and La Niña events.

The spatial distribution of dominant factors within the GWPCA-MGWR model can intuitively represent the range of spatial variation for both the warm pool and cold tongue. This is based on existing research indicating that the spatial variation in the Pacific warm pool and cold tongue influences the spatial distribution of yellowfin tuna catch [11,24,44]. For instance, studies have demonstrated that during El Niño events, the habitat of yellowfin tuna undergoes a corresponding spatial shift eastward with the expansion of the warm pool [24,62]. Consequently, when conducting an analysis and resource assessment of yellowfin tuna catch, it is crucial to fully consider the impact brought about by changes in the spatial range of the warm pool and cold tongue under varying ENSO events. However, the current zoning for a resource assessment of Pacific yellowfin tuna, as illustrated in Figure 1 [3], employs fixed rule boundaries. Such zoning rules are likely to overlook the influence of climate condition changes on the resource assessment model. Therefore, future spatial analysis and resource assessments of yellowfin tuna catch should consider using irregular dynamic boundaries for zoning based on the distribution of dominant environmental factors and their variations under different ENSO events. This approach would account for the impact brought about by the spatial variation in the Pacific warm pool and cold tongue, thereby facilitating a more accurate assessment of Pacific yellowfin tuna resources.

4.2. Spatial Association of Yellowfin Tuna Catch with Dominant Environmental Factors

This study examined the dominant environmental factors influencing the spatial distribution of Pacific yellowfin tuna catch. By employing the SABZ, we determined the response of yellowfin tuna catch to these dominant environmental factors, as detailed in Table 6. Our findings revealed that areas with a high concentration of yellowfin tuna catch were primarily sea areas dominated by seawater temperature and salinity. During El Niño events, there was a trend for areas with a high spatial correlation degree with yellowfin tuna catch to shift towards sea areas dominated by mesopelagic seawater temperature. Conversely, during La Niña events, areas with a high spatial correlation degree with medium–low yield shifted towards sea areas dominated by mesohaline salinity. The areas with a high concentration of yellowfin tuna catch were predominantly sea areas dominated by mesopelagic seawater temperature, which remained stable under different ENSO events. As the concentration of catch decreased, there was a gradual transition from sea areas with a high spatial correlation degree to those dominated by surface seawater temperature.

The spatial association between yellowfin tuna catch and dominant environmental factors may be influenced by the tuna’s biological characteristics, fishing methods, and other factors. Previous studies have demonstrated that yellowfin tuna exhibit a pronounced vertical movement behavior, spending most of their time in the water layer below the surface [1,6,67,68]. Adult yellowfin tuna are able and prefer to obtain food resources that juveniles cannot access in deeper water layers. Their daytime baiting water layer is primarily concentrated at a depth of 160–240 m. Research also indicates that yellowfin tuna regulate their own temperature and oxygen content by moving to deeper water layers [19,36,40,67]. This vertical movement of yellowfin tuna is primarily influenced and constrained by subsurface seawater temperature. Therefore, the impact of subsurface seawater temperature on the resource situation of yellowfin tuna is significant [37,69]. At the same time, many countries target adult yellowfin tuna for fishing [70], employing deep longline fishing methods [13,40,68]. Consequently, this study found that the high-yield area of yellowfin tuna longline fishing has a high degree of spatial association with the sea area where the dominant factor is mid-deep seawater temperature. In contrast, the low-yield area with a higher degree of spatial association with the sea area dominated by surface seawater temperature may be a comprehensive result of multiple factors such as yellowfin tuna living habits and fishing methods.

In conjunction with prior research, the occurrence of El Niño results in a change in the ocean’s thermal structure, which subsequently extends the habitat of yellowfin tuna vertically [71]. Consequently, areas with medium and low yellowfin tuna catch exhibit a higher spatial association degree than those dominated by deeper sea area change factors [24,44]. Conversely, during La Niña events, the penetration of the cold tongue into the western Pacific Ocean leads to a significant decrease in sea temperature and shallower mixed layer depth. Given the strong temperature dependence of yellowfin tuna, their fishing grounds shift towards areas with high seawater temperatures [22,24]. As a result, only medium- and low-yield areas exhibit a high degree of spatial association with salinity as the dominant environmental factor, while high-yield areas maintain a high degree of spatial association with areas influenced by deep-sea temperature. Based on these findings, the future selection of factors and weight assignment for yellowfin tuna resource assessment should be skewed towards medium and deep-sea temperature. Simultaneously, fishery management can further regulate the depth at which yellowfin tuna longline hooks are positioned.

4.3. Application of GWPCA-MGWR Model in Yellowfin Tuna Catch Study

In the correlation analysis of the spatial distribution of yellowfin tuna catch under varying ENSO events in the Pacific Ocean, the application of the GWPCA-MGWR model allows for a comprehensive consideration of the variances across each subregion within the study area. This model integrates multiple environmental factors influencing the spatial distribution of yellowfin tuna catch, facilitating a holistic macro-analysis of its catch in the Pacific Ocean. In this research, we employed the GWPCA-MGWR model to account for six distinct ENSO event backgrounds and 12 environmental factors impacting yellowfin tuna in the Pacific Ocean. This approach fully acknowledges the spatial heterogeneity among these factors, as detailed in Section 3.2 of this paper. Consequently, we identified the dominant environmental factor distribution under different ENSO events, as illustrated in Figure 5, and delved into its relationship with fluctuations in the Pacific warm pool and cold tongue, as well as the spatial distribution status of yellowfin tuna catch, as outlined in Section 4.1 and Section 4.2. Based on these findings, we constructed a robust prediction model, as depicted in Figure 6. Our results underscore the significant advantages of the GWPCA-MGWR model in studying yellowfin tuna catch in the Pacific Ocean.

