Marine fishery resource survey station adaptive optimization method based on space-time coupling model

Abstract

The invention discloses a marine fishery resource survey station adaptive optimization method based on a space-time coupling model. According to the method, a space-time continuous species distribution dynamic reference field is constructed by coupling a Bayesian hierarchical species distribution model and a regional ocean model system; performing ecological layering based on time, space and environment dimensions; and when the environment variables change significantly, a habitat suitability index is dynamically reconstructed, and the station layout is optimized again, so that the coverage of a key ecological hot spot area is ensured, and the in-gradient representativeness is maintained. The method has the advantages of fast environmental response, strong spatial representativeness, high adaptability and the like, and is suitable for scenes of resource assessment, ecological protection, biodiversity monitoring and the like.

Description

Technical Field

[0001] The present invention relates to the field of marine fishery resource monitoring and investigation technology, in particular to a station layout optimization method combining ecological modeling and marine numerical simulation, and specifically to a marine fishery resource survey station adaptive optimization method based on a spatiotemporal coupling model, belonging to the cross-technical category of marine ecological monitoring, fishery resource assessment and survey design optimization. Background Art

[0002] In current marine fishery resource assessments and surveys, the scientific nature and dynamic responsiveness of station placement directly impact the representativeness and accuracy of fishery resource monitoring. Traditional fishery resource surveys typically rely on fixed or random station placement methods, with simple random sampling, systematic sampling, cluster sampling, and stratified sampling being widely used. Stratified sampling, particularly common in my country's waters, is based on geographic characteristics such as water depth, bottom quality, estuary, and location. While these methods offer the advantages of ease of operation, they still have significant limitations in identifying ecologically sensitive areas, covering resource hotspots, and dynamically responding to environmental changes.

[0003] Specifically, several major issues with the current survey design include:

[0004] 1. Weak ability to depict spatial heterogeneity of resources: The existing stratification standards are mainly based on geographical factors and fail to reflect the coupling mechanism between resource concentration (hotspot) areas and environmental gradients. As a result, sampling points may miss ecological hotspots, affecting the representativeness of the survey.

[0005] 2. Lack of consideration of ecological processes: Traditional classification methods fail to reflect the driving effect of gradient changes in environmental factors (such as temperature and salinity) on species distribution, and thus cannot truly reflect the synergistic mechanism between resources and the environment.

[0006] 3. Static simulated distribution assumptions deviate from actual ecological dynamics: Kriging interpolation or statistical modeling methods are often used to simulate "real" distributions. However, most of these methods are based solely on historical static variables (geographic location, water depth, etc.), assuming that resource distribution is stable in time and space. This approach ignores the impact of seasonal fluctuations, climate events (such as El Niño), and sudden environmental disturbances on resource distribution, resulting in a distribution model that lacks foresight and adaptability.

[0007] 4. Lack of a rapid response mechanism to climate change and emergencies: When abnormal changes occur in the marine environment (such as sudden temperature rises or pollution incidents), the existing station layout is difficult to adjust in a timely manner, causing the survey to lag behind the ecological response, affecting the timeliness and accuracy of resource assessments.

[0008] In addition, against the backdrop of growing demand for multi-species surveys, traditional methods are unable to simultaneously take into account populations of different life history stages and ecological needs, especially during critical ecological periods such as spawning, feeding, and wintering. It is impossible to accurately identify species aggregation areas and effectively deploy stations.

[0009] Therefore, it is urgent to establish a survey station optimization method with dynamic modeling capabilities, ecological driving mechanism explanatory power and adaptive response capabilities, which can fully integrate environmental variables and species behavior, design survey plans based on the spatial-temporal-environmental three-dimensional characteristics of species distribution, improve survey efficiency and scientificity, and meet the high-quality development requirements of modern marine ecological protection and resource management. Summary of the Invention

[0010] The purpose of this invention is to overcome the problems of existing fishery resource surveys, such as the strong static nature of station design, insufficient coupling with ecological processes, and delayed response to environmental disturbances. This invention proposes an adaptive station optimization method that can dynamically adjust the survey layout according to environmental changes, while balancing ecological representativeness and operational efficiency. To achieve this, the present invention, based on the deep coupling of ecological modeling and physical simulation, constructs a dynamic reference field for species distribution that is continuous in time and space. On this basis, it designs a multi-dimensional ecological regional division system and introduces a dynamic station layout mechanism based on real-time environmental monitoring results, thereby enhancing the scientific nature and sensitivity of fishery resource surveys.

