A method and system for assessing the catchability of a lake fish resource

Abstract

The present application relates to the technical field of fishery resource assessment, and particularly relates to a lake fish resource catchability assessment method and an assessment system, comprising: step 1, data acquisition and preprocessing; step 2, based on environmental data, obtaining a resource existence sub-index, a spatial contact sub-index and an activity feeding sub-index, and then combining them into an empirical catchability factor; constructing a cross-lake training sample set, training a machine learning model using the cross-lake training sample set, predicting fish resource index values using the trained machine learning model, and converting the predicted fish resource index values into machine learning resource factors; step 3, obtaining a comprehensive catchability index based on the empirical catchability factor, the machine learning resource factor and the extrapolation control quantity, and obtaining a comprehensive resource index based on the resource existence sub-index, the machine learning resource factor and the extrapolation control quantity. The present application is at least beneficial to solve the problems of habitat and catchability confusion, unstable cross-lake prediction and inability to unify mechanism and data model.

Description

Technical Field

[0001] This invention belongs to the field of fishery resource assessment technology, and in particular relates to a method and system for assessing the catchability of lake fish resources. Background Technology

[0002] Lake fish resource assessment typically requires determining the spatial distribution of fish resources in different lake areas, seasons, and time periods. Common methods for lake fish resource assessment include manual or fishing gear sampling surveys, empirical modeling methods based on habitat suitability indices, spatial inversion methods based on remote sensing or water quality data, and machine learning prediction methods based on fish catch data and environmental data.

[0003] Manual or fishing gear sampling surveys can obtain measured data such as fish catch quantity, fish catch weight, and species composition. However, this method depends on the number of sampling sites and the sampling frequency, making it difficult to form resource simulation results for continuous spatial surfaces. For shallow lakes such as Yangcheng Lake, fish resources are affected by water depth, distance from the shore, dissolved oxygen, temperature, turbidity, chlorophyll, season, and time of day. A small number of sampling sites are insufficient to reflect the spatiotemporal differences within the lake area.

[0004] Empirical modeling methods based on habitat suitability indices typically convert environmental variables such as water depth, water temperature, dissolved oxygen, transparency, vegetation, or chlorophyll into suitability indices, which are then synthesized into habitat suitability. These methods have good ecological interpretability, but most models only express whether a location is suitable for the presence of fish, without further distinguishing whether fish resources are in a spatial layer that can be surveyed, caught, or utilized by recreational fisheries, or whether fish are actively active and feeding. Therefore, directly using habitat suitability as catchability or resource utilization intensity can easily lead to biased judgments such as fish being present but untouchable or fish being present but not feeding.

[0005] Machine learning-based fish resource prediction methods can fit the nonlinear relationship between fish abundance, biomass, or species richness and environmental factors using multi-lake fish catch data and environmental data. However, the water depth range, trophic status, water quality scale, sampling depth, sampling season, and missing data structure are often inconsistent between different lakes. If the model trained on other lakes is directly applied to the target lake, it is easy to produce problems such as inconsistent variable scales and extrapolation of training distribution. At the same time, conventional machine learning models usually only output predicted values, lacking constraints on the phased formation mechanism of fish resources and weight control on scenarios outside the training distribution.

[0006] Therefore, existing technologies have at least the following shortcomings: First, empirical habitat models typically do not break down fish catchability into stages such as resource availability, spatial accessibility, and active feeding, leading to inconsistencies between model output and actual catchability or resource utilization status; Second, cross-lake machine learning models lack unified variable transfer processing, missing encoding methods, and extrapolation risk control, resulting in insufficient prediction stability for target lakes; Third, they fail to integrate staged empirical mechanisms with cross-lake machine learning resource factors into a unified computational process, making it difficult to simultaneously consider ecological interpretability, cross-lake migration capability, and the ability to continuously output data from the target lake. Summary of the Invention

[0007] In view of this, the present invention aims to provide a method and system for assessing the catchability of lake fish resources, which at least helps to solve the problems of confusion between habitat and catchability, instability of cross-lake prediction, and the inability to unify and integrate mechanisms and data models.

[0008] To achieve the above objectives, the technical solution created by this invention is implemented as follows: This invention provides a method for assessing the catchability of fish resources in lakes, comprising: Step 1, data acquisition and preprocessing, including: taking the target lake area as the evaluation unit, collecting environmental data corresponding to the evaluation unit, and obtaining multivariate spatial features based on the environmental data; Step 2, constructing staged empirical sub-indices and cross-lake machine learning resource factors, wherein constructing staged empirical sub-indices includes: obtaining resource presence sub-indices based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors The construction of cross-lake machine learning resource factors includes: acquiring fish sampling event data and corresponding environmental data from multiple lake areas; constructing a cross-lake training sample set; training a machine learning model using the cross-lake training sample set; using the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of the evaluation units; and converting the predicted fish resource index values ​​into machine learning resource factors. Step 3, extrapolation control and fusion output, including: obtaining extrapolated control quantities based on environmental data from evaluation units and environmental data corresponding to cross-lake training sample sets. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

[0009] Furthermore, in step 2, which involves constructing a phased experience sub-index, a resource existence sub-index is obtained based on environmental data. include: ; Representing the Each resource has a suitability value for relevant environmental variables. The total number of environmental variables related to the existence of resources. It is a multi-factor composition function; In step 2, which involves constructing phased experience sub-indices, spatially accessible sub-indices are obtained based on environmental data. include: ; Represents the composition function of accessible terms. Indicates the distance from the shore. Represented as the lakebed slope term, Indicates the water depth at which the sample was taken; In step 2, which involves constructing staged experience sub-indices, activity-based feeding sub-indices are obtained based on environmental data. B include: B =WT×WDO×Wlight×Wcycle×Wprey; where WT represents the temperature response, WDO represents the dissolved oxygen response, Wlight represents the light response, Wcycle represents the seasonal or diurnal rhythm, and Wprey represents the food availability.

[0010] Furthermore, resource acquisition has sub-indices. In the process, the resources have relevant environmental variables including water depth, water temperature, dissolved oxygen, turbidity, and chlorophyll proxy variables; A piecewise trapezoidal suitability function is used to obtain the suitability values ​​of resource-existence-related environmental variables based on resource-existence-related environmental variables. The piecewise trapezoidal suitability function is as follows: ; , , , All of these are set thresholds. , This indicates that the lower threshold is completely unsuitable. This represents the lower limit threshold of the optimal range. This represents the upper limit threshold of the optimal range. This indicates that the upper limit threshold is completely unsuitable. hour, For the first Each resource has actual values ​​for relevant environmental variables; It is a weighted geometric mean function. ;in, Representing the Each resource has the weights of relevant environmental variables.

