Marine ecological management-oriented habitat suitability prediction method and system

Abstract

The invention relates to the technical field of marine ecological informatics, and discloses a habitat suitability prediction method and system for marine ecological management.The habitat suitability prediction method for marine ecological management.The habitat suitability prediction method for marine ecological management.The habitat suitability prediction method for marine ecological managementincludes the steps that traditional knowledge data of fishermen is obtained, and audio data is transcribed into a text format through a natural language processing technology; generating a traditional knowledge graph containing the corresponding relationship between the observation characteristics and the fishing results; carrying out canonical correlation analysis on surface observation characteristics and vortex physical parameters in the traditional knowledge graph, and converting traditional knowledge into quantitative expression of a vortex parameter space based on a mapping matrix; and adopting a random forest to integrate multi-source prediction results, taking a traditional knowledge prediction rule and a physical model calculation result as input features, and outputting a middle-layer fish habitat probability distribution diagram. The technical problem that traditional knowledge and a vortex physical model are difficult to effectively integrate for middle-layer fish habitat prediction is solved.

Description

Technical Field

[0001] This invention relates to the field of marine ecological informatics technology, and more specifically, to a method and system for predicting habitat suitability for marine ecological management. Background Technology

[0002] Mesophyll fish (200-1000 meters deep) are an important marine fishery resource, and their habitat distribution is closely related to the dynamic processes of oceanic mesoscale eddies. Traditionally, fishermen have relied on observing sea surface phenomena (such as changes in water color, foam line distribution, and seabird gatherings) to determine the location of deep-water fish schools. These surface signals are often correlated with the vertical transport processes of mesoscale eddies. However, fishermen's experiential knowledge is largely qualitative and lacks depth information, while eddy physics models, although capable of describing three-dimensional flow fields, lack biological behavioral response mechanisms. A semantic gap exists between these two knowledge systems, making it difficult to effectively integrate traditional knowledge with scientific observation data and accurately predict the habitat distribution of mesophyll fish. Summary of the Invention

[0003] This invention provides a habitat suitability prediction method and system for marine ecological management, which solves the technical problem that there is a semantic gap between two types of knowledge systems in related technologies, making it difficult to effectively integrate traditional knowledge and scientific observation data, and making it difficult to accurately predict the habitat distribution of mid-water fish.

[0004] This invention provides a habitat suitability prediction method for marine ecological management, comprising: We acquire traditional knowledge data of fishermen, transcribe audio data into text format using natural language processing technology, extract environmental cue vocabulary and spatial location descriptions using long short-term memory networks, and generate a traditional knowledge graph containing the correspondence between observation features and fishing results. Acquire satellite vortex observation data, use vortex automatic identification algorithm to detect vortex boundary and center position, calculate the three-dimensional velocity field and density field of the vortex based on quasi-geostrophic theory, and output vortex feature dataset containing parameters such as vortex intensity, radius, and vertical velocity; The mapping relationship between traditional knowledge and vortex parameters is calculated. Canonical correlation analysis is performed between the surface observation features in the traditional knowledge graph and the vortex physical parameters. The semantic-physical mapping matrix is ​​calculated, and the traditional knowledge is converted into a quantitative expression in the vortex parameter space based on the mapping matrix. Based on Bayesian inference and fusion of multi-source prediction information, the mapping results, fuzzy membership degree, environmental suitability index and transport flux distribution are input into the Bayesian inference network, and the posterior probability distribution is calculated through conditional probability update. By integrating prediction results through ensemble learning and combining multi-source prediction results using random forest, traditional knowledge prediction rules and physical model calculation results are used as input features to output a probability distribution map of mid-water fish habitats.

[0005] Furthermore, the construction of traditional knowledge graphs adopts the entity-relation-attribute triple representation method. The graph nodes include environmental observation entities and catch result entities, and the edges represent the causal relationship between observation features and catch results. Environmental feature words and catch quantity values ​​are extracted through named entity recognition, the grammatical relationship between words is identified using dependency parsing, and the mapping relationship between observation features and catch results is established through relation extraction algorithm. The confidence weight is calculated by the frequency and consistency of fishermen's descriptions.

[0006] Furthermore, the extraction of environmental cue vocabulary is achieved through word frequency statistics and semantic clustering algorithms. The semantic clustering algorithm adopts the K-means clustering method, and the input is a set of fishermen's descriptive text vectors processed by word embedding. The word embedding process adopts a pre-trained Word2Vec model. For the vectorization of text segments, the weighted average method of word vectors is adopted, and the weights are weighted by word frequency-inverse document frequency values ​​to highlight the importance of keywords.

[0007] Furthermore, the automatic vortex identification algorithm includes the following sub-steps: Spatial filtering is performed on the input sea surface height anomaly data field to remove noise less than twice the grid resolution; Calculate the sea surface height gradient for each grid point and identify local extrema points with gradients less than a threshold as potential vortex centers; Starting from each potential center, search for closed contour lines radially and calculate the area and equivalent radius within the contour lines; Choose a closed contour line that satisfies both radius and shape parameter constraints as the vortex boundary; Output the coordinates of the vortex center, the vortex radius, and the vortex intensity.

[0008] Furthermore, it includes the step of quantifying the uncertainty of traditional knowledge based on fuzzy logic: converting the qualitative description in traditional knowledge rules into membership functions of vortex features, defining a corresponding fuzzy set for each observed feature, quantifying the degree to which the feature value belongs to the fuzzy set through triangular or trapezoidal membership functions, and generating a knowledge representation that includes the range of uncertainty.