[0011] To achieve the above objectives, the present invention provides a method for adaptively optimizing marine fishery resource survey stations based on a spatiotemporal coupling model, the technical solution of which is as follows:

[0012] In one possible implementation, a method for adaptively optimizing marine fishery resource survey station locations based on a spatiotemporal coupling model is provided, aiming to improve the spatial representativeness and environmental responsiveness of fishery resource surveys. The method includes the following three core steps:

[0013] (1) Dynamic reference field construction: This step deeply couples the Bayesian hierarchical species distribution model (INLA) with the Regional Ocean Model System (ROMS), and uses data assimilation technology to integrate historical survey data and multi-source environmental factors (such as temperature, salinity, and chlorophyll concentration) to establish a dynamic reference field for species distribution that is both physically consistent and ecologically meaningful. This field has high temporal and spatial resolution and can dynamically reflect the changing trends of resource density, forming the basis for subsequent stratification and optimization operations.

[0014] (2) Three-dimensional ecoregional stratification design: To ensure that the sampling strategy reflects the characteristics of ecological processes, the ecoregions were divided using three factors: time dimension (e.g., life history stage division), spatial dimension (e.g., hotspot identification based on Getis-Ord Gi statistics), and environmental dimension (e.g., temperature and salinity gradient identification). The time dimension focuses on key stages of the target species, such as spawning, feeding, and overwintering; the spatial dimension uses the significance Z value to divide the degree of resource aggregation; and the environmental dimension uses the Sobel operator and Thorpe scale to identify ecologically sensitive physical boundaries.

[0015] (3) Adaptive station optimization mechanism: When the actual environmental conditions detected deviate significantly from those during the model construction period (e.g., temperature difference exceeding ±1.5°C), the benchmark reconstruction and station adjustment are automatically triggered. The optimization goal is to ensure that the survey stations cover more than 80% of the resource hotspot areas and deploy at least three stations within each environmental gradient zone to enhance the spatial representativeness and ecological adaptability of the data. Station selection is completed through stratified random sampling. The candidate point set is initialized on a predefined grid (e.g., 30′×30′). Evaluation indicators include relative estimation error and relative deviation.

[0016] In one possible implementation, the Bayesian hierarchical model consists of two sub-models:

[0017] Species presence / absence submodel: used to predict the probability of occurrence of target species in different areas, based on binomial distribution and log link function;

[0018] Species conditional abundance submodel: used to estimate species density within the occurrence area, using a Poisson distribution with a log link function;

[0019] Spatial autocorrelation is modeled through Gaussian Markov random fields to ensure the spatial continuity and rationality of the prediction results.

[0020] In one possible implementation, the data assimilation phase introduces different mechanisms to handle heterogeneous variables:

[0021] Temperature and salinity are assimilated in real time using particle filtering technology;

[0022] Chlorophyll concentration was combined with the NPZD ecological model for bio-physical coupling assimilation to improve the ecological accuracy of habitat prediction.

[0023] In one possible implementation, the three-dimensional hierarchical design is guided by ecological processes:

[0024] The spawning period in the temporal dimension is determined by gonad development indicators or the cumulative temperature threshold method;

[0025] Determined by combining feeding period with stomach contents analysis;

[0026] The overwintering period is identified based on overwintering cluster behavior.

[0027] In the spatial dimension (Getis-Ord Spatial statistical methods (such as statistic) are used to identify species distribution hotspots. By calculating the statistical significance of each spatial unit and its neighboring units, the study area is divided into core area, transition area and background area.

[0028]

[0029] Among them, w i,j is the spatial weight of position i and j, is the population mean, and S is the standard deviation. Size, divide the area into hotspot core areas transition zone and background area Achieve quantitative description of resource spatial heterogeneity.

[0030] In terms of environmental dimensions, the gradient fields of temperature and salinity are calculated using the Sobel operator, and the mutation points of the mixed layer depth are located using the Thorpe scaling method to identify gradient zones with ecological transition significance.

[0031] In one possible implementation, adaptive optimization is achieved through the following process:

[0032] Taking temperature as the main monitoring variable, set the deviation threshold of ±1.5℃ as the trigger condition;

[0033] Use the ROMS model to obtain updated environmental fields and input them into the INLA model to predict resource distribution;

[0034] Recalculate the Habitat Suitability Index (HSI) and reassess ecoregional divisions accordingly;

[0035] A stratified random sampling strategy is adopted to select stations from the candidate point set (e.g., 30′×30′ grid) to meet the dual constraints of “hotspot coverage + gradient representativeness”.

[0036] In one possible implementation, the technical configuration of the ROMS model includes:

[0037] The horizontal resolution is 1 to 3 kilometers;

[0038] The vertical division is into 20 to 30 layers of terrain-following coordinates;

[0039] The atmospheric forcing data source is the ERA5 reanalysis data, with a temporal resolution better than 6 hours.