[0011] Furthermore, access to spatially accessible sub-indices In the steps, the distance from the shore is calculated using a left-shoulder-type suitable function. The appropriate function for the left shoulder type is as follows: ,when hour, Represents the distance from the shore. This represents the optimal nearshore threshold. This represents the maximum operational offshore distance; The lakebed slope term is calculated using a piecewise trapezoidal suitable function based on the lakebed slope. ; The sampling water depth term is calculated using a piecewise trapezoidal suitable function based on the sampling water depth. ; Contactless term composition function as follows: A=λ+(1 l)[0.55A d +0.45(oh n A s +(1 oh n ) )] , l This represents the lower limit of the accessibility index. oh n Represents the weight of the lakebed slope term. 1 oh n Represents the weight of the sampled water depth term, 0≤ oh n ≤1.

[0012] Furthermore, obtain the activity ingestion index. In this step, the dissolved oxygen response term is calculated using a piecewise trapezoidal fit function based on dissolved oxygen. Based on seasonal or diurnal rhythms, piecewise trapezoidal appropriate functions are used to calculate seasonal or diurnal rhythm terms. Based on food availability, a piecewise trapezoidal suitability function is used to calculate the food availability term. ; ,in, , , , The solar altitude angle, For normalized turbidity proxy variables in the range of 0 to 1, To normalize the effective light intensity of the water surface, The optimal light level represents the light intensity at which fish are most suitable for activity. The standard deviation represents the range of suitable lighting conditions.

[0013] Furthermore, resources will exist in sub-indices. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors include: .

[0014] Furthermore, in step 2, the step of constructing a cross-lake training sample set: The continuous environmental variables in environmental data are converted into quantile features based on the following formula. , Representing the Quantile characteristics of continuous environmental variables Representing the A continuous environment variable, Representing the Empirical distribution function of continuous environmental variables; Set missing variables in environmental data as missing feature characteristics; The time-varying variables in the environmental data remain unchanged; Machine learning input features are constructed based on quantile features of continuous environmental variables, time-period variables, and missing data markers. ; Continuous environmental variables include: water depth, water temperature, dissolved oxygen, turbidity, chlorophyll proxy variables, pH, and conductivity. , Represents time period variables, Indicates the water depth quantile. Indicates the water temperature quantile. Indicates dissolved oxygen quantiles. Indicates turbidity quantiles. Indicates the quantiles of the chlorophyll proxy variable. Indicates conductivity quantiles, Indicates the acidity / alkalinity quantile. Indicates missing marker features; Fish sampling event data includes fish resource indicators, such as abundance, biomass, and species richness. The target transformation of these fish resource indicators is performed based on the following formula: ,in, As an indicator of fish resources, Represents abundance, Represents biomass, Represents species richness, These are the specific numerical values ​​for fish resource indicators. The target is the fish resource index after transformation.

[0015] Furthermore, in step 2, training the machine learning model using a cross-lake training sample set includes: using fish resource indicators after target transformation. and machine learning input features Train the machine learning model; the machine learning model is a random forest regression model, and the mathematical expression of the random forest regression model is as follows: ,in, Indicates the first A tree of return, This represents the number of regression trees; Step 2, using the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of the evaluation units, includes: inputting the multivariate spatial features as input features into the trained machine learning model, and the trained machine learning model outputting the predicted logarithm result. Based on the formula For the predicted logarithmic results Exponential reduction was performed to obtain predicted abundance, predicted biomass, and predicted species richness; In step 2, the predicted fish resource index values ​​are converted into machine learning resource factors. This includes: normalizing the predicted fish resource index values ​​to the corresponding resource factors based on the following formula: , This represents the predicted value of fish resource indicators. Representing the corresponding resource factors, Representing the The 5th percentile of the reference distribution of resource classes. Representing the The 95th percentile of resource-related indicators This is a truncation function; ,in, Represents abundance factor, The weights representing the abundance factors, Represents biomass factors, The weights representing biomass factors, Representative species factor, The weights of species factors.

[0016] Furthermore, the extrapolation control quantity in step 3 It characterizes the degree to which the environmental conditions of the current evaluation unit deviate from the environmental data corresponding to the cross-lake training sample set; ; This represents the total number of continuous environmental variables involved in the modeling. For the first The pre-defined low quantile boundaries for each continuous environmental variable. For the first The high quantile boundaries of each continuous environmental variable are preset. For the current evaluation unit, the first Quantile characteristics of continuous environmental variables For indicator functions; The lower quantile boundary is the 5th percentile, and the upper quantile boundary is the 95th percentile.

[0017] Furthermore, the empirically based catchability factor in step 3 Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index include: ; Step 3: Resource Existence Sub-Index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index include: ; in, For empirical factor weights, , The minimum weight for the empirical path, This represents the maximum weight of the empirical path.

[0018] This invention provides a lake fish resource catchability assessment system to implement the aforementioned lake fish resource catchability assessment method. The lake fish resource catchability assessment system includes: a data acquisition and preprocessing unit, used to collect environmental data corresponding to the target lake area as the evaluation unit, and stack the environmental data in a parameter layer under a unified spatial grid to obtain multivariate spatial features; and a staged empirical sub-index construction unit, used to obtain resource presence sub-indices based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors The cross-lake machine learning resource factor construction unit is used to acquire fish sampling event data and corresponding environmental data from multiple lake areas, construct a cross-lake training sample set, train a machine learning model using the cross-lake training sample set, and use the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of the evaluation unit. The predicted fish resource index values ​​are then converted into machine learning resource factors. The extrapolation control and fusion output unit is used to obtain extrapolated control quantities based on the environmental data from the evaluation unit and the environmental data corresponding to the cross-lake training sample set. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

