A method and system for predicting tuna longline fishing activity

Abstract

The application discloses a kind of tuna longline fishing activity prediction method and system, comprising the following steps: step S1, the fishing effort data of tuna and the marine environment data corresponding to the regional fishing effort data are acquired;Step S2, different time resolution and spatial resolution are constructed to construct the space-time sequence tensor data of different space-time scales;Step S3, ConvLSTM-Attention U-Net network model is constructed and trained, and the optimal space-time scale model is determined, wherein the time step M is greater than or equal to 3;Step S4, based on the optimal space-time scale model, the model prediction result is visualized and compared with the real fishing effort distribution;Step S5, the average absolute gradient value of variable under different space-time scales is calculated, and the importance of variable under different space-time scales is evaluated and visualized.The application can predict the position distribution and numerical size of fishing effort and make accurate prediction on tuna longline fishing activities in the mid-Pacific region.

Description

Technical Field

[0001] This invention relates to the field of fisheries and fishing grounds, and more particularly to a method and system for predicting tuna longline fishing activities. Background Technology

[0002] While AIS data can quantify historical fishing intensity, it cannot predict future trends. Medium- to long-term forecasts of fishing effort are particularly important for fisheries resource management. Research indicates that the distribution of fishing effort by high seas longline fishing vessels is influenced by environmental factors; by combining environmental data with fisheries observations, predictive models can reveal the predictability of fishing intensity.

[0003] Compared to fisheries forecasting, predicting fishing intensity presents greater challenges. Fishing activities are a dynamic and continuous process; the selection of fishing spots is influenced not only by environmental factors but also by previous fishing locations, resulting in a clear spatiotemporal correlation in fishing activities. Traditional forecasting methods have limitations in processing spatiotemporal data, mostly selecting only single-point environmental data for modeling, making it difficult to comprehensively reveal the spatiotemporal trends of fishing intensity.

[0004] In recent years, deep learning methods have demonstrated their potential in fishery prediction. Convolutional Long Short-Term Memory (ConvLSTM) networks combine the advantages of CNNs and LSTMs, enabling them to handle spatiotemporal data. While ConvLSTMs can capture local spatial dependencies, they are weaker at handling long-distance spatial dependencies. U-Net, on the other hand, excels in spatiotemporal prediction due to its ability to capture multi-scale spatial features. The Attention U-Net, combining U-Net with an attention mechanism, further improves the model's prediction accuracy. The combination of ConvLSTM and Attention U-Net can simultaneously analyze spatial and temporal features, enhancing the accuracy of fishing intensity prediction.

[0005] Existing research typically considers only environmental factors, neglecting the spatiotemporal evolution of fisheries data. However, historical fishing information can influence future fishing activities, making its inclusion in models crucial. Meanwhile, while deep neural networks offer high prediction accuracy, their interpretability is poor. Understanding the importance of variables within the network is just as important as prediction accuracy. Most existing variable selection methods in deep learning models are only applicable to shallow networks or have high computational complexity, while variable importance assessment methods based on backpropagation offer better feasibility and scientific validity.

[0006] The central and western Pacific is the world's largest tuna fishing ground, with a catch of approximately 2.7 million tons in 2022, accounting for 54% of the global total. Over the past 50 years, global tuna populations have declined by 60%, making their conservation and sustainable development a key international concern. Tuna longline fishing covers a vast area and exerts the greatest pressure on marine ecosystems. Therefore, achieving refined forecasting of tuna longline fishing activities in the central and western Pacific and clarifying the importance of variables in fishing activity forecasting is of great significance for ensuring the sustainable development of marine fishery resources. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for predicting tuna longline fishing activity, in order to solve the problems mentioned in the background art.

[0008] To achieve the above-mentioned objective, one aspect of the present invention provides a method for predicting tuna longline fishing activity, comprising the following steps:

[0009] Step S1: Obtain fishing effort data for tuna in the central and western Pacific region from 2017 to 2023 and the corresponding marine environmental data.