[0040] In one possible implementation, this method is applicable to various application scenarios such as marine fishery resource assessment, ecological protection area planning, and marine biodiversity monitoring, and has strong promotion value and practical significance.

[0041] Based on the above technical solution, the "adaptive optimization method for marine fishery resource survey stations based on spatiotemporal coupling model" provided by the present invention introduces three major technical innovations in the field of fishery resource survey design, namely physical-ecological coupling, ecological process stratification and real-time response mechanism. It systematically solves the key problems existing in traditional fishery survey design methods, such as static station distribution, poor ecological representativeness and delayed environmental response, and has significant technological advancement and practical application value.

[0042] First, the dynamic reference field construction method proposed in this invention breaks through the limitations of previous reliance on static sample data or geographical factors for species distribution modeling, and for the first time realizes the deep coupling of the Regional Ocean Model System (ROMS) and the Bayesian Hierarchical Species Distribution Model (INLA). By introducing a data assimilation mechanism, integrating satellite remote sensing, buoy observations, and historical survey data, and strengthening parameter consistency and spatial continuity through filtering and ecological models, a species distribution prediction field with high resolution, strong timeliness, and complete ecological logic is generated. This dynamic reference field can not only be used for the optimization of current survey sites, but also serve for resource change warnings and trend simulations in the future, significantly improving the foresight of resource management.

[0043] Secondly, the three-dimensional ecological stratification system constructed by this invention has achieved a fundamental transformation of sampling design from "geographically driven" to "ecologically driven". In the temporal dimension, the monitoring window is dynamically divided according to the life history stages of species, which enhances the representativeness of sampling during key periods such as spawning period, feeding period, and wintering period. In the spatial dimension, the system introduces (Getis-Ord Spatial statistical methods (using statistic data) identify resource hotspots and scientifically delineate high-concentration and background areas, improving the station's ability to capture resource spatial heterogeneity. In the environmental dimension, ecological boundaries are identified based on temperature-salinity gradients and mixed layer structure, addressing the shortcomings of traditional methods in identifying physical ecological transition zones. A collaborative three-dimensional stratified design provides an ecological foundation for station optimization, ensuring the systematic and completeness of sampling results.

[0044] Again, the adaptive station optimization mechanism introduced in the present invention has dynamic perception and real-time response capabilities. When external environmental conditions such as temperature deviate significantly from the model set value (such as more than ±1.5°C), the system can automatically rebuild the dynamic reference field and adjust the sampling layout to ensure that the survey design evolves synchronously with the actual ecology. The optimization algorithm comprehensively considers the dual constraints of ecological hotspot coverage and environmental gradient representativeness, and through stratified random sampling, it concentrates on screening the optimal station layout scheme at candidate grid points to further improve the accuracy and stability of resource density estimation. In actual simulations, under the premise of reducing the number of stations, the optimization mechanism has better monitoring indicator error control than the traditional equidistant layout method, showing higher efficiency and reliability.

[0045] Furthermore, the present invention boasts excellent versatility and scalability. Its framework is independent of specific regional parameter settings and can adapt to fishery surveys in different sea areas, targeting different species, and over varying timescales. Its modular model structure and standardized data input facilitate integration with existing operational numerical forecasting systems, making it applicable in scenarios such as national and regional fishery resource assessments, ecological protection zone demarcation, and marine ranch planning.

[0046] In summary, the present invention organically integrates ecological modeling methods (such as the Bayesian hierarchical species distribution model) and marine numerical simulation systems (such as the ROMS-ecological coupling model) for the first time, and proposes an adaptive optimization method for fishery resource survey stations for multiple species, multiple time periods, and multiple ecological gradients. Compared with the existing station design methods based on spatial uniformity or empirical regional division, the present invention establishes a station optimization process with real-time response capabilities through dynamic benchmark field-driven suitability assessment, spatial hotspot identification, and ecological zoning modeling, breaking through the technical bottlenecks of "static division, single-scale consideration, and low environmental sensitivity" in traditional methods. This method has a high degree of ecological representativeness, environmental adaptability, and algorithm scalability, and can be widely applied to survey tasks in different sea areas and multiple target species. It enhances the scientificity and accuracy of resource assessment while improving survey efficiency, and has important theoretical value and application prospects. DETAILED DESCRIPTION

[0047] In order to better understand the adaptive optimization method for marine fishery resource survey stations based on a spatiotemporal coupling model proposed in the present invention, the technical implementation process of the present invention is described in detail below in combination with typical application scenarios.