[0019] Compared with existing technologies, this invention achieves the following beneficial effects: To address the problems of conflation between habitat suitability and catchability in lake fish resource simulation methods, insufficient stability of cross-lake machine learning migration predictions, and the lack of unified technical constraints in the simple parallel application of empirical and machine learning models, this invention provides a lake fish resource catchability assessment method and system for predicting lake fish resources and catchability in cross-lake migration scenarios. It employs a staged empirical sub-index (resource presence sub-index). Space accessibility index and activity-based food intake index ), cross-lake quantile transfer learning resource factor Furthermore, an extrapolation-controlled fusion method for assessing lake fish resources and catchability integrates empirical models and machine learning. Taking the evaluation unit as the object, it first calculates three stages of empirical sub-indices—resource presence, spatial accessibility, and active feeding—and synthesizes them into an empirical catchability factor. Then, the continuous environmental variables in the cross-lake training samples are converted into quantile features and combined with missing data to construct machine learning resource factors. Finally, the extrapolated control quantity is based on the evaluation unit relative to the training distribution. Adjusting the fusion weights of empirical factors and machine learning resource factors to output a comprehensive resource index or a comprehensive catchability index can help solve problems such as habitat confusion, unstable cross-lake prediction, and the inability to unify and integrate mechanisms and data models. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 A schematic flowchart illustrating the lake fish resource catchability assessment method described in this embodiment of the invention; Figure 2 A schematic diagram of parameter layer stacking input as described in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the stacking of multi-temporal remote sensing scenes as described in an embodiment of the present invention; Figure 4 This is a schematic diagram of the feature stack transformation of the evaluation unit described in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0023] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0024] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0025] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] refer to Figure 1 to Figure 4 This invention provides a method for assessing the catchability of fish resources in lakes, comprising: Step 1, data acquisition and preprocessing, including: taking a target lake area as the evaluation unit, which can be a lake grid after a lake has been divided into multiple lake grids; collecting environmental data corresponding to the evaluation unit; and stacking the environmental data in a unified spatial grid to obtain multivariate spatial features; Step 2, constructing staged empirical sub-indices and cross-lake machine learning resource factors, wherein constructing staged empirical sub-indices includes: obtaining resource presence sub-indices based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors Empirical capture factor Within the representative evaluation unit, a comprehensive assessment of the ecological mechanisms reveals the inherent potential of fish species to be caught / surveyed, and a resource existence sub-index is established. Spatial accessibility sub-index is used to measure whether the basic habitat for fish meets the standards. The active feeding index indicates whether a water area is accessible for fishing or survey operations. Characterizing fish activity and foraging activity, and their susceptibility to fishing gear; constructing cross-lake machine learning resource factors includes: acquiring fish sampling event data and corresponding environmental data from multiple lake areas, constructing a cross-lake training sample set, training a machine learning model using the cross-lake training sample set, using the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of evaluation units, and converting the predicted fish resource index values ​​into machine learning resource factors. Step 3, extrapolation control and fusion output, including: obtaining extrapolated control quantities based on environmental data from evaluation units and environmental data corresponding to cross-lake training sample sets. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

[0027] This invention also provides a lake fish resource catchability assessment system for implementing the aforementioned lake fish resource catchability assessment method. The lake fish resource catchability assessment system includes: a data acquisition and preprocessing unit, used to collect environmental data corresponding to the target lake area as the evaluation unit, and stack the environmental data in a parameter layer under a unified spatial grid to obtain multivariate spatial features; and a staged empirical sub-index construction unit, used to obtain resource presence sub-indices based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors The cross-lake machine learning resource factor construction unit is used to acquire fish sampling event data and corresponding environmental data from multiple lake areas, construct a cross-lake training sample set, train a machine learning model using the cross-lake training sample set, and use the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of the evaluation unit. The predicted fish resource index values ​​are then converted into machine learning resource factors. The extrapolation control and fusion output unit is used to obtain extrapolated control quantities based on the environmental data from the evaluation unit and the environmental data corresponding to the cross-lake training sample set. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

[0028] refer to Figure 1The data acquisition and preprocessing unit is used to implement the data acquisition and preprocessing in step 1. The data acquisition and preprocessing unit is connected to the staged empirical sub-index construction unit and the cross-lake machine learning resource factor construction unit. The outputs of the staged empirical sub-index construction unit and the cross-lake machine learning resource factor construction unit are jointly fed into the extrapolation control and fusion output unit.

[0029] In some embodiments, environmental data may include remote sensing data, lake geometry data, water quality data, fish resource data, and time-period data. Remote sensing data may include: multispectral remote sensing data (Sentinel-2), water body mask (effective water surface area of ​​the lake area), normalized chlorophyll index, normalized turbidity index, normalized water body index, improved normalized water body index, etc.; lake geometry data may include static water depth, lakebed elevation, lakebed slope, distance from shore, nearshore zone width, lake boundary vector data, etc.; water quality data may include: water temperature, dissolved oxygen, turbidity or transparency, chlorophyll, pH, conductivity, nutrient or productivity proxy variables, etc.; fish resource data may include sampling location, sampling time, sampling season, sampling water depth, fishing gear type, number of fish caught, fish weight, number of fish per unit area, weight per unit area, species composition, and number of species, etc.; time-period data may include sampling date, season, dawn, daytime, dusk, nighttime, solar altitude angle, etc.

[0030] In some embodiments, the data acquisition and preprocessing in step 1 further includes preprocessing the acquired data. Preprocessing includes outlier removal, spatial coordinate unification, temporal matching, spatial correlation, and variable normalization. For example, when using a lake grid as the evaluation unit, remote sensing data, water depth, slope, and distance from the shore are resampled to a unified grid. Water quality profile data are correlated according to sampling depth or approximate fish sampling depth. Fish catch data are correlated according to sampling location, sampling date, and season.

[0031] like Figure 2 As shown, various environmental data are stacked in a unified spatial grid to obtain multivariate spatial features. This involves unifying the target lake's remote sensing scene, water mask, water depth, distance from shore, water temperature, dissolved oxygen, turbidity or transparency, normalized chlorophyll index, pH, and conductivity proxy variables into a single spatial grid, forming a model input feature stack. This feature stack is then used for calculating staged empirical sub-indices and cross-lake machine learning resource factors. (Reference) Figure 3 The remote sensing scenes from different dates are organized into a time series input according to year, season and time period. Each remote sensing scene can derive corresponding water body mask and water quality or productivity proxy parameters, thereby supporting the output of resource index or catchability index of target lake scene by scene.

[0032] Furthermore, in step 2, which involves constructing a phased experience sub-index, a resource existence sub-index is obtained based on environmental data. include: ; Representing the Each resource has a suitability value for relevant environmental variables. The total number of environmental variables related to the existence of resources. It is a multi-factor composition function; it is understandable that computational resources have sub-indices. That is, for the evaluation unit, the environmental data reflecting the conditions for the existence of fish resources are converted into a single-factor suitability value between 0 and 1.