[0010] Step S2: Construct spatiotemporal sequence tensor data of different spatiotemporal scales according to different temporal and spatial resolutions;

[0011] Step S3: ConvLSTM-Attention U-Net network model is constructed. At different spatiotemporal scales, the fishing effort data and marine environmental data of the first M time steps are used to predict the distribution of fishing activities in the next time step. The ConvLSTM-Attention U-Net network is trained to learn the spatiotemporal information in historical fishing effort data and environmental variable data, as well as the complex mapping relationship between environmental variables and fishing effort distribution. The optimal spatiotemporal scale model is determined by comparing the prediction performance of the model at different spatiotemporal scales, wherein the time step M is greater than or equal to 3.

[0012] Step S4: Based on the optimal spatiotemporal scale model, visualize the model prediction results and compare them with the actual distribution of fishing effort.

[0013] Step S5: Calculate the mean absolute gradient of the variables at different spatiotemporal scales, and evaluate and visualize the importance of the variables at different spatiotemporal scales.

[0014] Furthermore, the fishing effort data in step S1 includes the operation date, operation location, and fishing effort; the environmental factor data includes sea surface temperature, dissolved oxygen concentration at a depth of 100 meters, sea temperature at a depth of 400 meters, chlorophyll a concentration, dissolved oxygen concentration at a depth of 400 meters, and salinity at a depth of 100 meters.

[0015] Furthermore, in step S2, the processing method for fishing effort and marine environmental variables at different spatiotemporal scales is to calculate the daily average value of fishing effort for each grid at each spatiotemporal scale as the fishing effort value for that grid, and to calculate the overall average value of environmental data within each grid as the environmental variable value for that grid.

[0016] Furthermore, the ConvLSTM-Attention U-Net model structure combines the ConvLSTM model and the Attention U-Net model to capture long-term temporal features of data and process spatial features at different scales, enabling refined prediction of fishing activities.

[0017] Furthermore, the ConvLSTM model excels at learning long-term spatiotemporal features in data, and its calculation method is as follows:

[0018] ,

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] in, , , , and These represent the input gate, forget gate, memory unit, output gate, and hidden state, respectively. This is the input for the current time step. It is the hidden state of the previous time step. It is the memory unit of the previous time step. express Activation function This represents the hyperbolic tangent activation function. This represents the convolution operation. Representing a matrix product, and This represents the weight matrix of the input gate. and The weight matrix represents the forget gate. and The weight matrix represents the memory unit. and This represents the weight matrix of the output gate. and Indicates bias.

[0024] Furthermore, the time step M is set to 5.

[0025] Furthermore, at different spatiotemporal scales, the distribution of fishing activities in the next time step is predicted using fishing effort data and marine environmental data from the first M time steps. The ConvLSTM-Attention U-Net network is trained to extract spatiotemporal features from the data. The optimal spatiotemporal scale model is selected based on the general evaluation metric F1 score and the length of the time series. The formula for calculating the F1 score is as follows:

[0026] ,

[0027] ,

[0028] ,

[0029] in, This represents the number of grids where fishing effort actually exists and is predicted to have fishing effort. This represents the number of grids where fishing effort actually exists but is predicted to be absent. This represents the number of grids that are predicted to have fishing effort but do not actually have it.

[0030] Furthermore, the spatiotemporal resolution of the ConvLSTM-Attention U-Net model is set to 1°×1° and 30 days, and the spatiotemporal scale prediction results are visualized to be closest to the actual distribution of fishing effort.

[0031] Furthermore, the ConvLSTM-Attention U-Net model shows that historical fishing effort data has the highest importance across all spatiotemporal scales, while the importance of the six marine environmental variables is relatively similar with no significant differences.

[0032] Another aspect of the present invention provides a prediction system for tuna longline fishing activities in the central and western Pacific Ocean based on claims 1 to 9, comprising a data acquisition module, a data construction module, a model building module, a prediction comparison module, and a prediction evaluation module, wherein:

[0033] The data acquisition module is used to acquire data on tuna fishing efforts in the central and western Pacific region from 2017 to 2023, as well as corresponding marine environmental data.

