Method for forming continuous spatio-temporal data sequence based on species distribution model and environmental data supplementary time section

Abstract

The invention relates to a method for forming a continuous spatio-temporal data sequence based on a species distribution model and an environmental data supplementary time section, and relates to the technical field of fishery resource investigation and evaluation and ecological system modeling and management. The method comprises the following steps: collecting fixed-point survey data and environmental condition data; establishing a resource-environment relation model; acquiring environment data at a target time point and speculating a fishery resource distribution state based on the model; speculating resource distribution characteristics in the middle time period through interpolation between the two investigation time points; adjusting environmental adaptation feature optimization prediction based on the speculation result; and finally, integrating the speculated data and the measured data to generate a minute-scale, hour-scale or daily-scale continuous three-dimensional resource distribution data set. According to the method, the temporal-spatial resolution and continuity of fishery resource distribution data can be remarkably improved, the ecological mechanism support of prediction is enhanced, and the method is widely applied to fishery resource dynamic monitoring, ecological system evaluation, protection division formulation and fishery management based on an ecological system.

Description

Technical Field

[0001] The present invention relates to the technical fields of fishery resource investigation and assessment, ecosystem modeling and management, and in particular to a method for forming a continuous spatiotemporal fishery resource distribution data sequence based on a species distribution model and multi-source environmental data supplemented time sections, belonging to the technical fields of ecosystem modeling and dynamic resource monitoring. Background Art

[0002] In the modern fishery resource survey and management system, accurate and continuous resource distribution data are of great significance for scientific research, ecological assessment and sustainable fishery management decision-making. However, the current scientific fishery resource survey system generally suffers from the problem of insufficient spatiotemporal data continuity. Traditional survey methods mainly rely on ship-based fixed-point sampling, usually adopting a quarterly survey model, and only collect data at four time points each year, resulting in extremely sparse data within the annual scale. This low-frequency sampling strategy cannot accurately reflect the dynamic distribution characteristics of fishery resource populations under driving factors such as seasonal changes and short-term environmental fluctuations, limiting the accuracy of resource dynamics research and ecosystem health status assessment, and also affecting the scientific formulation of Ecosystem-Based Fisheries Management (EBFM) strategies.

[0003] To address the shortcomings of traditional survey data, some research institutions have introduced non-systematic observation methods, such as vessel monitoring systems (VMS), in an attempt to expand data coverage. However, due to random sample collection, inconsistent observation standards, and insufficient data systematization, resource distribution information constructed based on such data still suffers from low spatiotemporal resolution, poor data consistency, and limited comparability.

[0004] Faced with these limitations, the research community has gradually explored methods based on coupled ecological and environmental factor modeling. By analyzing the response relationship between species distribution and environmental parameters (such as sea surface temperature, chlorophyll concentration, and current velocity), predictive models are established. These models are then combined with satellite remote sensing data, fixed-point observation data, and ocean numerical model outputs to improve the spatiotemporal resolution of environmental variables. However, existing methods still have significant problems in practical application, including:

[0005] There is a loss of accuracy in the process of fusing multi-source environmental data, and there is a lack of a unified and standardized data processing process;

[0006] Fishery resource distribution prediction models are mostly based on empirical statistics, with insufficient description of ecological mechanisms and difficulty in fully reflecting the complex dynamic relationship between species and environmental factors.

[0007] The model verification system is still imperfect, which limits the reliability and application value of the prediction results.

[0008] The main defects of the existing technology are summarized as follows:

[0009] The sampling time resolution is low, resulting in insufficient spatiotemporal continuity of the data;

[0010] The lack of multi-source data fusion standards affects the prediction accuracy;

[0011] The characterization of resource-environment response relationships is incomplete, and the ecological mechanism support is weak;

[0012] The prediction model lacks sufficient verification, which affects the actual application effect.