[0033] In step 2, which involves constructing phased experience sub-indices, spatially accessible sub-indices are obtained based on environmental data. include: ; Represents the composition function of accessible terms. Indicates the distance from the shore. Represented as the lakebed slope term, Indicates the water depth at which the sample was taken; In step 2, which involves constructing staged experience sub-indices, activity-based feeding sub-indices are obtained based on environmental data. B include: B =WT×WDO×Wlight×Wcycle×Wprey; where WT represents the temperature response, WDO represents the dissolved oxygen response, Wlight represents the light response, Wcycle represents the seasonal or diurnal rhythm, and Wprey represents the food availability.

[0034] Furthermore, resource acquisition has sub-indices. In the process, the relevant environmental variables for resources include water depth, water temperature, dissolved oxygen, turbidity, and chlorophyll proxy variables.

[0035] In some embodiments, a piecewise trapezoidal suitability function is used to obtain the suitability values ​​of resource-existence-related environmental variables based on resource-existence-related environmental variables. The piecewise trapezoidal suitability function is as follows: ; , , , All of these are set thresholds. , This indicates that the lower threshold is completely unsuitable. This represents the lower limit threshold of the optimal range. This represents the upper limit threshold of the optimal range. This indicates that the upper limit threshold is completely unsuitable, when hour, For the first Each resource has actual values ​​for relevant environmental variables; It is a weighted geometric mean function. ;in, Representing the Each resource has the weights of relevant environmental variables.

[0036] It is understood that in some embodiments, the selection of resource-related environmental variables, the method of obtaining the suitability values ​​of resource-related environmental variables, and the multi-factor synthesis function may also be other.

[0037] Furthermore, access to spatially accessible sub-indices In the steps, the distance from the shore is calculated using a left-shoulder-type suitable function. The appropriate function for the left shoulder type is as follows: ,when hour, Represents the distance from the shore. This represents the optimal nearshore threshold. This represents the maximum operational offshore distance, of which, A ∈[0,1], A The closer to 1, the easier it is to catch / survey fish in that area.

[0038] In some embodiments, the lakebed slope term is calculated using a piecewise trapezoidal fit function based on the lakebed slope. , , For the specific lakebed slope value, This is the lower limit threshold for slope. To provide a suitable left boundary for the interval, To provide a suitable right boundary for the interval, The upper limit threshold for slope is given. For details on the piecewise trapezoidal suitability function, please refer to the aforementioned piecewise trapezoidal suitability function provided by the relevant environmental variables for obtaining resource availability. It will not be elaborated here.

[0039] In some embodiments, the sampling water depth term is calculated using a piecewise trapezoidal suitability function based on the sampling water depth. , , Represents the water depth at the sampling point. This represents the lower limit threshold of water depth. Represents the left boundary of the suitable water depth. Represents the right boundary of the suitable water depth. The upper limit threshold for water depth is represented by the piecewise trapezoidal suitability function. For details on the piecewise trapezoidal suitability function provided by the aforementioned resource availability-related environmental variable suitability values, it will not be elaborated upon here.

[0040] In some embodiments, the accessible item composition function as follows: A=λ+(1 l)[0.55A d +0.45(oh n A s +(1 oh n ) )] , l This represents the lower limit of the accessibility index. oh n Represents the weight of the lakebed slope term. 1 oh n Represents the weight of the sampled water depth term, 0≤ oh n ≤1. When sampling water depth data is missing, Assign a preset neutral value, which is used to characterize the neutral accessibility level under conditions where water depth information is missing.

[0041] Furthermore, obtain the activity ingestion index. In this step, the dissolved oxygen response term is calculated using a piecewise trapezoidal fit function based on dissolved oxygen. Based on seasonal or diurnal rhythms, piecewise trapezoidal appropriate functions are used to calculate seasonal or diurnal rhythm terms. Based on food availability, a piecewise trapezoidal suitability function is used to calculate the food availability term. For details on the piecewise trapezoidal suitability function, please refer to the aforementioned piecewise trapezoidal suitability function provided by the relevant environmental variables for obtaining resources; it will not be elaborated upon here.

[0042] It should be noted that, Values ​​can be assigned based on the season and time of day, or they can be calibrated based on the historical fish catch distribution of the target lake.

[0043] In some embodiments, ,in, , , , The solar altitude angle, For normalized turbidity proxy variables in the range of 0 to 1, To normalize the effective light intensity of the water surface, The optimal light level represents the light intensity at which fish are most suitable for activity. The standard deviation represents the range of suitable lighting conditions.

[0044] Furthermore, resources will exist in sub-indices. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors include: When it is necessary to output the comprehensive catchability index At that time, step 1 outputs the empirical catchability factor. When it is necessary to output a comprehensive resource index At that time, the output resource in step 1 has a sub-index. .

[0045] Furthermore, in step 2, the step of constructing a cross-lake training sample set: The continuous environmental variables in environmental data are converted into quantile features based on the following formula. , Representing the Quantile characteristics of continuous environmental variables Representing the A continuous environment variable, Representing the The empirical distribution function of continuous environmental variables; missing variables in the environmental data are set as missing data markers; time-period variables in the environmental data remain unchanged; machine learning input features are constructed based on the quantile features of continuous environmental variables, time-period variables, and missing data markers. Continuous environmental variables include: water depth, water temperature, dissolved oxygen, turbidity, chlorophyll proxy variable, pH, and electrical conductivity.

[0046] , Represents time period variables, Indicates the water depth quantile. Indicates the water temperature quantile. Indicates dissolved oxygen quantiles. Indicates turbidity quantiles. Indicates the quantiles of the chlorophyll proxy variable. Indicates conductivity quantiles, Indicates the acidity / alkalinity quantile. This indicates the absence of key features; for example, if a sample lacks pH and conductivity, but all others are complete, then... .

[0047] In some embodiments, fish sampling event data includes fish resource indicators, which include abundance, biomass, and species richness. The fish resource indicators are then transformed based on the following formula: ,in, As an indicator of fish resources, Abundance refers to the number of fish per unit area. Represents biomass, which is the total weight of fish per unit area. Represents species richness, which is the number of fish species caught. These are the specific numerical values ​​for fish resource indicators. The target is the fish resource index after transformation.