[0034] The data construction module is used to construct spatiotemporal sequence tensor data at different spatiotemporal scales according to different temporal and spatial resolutions;

[0035] The model building module is used to construct a ConvLSTM-Attention U-Net network model. At different spatiotemporal scales, it uses fishing effort data and marine environmental data from the first M time steps to predict the distribution of fishing activities in the next time step. The ConvLSTM-Attention U-Net network is trained to learn the spatiotemporal information in historical fishing effort data and environmental variable data, as well as the complex mapping relationship between environmental variables and the distribution of fishing effort. The optimal spatiotemporal scale model is determined by comparing the prediction performance of the model at different spatiotemporal scales, where the time step M is greater than or equal to 3.

[0036] The prediction comparison module is used to visualize the prediction results of the model based on the optimal spatiotemporal scale model and compare them with the actual distribution of fishing effort.

[0037] The prediction and evaluation module is used to calculate the mean absolute gradient of variables at different spatiotemporal scales, and to evaluate and visualize the importance of variables at different spatiotemporal scales.

[0038] Compared with existing technologies, this system and method have the following advantages:

[0039] This invention constructs a predictive model for tuna longline fishing effort in the central and western Pacific Ocean at different spatiotemporal scales, capable of predicting the geographical distribution and magnitude of fishing effort. By comparing model performance, the optimal spatiotemporal scale is determined, enabling more accurate predictions of tuna longline fishing activities in the central and western Pacific. This accurate prediction not only provides a scientific basis for fisheries policy formulation but also supports the effective implementation of policies, thereby promoting the long-term sustainability of marine resources and fishing activities. Furthermore, by calculating the average absolute gradient values ​​of various environmental variables, the impact of each variable on the model's predictive performance can be quantitatively assessed, thus identifying key variables. This process helps improve the overall efficiency of the model, reduces the need for collecting non-critical data, and lowers the cost and time investment in data acquisition. Attached Figure Description Detailed Implementation

[0040] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] like Figure 1 The diagram shown is a flowchart of the method of the present invention. An embodiment of the present invention provides a short-to-medium term prediction method for tuna longline fishing effort, with the specific steps as follows:

[0042] Step S1, the implementation of the present invention relates to a method for constructing a prediction model for tuna longline fishing activities in the central and western Pacific Ocean based on deep learning, acquiring fishing effort data for tuna in the central and western Pacific Ocean from 2017 to 2023 and corresponding marine environmental data (sea surface temperature (SST), dissolved oxygen concentration at 100 meters deep (O2100), sea temperature at 400 meters deep (Tem400), chlorophyll a concentration (Chl-a), dissolved oxygen concentration at 400 meters deep (O2400), and salinity at 100 meters deep (Salt200)).

[0043] Step S2 involves dividing the region into spatial grids with spatial resolutions of 0.5°×0.5° and 1°×1°, and setting the temporal resolutions to 1 day, 3 days, 7 days, 15 days, and 30 days. These different temporal and spatial resolutions combine to form 10 spatiotemporal scales. The processing of fishing effort and marine environmental variables at different spatiotemporal scales is as follows: the daily average fishing effort for each grid at each spatiotemporal scale is calculated as the fishing effort value for that grid; the overall mean of environmental data within each grid is calculated as the environmental variable value for that grid.

[0044] Step S3: At different spatiotemporal scales, using the fishing effort data and marine environmental data from the first M (M≥3) time steps, predict the distribution of fishing activities in the next time step. Train the ConvLSTM-Attention U-Net network to capture the spatiotemporal information in the fishing effort data and environmental variable data, as well as the complex mapping relationship between environmental variables and the distribution of fishing effort, including:

[0045] The ConvLSTM model excels at learning long-term spatiotemporal features in data, and its calculation method is as follows:

[0046] ,

[0047] ,

[0048] ,

[0049] ,

[0050] ,

[0051] in, , , , and These represent the input gate, forget gate, memory unit, output gate, and hidden state, respectively. This is the input for the current time step. It is the hidden state of the previous time step. It is the memory unit of the previous time step. express Activation function This represents the hyperbolic tangent activation function. This represents the convolution operation. Representing a matrix Product (multiplication of corresponding elements), and This represents the weight matrix of the input gate. and The weight matrix represents the forget gate. and The weight matrix represents the memory unit. and This represents the weight matrix of the output gate. and Indicates bias.