The GWPCA-MGWR model effectively captures the variability in different subregions, offering more reasonable boundaries for natural zoning from a comprehensive standpoint. Previous research has indicated that due to the intricate dynamics of yellowfin tuna biology and the marine environment in the Pacific Ocean, significant disparities exist in the suitability and tolerance of yellowfin tuna across various sub-sea areas. This leads to multiple discontinuous hotspots in the spatial distribution of yellowfin tuna catches in the Pacific Ocean, with marked differences in environmental conditions between these regions. Consequently, most existing studies on Pacific yellowfin tuna tend to focus on smaller areas, such as an adjacent island or a specific commercial fleet’s passing zone, or they artificially segment them into several smaller areas for separate research during resource assessments. This approach isolates the spatial connection between yellowfin tuna in the Pacific Ocean, and the division basis is not a naturally occurring rule shaped by environmental changes impacting yellowfin tuna in the region. In this study, we utilized the GWPCA-MGWR model to analyze the overall environmental change situation in the Pacific Ocean from a macro perspective, thereby deriving clear boundaries for dominant environmental factor changes under different ENSO events. These boundaries are irregular, natural, and subject to change. The judicious application of the GWPCA-MGWR model to Pacific yellowfin tuna research can provide a more scientifically sound division basis for related studies.

The GWPCA-MGWR model leverages the benefits of principal component analysis, accounts for spatial heterogeneity among factor relationships, and constructs a robust regression model. In conducting research related to yellowfin tuna, addressing multicollinearity among marine environmental factors is crucial prior to model construction. Existing studies predominantly employ the variance inflation factor (VIF) method to eliminate collinear factors [20,69]. However, these eliminated environmental factors may still significantly influence yellowfin tuna catch. To counteract this, some researchers have turned to traditional principal component analysis to mitigate multicollinearity [17]. The GWPCA-MGWR model adeptly utilizes this traditional approach to exclude multicollinearity between environmental factors while considering their spatial heterogeneity. This is reflected in the score value of the principal component. By integrating this score value with the MGWR model’s capability to operate in multidimensional space scales, a pertinent regression model is constructed, as illustrated in Figure 6. This model can precisely forecast trends in yellowfin tuna catch variations.

5. Conclusions and Prospects

Given the complexities of marine environmental patterns and significant constraints in macro-level studies of Pacific Ocean yellowfin tuna resources under varying ENSO events, this study innovatively integrates the geographically weighted principal component analysis (GWPCA) model with a multiscale geographically weighted regression model (MGWR). Through an analysis of the spatial variation in dominant environmental factors impacting yellowfin tuna catch under different ENSO events, we discovered that these variations are linked to the spatial movements of the Pacific warm pool and cold tongue. This insight further elucidates the spatial variation area and distinct boundaries of environmental dominant factors under various ENSO events. This study innovatively integrates the SABZ method with the GWPCA method, aligning changes in yellowfin tuna catch volume with spatial variations in dominant environmental factors. It investigates the degree of spatial correlation between yellowfin tuna catch volume and these dominant environmental factors, as well as their fluctuations under varying ENSO events. The findings indicate that there are differences in the dominant environmental factors in sea areas that yield a higher degree of spatial correlation with varying catch levels. Specifically, areas with high catch exhibit a higher spatial correlation with sea areas dominated by meso-deep-sea temperature compared to those with low catch. Conversely, areas with low catch are primarily located in sea areas dominated by surface sea temperature. This study robustly illustrates the logic and benefits of the GWPCA-MGWR model within the context of tuna fishery research. It advocates for the inclusion of dynamic natural boundaries in future evaluations of yellowfin tuna resources in the Pacific Ocean. Approaching from a macro perspective, this paper addresses the existing gap in comprehensive research on Pacific yellowfin tuna and offers valuable insights and references for resource assessment and fishery management pertaining to Pacific yellowfin tuna.

Given that this study averaged environmental and fishery data across various ENSO events, the GWPCA-MGWR model employed did not account for temporal variations in the marine environment and yellowfin tuna resources. It is recommended that subsequent research build upon this foundation to account for time series effects, thereby elucidating the spatio-temporal fluctuations of Pacific yellowfin tuna in response to shifts in dominant environmental factors. Concurrently, this study coarsened the data in both spatial and temporal resolution to enhance model adaptability. However, this approach may result in an incomplete representation of the nuanced influence of environmental variables on fishery data. At the same time, while the GWPCA-MGWR model adeptly captures changes in these dominant environmental factors, there remains a need for refinement in the method of selecting these factors for future studies.