Pongshan fishery fishing area forecasting system based on multiple ocean elements
By constructing a multi-oceanic-factor fishing ground forecasting system, and utilizing the MITgcm and Darwin ecological models combined with the XGboost model, the problem of lagging analysis of fishing area location and quantity was solved. This enabled dynamic, accurate forecasting and automated operation of fishing grounds, supported fishery risk early warning, and improved the system's practicality and user experience.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- QINGDAO INTELLIGENT BLUE OCEAN ENG RES INST CO LTD
- Filing Date
- 2025-12-24
- Publication Date
- 2026-05-26
AI Technical Summary
In existing technologies, fishermen rely on experience or single pieces of information when searching for fishing areas, which leads to a lag in the analysis of the location and quantity of fishing areas. This makes it impossible to dynamically and accurately indicate the location of the Zhoushan fishing grounds and lacks a comprehensive assessment of multiple marine factors.
A fishing area forecasting system for the Zhoushan fishing grounds based on multiple marine elements was constructed, including modules for data acquisition, preprocessing, fishing area forecasting model library, data refinement, and artificial intelligence model training. The system utilizes the MITgcm and Darwin ecological models for physical-ecological coupling, combines the XGboost model for probabilistic prediction of fishing areas, and displays the results through WebGIS technology.
It enables direct, dynamic, and accurate analysis of the location and quantity of fishing grounds, improves the practicality of fishing area forecasting, supports automated operation and personalized queries, provides fishery risk warnings, and enhances the system's operational efficiency and user-friendliness.
Smart Images
Figure CN122087302A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marine ecological forecasting technology, and more specifically, to a forecasting system for the Zhoushan fishing grounds and fishing areas based on multiple marine elements. Background Technology
[0002] In recent years, the high-trophic-level and high-economic-value fish resources of the Zhoushan fishing grounds, historically represented by the "four major fish products"—large yellow croaker, small yellow croaker, ribbonfish, and squid—have severely declined due to long-term overfishing and environmental changes. During the fishing season, large numbers of fishing boats gather in areas identified through experience, leading to excessive fishing intensity in some sea areas and further increasing the risk of decline.
[0003] Fishermen primarily rely on two methods to locate fishing grounds: First, historical experience and "fishing season" patterns: relying on the captain's long-accumulated experience regarding water temperature, currents, and seasons. Second, scattered marine environmental information: some fishing boats purchase commercially available satellite remote sensing images of sea surface temperature, chlorophyll concentration, etc.
[0004] However, this experience or information is often singular and outdated. It lacks analysis of the location and quantity of fish occurrences; it lacks comprehensive assessment of the impact of multiple marine factors on fishing areas, and can only serve as a reference, unable to directly, dynamically, and accurately indicate the location of fish occurrences in the Zhoushan fishing grounds.
[0005] Therefore, in view of this, we will study and improve the existing structure to provide a forecasting system for the Zhoushan fishing grounds and fishing areas based on multiple marine elements, in order to achieve a more practical value. Summary of the Invention
[0006] 1. Technical problems to be solved To address the problems existing in the prior art, the purpose of this invention is to provide a Zhoushan fishing ground forecasting system based on multiple marine elements. This system can avoid the drawbacks of relying on experience or relying on single, outdated information, complete the analysis of the location and quantity of fish occurrences, increase the comprehensive assessment of the impact of multiple marine elements on fishing areas, realize the automated operation of forecasts, and directly, dynamically, and accurately indicate the location of fish occurrences in the Zhoushan fishing grounds, thereby improving the practical effectiveness of the system.
[0007] 2. Technical Solution To solve the above problems, the present invention adopts the following technical solution.
[0008] The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements includes: The data acquisition module is used to acquire or periodically update meteorological data, marine physical data, marine ecological data, historical catch statistics and tidal boundary data related to the fishing ground environment from multiple data sources in real time; The data preprocessing module is connected to the data acquisition module to preprocess marine physical and marine ecological elements; A fishing area forecasting model library is used to couple the MITgcm and Darwin ecological models to determine the latitude and longitude range of the simulated sea area (28°~33°N) and (120°E~125°E), with a grid spacing of 500m. Various parameters in the model are set to forecast the physical and ecological elements of the Zhoushan fishing grounds. The data refinement module is connected to the data preprocessing module and the fishing area forecasting model library, and is used for data assimilation, data verification and identification of marine dynamic processes. The artificial intelligence model training module collects and classifies multiple types of data, and finally establishes a probability model of fishing areas. The results output and operation module outputs a heat map of the probability of fishery occurrence within the Zhoushan fishing grounds for the forecast date, and deploys an automated program for automated operation.