[0052] ConvLSTM models rely on convolutional layers to capture spatial dependencies, but these dependencies are local and inefficient. Attention U-Net, through its encoder-decoder structure, can capture multi-scale spatial features, and it introduces an attention mechanism that enhances the model's ability to focus on important regions.

[0053] The model structure of ConvLSTM-Attention U-Net is as follows: Figure 2 As shown, it combines the advantages of ConvLSTM model and Attention U-Net, which can capture long-term temporal features of data and process spatial features at different scales, thus enabling effective and refined prediction of fishing activities.

[0054] We set M=5, at which point the model performs well and is highly efficient. Then, we trained the ConvLSTM-Attention U-Net model on each of these 10 datasets. After training, we set the positions in the predicted results where the fishing effort is greater than 0 to 1, and the other positions to 0. Next, we calculated the F1 score of the model. The F1 score was used to evaluate the predictive performance of the ConvLSTM-Attention U-Net model at different spatiotemporal scales, and the spatiotemporal scale with the highest F1 score was selected as the optimal spatiotemporal scale. The formula for calculating the F1 score is as follows:

[0055] ,

[0056] ,

[0057] ,

[0058] in, This represents the number of grids where fishing effort actually exists and is predicted to have fishing effort. This represents the number of grids where fishing effort actually exists but is predicted to be absent. This represents the number of grids that are predicted to have fishing effort but do not actually have it.

[0059] Step S4: Determine the optimal prediction performance at the optimal spatiotemporal scale.

[0060] The optimal spatiotemporal scale for the ConvLSTM-Attention U-Net model is 1°×1°, 30 days. The ConvLSTM-Attention U-Net model achieves an F1 score of 0.769 at this optimal spatiotemporal scale. Therefore, the ConvLSTM-Attention U-Net model at a spatiotemporal scale of 1°×1°, 30 days is the optimal prediction method for tuna longline fishing activities in the central and western Pacific region.

[0061] Step S5 visualizes the prediction results and variable importance. Based on the standard gradient backpropagation algorithm, the mean absolute gradient value of each variable is calculated to evaluate and visualize the importance of the variables.

[0062] The model predictions and visualizations of the actual fishing effort distribution at the optimal spatiotemporal scale (1°×1°, 30 days) are shown below. Figure 3 As shown, the prediction results of the ConvLSTM-Attention U-Net model are similar to the actual distribution of fishing effort in terms of both the location and intensity of fishing activities, indicating that the model's prediction results are quite accurate. Figure 4 As shown, in terms of the percentage of the average absolute gradient value of the variables to the total gradient value, the fishing effort data is the most important at all time and space scales. The importance of the six marine environmental variables is relatively close, with no obvious distinction.

[0063] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting tuna longline fishing activity, characterized in that, Includes the following steps: Step S1: Obtain fishing effort data for tuna in the central and western Pacific Ocean from 2017 to 2023 and the corresponding marine environmental data. The fishing effort data in Step S1 includes the operation date, operation location, and fishing effort. The environmental factor data includes sea surface temperature, dissolved oxygen concentration at a depth of 100 meters, sea temperature at a depth of 400 meters, chlorophyll a concentration, dissolved oxygen concentration at a depth of 400 meters, and salinity at a depth of 100 meters. Step S2: Construct spatiotemporal sequence tensor data of different spatiotemporal scales according to different temporal and spatial resolutions; Step S3: ConvLSTM-Attention U-Net network model is constructed. At different spatiotemporal scales, the fishing effort data and marine environmental data of the first M time steps are used to predict the distribution of fishing activities in the next time step. The ConvLSTM-Attention U-Net network model is trained to learn the spatiotemporal information in historical fishing effort data and environmental variable data, as well as the complex mapping relationship between environmental variables and the distribution of fishing effort. The optimal spatiotemporal scale model is determined by comparing the prediction performance of the model at different spatiotemporal scales, wherein the time step M is greater than or equal to 3. Step S4: Based on the optimal spatiotemporal scale model, visualize the model prediction results and compare them with the actual distribution of fishing effort. Step S5: Calculate the mean absolute gradient of the variables at different spatiotemporal scales, and evaluate and visualize the importance of the variables at different spatiotemporal scales.