[0048] Furthermore, in step 2, training the machine learning model using a cross-lake training sample set includes: using fish resource indicators after target transformation. and machine learning input features The machine learning model is trained; the machine learning model is a random forest regression model, and the mathematical expression of the random forest regression model is as follows: ,in, Indicates the first A tree of return, This represents the number of regression trees.

[0049] In some embodiments, step 2, obtaining the predicted fish resource index value based on the multivariate spatial features of the evaluation unit using the trained machine learning model, includes: inputting the multivariate spatial features as input features into the trained machine learning model, and the trained machine learning model outputting the predicted logarithm result. Based on the formula For the predicted logarithmic results By performing index reduction, predicted abundance, predicted biomass, and predicted species richness are obtained.

[0050] It should be noted that the environmental parameters involved in the multivariate spatial features input to the machine learning model need to be consistent with the environmental parameters involved in the environmental variables in the training dataset, and also need to be converted into input features that are consistent with the form of the environmental variables in the training dataset.

[0051] In some embodiments, in step 2, the predicted fish resource index value is converted into a machine learning resource factor. This includes: normalizing the predicted fish resource index values ​​to the corresponding resource factors based on the following formula: , This represents the predicted value of fish resource indicators. Representing the corresponding resource factors, Representing the The 5th percentile of the reference distribution of resource classes. Representing the The 95th percentile of resource-related indicators For truncation function, if If less than 0, force the value to be 0. If it is greater than 1, then it will be forcibly assigned the value 1. If the value is an intermediate value, the original calculation result is retained; ,in, Represents abundance factor, The weights representing the abundance factors, Represents biomass factors, The weights representing biomass factors, Representative species factor, The weights represent species factors. This is done to address the inconsistency in the output scale between empirical factors and machine learning resource factors.

[0052] It should be noted that the machine learning resource factor The relative fish resource index is predicted by the trained machine learning model and then normalized to the 0-1 range. It represents the nonlinear relationship between the environment and fish learned from big data from multiple lakes and provides a comprehensive score for the fish abundance, biomass and species richness of the current evaluation unit.

[0053] Furthermore, the aforementioned machine learning model is merely an example; other regression models can also be used as machine learning models.

[0054] Furthermore, the extrapolation control quantity in step 3 This characterizes the degree to which the environmental conditions of the current evaluation unit deviate from the environmental data corresponding to the cross-lake training sample set. Extrapolated control quantities can be calculated based on whether the environmental variables of the evaluation unit exceed the training set distribution or the preset quantile boundaries. The details are as follows: ; The total number of continuous environmental variables involved in the modeling. For the first The pre-defined low quantile boundaries for each continuous environmental variable. For the first The high quantile boundaries of each continuous environmental variable are preset. For the current evaluation unit, the first Quantile characteristics of continuous environmental variables is an indicator function; where the lower quantile boundary is the 5th quantile and the upper quantile boundary is the 95th quantile.

[0055] Extrapolation control quantity The value is in the range of 0 to 1, and the extrapolated control quantity The larger the value, the more significantly the scene of the evaluation unit deviates from the training distribution.

[0056] Furthermore, the empirically based catchability factor in step 3 Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index include: ; Step 3: Resource Existence Sub-Index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index include: ;in, For empirical factor weights, That is, using a linear weighting function. The minimum weight for the empirical path, This represents the maximum weight of the empirical path.

[0057] In a preferred embodiment, Within the range of 0.2 to 0.4, Within the range of 0.6 to 0.8, when the data corresponding to the evaluation unit is insufficient or the extrapolation degree of the evaluation unit is high, the value should be increased. When the evaluation unit has sufficient measured fish resource data and the degree of extrapolation is low, the level of extrapolation should be reduced. .

[0058] for In addition to linear weighting functions, monotonically increasing functions such as the Logistic function can also be used.

[0059] The comprehensive catchability index can be obtained based on actual application needs. and / or comprehensive resource index .

[0060] In some embodiments, the comprehensive catchability index is obtained. and / or comprehensive resource index Then, a tiered approach can be adopted based on the comprehensive catchability index. and / or comprehensive resource index Determine the grade of fish resources and / or catchability, for example, using the grading method shown in the table below:

[0061] Where I represents the comprehensive catchability index. Or comprehensive resource index .

[0062] The lake fish resource catchability assessment method provided by this invention decomposes fish catchability into resource presence sub-indices. Space accessibility index and activity-based food intake index Calculate and synthesize the empirical catchability factor separately. Therefore, this invention can distinguish between different states such as suitable but inaccessible habitats and resources present but not within the feeding window, avoiding the direct equation of a single HSI (Habitat Suitability Index) result with a catchability result. Before cross-lake machine learning modeling, this invention uses the empirical distribution function of the training set for quantile normalization and sets missing variables as missing markers. Therefore, variables such as water depth, temperature, dissolved oxygen, turbidity, productivity, conductivity, and pH across different lakes can be expressed in a unified feature space, reducing migration errors caused by directly using the original units. After the machine learning output, this invention further converts the prediction results into relative resource factors between 0 and 1, and introduces an extrapolation control quantity and a dynamic fusion weight determined by the extrapolation control quantity in the target lake prediction stage. the(X) Therefore, this invention avoids the unconstrained direct use of machine learning results from regions outside the training distribution, which helps reduce the extrapolation risk of cross-lake migration prediction and improves output stability. This invention integrates staged empirical factors and machine learning resource factors in the same evaluation unit, which not only preserves the constraints of ecological mechanisms such as temperature, dissolved oxygen, distance from shore, light, and food availability on the results, but also uses measured fish resource data from multiple lakes to supplement nonlinear relationships that are difficult for empirical models to express, thereby forming a spatially continuous resource map and catchability distribution map of the target lake. Since this invention outputs intermediate results such as resource existence, spatial accessibility, active feeding, machine learning resources, and comprehensive output, when the comprehensive results change, the specific limiting links can be traced, making it easy to determine whether the limiting factors come from habitat conditions, spatial accessibility, active feeding windows, or cross-lake resource relationships.

[0063] The lake fish resource catchability assessment method provided by this invention is used to simulate and assess lake fish resources and catchability. In a specific example, the specific steps of the lake fish resource catchability assessment method are as follows: Step 1, Data Acquisition and Preprocessing: The acquired data and acquisition methods are shown in the table below.