[0009] Furthermore, the data acquired by the data acquisition module specifically includes: Meteorological data: ECMWF weather forecast data, WOA climatological average data; Ocean physical data: GLORYS ocean forecast data, GEBCO topographic data; Marine ecological data: OCC-CI satellite remote sensing chlorophyll data; Historical catch statistics: Fishing vessel location information and corresponding catch data obtained from the fishing vessel's Beidou / VMS system; Tidal boundary data: The boundary tidal forcing is calculated using the TPXO model and the OTPS tool to generate tidal current velocity data.
[0010] Furthermore, when the data preprocessing module performs preprocessing operations: For marine physical elements: the physical data acquired by the data acquisition module are processed by cubic spline interpolation to the target model resolution and missing values are supplemented; For marine ecological elements: For satellite chlorophyll data, the vertical chlorophyll distribution is calculated based on empirical formulas, missing values are supplemented based on diffusion algorithms, and dissolved oxygen and phytoplankton content are estimated based on chlorophyll concentration; The nutrient data for phosphate and nitrate were obtained from WOA climatological average data, and the runoff nutrient input was obtained from the "China River Sediment Bulletin". The data were also preprocessed by interpolation and missing value filling.
[0011] Furthermore, the models stored in the fishing area forecasting model library include at least: The habitat suitability index model constructs a suitability function for a target fish species to various marine environmental factors and generates a comprehensive suitability index map by weighted superposition. Data-driven models based on machine learning are trained using historical catch data and concurrent marine environmental data. Model types include random forest, gradient boosting tree, or neural network. Ocean dynamics-ecology coupled models predict the location and changes of fishing grounds by simulating the interaction between key physical processes and basic productivity.
[0012] Furthermore, in the data refinement processing module: When performing data assimilation, satellite observation data, including sea surface temperature, sea surface height, and sea surface chlorophyll concentration, are assimilated to optimize the accuracy of simulation results; When conducting data verification and error analysis, the forecast results and satellite observation results are verified to ensure the accuracy of the forecast; When identifying ocean dynamic processes, mesoscale eddies and upwellings are identified based on ocean currents, and fronts are identified based on sea surface temperature.
[0013] Furthermore, the fishing area distribution probability map output by the data refinement processing module includes the potential fishing area ranges at different probability levels, along with corresponding forecast lead times and confidence level descriptions.
[0014] Furthermore, in the artificial intelligence model training module: Historical commercial fishing data were collected, and data on sea surface temperature, sea surface height, chlorophyll concentration, ocean fronts, mesoscale eddies, and upwelling were divided into training, validation, and test sets. The XGboost model was used for training to obtain the correlation coefficient weights between fishing data and marine elements, and a probability model of fishing areas was established.
[0015] Furthermore, in the result output and execution module: When outputting the results, the month, latitude and longitude, model forecast data, and frontal, mesoscale eddy and upwelling data identified by the model are input into the fishing area probability model, and the probability heat map of the fishing grounds within the Zhoushan fishing grounds is output for the forecast date. During automated operation, the automated program is deployed to periodically acquire data, preprocess the data, create input files, start the model, store / clean the results, import the regional forecast probability model to output the probability map of fishing area occurrence, and realize automated operation; The output and execution module supports result display based on WebGIS technology. Users can zoom in and out on the interactive map, query forecast details for different locations, and customize and export results according to conditions such as fish species, date, and forecast time.
[0016] Furthermore, the Zhoushan fishing grounds and fishing area forecasting system also includes: The early warning module, connected to the data refinement module, is used to generate and issue fishery disaster early warning information when the forecast model identifies extreme marine environmental conditions that may have an adverse impact on fishery production or marine ecosystems. Adverse conditions include hypoxia, harmful algal blooms, and strong winds and waves.