[0009] Furthermore, the method includes the following steps for obtaining Argo buoy data to analyze the environmental suitability of the mid-layer water: obtaining vertical profile data of temperature, salinity, and dissolved oxygen from the Argo buoy database, interpolating the profile data into a cylindrical coordinate system with the vortex center as the origin, assessing the environmental suitability index of each depth layer based on the physiological requirement threshold of fish, and calculating the environmental suitability index by multiplying the temperature suitability index, salinity suitability index, and dissolved oxygen suitability index.

[0010] Furthermore, the method includes the following steps: using a particle tracking algorithm to simulate the vertical material transport process: based on the three-dimensional velocity field of the vortex, the Lagrange particle tracking method is used to simulate the transport path of nutrients from the sea surface to the middle layer, the equation of motion is solved by numerical integration, the fourth-order Runge-Kutta method is used to calculate the particle position update, the time distribution of particles from the surface to different depths is calculated, and a spatiotemporal distribution map of vertical transport flux is generated.

[0011] Furthermore, the input to the canonical correlation analysis method consists of two sets of variable matrices: the traditional observation feature matrix and the vortex physical parameter matrix. Due to the difference in dimensions, the two sets of matrices need to be preprocessed by Z-score standardization before calculating the correlation. The mapping relationship is obtained by solving the eigenvalue problem, and the output is the canonical correlation coefficient and the corresponding canonical variable coefficient matrix.

[0012] Furthermore, the random forest consists of multiple decision trees, each constructed through recursive splitting nodes. The splitting criterion is Gini impurity. The input is the fused feature vector, which contains multi-dimensional features such as traditional knowledge mapping values, environmental suitability index, and transport flux. Before inputting the feature vector into the random forest, it needs to be standardized and preprocessed. The output is the probability of habitat existence.

[0013] This invention provides a habitat suitability prediction system for marine ecological management, comprising: The traditional knowledge acquisition module is used to acquire fishermen's traditional knowledge data and generate structured knowledge representations; The marine data processing module is used to acquire and process satellite vortex data and marine environmental data; The semantic mapping module is used to establish the mapping relationship between traditional knowledge and physical parameters; The multi-source fusion module is used to fuse multiple predictive information sources; The prediction output module is used to generate habitat distribution prediction results. The beneficial effects of this invention are as follows: This invention establishes a quantitative connection between fishermen's traditional observation characteristics and eddy physics parameters by calculating a semantic-physical mapping matrix, overcoming the semantic gap between traditional knowledge and quantitative scientific models. It also transforms qualitative descriptions into quantitative expressions with uncertain ranges through fuzzy logic methods, overcoming the application difficulties caused by the fuzziness of traditional knowledge. Furthermore, by mapping traditional knowledge to the eddy parameter space and combining it with the vertical transport calculation of quasi-geostrophic dynamics, it overcomes the limitation of lacking depth information in surface observations, thus solving the technical problem of the difficulty in effectively integrating traditional knowledge and eddy physics models for the prediction of mid-water fish habitats. Attached Figure Description

[0014] Figure 1 This is a flowchart of a habitat suitability prediction method for marine ecological management according to the present invention; Figure 2 This invention relates to a traditional knowledge graph relationship network diagram; Figure 3 This is a scatter plot of the vortex physical parameter distribution of the present invention; Figure 4 This is a heatmap of the semantic-physical mapping relationship of the present invention; Figure 5 This is a line graph showing the environmental suitability depth profile of the present invention; Figure 6 This is the Sankey diagram of the vertical transport process of particles in this invention. Detailed Implementation

[0015] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0016] At least one embodiment of the present invention discloses a habitat suitability prediction method for marine ecological management, such as Figure 1 As shown, it includes the following steps: Step 1: Acquire traditional knowledge data of fishermen and generate structured knowledge representations; Oral information about sea surface observation rules and corresponding catch records was collected from fishermen. The audio data was transcribed into text format, and the qualitative ratings (such as "high yield", "medium yield", "low yield") in the fishermen's descriptions were numerically encoded to convert the categorical data into numerical data: high yield = 3, medium yield = 2, low yield = 1. Environmental cue words and spatial orientation descriptions were extracted using a Long Short-Term Memory (LSTM) network to generate a traditional knowledge graph containing the correspondence between observation features and catch results.

[0017] Furthermore, the construction of traditional knowledge graphs adopts the entity-relation-attribute triple representation method. Graph nodes include environmental observation entities (such as "dark water color" and "seabirds gathering") and catch result entities (such as "high yield" and "low yield"), with edges representing the causal relationship between observation features and catch results. The specific construction process is as follows: first, environmental feature words and catch quantity values ​​are extracted through named entity recognition; then, dependency parsing is used to identify the grammatical relationships between words; finally, a relation extraction algorithm is used to establish the mapping relationship between observation features and catch results. The confidence weight is calculated by the frequency and consistency of the fishermen's descriptions.

[0018] Furthermore, the specific implementation method of named entity recognition adopts a sequence labeling model based on conditional random fields (CRF), with the input being the segmented text sequence. The output is the corresponding label sequence. The tags adopt the BIO annotation system: B-ENV represents the beginning of an environmental feature entity, I-ENV represents the interior of an environmental feature entity, B-CATCH represents the beginning of a catch result entity, I-CATCH represents the interior of a catch result entity, and O represents a non-entity word; the CRF model maximizes the conditional probability. The training process involves feature functions including lexical features, part-of-speech features, and contextual features.