[0064] Data preprocessing was performed as follows: outlier removal, coordinate system unification, temporal matching, and spatial correlation were applied to the collected data. When using the lake area grid as the evaluation unit, remote sensing data, water depth, slope, and distance from the shore were directly resampled to a unified grid. Water quality profile data were correlated according to sampling depth or approximate fish sampling depth. Fish catch data were correlated according to sampling location, sampling date, and season. The processed data were then used to construct models for building staged empirical sub-indices and models for building cross-lake machine learning resource factors.

[0065] Step 2: Calculate the phased empirical sub-index. Taking a specific lake as an example, 34 Sentinel-2 remote sensing scenes were used, including 14 scenes from 2024 and 20 scenes from 2025. The static water depth was calculated using static water depth data constructed based on the lake's historical lakebed topographic raster data. The lake area boundary was determined by multiple water body masks and water depth constraints. The resource existence sub-index was then calculated. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors .

[0066] Calculations showed that all 34 empirical models for the lake successfully ran, and the resources exhibited sub-indices. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors All values ​​are within the range of 0 to 1. Among the 16 scenes corresponding to spring and autumn, the mean captureability at dawn and dusk is higher than the mean captureability during the day. In the whole sample, the mean captureability of dusk for scenes with higher mean captureability in autumn is 0.283.

[0067] Subsequently, machine learning model training was conducted. Specifically, existing multi-lake training data was used to construct cross-lake training samples, totaling 2682 samples. Features included season, water depth, water temperature, dissolved oxygen, turbidity, productivity, conductivity, pH, and missing data. Target variables included abundance, biomass, and species richness. The preferred machine learning model was a random forest regression model. The results of the five-fold cross-validation are as follows:

[0068] The sample size, model type, and cross-validation metrics mentioned above are used to illustrate that the machine learning resource factors in this embodiment have a training and validation basis, and do not constitute a limitation on the scope of protection of this invention.

[0069] In some implementations, four sets of comparative verifications can be set up: empirical path output only, machine learning path output only, fixed-weight fusion output, and output based on extrapolation control variables. The dynamic fusion output is compared with the coefficient of determination, root mean square error, mean absolute error, out-of-distribution scene error, and spatial result volatility to verify the effect of staged empirical constraints and extrapolation control on the stability of cross-lake migration. It is understood that the above comparative verification is used to illustrate the verification path of this invention and does not constitute a limitation on the scope of protection of this invention.

[0070] Next, the proxy variables of water depth, temperature, dissolved oxygen, turbidity, productivity, conductivity, and pH from the lake remote sensing inversion were converted into the quantile space of the training set and input into the trained machine learning model to obtain resource prediction results for 34 lake scenes. The migration prediction results show that the scene with a higher predicted abundance relative to the mean resource factor has a predicted abundance relative resource factor mean of 0.872; the scene with a higher predicted biomass relative to the mean resource factor has a predicted biomass relative resource factor mean of 0.830; and the scene with a higher predicted species richness relative to the mean resource factor has a predicted species richness relative resource factor mean of 0.879.

[0071] In this embodiment, the output of the machine learning model is first converted into a machine learning resource factor of 0 to 1 according to a reference distribution. Then, it enters the integration step with the experience path, thereby ensuring the resource factor of machine learning. With empirical capture factor It is within the same dimension.

[0072] Subsequently, on-site verification and model reliability demonstration can be conducted. Specifically, in some embodiments, after completing the prediction of 34 scenarios, on-site verification sampling points can be set up in high-value, medium-value, and low-value areas respectively, and measured data such as fish catch quantity, fish catch weight, species count, acoustic density, water temperature, dissolved oxygen, transparency, or turbidity can be obtained simultaneously in the corresponding season and time period; the on-site data and the resource existence sub-index output by the model are then compared. Machine learning resource factors Comprehensive Resource Index Or a comprehensive capture index Corresponding comparisons were conducted to examine the model's ability to represent spatial distribution, hierarchical ranking, and time-series differences.

[0073] During field verification, the reliability of the model can be demonstrated from three levels: First, compare the consistency between the model output level and the on-site fish catch or acoustic density level; second, compare whether the relative high and low relationships between dawn / dusk and daytime scenes are consistent with the on-site survey results; third, analyze whether the deviation between the model and the measured results mainly comes from the resource presence sub-index, spatial accessibility sub-index, active feeding sub-index, or cross-lake machine learning resource factors when deviations occur. This is used to identify the source of error and call back the parameters of the calculation formula.

[0074] After supplementing the measured data, further verification indicators such as correlation coefficient, root mean square error, mean absolute error, grade consistency rate, and high value area hit rate can be given. When the comprehensive index F is consistent with the field survey results in terms of spatial distribution, grade ranking, and time period variation, the reliability of the method of the present invention for simulating fish resources and catchability in Yangcheng Lake can be verified.

[0075] Step 3, extrapolation control and fusion output, specifically, can be based on empirical capture factors. Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index For scenarios where the actual fish sample size in the target lake is insufficient or the extrapolation is highly significant, improving... the(X) For scenarios where measured fish resource data for the target lake is relatively abundant and the degree of extrapolation is low, the reduction of... the(X) The five-level grading standard determines the grades as high, relatively high, medium, relatively low, and low.

[0076] The output includes the spatial distribution results of resource existence sub-index, spatial accessibility sub-index, active feeding sub-index, machine learning resource factor, and fusion index. Based on the relative magnitude of each sub-index, it can be determined whether the limiting factors mainly come from habitat conditions, spatial accessibility, active feeding window, or cross-lake resource relationships. Based on the determination results, it can further provide a reference for resource monitoring, stock enhancement and release, fishing survey deployment, or selection of recreational fishing periods.

[0077] It should be noted that the data for each lake in the multi-lake training dataset includes, or can be converted into, indicators such as water depth, temperature, dissolved oxygen, turbidity or transparency, productivity or chlorophyll proxies, electrical conductivity, pH, season, and fish catch. While the remote sensing scene of the evaluation unit is not entirely equivalent to the in-situ sampling profile, it can provide isomorphic variables such as water body masks, temperature proxies, turbidity proxies, chlorophyll or NDCI proxies, spatial water depth, and distance from the shore. Therefore, these variables can be incorporated into the same machine learning feature framework.