[0017] 3. Beneficial Effects Compared with the prior art, the advantages of this invention are: This scheme builds an ultra-high resolution physical-ecological coupled model based on the MITgcm circulation model and the Darwin ecological model to predict the dynamic and ecological processes of the Zhoushan fishing grounds. The resolution reaches 500m and can distinguish mesoscale and sub-mesoscale processes in the ocean. Then, based on historical fishing data and satellite observation and reanalysis data from previous years, the XGboost artificial intelligence model was trained to obtain the correlation weights of various marine elements. Fronts, mesoscale eddies and upwellings were identified based on the predicted ocean temperature, sea surface height, ocean currents and chlorophyll. Then, based on the data of sea surface temperature, sea surface height, fronts, mesoscale eddies and upwellings and the weights of various marine elements obtained from the training, a probability model of fishing grounds and fishing areas was built. This avoids the drawbacks of experience or single and lagging information, and completes the analysis of the location and quantity of fish appearance. It also adds a comprehensive judgment on the impact of multiple marine elements on fishing areas. Then, by deploying automated programs to periodically acquire data, preprocess the data, create input files, start the model, and store / clean the results, the system imports the regional forecast probability model to output a probability map of fishing area occurrence, thus achieving automated operation. This can directly, dynamically, and accurately indicate the location of fishing areas in the Zhoushan fishing grounds, improving the system's practical effectiveness. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating the operation of the Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements in this invention. Figure 2 This is a schematic diagram illustrating the error analysis between model-predicted chlorophyll and satellite observations in this invention; Figure 3 This is a schematic diagram of the frontal region identified based on the sea surface temperature on August 6, 2023, in this invention. Figure 4 This is a schematic diagram of a heat map showing the predicted probability of mackerel fishing area occurrence on a certain day in 2025, as described in this invention. Detailed Implementation
[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Example
[0020] Please see Figure 1 The Zhoushan fishing ground forecasting system, based on multiple marine elements, includes: The data acquisition module is used to acquire or periodically update meteorological data, marine physical data, marine ecological data, historical catch statistics and tidal boundary data related to the fishing ground environment from multiple data sources in real time; The data preprocessing module is connected to the data acquisition module to preprocess marine physical and marine ecological elements; A fishing area forecasting model library is used to couple the MITgcm and Darwin ecological models to determine the latitude and longitude range of the simulated sea area (28°~33°N) and (120°E~125°E), with a grid spacing of 500m. Various parameters in the model are set to forecast the physical and ecological elements of the Zhoushan fishing grounds. The data refinement module is connected to the data preprocessing module and the fishing area forecasting model library, and is used for data assimilation, data verification and identification of marine dynamic processes. The artificial intelligence model training module collects and classifies multiple types of data, and finally establishes a probability model of fishing areas. The results output and operation module outputs a heat map of the probability of fishery occurrence within the Zhoushan fishing grounds for the forecast date, and deploys an automated program for automated operation.
[0021] The overall architecture of the Zhoushan fishing grounds and fishing area forecasting system was defined, including a complete module chain from data acquisition to result output. A closed-loop, automated forecasting workflow was constructed.
[0022] By systematically integrating multi-source data acquisition, physical-ecological coupled numerical forecasting, data assimilation and verification, artificial intelligence models, and automated output, an integrated and operational forecasting capability for the Zhoushan fishing grounds has been achieved, encompassing environmental simulation and fishing condition inference.
[0023] The modules are connected sequentially, clarifying the data flow and functional logic, thus ensuring the integrity and reliability of the forecast results.
[0024] Example 2: Based on the above embodiment 1, further description is provided.
[0025] Specifically, the data acquired by the data acquisition module includes: Meteorological data: ECMWF weather forecast data, WOA climatological average data; Ocean physical data: GLORYS ocean forecast data, GEBCO topographic data; Marine ecological data: OCC-CI satellite remote sensing chlorophyll data; Historical catch statistics: Fishing vessel location information and corresponding catch data obtained from the fishing vessel's Beidou / VMS system; Tidal boundary data: The boundary tidal forcing is calculated using the TPXO model and the OTPS tool to generate tidal current velocity data.
[0026] By integrating multi-source, heterogeneous climate, marine environmental, and fisheries production data, a comprehensive and reliable data foundation is provided for accurate forecasting.