[0048] It should be understood that the core concept of this invention is to integrate ecological process modeling with physical ocean simulation, construct a dynamic distribution reference field based on historical data and real-time assimilation results, and dynamically optimize the spatiotemporal configuration of survey stations by combining multidimensional ecological zoning methods with environmental response mechanisms. This method aims to improve the scientificity, representativeness, and adaptability of fishery survey station design, and is particularly suitable for scenarios where the distribution of target populations is significantly affected by climate and the environment, the spatial pattern is complex, and survey resources are limited.

[0049] This implementation uses a specific marine area as the subject, relying on real fishery survey data and environmental observation data, to gradually illustrate the entire process of model construction, data processing, ecological zoning, and station optimization, demonstrating the applicability and effectiveness of the present invention in actual survey tasks. However, it should be noted that the following examples are only preferred modes of the present invention, and any replacement or adjustment of the model structure, algorithm parameters, or data source without departing from the essence of the present invention falls within the scope of protection of the present invention.

[0050] The following describes in detail the proposed method for adaptive optimization of marine fishery resource survey stations based on a spatiotemporal coupling model, in conjunction with actual marine fishery survey scenarios. This implementation uses a key fishing ground in the East my country Sea as an example, and the specific process is as follows:

[0051] During the survey preparation phase, we first collected fishery resource survey data covering the spawning, feeding, and wintering periods. We also acquired environmental observation data for the corresponding time periods, including sea surface temperature, salinity, chlorophyll a concentration, wind speed, and other multi-source information. This data served as the input dataset for the coupled prediction model, providing support for subsequent modeling and optimization.

[0052] In terms of model construction, the Integrated Nested Laplace Approximation (INLA) method was combined with the Regional Ocean Modeling System (ROMS) to construct an ecologically and physically integrated species distribution prediction model. The INLA method was used to approximate Bayesian inference in the latent Gaussian model, with the goal of achieving fast and accurate calculation of the posterior joint distribution. The model assumptions are as follows:

[0053] Suppose the observation vector is y = (y1,…,yn), the Gaussian random field is x = (x1,…,xn), the hyperparameter is θ = (θ1,…,θk)(k∈N), μ(θ) and Q(θ) represent the mean vector and precision matrix respectively, then the joint distribution satisfies:

[0054]

[0055] Its linear prediction term satisfies:

[0056]

[0057] Among them, α is the intercept, β k is the covariate z ik The coefficient, f j Indicates the covariate u ij is the random effect of the domain, ε i is the error term. In the present invention, f j It is mainly used to express spatial (or spatiotemporal) structures, and usually quantifies spatial autocorrelation through Gaussian Markov Random Field (GMRF) modeling.

[0058] The INLA model structure consists of two sub-models: a binomial distribution model for species presence / absence, and a Poisson distribution model for abundance under conditional presence. Both models use a logarithmic link function. Furthermore, to account for zero inflation and spatial heterogeneity during sampling, a spatial random term is introduced to enhance the confidence of the predictions.

[0059] The environmental covariates for these models are derived from the ROMS output field. The ROMS model has a spatial horizontal resolution of 1–3 km and uses terrain-following coordinates (sigma coordinates) to create 20–30 vertical layers, with emphasis on the thermocline and surface regions. The initial physical state is set using global reanalysis data, with boundary conditions provided by global models such as HYCOM or GLORYS. Atmospheric forcing data are provided by the ERA5 dataset, with a temporal resolution of 6 hours. Chlorophyll concentration is introduced into the NPZD ecological model as an indicator of ecological processes. The ROMS and NPZD modules jointly construct an integrated physical-ecological simulation system.

[0060] During the data assimilation process, temperature and salinity variables are physically corrected using a particle filter algorithm, while chlorophyll concentration is coupled with remote sensing data through the NPZD model. Short-term forecast analysis is performed every six hours, and the physical plausibility of frontal and mixed layer changes is assessed based on the assimilated incremental fields.

[0061] Based on the above modeling and assimilation results, a high-temporal and spatial resolution species distribution dynamic benchmark field is generated. Then, the Habitat Suitability Index (HSI) is calculated by combining the predicted resource abundance A and its extreme value:

[0062]

[0063] Among them, A min With A maxThe HSI is used to measure the resource suitability of different grid cells and serves as the basis for subsequent ecological evaluation for stratification and optimization.

[0064] After completing the benchmark site construction, the three-dimensional ecological stratification phase begins. First, in the temporal dimension, the year is divided into spawning, feeding, and overwintering periods based on the life history stages of the target species. Identification methods include: determining the spawning period using gonadal development classification or cumulative temperature thresholds; identifying the feeding period based on the peak timing of zooplankton; and determining the overwintering period based on winter aggregation behavior.