[0013] Therefore, there is an urgent need to develop a new method that can make full use of multi-source high-resolution environmental data, systematically establish a dynamic response mechanism between resources and the environment, improve the scientificity and accuracy of predictions, and continuously reconstruct the distribution characteristics of fishery resources in time and space, so as to effectively support practical application needs such as dynamic monitoring of fishery resources, ecosystem health assessment, protection zone zoning, and ecosystem-based fishery management. Summary of the Invention

[0014] In order to address the problems of low sampling time resolution, insufficient spatiotemporal continuity of data, insufficient resource-environment response modeling, and lack of multi-source environmental data fusion processing in the existing fishery resource survey system, the present invention provides a method for forming a continuous spatiotemporal data sequence based on a species distribution model (SDM) and supplementary time sections of environmental data.

[0015] The method of the present invention comprises the steps of:

[0016] Step 1. Obtain fixed-point survey data and environmental condition data

[0017] Through scientific survey vessels or fishing vessels, using standardized sampling equipment and processes, fishery resource distribution data are collected at specific times and locations, and environmental condition data from in situ observations, remote sensing observations and ocean numerical model outputs are simultaneously collected to provide high-quality input for modeling.

[0018] Step 2. Establish resource-environment relationship model

[0019] Use species distribution modeling methods to analyze the response relationship between fishery resources and environmental factors, screen key environmental variables, and construct a mathematical model that can quantify the relationship between resource distribution and environmental factors.

[0020] Step 3. Obtain environmental condition data at the target time point

[0021] Collect historical observation data or numerical model deduction data, preprocess and standardize the data to ensure consistency of time scale, spatial scale and variable unit.

[0022] Step 4. Estimate the spatial distribution of fishery resources at the target time point

[0023] The environmental data at the target time point is input into the resource-environment relationship model to infer the spatial distribution status of fishery resources at that time point, thereby realizing the dynamic prediction of resources at unsurveyed time points.

[0024] Step 5. Estimate resource distribution characteristics at intermediate time points

[0025] When there are two survey time points, the resource distribution characteristics of the intermediate time point are estimated through interpolation or fitting methods based on the changing patterns of environmental factors and combined with historical change characteristics.

[0026] Step 6. Optimize model environment adaptation features

[0027] Adjust model parameters based on the inference results, combine with the latest environmental data at the target time point, optimize resource distribution prediction results, and enhance the ability to respond to environmental changes.

[0028] Step 7. Generate a continuous three-dimensional resource distribution dataset

[0029] Integrate inferred data with measured survey data to generate a continuous three-dimensional fishery resource distribution dataset at the minute, hour or daily scale according to a unified time step and spatial grid.

[0030] Preferably, in the present invention:

[0031] Environmental condition data including, but not limited to, sea surface temperature, salinity, dissolved oxygen, chlorophyll concentration, water depth, current velocity, or pH;

[0032] The resource-environment relationship model can be selected from the maximum entropy model (MaxEnt), artificial neural network (ANN), classification tree analysis (CTA), random forest (RF) or habitat suitability index model (HSI);

[0033] Fishing vessel operation production data can be introduced into the inference process as external validation data for cross-validation and model optimization;

[0034] Time series data supplementation should preferably use spatiotemporal interpolation algorithms, including spline interpolation or Kriging interpolation;

[0035] The quality of continuous spatiotemporal datasets is verified through error propagation analysis and uncertainty quantification methods;

[0036] Multi-source environmental data is homogenized using fusion technology to improve data consistency and prediction accuracy.

[0037] The present invention has the following beneficial effects:

[0038] Significantly improve the data continuity and temporal and spatial resolution of dynamic monitoring of fishery resources;

[0039] Enhance the scientific nature and accuracy of prediction models through ecological mechanism modeling;

[0040] Improve data consistency through multi-source data standardization;

[0041] Introducing error propagation analysis and uncertainty quantification to ensure prediction reliability;

[0042] It has broad application prospects and can support dynamic resource monitoring, ecological assessment and ecosystem-based fishery management. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the present invention is further described below in conjunction with the accompanying drawings, but the accompanying drawings are only used for illustration and do not constitute a limitation on the scope of protection of the present invention.