[0027] Specifically, when the data preprocessing module performs preprocessing tasks: For marine physical elements: the physical data acquired by the data acquisition module are processed by cubic spline interpolation to the target model resolution and missing values are supplemented; For marine ecological elements: For satellite chlorophyll data, the vertical chlorophyll distribution is calculated based on empirical formulas, missing values are supplemented based on diffusion algorithms, and dissolved oxygen and phytoplankton content are estimated based on chlorophyll concentration; The nutrient data for phosphate and nitrate were obtained from WOA climatological average data, and the runoff nutrient input was obtained from the "China River Sediment Bulletin". The data were also preprocessed by interpolation and missing value filling.
[0028] This study addresses the usability issue of multi-source, heterogeneous marine environmental data. Through interpolation of physical elements and empirical formula extrapolation and missing value imputation for ecological elements, data from different sources, at different resolutions, and with missing values are uniformly processed into a standardized data format that is spatiotemporally continuous and complete, as required by the model. This provides a high-quality, consistent input data foundation for subsequent numerical simulations and artificial intelligence analysis, and is a primary technical step in ensuring the forecast accuracy of the entire system.
[0029] Specifically, the models stored in the fishing area forecasting model library include at least: The habitat suitability index model constructs a suitability function for a target fish species to various marine environmental factors and generates a comprehensive suitability index map by weighted superposition. Data-driven models based on machine learning are trained using historical catch data and concurrent marine environmental data. Model types include random forest, gradient boosting tree, or neural network. Ocean dynamics-ecology coupled models predict the location and changes of fishing grounds by simulating the interaction between key physical processes and basic productivity.
[0030] Multiple models can coexist, and can be selected or combined according to different fish species and data completeness to improve the accuracy and robustness of forecasts.
[0031] Specifically, in the data refinement module: When performing data assimilation, satellite observation data, including sea surface temperature, sea surface height, and sea surface chlorophyll concentration, are assimilated to optimize the accuracy of simulation results.
[0032] By assimilating observational data, the initial field and simulation results of the model are optimized and corrected, directly improving the forecast accuracy of physical and ecological elements.
[0033] When conducting data verification and error analysis, the forecast results and satellite observation results are verified to ensure the accuracy of the forecast; By verifying data and analyzing errors, we conduct quality control on forecast results to ensure the reliability of the output products.
[0034] When identifying ocean dynamic processes, mesoscale eddies and upwellings are identified based on ocean currents, and fronts are identified based on sea surface temperature.
[0035] By identifying key ocean dynamic processes such as fronts and mesoscale eddies, raw environmental field data are transformed into fisheries oceanographic features that directly indicate the formation of fishing grounds, making the forecast results more interpretable and applicable in fisheries science.
[0036] Specifically, the data refinement module outputs a probability map of fishing area distribution, which includes the potential fishing area ranges of different probability levels (such as high, medium, and low), along with corresponding forecast lead time and confidence level descriptions.
[0037] Probabilistic forecasts are better suited to the decision-making needs of fisheries production and help users assess risks and benefits.
[0038] Specifically, in the artificial intelligence model training module: Historical commercial fishing data were collected, and data on sea surface temperature, sea surface height, chlorophyll concentration, ocean fronts, mesoscale eddies, and upwelling were divided into training, validation, and test sets. The XGboost model was used for training to obtain the correlation coefficient weights between fishing data and marine elements, and a probability model of fishing areas was established.
[0039] A data-driven probabilistic prediction model for fishing grounds was established. By collecting historical fishing data and various marine environmental characteristic data, and training the model using machine learning algorithms such as XGBoost, the model can automatically learn and quantify the complex nonlinear relationships and relative importance (weights) between various marine elements (such as temperature, chlorophyll, eddies, etc.) and fishing ground formation.
[0040] Ultimately, the model can output the probability of future fishing grounds based on real-time or forecasted marine environmental characteristics, achieving a key leap from "environmental forecasting" to "fishery situation forecasting," and the method is flexible and can be continuously optimized as data accumulates.