2. The method for predicting tuna longline fishing activity according to claim 1, characterized in that, In step S2, the processing method for fishing effort and marine environmental variables at different spatiotemporal scales is to calculate the daily average value of fishing effort for each grid at each spatiotemporal scale as the fishing effort value for that grid, and to calculate the overall average value of environmental data within each grid as the environmental variable value for that grid.

3. The method for predicting tuna longline fishing activity according to claim 1, characterized in that, The ConvLSTM-Attention U-Net network model combines the ConvLSTM and Attention U-Net models to capture long-term temporal features of data and process spatial features at different scales, enabling refined prediction of fishing activities.

4. The method for predicting tuna longline fishing activity according to claim 3, characterized in that, The ConvLSTM model excels at learning long-term spatiotemporal features in data, and its calculation method is as follows: ， ， ， ， ， in, , , , and These represent the input gate, forget gate, memory unit, output gate, and hidden state, respectively. This is the input for the current time step. It is the hidden state of the previous time step. It is the memory unit of the previous time step. express Activation function This represents the hyperbolic tangent activation function. This represents the convolution operation. Representing a matrix product, and This represents the weight matrix of the input gate. and The weight matrix represents the forget gate. and The weight matrix represents the memory unit. and This represents the weight matrix of the output gate. and Indicates bias.

5. The method for predicting tuna longline fishing activity according to claim 1, characterized in that, The time step M is set to 5.

6. The method for predicting tuna longline fishing activity according to claim 1, characterized in that, The method involves using fishing effort data and marine environmental data from the first M time steps at different spatiotemporal scales to predict the distribution of fishing activities in the next time step. A ConvLSTM-Attention U-Net network model is trained to extract spatiotemporal features from the data. The optimal spatiotemporal scale model is selected based on the general evaluation metric F1 score and the length of the time series. The formula for calculating the F1 score is as follows: ， ， ， in, This represents the number of grids where fishing effort actually exists and is predicted to have fishing effort. This represents the number of grids where fishing effort actually exists but is predicted to be absent. This represents the number of grids that are predicted to have fishing effort but do not actually have it.

7. The method for predicting tuna longline fishing activity according to claim 1, characterized in that, The spatiotemporal resolution of the ConvLSTM-Attention U-Net network model is set to 1°×1° and 30 days.

8. A prediction system for the tuna longline fishing activity prediction method according to any one of claims 1 to 7, characterized in that, It includes a data acquisition module, a data construction module, a model building module, a prediction comparison module, and a prediction evaluation module, among which: The data acquisition module is used to acquire data on tuna fishing efforts in the central and western Pacific region from 2017 to 2023, as well as corresponding marine environmental data. The data construction module is used to construct spatiotemporal sequence tensor data at different spatiotemporal scales according to different temporal and spatial resolutions; The model building module is used to construct a ConvLSTM-Attention U-Net network model. At different spatiotemporal scales, it uses fishing effort data and marine environmental data from the first M time steps to predict the distribution of fishing activities in the next time step. The ConvLSTM-Attention U-Net network model is trained to learn the spatiotemporal information in historical fishing effort data and environmental variable data, as well as the complex mapping relationship between environmental variables and the distribution of fishing effort. The optimal spatiotemporal scale model is determined by comparing the prediction performance of the model at different spatiotemporal scales, where the time step M is greater than or equal to 3. The prediction comparison module is used to visualize the prediction results of the model based on the optimal spatiotemporal scale model and compare them with the actual distribution of fishing effort. The prediction and evaluation module is used to calculate the average absolute gradient of variables at different spatiotemporal scales, and to evaluate and visualize the importance of variables at different spatiotemporal scales.

Images