[0044] Figure 1 This is a schematic diagram of the biological spatial distribution model of resource and environment coupling described in an embodiment of the present invention, showing the response relationship and modeling process between the distribution characteristics of fishery resources and environmental factors (including sea surface temperature, salinity, dissolved oxygen, chlorophyll concentration, etc.).

[0045] Figure 2 This is a flowchart of an embodiment of the present invention for inferring continuous time series data based on spatial distribution characteristic data at a single time point, illustrating the method steps of how to use environmental data at a target time point, combined with an established resource-environment relationship model, to predict forward or review backward to supplement the spatial distribution status of fishery resources at consecutive time points.

[0046] Figure 3 This is a flowchart of the method of inferring continuous distribution characteristic data based on spatial distribution characteristic data at two time points as described in an embodiment of the present invention. It shows the technical process of using interpolation methods to infer the gradual process of biological adaptation to the environment based on the known characteristics of environmental factor changes at two survey time points, and combining relevant environmental data to supplement the resource distribution characteristics at intermediate time points. DETAILED DESCRIPTION

[0047] The present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the present invention is not limited to the following specific embodiments, and any modifications, equivalent substitutions, and improvements made within the spirit and substance of the present invention should be included within the scope of protection of the present invention.

[0048] like Figure 1-3The present invention's method for spatial and temporal continuity of fishery resources based on species distribution models begins with a scientific survey vessel equipped with a CTD profiler and standard sampling gear. Observation stations are deployed in a 0.5°×0.5° grid across the study area to simultaneously collect data on fishery resource abundance and environmental parameters such as sea surface temperature and salinity. Environmental data is collected simultaneously from in-situ observation equipment, MODIS satellite L2 products, and daily-scale data output by the HYCOM model. All data undergo radiometric calibration and geometric correction, and are then verified by inter-laboratory comparison using the ISO17025 standard to ensure measurement errors are within the technical specifications of ±0.02°C for salinity and ±0.1°C for temperature.

[0049] During the resource-environment relationship modeling phase, collected fishery resource data and environmental factors filtered by Spearman correlation were input into the MaxEnt model. A Bayesian optimization algorithm automatically adjusted the feature multiplier and regularization coefficient. A 5-fold cross-validation approach was used to calculate the AUC for each iteration, with training terminated when the AUC fluctuation for 20 consecutive iterations was less than 0.01. To enhance model generalization, the environmental mean of the previous seven days was introduced as a time-lagged feature. A random forest and XGBoost model integration system was constructed. Finally, a stacking strategy was used to input the prediction results of each model into a logistic regression meta-model, forming a composite predictor with uncertainty quantification capabilities.

[0050] When inferring resource distribution at historical or future points in time, the environmental field at the target time point is extracted from the ECCO reanalysis dataset. Missing data are then spatially and temporally filled using the DINEOF method. Its core algorithm constructs empirical orthogonal functions through singular value decomposition, meeting the technical requirement of a reconstruction error of RMSE ≤ 0.3. The processed environmental data is fed into the previously trained ensemble model. Monte Carlo simulations are performed simultaneously with the output of the resource probability distribution map. 95% confidence intervals are calculated using 1000 random perturbations, resulting in a forecast with error bands.

[0051] For data reconstruction during the survey interval, based on the changing trends of environmental data between the two measured time points, cubic spline interpolation was first used to generate the intermediate time series of temperature and salinity fields. Kriging interpolation was then used to dynamically propagate the spatial variogram parameters to each interpolation point. Specifically, when a mesoscale eddy was detected (determined by an Okubo-Weiss parameter < -0.2), the eddy rotation angle was additionally incorporated into the interpolation process as a constraint to ensure the rationality of the physical process. The resulting continuous dataset was organized as a three-dimensional NetCDF file according to the CF-1.8 standard. The temporal resolution reached daily scales, and the spatial resolution used a 0.1° × 0.1° grid in the WGS84 coordinate system. Each grid cell stores the mean and standard deviation of the resource density.