[0019] Furthermore, the specific implementation method of dependency parsing adopts a graph-based dependency parsing algorithm, with the input being a sequence of words. The output is a dependency tree. ,in For a set of word nodes, It is a set of dependency arcs; the algorithm calculates the dependency relationship score matrix between words. ,in Words As a word The scores of the governing words are calculated, and then the maximum spanning tree algorithm (such as Edmonds' algorithm) is used to find the optimal dependency tree, ensuring that each word (except the root node) has one and only one governing word.

[0020] Furthermore, the specific implementation method of the relation extraction algorithm adopts a rule-based approach based on pattern matching combined with a machine learning classifier. First, a set of language patterns representing causal relationships is defined. The algorithm identifies causal conjunction patterns such as "cause," "lead," and "cause," and then searches the dependency syntax tree for paths that match these patterns. For matched entity pairs... The Support Vector Machine (SVM) classifier is used to determine whether a causal relationship exists. The input features of the classifier include the syntactic distance between entities, dependency path type, and contextual lexical features.

[0021] Furthermore, the specific calculation method for confidence weights is as follows:

[0022] in The proportion of relationship frequency. This represents the number of times the relationship appears in all the fishermen's descriptions. This represents the total number of relationships. The consistency coefficient, and These represent the mean and standard deviation of different fishermen's ratings of the same relationship, respectively; weighting coefficients. , This requires comprehensive consideration of both frequency and consistency.

[0023] The aforementioned semantic clustering algorithm employs the K-means clustering method, and its input is a set of fishermen's descriptive text vectors processed by word embedding. ,in For the first Vector representation of a text segment; Furthermore, word embedding processing employs a pre-trained Word2Vec model, with the input being a segmented sequence of text describing fishermen, and the output being... The word vector representation is 3D; for the vectorization of text segments, a weighted average method of word vectors is used: ,in For the first The number of words in a text segment. For the first The TF-IDF weights of each word , For the corresponding word vectors, the weights are weighted using TF-IDF values ​​to highlight the importance of the keywords.

[0024] Before clustering, the word embedding vectors are preprocessed using L2 normalization, i.e. To eliminate the impact of vector length differences on clustering results; the output is Cluster centers and the category label for each text segment; the algorithm iteratively optimizes the objective function. To determine the optimal clustering result.

[0025] Furthermore, the convergence condition of the K-means algorithm is set as follows: when the distance the cluster centers move in two consecutive iterations satisfies... The algorithm converges in time, where The convergence threshold, Set the number of iterations; also set the maximum number of iterations. To avoid infinite loops, the cluster center update formula is: ,in For the first In the nth iteration, it is assigned to the ... A set of samples for each cluster.

[0026] Furthermore, cluster numbers The contour coefficient method was used to determine the value. Calculations are different Profile coefficient at value ,in For the sample The average distance to other samples of the same type For the sample The average distance to the nearest outlier sample; select the sample that maximizes the silhouette coefficient. The value is used as the optimal cluster number, and its range is constrained. Ensure that each cluster contains at least 5 samples.

[0027] Furthermore, the profile coefficient The constraint range is ,in A value close to 1 indicates good clustering performance (samples are close to their own kind and large to their dissimilar kind). A value close to 0 indicates that the cluster boundaries are ambiguous (the sample is near the cluster boundary). A value close to -1 indicates poor clustering performance (samples may be assigned to incorrect clusters), and is typically chosen as a low-value cluster. of The value is used as the number of effective clusters.

[0028] The aforementioned LSTM network is used for text sequence feature extraction, and its input layer receives a sequence of word embedding vectors. ,in For word embedding dimension, For each time step in the sequence, the output layer outputs the hidden state at each time step. ,in The hidden layer dimension is used; supervised learning mode is used during training, employing the cross-entropy loss function from the sequence labeling task.

[0029] in For the sample size, For the number of categories, For real labels, To predict probabilities, the Adam optimizer is used as the optimization strategy, with a learning rate set to 0.001. During training, a linear mapping layer is used to map the hidden states of the LSTM to the label space, i.e. ,in This is the weight matrix. This is the bias vector; the softmax probability distribution output by the LSTM network needs to be decoded into specific feature labels using the argmax function, i.e. Then, the numerical labels are converted into corresponding environmental feature words through a predefined label mapping table (such as label 1 corresponding to "dark water color", label 2 corresponding to "dense foam zone", etc.).

[0030] Step 2: Acquire satellite vortex observation data and calculate vortex physical parameters; Sea surface height anomaly data are obtained from satellite altimeters. The vortex boundary and center position are detected using an automatic vortex identification algorithm. The three-dimensional velocity field and density field of the vortex are calculated based on the quasi-geostrophic theory. The output is a vortex feature dataset containing parameters such as vortex intensity, radius, and vertical velocity.

[0031] Furthermore, the three-dimensional velocity field calculation method based on quasi-geostrophic theory is as follows: the horizontal geostrophic velocity component is obtained through the sea surface height anomaly gradient calculation. ,

[0032] in This represents an anomaly in sea level height (unit: m). It is the acceleration due to gravity. For Coriolis parameters, The angular velocity of Earth's rotation. Latitude; Vertical velocity Calculated using the quasi-geostrophic vorticity equation: ,in Stream function (unit: Solving the equation of the ellipse get, Geostrophic vorticity (unit: ).

[0033] Furthermore, the time partial derivative term The calculation uses the finite difference method, and the time window is set to... sky, The following results were obtained from the stream function field calculations at two consecutive time points: ,in At the current moment, the sampling interval for the time series data is on a daily scale to ensure the capture of the quasi-steady-state characteristics of mesoscale eddies; the range of vertical velocity values ​​is constrained as follows. Outliers outside this range are processed using median filtering.