[0065] In the spatial dimension, Getis-Ord Spatial statistical methods are used to identify resource hotspots and calculate the significance value of each grid cell and its neighboring cells. value:

[0066]

[0067] Among them, w i,j is the spatial weight of position i and j, is the overall mean, S is the standard deviation (S is the standard deviation of the global attribute variable). Size, divide the area into hotspot core areas transition zone and background area Achieve quantitative description of resource spatial heterogeneity.

[0068] In the environmental dimension, in order to identify key ecological gradient zones, the Sobel operator is used to calculate the horizontal gradient fields of temperature and salinity, and the Thorpe scaling method is used to analyze the sudden changes in the mixed layer depth to define the jump layer structure and frontal boundary and capture the location of the ecological transition zone.

[0069] Based on the three-dimensional zoning results described above, the present invention further designed an adaptive station optimization mechanism. When the sea surface temperature during the planned survey period deviates by more than ±1.5°C from the baseline value used when the model was constructed, a dynamic reconstruction process is triggered. The system uses the INLA and ROMS models for rapid re-evaluation, updating the HSI distribution and ecological stratification structure. Station placement is adjusted within two optimization constraints: first, the deployment area must cover at least 80% of the hotspot grid; second, at least three survey stations must be deployed within each environmental gradient zone.

[0070] The final station location optimization was achieved using a stratified random sampling algorithm. First, a 30′×30′ set of candidate grid points was established based on the ecoregion boundary. Subject to the aforementioned constraints, grid points with high HSI values ​​and strong representativeness of environmental gradients were prioritized for station placement. By introducing a resource abundance index (such as CPUE) as a sampling weight and comparing relative estimation error (Relative Estimation Error) and relative bias (Relative Bias) under different station numbers, the optimization scheme's comprehensive advantages in terms of accuracy, efficiency, and cost were verified.

[0071] Application Example 1:

[0072] In a specific embodiment, the practical application and technical verification of the method of the present invention were carried out in the East China Sea of ​​my country as the survey area.

[0073] First, data preparation and model initialization were performed. The survey team collected fishery resource survey data covering the past five years, covering different seasons. This data included the spatial distribution and abundance of target species during spawning, feeding, and wintering periods. They also acquired multi-source environmental observation data from the same period, including sea surface temperature (SST), salinity (SSS), chlorophyll a concentration, wind speed, and wind stress. All data were preprocessed into a unified format before serving as model input.

[0074] Subsequently, a Regional Ocean Numerical Model System (ROMS) was constructed and initialized, with a horizontal resolution of 1 to 3 kilometers for the simulated area, 20 to 30 vertical layers, and a σ coordinate system to accommodate complex bottom geomorphology. ROMS boundary conditions were provided by the HYCOM global reanalysis product, and the atmospheric forcing field was provided by the ERA5 dataset. The temporal resolution was controlled to within 6 hours. The coupled biogeochemical module employed the NPZD model to simulate key ecological factors such as chlorophyll concentration, thereby constructing a coupled physical-ecological model system with ecological dynamics.

[0075] Next, a dynamic baseline field was constructed. The temperature and salinity fields were assimilated using a particle filter method, and the chlorophyll concentration was assimilated using a bio-physical coupled assimilation method combined with the NPZD model. A Bayesian hierarchical species distribution model was constructed using the integrated nested Laplace approximation (INLA) algorithm, where:

[0076] The first layer is the occurrence probability submodel, which uses binomial distribution and logarithmic link function;

[0077] The second layer is the conditional abundance submodel, which uses a Poisson distribution and a logarithmic link function;

[0078] Gaussian Markov random fields are introduced in both layers to characterize spatial correlation.

[0079] After the model inference is completed, the maximum value A based on the predicted resource abundance max With the minimum value A min , calculate the habitat suitability index (HSI) for each predicted point:

[0080]

[0081] HSI is used as a quantitative indicator to measure the habitat suitability of target species in a specific grid, providing a basis for subsequent regional division.

[0082] In the three-dimensional ecological stratification design stage, the survey area was divided into three dimensions: time, space, and environment. In the time dimension, based on the ecological habits of the target species and historical observation data, the whole year was divided into spawning period (based on the level of gonad development and temperature accumulation), feeding period (combined with the peak biomass of zooplankton) and wintering period (referring to the characteristics of winter clustering behavior). In the spatial dimension, the Getis-Ord Statistical methods were used to identify resource hotspots. By calculating the local standardized score Z, the study area was divided into a core hotspot area (Z>1.96), a transition zone (-1.96≤Z≤1.96), and a background zone (Z<-1.96). In the environmental dimension, the Sobel operator was used to calculate the temperature and salinity gradient, and the Thorpe scaling method was used to identify the abrupt changes in the mixed layer depth to define the boundaries of the ecological gradient.