[0052] In order to verify the actual effect of this method, the 2022 East China Sea hairtail resource survey data was selected for comparative testing. Compared with the traditional quarterly survey results, the daily-scale data products generated by the present invention reduced the resource estimation error at the verification site from ±35% to ±12%, and successfully captured the northward shift of resource distribution caused by the swing of the Kuroshio front in spring. This feature was missed in conventional surveys due to insufficient temporal resolution. It was further proved by assimilating the AIS data of fishing vessels that the spatial matching degree between the high-yield areas predicted by the present invention and the actual operation hotspots reached 82%, which is significantly better than the 65% matching rate of the static model.

[0053] Application Examples

[0054] This example uses the East China Sea small yellow croaker (Larimichthys polyactis) population as the verification object, and the specific implementation process is as follows:

[0055] Data collection phase

[0056] Sixty observation stations were deployed in the central waters of the East China Sea (122°E-126°E, 27°N-32°N). The SBE 911plus CTD profiler was used to simultaneously collect environmental parameters such as temperature, salinity, and dissolved oxygen. At the same time, a standard bottom trawl (net opening area of 8m2) was used to collect the data. 2 , 2cm mesh) to obtain biological resource data. Specifically, to capture ecological transitions in the water mass convergence zone, 12 dynamic monitoring stations were deployed at the confluence of the Yangtze River's diluted water front and the Kuroshio branch. Sampling depth was precisely controlled within the bottom layer range of 0.5-5m, and the trawl speed was maintained within the technical specification of 2.5 knots ± 0.2 knots. Environmental data was simultaneously integrated with MODIS-Aqua satellite chlorophyll a remote sensing data (1km resolution) and ocean current fields output by the HYCOM model. The multi-source data was temporally and spatially aligned using UTC timestamps and WGS84 coordinate system.

[0057] Quality control

[0058] A three-level validation was performed on the collected raw data: first, outliers were automatically filtered based on the ocean physical threshold range (temperature 0-35°C, salinity 5-40 PSU); then, spatial outliers were identified through semivariogram analysis, and missing data were supplemented using a modified inverse distance weighting (IDW) method, in which the power exponent parameter was optimized to 2.3 through cross-validation; finally, satellite remote sensing products were bias-corrected using Argo float observation data, reducing the root mean square error of the chlorophyll a data to 0.12 mg / m 3 The processed dataset forms a spatiotemporally continuous 4D matrix (longitude × latitude × depth × time) with a temporal resolution of hours.

[0059] Model building process

[0060] When using the maximum entropy model (MaxEnt 3.4.1) to establish resource-environment response relationships, an innovative dynamic feature selection mechanism was introduced: the Pearson correlation coefficient matrix was calculated using a sliding time window, and principal component analysis (PCA) dimensionality reduction was automatically triggered when |r| between environmental factors was greater than 0.7, ensuring that the input variables always maintained orthogonality. During the model training phase, a Bayesian optimization algorithm was used to automatically adjust the regularization multiplier (search range 0.1-10). Five-fold cross-validation was performed and the AUC value was calculated for each iteration. Training was terminated when the AUC fluctuation for 15 consecutive iterations was less than 0.008. The final model achieved a prediction accuracy of AUC 0.89±0.02 on the test set, significantly outperforming the traditional static model's 0.76±0.05 (p<0.01, t-test).

[0061] Implementation of space-time continuity

[0062] For the vacant period from April to November 2022, the environmental data were first subjected to spatiotemporal fusion. Adaptive kriging interpolation was used in the spatial dimension, and the optimal interpolation radius (82 km in longitudinal direction and 64 km in latitudinal direction) was determined by fitting a variogram. In the temporal dimension, a triple spline function was constructed, incorporating the tidal period (12.4 hours) and the lunar period (29.5 days) as constraints. When the processed environmental field was fed into the trained model, an error propagation calculation was performed simultaneously. Monte Carlo simulations (1000 repetitions) were used to quantify the forecast uncertainty. The resulting resource probability distribution map had an error band within ±0.08 at a 95% confidence level.