[0041] Specifically, in the results output and execution modules: When outputting the results, the month, latitude and longitude, model forecast data, and frontal, mesoscale eddy and upwelling data identified by the model are input into the fishing area probability model, and the probability heat map of the fishing grounds within the Zhoushan fishing grounds is output for the forecast date. During automated operation, the automated program is deployed to periodically acquire data, preprocess the data, create input files, start the model, store / clean the results, import the regional forecast probability model to output the probability map of fishing area occurrence, and realize automated operation.
[0042] The forecast results are output in the form of a probability heatmap, which intuitively shows the probability of different sea areas becoming fishing grounds, making it easier for fishermen and fisheries managers to make risk decisions.
[0043] By deploying fully automated programs, unattended operations were achieved from data acquisition, processing, model running to result generation and publishing, greatly improving the system's operational efficiency and stability.
[0044] By supporting interactive display and query based on WebGIS, the system provides users with a convenient, intuitive, and personalized information retrieval experience, significantly improving the system's service capabilities and user-friendliness.
[0045] The results output and execution module supports result display based on WebGIS technology. Users can zoom in and out on interactive maps, query forecast details for different locations, and customize and export results according to conditions such as fish species, date, and forecast time.
[0046] It provides an intuitive and convenient user interaction experience, meeting the personalized information acquisition needs of different users.
[0047] Specifically, the Zhoushan fishing ground and fishing area forecasting system also includes: The early warning module, connected to the data refinement module, is used to generate and issue fishery disaster early warning information when the forecast model identifies extreme marine environmental conditions that may have an adverse impact on fishery production or marine ecosystems. Adverse conditions include hypoxia, harmful algal blooms, and strong winds and waves.
[0048] The system's service functions have been expanded to not only forecast fishing areas but also provide early warnings of risks, ensuring safe production in the fisheries sector.
[0049] Working principle: First, we acquired multi-source ocean data, including ECMWF meteorological forecast data, GEBCO topographic data, GLORYS ocean forecast data, and OCC-CI satellite remote sensing chlorophyll data, and then used the TPXO model OTPS tool to calculate the boundary tidal forcing. Then, the acquired data is processed by cubic spline interpolation to the target model resolution, and a physical element input file is created. For satellite chlorophyll data, the vertical chlorophyll distribution is calculated according to empirical formulas and interpolated to the model resolution along with other ecological data.
[0050] Then, a 500m resolution numerical model was built based on the MITgcm and Darwin ecological models to forecast marine physical and ecological elements. The model assimilates sea surface height, temperature and sea surface chlorophyll observed by satellite and performs error analysis on the model forecast data.
[0051] Then, based on the output forecast data, fronts, mesoscale eddies, and upwellings are identified, and these dynamic phenomena are quantified. Based on monthly, latitude and longitude, fishing data, and multivariate marine environmental data, an XGboost AI model is trained. The collected historical fishing data, along with ocean temperature, sea surface height, chlorophyll concentration, fronts, mesoscale eddies, and upwellings, are used to obtain the weights of each parameter, establishing a fishing area forecast probability model. The monthly, latitude and longitude data, as well as the model's predicted ocean temperature, sea surface height, chlorophyll concentration, fronts, mesoscale eddies, and upwellings data, are input into the fishing area forecast probability model. The output is the probability of the fishing area appearing within the latitude and longitude range of the Zhoushan fishing grounds on the forecast day.
[0052] Then, a heatmap of the probability of fish appearing in the Zhoushan fishing grounds (0-100%) is output for the forecast date. Subsequently, an automated program is deployed to periodically acquire data, preprocess the data, create input files, start the model, store / clean the results, import the regional forecast probability model, and output the probability map of fish appearing in the fishing area, thus achieving automated operation.
[0053] Example 3: Further description is provided based on the above embodiments 1 and 2.
[0054] Fish species exhibit significant differences in their spatial and temporal affinity for marine elements, which is also related to the fish's growth stage. Therefore, when constructing probabilistic models for fishing areas, separate models should be built for different fish species. The following describes the process of constructing a forecasting system for the mackerel fishing area in the Zhoushan fishing grounds.