[0083] Based on the three-dimensional zoning results, an adaptive station optimization mechanism was implemented. Using temperature as the key environmental variable, a trigger threshold of ±1.5°C was set. When the environmental deviation between the predicted survey period and the modeling period exceeded the threshold, the system automatically reconstructed the dynamic reference field and ecological stratification structure and used the INLA model to re-estimate the posterior distribution of species distribution. The following dual constraints were followed when optimizing station placement:

[0084] First, ensure that more than 80% of ecological hotspot grids are covered;

[0085] Second, at least three survey stations should be set up in each environmental gradient zone.

[0086] On this basis, a 30′×30′ spatial grid candidate point set was constructed, and the final survey sites were selected through a stratified random sampling algorithm, giving priority to grid locations with strong ecological representativeness and high resource density. At the same time, the number and distribution of sites were controlled, taking into account both survey accuracy and cost-effectiveness.

[0087] To evaluate the effectiveness of the optimization scheme, a survey accuracy evaluation index system was constructed. Relative estimation error (REE) and relative bias (RB) were used as evaluation indicators, and the proposed scheme was compared with the traditional equidistant survey method. Simulation results showed that, with the same number of stations, the optimized scheme achieved an average reduction of REE and RB by over 15%, and also achieved superior performance in terms of hotspot accuracy and ecological gradient zone coverage.

[0088] This example also verifies the practicality of this method in application scenarios such as red tide response, early warning of interannual changes in fishery resources, and dynamic demarcation of ecological red lines, reflecting its wide adaptability and engineering feasibility in data-driven ecological monitoring.

[0089] Application Example 2:

[0090] In another specific embodiment, the method of the present invention was applied to the nearshore fishing grounds of the southern Yellow Sea. This area is a key habitat and feeding ground for a variety of commercial fish species (such as small fish and yellow crucian carp). In recent years, it has been disrupted by frequent red tide outbreaks, resulting in significant fluctuations in resource distribution. To achieve dynamic monitoring of key populations in this area and efficiently deploy survey stations, a pilot application of the adaptive optimization method based on the spatiotemporal coupling model described in this invention was implemented.

[0091] First, survey data and environmental data were collected simultaneously. The research team retrieved fishery resource survey data from the same period (May to August) over the past three years, including trawl density of target fish, catch biomass, and zooplankton distribution. They also obtained high-frequency environmental observation data collected by nearshore buoys and drone remote sensing systems. The collected environmental variables included sea surface temperature (SST), salinity, chlorophyll a concentration, turbidity, wind patterns, and tidal currents.

[0092] Based on these data, a ROMS numerical model for the southern Yellow Sea was constructed. To accommodate the complex nearshore topography and shelf dynamics, the model grid resolution was set to 1 km horizontally, and a 24-layer σ coordinate system was used vertically, with refinement in the upper mixed layer. Boundary data were derived from the regionally nested North Yellow Sea HYCOM submodel, and the atmospheric forcing input was 6-hourly resolution ERA5 data provided by CMEMS. The NPZD module was tightly coupled with the physical field to describe the evolution of chlorophyll concentration and nutrients, providing dynamic boundary conditions for the ecological model.

[0093] After initializing the numerical model, a particle filter was introduced to assimilate sea temperature and salinity data. The biological field variable (chlorophyll a) was then synergistically corrected with the NPZD output to generate a realistic coupled physical-ecological state field. Furthermore, a hierarchical Bayesian distribution model was constructed using the INLA method, integrating environmental variables with historical species distribution data to obtain a probabilistic spatial distribution forecast for the target fish species within the region.

[0094] Specifically, to address abnormal sea surface temperature and chlorophyll disturbances caused by red tide events, this embodiment sets adaptive station placement thresholds at: sea surface temperature deviations exceeding 1.2°C from the baseline or an abnormal increase in chlorophyll concentration exceeding 50%. When these trigger conditions are met, the system automatically performs a dynamic baseline reconstruction, recalculates the Habitat Suitability Index (HSI), and re-demarcates ecological regions.

[0095] In terms of ecological stratification, the time dimension defines key monitoring periods based on the peak feeding intensity and spawning activity of target species; the spatial dimension uses the Gi statistical method to identify nearshore high-density habitat patches, and the results show that resource hotspots show significant linear clustering characteristics, with the core area concentrated near the estuary outlet zone and tidal mixing front; the environmental dimension uses the Sobel operator to identify the location of the temperature-salinity front and the nutrient mutation zone to construct stratification boundaries with ecological indicative significance.