[0034] The aforementioned automatic vortex identification algorithm includes the following sub-steps: Sub-step 2.1: Process the input sea surface height anomaly data field Perform spatial filtering to remove elements smaller than 10 ... The noise, of which For grid resolution; Furthermore, a Gaussian low-pass filter is used for noise determination, which is suitable for spatial scales smaller than [specific size]. High-frequency signals are suppressed, and the filter's cutoff frequency is set to... The amplitude of the signal retained after filtering is at least half of the original signal's amplitude. This ensures that effective signals are not excessively suppressed.

[0035] Furthermore, the specific implementation method of the Gaussian low-pass filter is as follows: Filter kernel function:

[0036] Standard deviation The filter window size is set to .

[0037] Filtering operations are implemented through convolution operations:

[0038] in Let be the window radius.

[0039] Sub-step 2.2: Calculate the sea surface height gradient at each grid point. Identify gradients less than a threshold The local extreme points are taken as potential vortex centers; Furthermore, gradient threshold The value of is determined using statistical methods, and the gradient field of the entire sea area is calculated. Standard deviation ,set up Gradient thresholding can effectively identify local extrema where the gradient is close to zero, while avoiding misidentifying noise points as vortex centers.

[0040] Furthermore, gradient threshold The constraint function is in the form of: The lower bound open interval ensures the threshold is positive, and the upper bound closed interval restricts the threshold from exceeding the standard deviation of the gradient field. When the value is too small, too many noise points will be mistaken for the vortex center. When the value is too large, the true vortex center will be missed, so set it to... As the empirically optimal value.

[0041] Sub-step 2.3: Starting from each potential center, search for closed contour lines radially and calculate the area within the contour lines. and equivalent radius ; Sub-step 2.4: Select the one that satisfies And shape parameters The closed contour lines serve as the vortex boundary, where The perimeter of the contour lines; Furthermore, shape parameters The constraint function is in the form of: The lower boundary Corresponding to a perfect circle ( ), Upper Realm Ensure that the contour lines are not excessively distorted, when A value close to 1 indicates that the vortex shape is close to circular. A value close to 2 indicates that the vortex shape is somewhat distorted but still acceptable. Contours exceeding 2 are judged to be non-vortex structures and are discarded.

[0042] Furthermore, the determination of the vortex radius constraint parameters is based on the deformation radius of the geostrophic theory, and the calculation... ,in It is the acceleration due to gravity. For Coriolis parameters, To set the vortex intensity; and The vortex radius constraint parameter range can effectively filter out excessively small noise structures and excessively large non-vortex structures.

[0043] Furthermore, the deformation radius The specific calculation method is as follows: First, calculate the Coriolis parameter. ,in The angular velocity of Earth's rotation. To study the average latitude of the sea area (in radians); then calculate the eddy intensity. absolute value Finally, through the formula Calculate the deformation radius, where The calculation results are in meters, which need to be converted to kilometers for radius constraints.

[0044] Sub-step 2.5: Output the coordinates of the vortex center vortex radius and vortex intensity, ,in The average sea level height at the boundary.

[0045] Furthermore, vortex intensity The unit is meters (m), representing the difference in sea surface height between the vortex center and its boundary, with a range of values ​​constrained to [value missing]. Positive values ​​represent anticyclonic vortices (warm vortices), and negative values ​​represent cyclonic vortices (cold vortices); vortex radius The unit is kilometers (km), and the range of values ​​is... ;when or At that time, the vortex structure was judged as noise rather than a real vortex and was discarded.

[0046] Step 3: Calculate the mapping relationship between traditional knowledge and eddy parameters to generate habitat prediction results; Correlation analysis is performed on the surface observation features in the traditional knowledge graph obtained in step 1 and the vortex physical parameters obtained in step 2 to calculate the semantic-physical mapping matrix. ,in Indicates the first The first traditional observation feature and the second The correlation coefficients between the physical parameters of each vortex are used to transform traditional knowledge into a quantitative expression of the vortex parameter space based on the mapping matrix. Combined with the vertical transport characteristics of the vortex, a probability distribution map of the mid-water fish habitat is output.

[0047] The aforementioned canonical correlation analysis method takes two sets of variable matrices as input: a traditional observation feature matrix and a standard observation feature matrix. and vortex physical parameter matrix ,in For the sample size, To observe the feature dimensions, For the physical parameter dimension; due to the significant dimensional and numerical range differences between traditional observation features (qualitative scoring, dimensionless values ​​1-3) and vortex physical parameters (intensity in meters, radius in kilometers, velocity in meters per second, etc.), directly calculating the correlation will lead to a dimensional mismatch problem. Therefore, Z-score standardization preprocessing must be performed on both sets of matrices before calculating the correlation: , ,in and These are the mean and standard deviation vectors, respectively. The standardized data are dimensionless with a mean of 0 and a standard deviation of 1, eliminating the influence of dimensional differences on correlation calculations. The output is the canonical correlation coefficient. and the corresponding canonical variable coefficient matrix , ,in ; By solving the eigenvalue problem Obtain the mapping relationship, where Let be the covariance matrix.

[0048] Furthermore, the method for calculating the covariance matrix is ​​as follows:

[0049]

[0050]

[0051]

[0052] Before solving the eigenvalue problem, the covariance matrix needs to be regularized by adding a small number of regularization terms. ( To avoid numerical instability.

[0053] Step 4: Quantify the range of uncertainty of traditional knowledge based on fuzzy logic; The qualitative descriptions in traditional knowledge rules are transformed into membership functions of vortex features. For each observed feature, a corresponding fuzzy set is defined, and triangular or trapezoidal membership functions are used. Quantization eigenvalues The degree to which it belongs to a fuzzy set, where This generates a knowledge representation that includes a range of uncertainties.