[0055] first step: Based on the coupled MITgcm model and Darwin ecological model, data were acquired within the latitude and longitude range (120°–125°, 28°–33°): Atmospheric forecast data were obtained from publicly available forecasts from the European Centre for Medium-Range Weather Forecasts (ECMWF), including 10m zonal / meridian wind fields, 2m air temperature and humidity, sea surface atmospheric pressure, longwave and shortwave radiation, and precipitation, with a resolution of 28km and a forecast lead time of 10 days. Topographic data were acquired using the latest version (15 arcsecond resolution) of GEBCO ocean topographic data. Ocean forecast data were obtained from the Copernicus Marine Environment Monitoring Service, with a resolution of 1 / 12° (approximately 10km) and a forecast lead time of 10 days: three-dimensional temperature, salinity, and current fields (50 layers from 0–1500m, with vertical resolution increasing with depth), and sea surface height anomaly (SLA) data. Sea surface chlorophyll data were obtained from OC-CCI, deep chlorophyll data were estimated using an algorithm, dissolved oxygen and phytoplankton abundance were estimated based on chlorophyll, other nutrient data were obtained from WOA climatological average data, and runoff nutrient input data were obtained from the "China River Sediment Bulletin". All the above data were interpolated to a 500m resolution, and corresponding input files were created according to the specified format.
[0056] Step Two: The TPXO model and OTPS tool are used to predict boundary tidal forcing and generate tidal current velocity data. The specific process is as follows: Ten tidal constituent names (M2, S2, N2, K2, K1, O1, P1, Q1, M4, MS4) are entered in the configuration file. Then, a file containing the location and time of the predicted points is generated. The `make` program is executed in the OTPS directory to generate the `predict_tide` executable file. Next, `predict_tide` is executed to output the tidal current data for the China Sea boundary. Then, cubic spline interpolation is used to match the model resolution. The tidal data is corrected to the East 8 time zone as the outer sea boundary condition, and finally, a boundary condition driving file is created.
[0057] The simulated sea area covers longitude 120°–125° and latitude 28°–33°, with a resolution of 500m and 40 vertical layers, increasing from a 1.5m interval at the sea surface towards the seabed, with a time step of 10s.
[0058] The time stepping method uses the Adams-Bashforth scheme:
[0059] The bottom boundary uses a cutting mesh to bridge terrain variations.
[0060] In the model's code folder, set the model's mesh spacing, computation kernel, and parameterization method. Similarly, select the parameterization method for the Darwin ecosystem model. Execute the following command in the build folder: .. / .. / .. / tools / genmake2 -mods .. / code -mpi -of .. / .. / .. / tools / build_options / liunx* The Linux system uses the local processor model to generate a Makefile, then executes the commands `makedepend` and `make -j 8` to start compilation and generate the `mitgcmuv` executable file.
[0061] Next, in the run file, link the initial field and forced field files created earlier to this folder, copy the generated migcmuv executable file to this folder, configure other model parameters, and then execute the following command, where num is the number of computation kernels: mpirun -np num . / mitgcmuv Step 3: Assimilate satellite observation data (sea surface height, temperature, and sea surface chlorophyll) to optimize the accuracy of simulation results.
[0062] The forecast data includes sea surface temperature, sea level, ocean current velocity, and chlorophyll concentration. To verify the accuracy of the forecast, an error analysis was performed between the forecast data and satellite observation data. Taking chlorophyll as an example, the figure shows the error analysis between the forecast daily average chlorophyll on August 6, 2025, and satellite observations (after interpolation). The RMSE is 2.167, the MAE is 1.090, and the R-squared value is 0.871.
[0063] Historical mackerel catch data, ocean satellite observation data, and reanalysis data were acquired and trained using the XGboost model. Specifically, historical fronts, mesoscale eddies, and upwellings were first identified based on historical ocean environmental data to quantify ocean dynamic processes in data form. Taking fronts as an example, the front identification algorithm was based on the Cayula and Cornillon (CCA) histogram algorithm, which is the most widely used detection algorithm in marine ecology and fisheries.
[0064] like Figure 2The image shows oceanic fronts identified based on sea surface temperature data from the 0.05°E ESA-CCI hybrid satellite system on August 6, 2023. Red indicates warm fronts, and blue indicates cold fronts. Since the impact of frontal effects on the ecological environment is currently unclear, warm and cold front regions are set to 1. Similarly, identified mesoscale eddies and upwelling regions are also set to 1. Temperature, sea surface height, and sea surface chlorophyll concentration data have been normalized.