[0096] In the final optimized station layout, the optimal station layout was selected from an initial 30′×30′ grid using a stratified random sampling method, while meeting the dual constraints of "80% hotspot coverage" and "≥3 stations per gradient zone." Comparison of the optimized results with a traditional uniformly spaced layout revealed that, with the same number of stations, the representativeness of the target species' habitat hotspots increased by 18%, the resource abundance error decreased by 12%, and the system was able to adaptively update the station layout within 48 hours after a red tide outbreak.

[0097] It can be seen that the present invention shows good sensitivity and responsiveness in small-scale, highly dynamic nearshore ecosystems, can effectively improve the scientific nature and adaptability of fishery surveys, and provides important technical support for monitoring and protecting fishery resources in response to environmental emergencies.

[0098] Application Example 3:

[0099] In another specific embodiment, the present invention is applied to the Taiwan Strait region, which is significantly influenced by multi-scale ocean circulation (such as monsoon currents, coastal currents, and strait exchange currents), has strong physical field instability, and has a complex variety of fishery resources with significant seasonal and regional distribution. To carry out cross-seasonal, multi-target species surveys while addressing challenges such as long survey cycles and high prediction uncertainty, this embodiment adopts the proposed method for adaptive optimization of survey stations based on a spatiotemporal coupling model to improve the scientific nature, foresight, and robustness of station placement.

[0100] The project implementation period covers April to October, including the three main seasonal stages of spring, summer and autumn. The target species are large pelagic fish and economic cephalopods. First, a ROMS regional model covering the entire Taiwan Strait was constructed with a horizontal resolution of 3 kilometers. A two-layer nesting mechanism was adopted, in which the outer boundary was connected to HYCOM, and the inner sub-area was used for high-precision prediction of key sea areas. A 30-layer σ coordinate system was used vertically, with a focus on strengthening the vertical accuracy of the pelagic to thermocline region. The atmospheric forcing field was derived from the ERA5 product with a 3-hour resolution. The biogeochemistry module used the NPZD advanced variant module, and introduced the sedimentary nutrient release mechanism to simulate the impact of bottom nitrogen and phosphorus loads on chlorophyll and plankton.

[0101] To account for seasonal fluctuations that may occur during long-term survey missions (such as the onset of the summer monsoon and vertical water mass disturbances caused by tropical cyclones), a multi-source fusion strategy was implemented for assimilating environmental variables. Sea surface temperature and salinity fields were fused with remote sensing data (such as MODIS and AVHRR) and Argo profiles using a particle filter. Plankton fields and chlorophyll concentrations were updated on a rolling basis using the NPZD model combined with shipborne sampling data. The assimilation system runs daily, producing forecast products for 3-7 days.

[0102] At the ecological model level, the INLA model was used to model each target species separately, constructing a multi-species, hierarchical joint distribution framework. The INLA model structure incorporates inter-species spatial random effects covariance terms within the standard GAM framework, enabling it to capture spatial covariation patterns among symbiotic populations. The predicted results for each species were superimposed and weighted averaged to form a comprehensive resource suitability index (Multi-species HSI). This index serves as a basis for comprehensive site optimization and ensures spatial representativeness in multi-target species surveys.

[0103] The calculation of the Habitat Suitability Index (HSI) is still based on the standardized formula:

[0104]

[0105] in, represents the predicted abundance of the mth species at position i, and are the maximum and minimum abundance of the species at all grid points respectively. The comprehensive index is obtained by:

[0106]

[0107] Among them, W m is the weight of the mth species, which is set according to its ecological status, economic importance and seasonal dynamics.

[0108] When optimizing station locations, stricter station placement constraints are set, taking into account the long survey period and unstable meteorological windows:

[0109] Coverage constraint: The cumulative coverage rate of the area with comprehensive HSI>0.6 must reach more than 90%;

[0110] Redundancy constraint: Two sets of redundant stations are required in the core hotspot area to replace in severe sea conditions;

[0111] Transmission constraints: The optimization results should be adapted to the ship's maneuvering path and fuel budget, and the spatial distribution scheme with the shortest route should be prioritized.

[0112] The optimization algorithm incorporates a combined spatial clustering and stratified sampling approach: First, candidate sites are preliminarily clustered using K-means based on ecological stratification results. Then, weighted stratified sampling is performed within each ecological subregion to ensure a balance between representativeness and operational feasibility. A comparison of the optimized station locations output by the system with the original voyage station layout shows that while the total number of stations is reduced by 12%, the average sampling time is reduced by 15%, and the error in species abundance estimates is kept within ±10%.