[0063] Verification and Application

[0064] VMS fishing vessel trajectory data (spatial accuracy ±100m) obtained through the East China Sea Fishery Administration Platform show that the spatial overlap between the high-probability areas (>0.7) predicted by this method and the actual catches of fishing sites >50kg reaches 82%, a 27 percentage point increase in matching rate compared to traditional quarterly survey data. When the generated NetCDF dataset was applied to fishery management decision-making, the following results were achieved: (1) the migration path of the yellow croaker spawning ground was accurately identified, with the spawning ground shifting 23km northeastward between June and July; (2) a stress response of a 15km southward shift in the resource distribution center was detected six days after Typhoon Meihua passed in September; and (3) a fishing quota scheme optimized based on continuous data reduced fishery conflicts by 42% that year.

[0065] In summary, through the application verification of this embodiment, the method proposed in the present invention to form a continuous spatiotemporal data sequence based on the species distribution model and the environmental data supplementary time section can systematically infer the distribution status of fishery resources at unsurveyed time points, and accurately reconstruct the dynamic change process of resources with high spatiotemporal resolution. By introducing multi-source environmental data fusion, ecological mechanism-driven modeling, spatiotemporal interpolation and error propagation analysis, the present invention effectively overcomes the problems of traditional survey data discontinuity, insufficient prediction accuracy and poor data consistency, and significantly improves the scientificity and reliability of fishery resource monitoring and ecological management. The above specific embodiments are only used to illustrate the technical solutions of the present invention, and cannot be used to limit the scope of protection of the present invention. All equivalent deformations or replacements based on the concept of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for forming a continuous spatiotemporal data sequence based on a species distribution model and environmental data supplemented with time sections, characterized in that: The following steps are involved: (1) Collecting fixed-point survey data and environmental condition data. The fixed-point survey data includes fishery resource distribution data collected by scientific research vessels or fishing vessels at specific times and locations. The environmental condition data includes data derived from in-situ observations, remote sensing observations, or ocean numerical model outputs; (2) Based on the fixed-point survey data and environmental condition data, a resource-environment relationship model is established. By using species distribution modeling methods, the response relationship between fishery resources and environmental factors is analyzed, the main environmental variables are screened, and a mathematical model of resource distribution and environmental factors is constructed; (3) obtaining environmental condition data at a target time point, including historical observation data or predicted data based on numerical model deduction, and preprocessing and standardizing the environmental condition data at the target time point; (4) Input the preprocessed and standardized environmental condition data at the target time point into the resource-environment relationship model to infer the spatial distribution of fishery resources at that time point; (5) In the case of two survey time points, based on the changing trends of the environmental condition data at the two survey time points, the model response environmental condition characteristics at the intermediate time point are inferred by the spatiotemporal interpolation algorithm; (6) Based on the inference results, adjust the environmental adaptation characteristic parameters of the resource-environment relationship model and optimize the spatial distribution prediction of fishery resources by combining the latest environmental condition data at the target time point; (7) Integrate the inferred data with the measured survey data to generate a continuous three-dimensional fishery resource distribution dataset at the minute, hour, or daily scale based on a unified time step and spatial grid.

2. The method according to claim 1, wherein The environmental condition data includes at least one or more of sea surface temperature, salinity, dissolved oxygen, chlorophyll concentration, water depth, flow rate and pH value.

3. The method according to claim 1, wherein The resource-environment relationship model includes at least one of a maximum entropy model (MaxEnt), an artificial neural network (ANN), a classification tree analysis (CTA), a random forest (RF) or a habitat suitability index model (HSI).

4. The method according to claim 1, wherein When inferring the spatial distribution of fishery resources at the target time point in step (4), fishing vessel operation production data are introduced as an external validation dataset for model cross-validation and parameter optimization.

5. The method according to claim 1, wherein The spatiotemporal interpolation algorithm used in step (5) includes spline interpolation or Kriging interpolation.

6. The method according to claim 1, wherein The continuous three-dimensional fishery resource distribution dataset generated in step (7) is quality verified through error propagation analysis and uncertainty quantification methods.

7. The method according to claim 1, wherein In the preprocessing and standardization of the environmental condition data at the target time point in step (3), multi-source data fusion technology is used, including spatial resampling, scale unification, data standardization and outlier correction.

Images