[0054] The aforementioned membership function Using the triangular membership function form: ,in , , The left, peak, and right boundary parameters of the triangular membership function are determined based on the eigenvalue range described by the fishermen; for the trapezoidal membership function form: ,in , , , For the four vertices of the trapezoid, when hour ,when or hour .

[0055] Furthermore, the method for determining the membership function parameters is as follows: First, perform statistical analysis on the qualitative scores in the fishermen's descriptions, and calculate the log-normal distribution parameters corresponding to each feature. and For the membership function of a triangle, set... , , For trapezoidal membership functions, set , , , The membership function parameter setting method can ensure that 95% of the observations fall within the support interval of the membership function.

[0056] It should be noted that the input vortex eigenvalues Before applying membership functions, min-max normalization preprocessing is required to scale physical parameters of different dimensions uniformly to a uniform scale. Interval: ,in and These are the minimum and maximum values ​​of the vortex characteristic, respectively, to ensure the effectiveness of the membership function calculation.

[0057] Step 5: Obtain Argo buoy data and analyze the suitability of the mid-water environment; Vertical profile data of temperature, salinity, and dissolved oxygen were obtained from the Argo buoy database. The profile data were interpolated into a cylindrical coordinate system with the vortex center as the origin to calculate the distribution of environmental parameters in the depth range of 200-1000 meters. The environmental suitability index of each depth layer was evaluated based on the physiological requirement threshold of fish.

[0058] Furthermore, the cylindrical coordinate interpolation method employs an inverse distance weighted interpolation algorithm for the vortex center coordinates. The environmental parameter values ​​at that location are calculated using the following formula: , in For the first Environmental parameter observations for each Argo site As weight, For the target point to the 1st The three-dimensional Euclidean distance between the stations, where Radial distance, It is the azimuth angle. For depth, To prevent division by zero errors in the smoothing term, the interpolation radius was set to 50km, ensuring that at least three observation stations participated in the interpolation for each grid point.

[0059] Furthermore, azimuth angle Constraint range Indicates from due east ( Begin by measuring counter-clockwise. radians (excluding) ), covering the entire 360-degree range, when When the value exceeds this range, a periodic constraint is applied using modulo operations: This ensures that the azimuth angle is always within the effective range.

[0060] Furthermore, the environmental suitability index is calculated based on the physiological requirement thresholds of mid-water fish, and the temperature suitability index...

[0061] in °C is the optimal temperature. °C represents the temperature tolerance. Salinity suitability index:

[0062] in psu, psu; Dissolved oxygen suitability index ,in mg / L, mg / L is the minimum requirement threshold; comprehensive suitability index .

[0063] It should be noted that due to the significant differences in the dimensions and numerical ranges of environmental parameters such as temperature (unit: °C, numerical range: approximately 2-8 °C), salinity (unit: psu, numerical range: approximately 33-36 psu), and dissolved oxygen (unit: mg / L, numerical range: approximately 2-8 mg / L), directly performing multiplication operations to calculate the comprehensive suitability index is not feasible. This can lead to a mismatch in dimensions. Therefore, Z-score standardization preprocessing is required for each parameter before calculating the environmental suitability index. , , ,in and These are the mean and standard deviation of each parameter, respectively. The standardized parameters are all dimensionless, eliminating the impact of dimensional differences on suitability assessment.

[0064] Step 6: Simulate the vertical mass transport process using the particle tracking algorithm; Based on the vortex three-dimensional velocity field calculated in step 2, the Lagrange particle tracking method was used to simulate the transport path of nutrients from the sea surface to the middle layer, and the time distribution of particles reaching different depths from the surface was calculated. ,in To the depth, a spatiotemporal distribution map of vertical transport flux is generated.

[0065] The input to the aforementioned Lagrange particle tracking method is a three-dimensional velocity field. and initial particle position set ,in The output is the particle count; the output is the particle trajectory at each time step. and time to reach depth The algorithm solves the equations of motion through numerical integration. The fourth-order Runge-Kutta method is used to calculate particle position updates: ,in This is an estimate of the intermediate speed.

[0066] Furthermore, the intermediate velocity estimate in the fourth-order Runge-Kutta method is calculated as follows:

[0067]

[0068]

[0069]

[0070] Time step The selection must meet the CFL condition, that is... To ensure numerical stability.

[0071] Furthermore, the constraint function form of the CFL condition is as follows: ,in , For CFL numbers, usually taken as To ensure numerical stability, when Exceeding this upper bound will lead to numerical instability and calculation divergence.

[0072] The aforementioned time distribution function Defined as: ,in To reach depth The total number of particles, For the first A particle from the sea surface Depth of reach Time, by judging conditions To determine whether the particle has reached the target depth, This is the depth tolerance threshold.

[0073] Furthermore, the time integration interval is set as follows: Heavens, particle tracking from Start to End, time step is Hours, total number of steps calculated step; For particles that do not reach the target depth within the time window, set As a truncation treatment; The effective range constraint of arrival time is Average transport time The unit is days, and the range of values ​​is [missing information]. sky.

[0074] Furthermore, depth tolerance threshold The value is determined based on the vertical grid resolution, and is set to... ,in The vertical grid spacing and depth tolerance threshold settings ensure that the accuracy of particle arrival judgment matches the vertical resolution of the model, avoiding particle omissions due to too small a tolerance or inaccurate depth judgment due to too large a tolerance.