[0078] Different lakes vary in water depth, trophic status, transparency, and sampling regime. To reduce the dimensional differences across lakes, this invention does not directly use the original dimensions for training. Instead, it uses the empirical distribution function of the training set for quantile normalization and sets a missing value for each variable. During actual prediction, the model simultaneously outputs extrapolated control values. When the target scene variable deviates significantly from the training distribution, the machine learning weights are reduced or the empirical model becomes the constraint.

[0079] The unified training data employed five-fold cross-validation. The cross-validation determination coefficients R² of the random forest regression model on the three objectives of abundance, biomass, and species richness were 0.828, 0.923, and 0.744, respectively, and the Spearman correlation coefficients were 0.815, 0.862, and 0.818, respectively, indicating that the machine learning resource factors in this implementation method have a training and validation basis.

[0080] To verify the stability of the technical solution of this invention, four sets of comparisons can be set up: output of empirical model alone, output of machine learning model alone, fixed weight fusion without extrapolation control, and dynamic fusion with extrapolation control. The cross-validation error, out-of-distribution scene error and spatial result volatility are compared. This comparison is used to illustrate the stability improvement brought about by the combined effect of staged empirical constraints, quantile migration and extrapolation control.

[0081] Fish resources are constrained by multiple essential environmental factors. When a single factor is extremely low, other factors cannot fully compensate for it. Weighted geometric mean can reflect the effect of limiting factors. That is, when any key SI approaches 0, the comprehensive resource existence sub-index decreases significantly. The present invention uses weighted geometric mean to synthesize water depth, water temperature, dissolved oxygen, turbidity and vegetation or chlorophyll factors, which is reasonable.

[0082] Whether lake fish resources can be surveyed, caught, or utilized for recreational fisheries depends not only on the presence of fish, but also on whether the target resources are located within the shoreline reachable range, the operational depth, or the corresponding operational layer. This invention uses the distance from the shore, the slope, and the layer to calculate the spatial accessibility index, and avoids unreasonable extreme values ​​of the spatial accessibility factor when local information is lacking by using the lower limit term and the neutral assignment when the sampling water depth term is missing, which meets the engineering requirements of lake survey and utilization scenarios.

[0083] Lake fish catchability depends not only on the presence of fish but also on whether they are in a window of active feeding and contact with fishing gear. Related technologies indicate that environmental variables influence fish feeding behavior and bait-gear performance. Low light, temperature, lunar phase, and weather conditions affect the catchability of pikeperch and pike. The mere encounter with bait is insufficient to determine whether a fish is caught; fine-scale feeding behavior after contact remains crucial. Light and temperature jointly determine the thermo-photonic environment of pikeperch. Therefore, this invention decomposes the active feeding sub-index into temperature, dissolved oxygen, light intensity, seasonal rhythms, and food availability, consistent with the behavioral mechanisms of lake fish.

[0084] This invention differs from traditional technologies in the following main ways: First, it breaks down the fish catchability formation process into three empirical sub-indices: resource availability, spatial accessibility, and active feeding, and then calculates and synthesizes them into a staged empirical factor. Second, this invention converts continuous environmental variables from different lakes into quantile features using the empirical distribution function of the training set, and combines this with missing data features to form a unified input space across lakes. Third, it converts machine learning predictions into machine learning resource factors between 0 and 1. And calculate the extrapolated control quantity of the target lake evaluation unit relative to the training set distribution. Then, based on the extrapolated control quantity Dynamically determine the weights of empirical factors Fourth, integrate staged experience factors and machine learning resource factors within the same evaluation unit to form a comprehensive resource index or comprehensive catchability index; Fifth, output intermediate indices such as resource existence, spatial accessibility, active feeding, machine learning resources, and comprehensive output to support the tracing of limiting factors.

[0085] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0086] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for assessing the catchability of lake fish resources, characterized in that, include: Step 1, data acquisition and preprocessing, includes: taking the target lake area as the evaluation unit, collecting environmental data corresponding to the evaluation unit, and obtaining multivariate spatial features based on the environmental data; Step 2: Construct staged empirical sub-indices and cross-lake machine learning resource factors. The construction of staged empirical sub-indices includes: obtaining resource existence sub-indices based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors ; Constructing cross-lake machine learning resource factors involves: acquiring fish sampling event data and corresponding environmental data from multiple lake areas; building a cross-lake training sample set; training a machine learning model using the cross-lake training sample set; using the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of evaluation units; and converting the predicted fish resource index values ​​into machine learning resource factors. ; Step 3, extrapolation control and fusion output, includes: obtaining extrapolated control quantities based on environmental data from evaluation units and environmental data corresponding to cross-lake training sample sets. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

2. The method for assessing the catchability of lake fish resources according to claim 1, characterized in that, In step 2, which involves constructing phased experience sub-indices, a resource existence sub-index is obtained based on environmental data. include: ; Representing the Each resource has a suitability value for relevant environmental variables. , The total number of environmental variables related to the existence of resources. It is a multi-factor composition function; In step 2, which involves constructing phased experience sub-indices, spatially accessible sub-indices are obtained based on environmental data. include: ; Represents the composition function of accessible terms. Indicates the distance from the shore. Represented as the lakebed slope term, Indicates the water depth at which the sample was taken; In step 2, which involves constructing staged experience sub-indices, activity-based feeding sub-indices are obtained based on environmental data. B include: B =WT×WDO×Wlight×Wcycle×Wprey; where WT represents the temperature response, WDO represents the dissolved oxygen response, Wlight represents the light response, Wcycle represents the seasonal or diurnal rhythm, and Wprey represents the food availability.

3. The method for assessing the catchability of lake fish resources according to claim 2, characterized in that, Resource acquisition has sub-index In the process, the resources have relevant environmental variables including water depth, water temperature, dissolved oxygen, turbidity, and chlorophyll proxy variables; A piecewise trapezoidal suitability function is used to obtain the suitability values ​​of resource-existence-related environmental variables based on resource-existence-related environmental variables. The piecewise trapezoidal suitability function is as follows: ； , , , All of these are set thresholds. , This indicates that the lower threshold is completely unsuitable. This represents the lower limit threshold of the optimal range. This represents the upper limit threshold of the optimal range. This indicates that the upper limit threshold is completely unsuitable, when hour, For the first Each resource has actual values ​​for relevant environmental variables; It is a weighted geometric mean function. ;in, Representing the Each resource has the weights of relevant environmental variables.