[0113] The present invention achieves high-precision adaptive layout of survey stations in the complex, highly dynamic, and multi-species co-inhabiting area of ​​the Taiwan Strait, demonstrating strong system adaptability, spatial generalization capability, and multi-objective balancing capability. It is particularly suitable for future national-level long-term monitoring plans for fishery resources, coordinated optimization of R / V survey voyages, and standardized station design requirements for related international joint survey projects.

[0114] In summary, this paper provides an adaptive optimization method for marine fishery resource survey station locations based on a spatiotemporal coupling model. This method fully integrates ocean physical processes and biological and ecological characteristics. By constructing a high-precision dynamic reference field, achieving multidimensional ecological stratification, and implementing adaptive station placement based on environmental triggering mechanisms, it effectively enhances the scientific nature, systematicity, and real-time responsiveness of fishery resource surveys. This method overcomes the limitations of traditional station design, which relies on static assumptions and has poor adaptability to environmental changes. It enables flexible deployment and dynamic optimization in multi-species, multi-region, and multi-scale application scenarios, demonstrating excellent versatility and engineering feasibility.

[0115] The technical solution described in the present invention is not only suitable for fishery resource assessment and dynamic monitoring, but can also be widely used in ecological protection zone planning, marine environmental impact assessment, fishery management early warning system construction and other fields, and has significant application prospects and promotion value.

[0116] Those skilled in the art should understand that, without departing from the essential spirit of the present invention, any equivalent replacement or modification of the specific structure, model details, and parameter settings should be deemed to fall within the scope of protection of the present invention.

Claims

1. A method for adaptive optimization of marine fishery resource survey stations based on a spatiotemporal coupling model, characterized in that: The following steps are involved: (1) Dynamic reference field construction: coupling the Bayesian hierarchical species distribution model (INLA) with the Regional Ocean Model System (ROMS), integrating historical fishery survey data and environmental variables through data assimilation technology, and generating a spatiotemporally continuous dynamic reference field for species distribution; (2) Three-dimensional hierarchical design: ecological regions are divided based on the temporal dimension (life history stage), spatial dimension (hot spot analysis), and environmental dimension (gradient zone characteristics); (3) Adaptive station optimization: When the change of environmental variables exceeds the preset threshold, the habitat suitability index is recalculated and the station layout is adjusted to ensure that more than 80% of the ecological hotspot areas are covered and at least three stations are deployed in each environmental gradient zone.

2. The method according to claim 1, characterized in that The Bayesian hierarchical species distribution model in step (1) includes: Species presence / absence submodel: predicts the probability of occurrence based on a binomial distribution and a log link function; Species conditional abundance submodel: predicts abundance values ​​based on Poisson distribution and log link function; The model quantifies spatial autocorrelation via a Gaussian Markov random field.

3. The method according to claim 1, characterized in that The data assimilation techniques described in step (1) include: Particle filter assimilation is used for temperature and salinity; The NPZD ecological model was used to perform bio-physical coupled assimilation on chlorophyll concentration.

4. The method according to claim 1, wherein The temporal life history stage division in step (2) includes: Spawning period: identified by gonadal development characteristics or cumulative temperature threshold method; Feeding period: determined in combination with the changes in plankton biomass; Wintering period: delineated based on winter clustering behavior.

5. The method according to claim 1, wherein The hot spot analysis of the spatial dimension in step (2) is performed using (Getis-Ord Statistics) Statistical methods, including: Core area: Transition Zone: Background area:

6. The method according to claim 1, wherein The identification of the gradient band of the environmental dimension in step (2) includes: Use the Sobel operator to calculate the temperature and salinity gradient fields; The Thorpe scale method is used to locate the depth mutation point of the mixed layer.

7. The method according to claim 1, characterized in that The triggering condition for the environmental variable change in step (3) is: The average deviation of the temperature during the planned survey period from that at the time of model construction exceeded ±1.5°C.

8. The method according to claim 1, characterized in that The site optimization in step (3) is achieved through stratified random sampling, specifically including: Initialize the candidate point set of 30′×30′ grid; The resource index is used as an indicator, and the accuracy is evaluated through relative estimation error and relative deviation.

9. The method according to claim 1, characterized in that The configuration of the Regional Ocean Model System (ROMS) includes: Horizontal resolution of 1-3 km, vertical resolution of 20-30 layers of terrain following coordinates; The atmospheric forcing field uses ERA5 reanalysis data with a temporal resolution of ≤6 hours.

10. Use of the method according to any one of claims 1 to 9 in marine fishery resource assessment, ecological protection area planning or biodiversity monitoring.

Images