[0075] Step 7: Fuse multi-source prediction information based on Bayesian inference; The mapping results obtained in step 3, the fuzzy membership degrees obtained in step 4, the environmental suitability index obtained in step 5, and the transport flux distribution obtained in step 6 are input into the Bayesian inference network, and the posterior probability distribution is calculated through conditional probability updates. ,in This indicates the assumption of habitat existence. Indicates the first Based on similar evidence, output the integrated habitat probability assessment results.

[0076] The aforementioned Bayesian inference network consists of nodes and directed edges, with nodes including evidence nodes. and hypothetical nodes Directed edges represent conditional dependencies; the input to the evidence nodes are the feature values ​​of each prediction source. ,in For the first The dimensions of evidence; due to the differences in dimensions and numerical ranges of the features of various prediction sources (mapping correlation coefficient, fuzzy membership degree, environmental suitability index, transport flux), all features need to be min-max normalized before being input into the Bayesian network, and uniformly scaled to a certain value. Intervals; the network calculates the posterior probability using Bayes' theorem: The training method employs maximum likelihood estimation, which estimates the conditional probability table using historical observation data. and prior probability The loss function is the negative log-likelihood. ,in The number of training samples, indicated by the superscript. Indicates the first One sample; the optimization strategy uses the expectation-maximization (EM) algorithm to iteratively update the network parameters until convergence.

[0077] Furthermore, the specific implementation method of maximum likelihood estimation is as follows: For discrete evidence variables conditional probability Calculated through frequency statistics: ,in In the training data and The number of samples, for The total number of samples; for continuous evidence variables, assume they follow a Gaussian distribution. The parameters are estimated using the sample mean and variance: , .

[0078] Furthermore, the convergence criterion for the EM algorithm is set as follows: when the difference between the log-likelihood function values ​​of two consecutive iterations satisfies... The algorithm converges in time, where The convergence threshold, Set the number of EM iterations; also set the maximum number of iterations. To avoid an infinite loop; The E-step estimates the posterior distribution of the latent variables using the current parameters, and the M-step updates the parameters by maximizing the expected log-likelihood. ,in For observation data, As a latent variable, These are network parameters.

[0079] It should be noted that the posterior probability output by the Bayesian inference network... For continuous numerical values ​​(range [0,1]), a probability threshold is needed to determine whether the decoding process converts them into discrete habitat prediction results. when If the habitat is deemed suitable (label=1), it is deemed unsuitable (label=0). The threshold is used for determination; at the same time, the original probability value is retained as a quantitative assessment of habitat suitability, providing continuous input features for subsequent ensemble learning.

[0080] Furthermore, determine the threshold. The selection is based on the principle of maximizing classification accuracy. The decision threshold minimizes the overall error rate of Type I error (misclassifying suitable habitats as unsuitable) and Type II error (misclassifying unsuitable habitats as suitable), which is in line with the minimization of total risk criterion in Bayesian decision theory. When prior probability Furthermore, when the costs of the two types of errors are equal, 0.5 is the theoretically optimal threshold.

[0081] Step 8: Integrate the prediction results using ensemble learning and output the final habitat distribution map; Random forest is used to integrate multi-source prediction results. Traditional knowledge prediction rules and physical model calculation results are used as input features. The results of each prediction source are integrated through weighted voting to output a probability distribution map of mid-water fish habitat in the depth range of 200-1000 meters, with probability values ​​ranging from [0,1].

[0082] The aforementioned random forest consists of multiple decision trees, each decision tree The splitting criterion is Gini impurity, constructed recursively through splitting nodes. ,in Category in the node The sample ratio; the input is the fused feature vector. It includes traditional knowledge mapping values ​​(dimensionless correlation coefficient), environmental suitability index (dimensionless probability value), and transport flux (unit). )wait For multidimensional features, since each feature comes from a different source and has different dimensions (correlation coefficients and probability values ​​are dimensionless, while transport flux has dimensions), the feature vectors need to be preprocessed using Z-score standardization before being input into the random forest. Eliminate the impact of differences in dimensions and numerical ranges on decision tree splitting; output is the probability of habitat presence. ,in The number of decision trees; the training method uses bootstrap sampling, with each tree using randomly selected... One sample and Training is performed using 10 features; the loss function is the logarithmic loss for the classification task. ,in The labels are real; the optimization strategy is a greedy algorithm that selects the optimal splitting feature and threshold at each node.

[0083] Furthermore, the specific implementation method of bootstrap sampling is as follows: For the original training set Each decision tree training set Generated through random sampling with replacement, the sampling process is as follows: From Random selection Generate samples (repeated samples allowed) ,in For random indexes; and simultaneously, when splitting at each node, from Randomly selected from the features Each feature is used as a candidate splitting feature to ensure the diversity of the decision tree.

[0084] The aforementioned decision tree function Defined as: starting from the root node, for the input feature vector Perform layer-by-layer judgment at each internal node. According to the conditions of division Decide whether to move to the left or right subtree, where This is the index for the splitting features of internal nodes. This is the splitting threshold; when a leaf node is reached... At that time, output the predicted value of the leaf node. ,in and These represent the number of positive and negative samples in the leaf nodes, respectively.

[0085] It should be noted that the habitat existence probability output by Random Forest is... It needs to be converted into a complete habitat distribution map using spatial interpolation methods: for each spatial grid point within the study area ,in Indicates the first At each layer depth, the habitat probability is calculated based on the feature vector of that point. Then, a continuous three-dimensional habitat probability distribution field is generated using bilinear interpolation. The final output is a habitat suitability distribution map for each layer within a depth range of 200-1000 meters.