[0065] Step 4: Monthly, latitude and longitude, sea surface temperature, sea surface height, chlorophyll concentration, as well as identified fronts, mesoscale eddies, and upwelling data are used as feature variables and aligned and integrated with the target variable of catch yield.
[0066] Further, these features will be expanded to include delayed chlorophyll concentration, temperature, and sea level height; delayed fronts and mesoscale eddies; and gradients of chlorophyll concentration, temperature, and sea level height, for a total of 32 features.
[0067] The XGBoost model is trained by dividing the data into training, validation, and test sets. XGBoost supports distributed deployment, can handle massive amounts of environmental monitoring data, has a built-in cross-validation mechanism to avoid bias from manual parameter tuning, provides multiple feature importance evaluation metrics (gain, coverage, frequency) to enhance model interpretability, and automatically learns the complex nonlinear relationship between environmental features and fishing results through gradient boosting decision trees.
[0068] After the training set is input into the model, the optimal weights of the initial leaf nodes are first calculated. Gradient statistics are then calculated, and in each iteration, the algorithm calculates the first and second gradients of the loss function. A weighted quantile sketch algorithm is further employed to pre-sort and block-process the marine environmental features, enabling efficient split point finding. For each feature splitting candidate point, the splitting gain is calculated. L1 and L2 dual regularization terms are introduced to control model complexity.
[0069] The number of base learners was set to 1000, the maximum tree depth to 6, 3-fold cross-validation was used, the learning rate was 0.05, and the number of early stopping rounds was set to 50. Training began, and the optimal model was saved. After training, the model output the global importance percentage of each environmental factor, establishing a probabilistic model for mackerel fishing areas. Validation was successful. The top 5 features by weight are: delayed chlorophyll concentration, month, latitude and longitude, delayed temperature, and delayed sea level. Chlorophyll, temperature, and sea level correspond to fish feeding areas, month, and seasonality and regionality of latitude and longitude.
[0070] The overall average accuracy (ACC) of the test set in different months was 83.92%, indicating that the predicted locations were relatively accurate.
[0071] Figure 3 As shown, specifically, based on the validated forecasts mentioned above—sea surface temperature, sea level, ocean current velocity, and chlorophyll concentration—a recognition algorithm identifies the predicted fronts, mesoscale eddies, and upwellings. The sea surface temperature, sea level, ocean current velocity, chlorophyll concentration, fronts, mesoscale eddies, and upwelling data are then input into the mackerel fishing area probability model to obtain the mackerel fishing area probability distribution, and a heatmap of the fishing area probability distribution is plotted.
[0072] Figure 4 The image shows a heat map of the probability forecast for mackerel fishing areas in the Zhoushan fishing grounds on a certain day in 2025. Areas with a probability of occurrence below 50% were removed, while areas with a probability above 50% were considered likely to be fishing areas.
[0073] Note that 50% is the threshold. In actual use, it is necessary to consider the specific needs of fishermen regarding catch volume and fuel consumption of fishing boats: if fishermen expect a larger catch volume, the threshold should be appropriately lowered; if fishermen expect lower fuel consumption, the threshold should be increased accordingly.
[0074] Specifically, the program is written to implement the following process: data acquisition → data preprocessing → importing into the numerical forecasting model → outputting ocean element forecast results → identifying fronts, mesoscale eddies, and upwellings based on the results → importing forecast data into the fishing area probability model → outputting the Zhoushan fishing ground fishing area probability forecast results.
[0075] The above description is merely a preferred embodiment of the present invention; however, the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concepts, should be covered within the scope of protection of the present invention.
Claims
1. A fishing ground and fishing area forecasting system based on multiple marine elements in Zhoushan fishing grounds, characterized by: The Zhoushan fishing grounds and fishing area forecasting system includes: The data acquisition module is used to acquire or periodically update meteorological data, marine physical data, marine ecological data, historical catch statistics and tidal boundary data related to the fishing ground environment from multiple data sources in real time; The data preprocessing module is connected to the data acquisition module to preprocess marine physical and marine ecological elements; A fishing area forecasting model library is used to couple the MITgcm and Darwin ecological models to determine the latitude and longitude range of the simulated sea area (28°~33°N) and (120°E~125°E), with a grid spacing of 500m. Various parameters in the model are set to forecast the physical and ecological elements of the Zhoushan fishing grounds. The data refinement module is connected to the data preprocessing module and the fishing area forecasting model library, and is used for data assimilation, data verification and identification of marine dynamic processes. The artificial intelligence model training module collects and classifies multiple types of data, and finally establishes a probability model of fishing areas. The results output and operation module outputs a heat map of the probability of fishery occurrence within the Zhoushan fishing grounds for the forecast date, and deploys an automated program for automated operation.
2. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: The data acquired by the data acquisition module specifically includes: Meteorological data: ECMWF weather forecast data, WOA climatological average data; Ocean physical data: GLORYS ocean forecast data, GEBCO topographic data; Marine ecological data: OCC-CI satellite remote sensing chlorophyll data; Historical catch statistics: Fishing vessel location information and corresponding catch data obtained from the fishing vessel's Beidou / VMS system; Tidal boundary data: The boundary tidal forcing is calculated using the TPXO model and the OTPS tool to generate tidal current velocity data.
3. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: When the data preprocessing module performs preprocessing operations: For marine physical elements: the physical data acquired by the data acquisition module are processed by cubic spline interpolation to the target model resolution and missing values are supplemented; For marine ecological elements: For satellite chlorophyll data, the vertical chlorophyll distribution is calculated based on empirical formulas, missing values are supplemented based on diffusion algorithms, and dissolved oxygen and phytoplankton content are estimated based on chlorophyll concentration; The nutrient data for phosphate and nitrate were obtained from WOA climatological average data, and the runoff nutrient input was obtained from the "China River Sediment Bulletin". The data were also preprocessed by interpolation and missing value filling.
4. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: The models stored in the fishing area forecasting model library include at least: The habitat suitability index model constructs a suitability function for a target fish species to various marine environmental factors and generates a comprehensive suitability index map by weighted superposition. Data-driven models based on machine learning are trained using historical catch data and concurrent marine environmental data. Model types include random forest, gradient boosting tree, or neural network. Ocean dynamics-ecology coupled models predict the location and changes of fishing grounds by simulating the interaction between key physical processes and basic productivity.
5. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: In the data refinement processing module: When performing data assimilation, satellite observation data, including sea surface temperature, sea surface height, and sea surface chlorophyll concentration, are assimilated to optimize the accuracy of simulation results; When conducting data verification and error analysis, the forecast results and satellite observation results are verified to ensure the accuracy of the forecast; When identifying ocean dynamic processes, mesoscale eddies and upwellings are identified based on ocean currents, and fronts are identified based on sea surface temperature.
6. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: The data refinement module outputs a probability map of fishing area distribution, which includes the potential fishing area ranges at different probability levels, along with corresponding forecast lead times and confidence level descriptions.
7. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: In the artificial intelligence model training module: Historical commercial fishing data were collected, and data on sea surface temperature, sea surface height, chlorophyll concentration, ocean fronts, mesoscale eddies, and upwelling were divided into training, validation, and test sets. The XGboost model was used for training to obtain the correlation coefficient weights between fishing data and marine elements, and a probability model of fishing areas was established.
8. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: In the result output and execution module: When outputting the results, the month, latitude and longitude, model forecast data, and frontal, mesoscale eddy and upwelling data identified by the model are input into the fishing area probability model, and the probability heat map of the fishing grounds within the Zhoushan fishing grounds is output for the forecast date. During automated operation, the automated program is deployed to periodically acquire data, preprocess the data, create input files, start the model, store / clean the results, import the regional forecast probability model to output the probability map of fishing area occurrence, and realize automated operation; The output and execution module supports result display based on WebGIS technology. Users can zoom in and out on the interactive map, query forecast details for different locations, and customize and export results according to conditions such as fish species, date, and forecast time.
9. The Zhoushan fishing ground and fishing area forecasting system based on multiple marine elements according to claim 1, characterized in that: The Zhoushan fishing grounds and fishing area forecasting system also includes: The early warning module, connected to the data refinement module, is used to generate and issue fishery disaster early warning information when the forecast model identifies extreme marine environmental conditions that may have an adverse impact on fishery production or marine ecosystems. Adverse conditions include hypoxia, harmful algal blooms, and strong winds and waves.