4. The method for assessing the catchability of lake fish resources according to claim 2, characterized in that, Acquiring Spatial Accessible Sub-Index In the steps, the distance from the shore is calculated using a left-shoulder-type suitable function. The appropriate function for the left shoulder type is as follows: ,when hour, Represents the distance from the shore. This represents the optimal nearshore threshold. This represents the maximum operational offshore distance; The lakebed slope term is calculated using a piecewise trapezoidal suitable function based on the lakebed slope. ; The sampling water depth term is calculated using a piecewise trapezoidal suitability function based on the sampling water depth. ; Contactless term composition function as follows: A=λ+(1 λ)[0.55A d +0.45(ω n A s +(1 ω n ) )] , λ This represents the lower limit of the accessibility index. ω n Represents the weight of the lakebed slope term. 1 ω n Represents the weight of the sampled water depth term, 0≤ ω n ≤1.

5. The method for assessing the catchability of lake fish resources according to claim 2, characterized in that, Obtain the activity ingestion index In this step, the dissolved oxygen response term is calculated using a piecewise trapezoidal fit function based on dissolved oxygen. Based on seasonal or diurnal rhythms, piecewise trapezoidal appropriate functions are used to calculate seasonal or diurnal rhythm terms. Based on food availability, a piecewise trapezoidal suitability function is used to calculate the food availability term. ; ,in, , , , The solar altitude angle, For normalized turbidity proxy variables in the range of 0 to 1, To normalize the effective light intensity of the water surface, The optimal light level represents the most suitable light intensity for fish activity. The standard deviation represents the range of suitable lighting conditions.

6. The method for assessing the catchability of lake fish resources according to claim 1, characterized in that, In step 2, which involves constructing a training sample set across lakes: The continuous environmental variables in environmental data are converted into quantile features based on the following formula. , Representing the Quantile characteristics of continuous environmental variables Representing the A continuous environment variable, Representing the Empirical distribution function of continuous environmental variables; Set missing variables in environmental data as missing feature characteristics; The time-varying variables in the environmental data remain unchanged; Machine learning input features are constructed based on quantile features of continuous environmental variables, time-period variables, and missing data markers. ; Continuous environmental variables include: water depth, water temperature, dissolved oxygen, turbidity, chlorophyll proxy variables, pH, and conductivity. , Represents time period variables, Indicates the water depth quantile. Indicates the water temperature quantile. Indicates dissolved oxygen quantiles. Indicates turbidity quantiles. Indicates the quantiles of the chlorophyll surrogate variable. Indicates conductivity quantiles, Indicates the acidity / alkalinity quantile. Indicates missing marker features; Fish sampling event data includes fish resource indicators, such as abundance, biomass, and species richness. The target transformation of these fish resource indicators is performed based on the following formula: ,in, As an indicator of fish resources, Represents abundance, Represents biomass, Represents species richness, These are the specific numerical values ​​for fish resource indicators. The target is the fish resource index after transformation.

7. The method for assessing the catchability of lake fish resources according to claim 6, characterized in that, Step 2, training the machine learning model using a cross-lake training sample set, includes: using fish resource indicators after target transformation. and machine learning input features Train the machine learning model; the machine learning model is a random forest regression model, and the mathematical expression of the random forest regression model is as follows: ,in, Indicates the first A tree of return, This represents the number of regression trees; Step 2, using the trained machine learning model to obtain predicted fish resource index values ​​based on the multivariate spatial features of the evaluation units, includes: inputting the multivariate spatial features as input features into the trained machine learning model, and the trained machine learning model outputting the predicted logarithm result. Based on the formula For the predicted logarithmic results Exponential reduction was performed to obtain predicted abundance, predicted biomass, and predicted species richness; In step 2, the predicted fish resource index values ​​are converted into machine learning resource factors. This includes: normalizing the predicted fish resource index values ​​to the corresponding resource factors based on the following formula: , This represents the predicted value of fish resource indicators. Representing the corresponding resource factors, Representing the The 5th percentile of the reference distribution of resource classes. Representing the The 95th percentile of resource-related indicators This is a truncation function; ,in, Represents abundance factor, The weights representing the abundance factors, Represents biomass factors, The weights representing biomass factors, Representative species factor, The weights of species factors.

8. The method for assessing the catchability of lake fish resources according to claim 1, characterized in that, Extrapolation control quantity in step 3 It characterizes the degree to which the environmental conditions of the current evaluation unit deviate from the environmental data corresponding to the cross-lake training sample set; ; This represents the total number of continuous environmental variables involved in the modeling. For the first The pre-defined low quantile boundaries for each continuous environmental variable. For the first The high quantile boundaries of each continuous environmental variable are preset. The current evaluation unit is the Quantile characteristics of continuous environmental variables For indicator functions; The lower quantile boundary is the 5th percentile, and the upper quantile boundary is the 95th percentile.

9. The method for assessing the catchability of lake fish resources according to claim 1, characterized in that, Step 3: Empirical Capacity Factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index include: ; Step 3: Resource Existence Sub-Index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index include: ; in, For empirical factor weights, , The minimum weight of the empirical path, This represents the maximum weight of the empirical path.

10. A system for assessing the catchability of lake fish resources, characterized in that, For implementing the lake fish resource catchability assessment method according to any one of claims 1 to 9, the lake fish resource catchability assessment system comprises: The data acquisition and preprocessing unit is used to take the target lake area as the evaluation unit, collect environmental data corresponding to the evaluation unit, and stack the environmental data in a parameter layer under a unified spatial grid to obtain multivariable spatial features. Phased experience sub-index construction unit, used to obtain resource existence sub-index based on environmental data. Space accessibility index and activity-based food intake index The resources will be stored in the sub-index. Space accessibility index and activity-based food intake index Synthesized into empirically traversable factors ; A cross-lake machine learning resource factor construction unit is used to acquire fish sampling event data and corresponding environmental data from multiple lake areas, construct a cross-lake training sample set, train a machine learning model using this set, and then use the trained model to predict fish resource index values ​​based on the multivariate spatial features of the evaluation unit. These predicted fish resource index values ​​are then converted into machine learning resource factors. ; The extrapolation control and fusion output unit is used to obtain extrapolated control quantities based on environmental data from the evaluation unit and environmental data corresponding to the cross-lake training sample set. Based on empirical capture factor Machine learning resource factors and extrapolation control quantity Obtain the comprehensive catchability index Based on the resource existence sub-index Machine learning resource factors and extrapolation control quantity Obtain Comprehensive Resource Index .

Images