[0086] Furthermore, the specific implementation of the bilinear interpolation method is as follows: for the target point First, determine its position in the grid, then find the four neighboring grid points surrounding that point. , , , and their corresponding probability values , , , ; Calculate the interpolation weights: , The probability value of the target point is calculated using the bilinear interpolation formula: The bilinear interpolation method ensures that the generated probability distribution field is spatially continuous and smooth.

[0087] Furthermore, the constraint range for the bilinear interpolation weights is: When the target point When located inside a grid cell, the weights satisfy the normalization condition: This ensures that the probability values ​​of the interpolation results remain within the valid range. Within the grid, the weight is 1 when the target point is exactly located on a grid node, and 0 for all other weights.

[0088] This implementation example focuses on the western Pacific Kuroshio Current extension (35°N–40°N, 140°E–150°E) to predict habitat suitability for mid-water tuna (dwelling depths of 200–1000 meters). This area is characterized by frequent eddy activity and is an important fishing ground, with local fishermen possessing extensive traditional observation experience. The study took place in April 2023, and data sources included oral records from local fishermen, Jason-3 satellite altimeter data, and Argo buoy network observations.

[0089] Step 1: Results of traditional knowledge structuring processing; The text describing the fishermen's experiences was processed using an LSTM network to extract standardized environmental cue words. A traditional knowledge graph was then constructed using K-means clustering analysis, employing entity-relationship-attribute triples. This indicates the correspondence between observed characteristics and fish catch results.

[0090] Figure 2It showcases the results of the structured processing of fishermen's traditional knowledge and visualizes the causal relationship network between characteristic entities and catch result entities.

[0091] Step 2: Calculate the vortex physical parameters; Based on satellite altimeter data, a three-dimensional velocity field is calculated using an automatic vortex identification algorithm and quasi-geostrophic theory.

[0092] Figure 3 This demonstrates the distribution characteristics of vortex physical parameters observed by the satellite in step 2.

[0093] Step 3: Calculate the mapping relationship between traditional knowledge and eddy parameters to generate habitat prediction results; Canonical correlation analysis was used to calculate the mapping matrix between traditional observation features and vortex physical parameters. And based on the mapping relationship, the probability distribution of mid-water fish habitats is generated.

[0094] Figure 4 The mapping matrix M between traditional observation features and vortex physical parameters is shown, with the color intensity indicating the strength of the correlation coefficient.

[0095] Step 4: Quantify the uncertainty of traditional knowledge based on fuzzy logic; Using triangle membership functions Quantify the uncertainty of traditional knowledge.

[0096] Step 5: Obtain Argo buoy data and analyze the suitability of the mid-water environment; Environmental suitability indices for each depth layer were calculated based on Argo buoy data. The temperature suitability index was calculated using... calculate.

[0097] Figure 5 This study demonstrates the changes in environmental suitability indices at different depths, including three components: temperature, salinity, and dissolved oxygen, as well as a comprehensive suitability index. It reveals the distribution patterns of environmental preferences among mid-water fish species at depths ranging from 200 to 1000 meters.

[0098] Step 6: Simulate the vertical mass transport process using the particle tracking algorithm; Based on the vortex three-dimensional velocity field calculated in step 2, the Lagrange particle tracking method was used to simulate the transport path of nutrients from the sea surface to the middle layer, and the time distribution of particles reaching different depths was calculated. .

[0099] Figure 6 This study demonstrates the transport flow direction and flux distribution of Lagrange particles from the sea surface to different depths, with the width of the flow band representing the magnitude of the transport flux. It also reveals the hierarchical characteristics of eddy vertical transport and the efficiency of nutrient transfer.

[0100] Step 7: Fuse multi-source prediction information based on Bayesian inference; The mapping results, fuzzy membership degrees, environmental suitability index, and transport flux distribution are input into the Bayesian inference network, and the posterior probability distribution is calculated through conditional probability updates. .

[0101] Step 8: Integrate the prediction results using ensemble learning; The random forest is used to integrate the multi-source prediction results and output the final habitat probability distribution map.

[0102] This implementation example successfully transforms fishermen's traditional observational experience into a quantitative knowledge graph triple representation: The causal relationship between "abnormal water color changes" and "high yield" reached a confidence level of 0.85, while the confidence level for "warmer water temperature" reached 0.82. A semantic-physical mapping matrix was established through canonical correlation analysis. Correlation coefficient Between 0.66 and 0.83, the correlation coefficient between "warmer water temperature phenomenon" and vortex parameters was the highest (0.83).

[0103] The uncertainty quantification results of fuzzy logic processing show that the membership degree of traditional knowledge The membership values ​​ranged from 0.67 to 0.92, with the "warmer water temperature phenomenon" exhibiting the highest membership (0.92), effectively quantifying the reliability of empirical knowledge. Particle tracking results indicated that the average time for nutrients to reach a depth of 400 meters from the sea surface was 7.8 days, with a transport flux of [missing information]. The transport efficiency reached 0.85, verifying the effectiveness of vortex vertical transport.

[0104] After Bayesian inference fuses information from multiple sources, the posterior probability... Between 0.59 and 0.79, the probability of habitat presence was highest at the location (36.8°N, 147.3°E) (0.79). The final ensemble learning results showed that highly suitable habitats accounted for 23.5% of the studied sea area, mainly distributed at a depth of 400 meters, which highly coincided with the downflow region of the anticyclonic vortex, verifying the intrinsic connection between the traditional observational pattern of "fish gathering in areas with dark water color" and the vortex's physical mechanism.

[0105] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A habitat suitability prediction method for marine ecological management, characterized in that, Includes the following steps: We acquire traditional knowledge data of fishermen, transcribe audio data into text format using natural language processing technology, extract environmental cue vocabulary and spatial location descriptions using long short-term memory networks, and generate a traditional knowledge graph containing the correspondence between observation features and fishing results. Acquire satellite vortex observation data, use vortex automatic identification algorithm to detect vortex boundary and center position, calculate the three-dimensional velocity field and density field of the vortex based on quasi-geostrophic theory, and output vortex feature dataset containing parameters such as vortex intensity, radius, and vertical velocity; The mapping relationship between traditional knowledge and vortex parameters is calculated. Canonical correlation analysis is performed between the surface observation features in the traditional knowledge graph and the vortex physical parameters. The semantic-physical mapping matrix is ​​calculated, and the traditional knowledge is converted into a quantitative expression in the vortex parameter space based on the mapping matrix. Based on Bayesian inference and fusion of multi-source prediction information, the mapping results, fuzzy membership degree, environmental suitability index and transport flux distribution are input into the Bayesian inference network, and the posterior probability distribution is calculated through conditional probability update. By integrating prediction results through ensemble learning and combining multi-source prediction results using random forest, traditional knowledge prediction rules and physical model calculation results are used as input features to output a probability distribution map of mid-water fish habitats.

2. The method according to claim 1, characterized in that, The traditional knowledge graph is constructed using an entity-relation-attribute triple representation. Graph nodes include environmental observation entities and catch result entities, and edges represent the causal relationship between observation features and catch results. Environmental feature words and catch quantity values ​​are extracted through named entity recognition. Dependency parsing is used to identify the grammatical relationships between words. A relation extraction algorithm is used to establish the mapping relationship between observation features and catch results. Confidence weights are calculated based on the frequency and consistency of fishermen's descriptions.

3. The method according to claim 1, characterized in that, The extraction of environmental cue vocabulary is achieved through word frequency statistics and semantic clustering algorithms. The semantic clustering algorithm adopts the K-means clustering method. The input is a set of fishermen's descriptive text vectors after word embedding. The word embedding process adopts a pre-trained Word2Vec model. For the vectorization of text segments, the weighted average method of word vectors is adopted. The weights are weighted by word frequency-inverse document frequency values ​​to highlight the importance of keywords.

4. The method according to claim 1, characterized in that, The automatic vortex identification algorithm includes the following sub-steps: Spatial filtering is performed on the input sea surface height anomaly data field to remove noise less than twice the grid resolution; Calculate the sea surface height gradient for each grid point and identify local extrema points with gradients less than a threshold as potential vortex centers; Starting from each potential center, search for closed contour lines radially and calculate the area and equivalent radius within the contour lines; Choose a closed contour line that satisfies both radius and shape parameter constraints as the vortex boundary; Output the coordinates of the vortex center, the vortex radius, and the vortex intensity.

5. The method according to claim 1, characterized in that, It also includes the step of quantifying the uncertainty of traditional knowledge based on fuzzy logic: converting the qualitative description in traditional knowledge rules into membership functions of vortex features, defining the corresponding fuzzy set for each observed feature, quantifying the degree to which the feature value belongs to the fuzzy set through triangular or trapezoidal membership functions, and generating a knowledge representation that includes the range of uncertainty.

6. The method according to claim 1, characterized in that, It also includes the steps of obtaining Argo buoy data to analyze the environmental suitability of the middle water layer: obtaining vertical profile data of temperature, salinity, and dissolved oxygen from the Argo buoy database, interpolating the profile data into a cylindrical coordinate system with the vortex center as the origin, evaluating the environmental suitability index of each depth layer based on the physiological demand threshold of fish, and calculating the environmental suitability index by multiplying the temperature suitability index, salinity suitability index, and dissolved oxygen suitability index.

7. The method according to claim 1, characterized in that, It also includes the step of simulating the vertical material transport process using a particle tracking algorithm: based on the three-dimensional velocity field of the vortex, the Lagrange particle tracking method is used to simulate the transport path of nutrients from the sea surface to the middle layer, the motion equations are solved by numerical integration, the fourth-order Runge-Kutta method is used to calculate the particle position update, the time distribution of particles from the surface to different depths is calculated, and a spatiotemporal distribution map of vertical transport flux is generated.

8. The method according to claim 1, characterized in that, The input to the canonical correlation analysis method consists of two sets of variable matrices: the traditional observation feature matrix and the vortex physical parameter matrix. Due to the difference in dimensions, the two sets of matrices need to be preprocessed by Z-score standardization before calculating the correlation. The mapping relationship is obtained by solving the eigenvalue problem, and the output is the canonical correlation coefficient and the corresponding canonical variable coefficient matrix.

9. The method according to claim 1, characterized in that, The random forest consists of multiple decision trees, each constructed through recursive splitting nodes. The splitting criterion is Gini impurity. The input is a fused feature vector containing traditional knowledge mapping values, environmental suitability index, and transport flux. The feature vector needs to be standardized and preprocessed before being input into the random forest. The output is the probability of habitat existence.

10. A habitat suitability prediction system for marine ecological management, used to perform the method according to any one of claims 1-9, characterized in that, include: The traditional knowledge acquisition module is used to acquire fishermen's traditional knowledge data and generate structured knowledge representations; The marine data processing module is used to acquire and process satellite vortex data and marine environmental data; The semantic mapping module is used to establish the mapping relationship between traditional knowledge and physical parameters; The multi-source fusion module is used to fuse multiple predictive information sources; The prediction output module is used to generate habitat distribution prediction results.

Images