A method for quantitatively estimating lake fishery resources based on multi-dimensional environmental factors

Abstract

The application discloses a method for quantitatively estimating lake fishery resources based on multi-dimensional environmental factors. Based on the publicly obtained remote sensing data and a small amount of measured data, a remote sensing model method for quantitatively estimating lake fishery resources based on multi-dimensional (hydrology, water quality and water ecology) environmental factors is constructed by using a machine learning intelligent algorithm. The method uses the lake area, water level and water volume extracted from multi-source remote sensing data as the hydrological factors of model construction; uses the transparency and suspended matter of the lake and other water quality information remotely sensed based on remote sensing as the water quality factors of model construction; and uses the satellite remote sensing inversion parameters combined with the lake water depth information to deduce the potential spawning ground and habitat distribution of fish as the water ecological factors of model construction. Finally, a set of technical method for quantitatively estimating and predicting the lake fishery resources by using the machine learning intelligent algorithm to integrate multi-dimensional environmental factors is constructed. The method can be used for spatial simulation in the whole lake area, and can be popularized to other regional lakes.

Description

Technical Field

[0001] This invention belongs to the field of resource and environmental remote sensing, and specifically relates to a method for quantitatively estimating lake fishery resources based on multidimensional environmental factors. Background Technology

[0002] Fishery resources, as an important component of natural resources, provide humans with direct edible protein; therefore, their sustainable utilization is crucial to people's livelihoods and food security. Lakes, as vital sources of freshwater fishery resources, play a significant role in simulating the impact of lake environmental changes on fishery resource levels, which is of great practical importance for maintaining lake resource environmental service functions and ecosystem stability.

[0003] Currently, domestic and international scholars and stakeholders primarily rely on manual surveys based on single-point cases when investigating lake fishery resources. For example, using expensive underwater acoustic detection or fish tagging techniques for fishery resource and habitat assessment requires significant resources, manpower, and time, and fails to reflect the overall condition and spatiotemporal differences across the entire lake area, thus limiting its widespread application on a large scale. With the rapid development of remote sensing technology, the methods and theories for investigating lake fish habitats and fishery resource characteristics have made leaps and bounds in terms of advancement and diversification. International research has shown that remote sensing technology is widely used in lake fishery resource assessment and analysis, spatial distribution, fish habitat evaluation, and fish behavior monitoring and forecasting. Domestic research focuses more on using remote sensing technology to monitor lake net cage aquaculture. While remote sensing technology has advantages in the areas of typical fish habitats and fishery resources, enabling faster and more effective acquisition of dynamic characteristics of fishery resources across dense temporal scales and broad spatial distributions, it still has enormous application potential in reflecting habitat characteristics (spawning, foraging, migration, and habitat) that directly affect fishery resource quantities.

[0004] In summary, lake environmental information is crucial for influencing lake fishery resources. Existing survey methods cannot be used to measure the entire lake area. Therefore, in the context of protecting fishery resources and promoting sustainable development, this paper proposes a method for quantitatively estimating lake fishery resources based on multidimensional environmental factors. This method has significant application value for the management and protection of lake fishery resources and the maintenance of fish population diversity. Summary of the Invention

[0005] The purpose of this invention is to overcome the bottlenecks in traditional methods for investigating aquatic habitats and fishery resources, and to propose a method for quantitatively estimating lake fishery resources based on multidimensional environmental factors (hydrological, water quality, and aquatic ecology). This method integrates remote sensing monitoring and machine learning intelligent algorithm simulation technology to achieve quantitative simulation of the impact of multidimensional lake aquatic environment information on fishery resources.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solution: Methods for quantitatively estimating lake fishery resources based on multidimensional environmental factors include: The hydrological factors, water quality factors, and aquatic ecological factors of the lake environment were obtained using satellite data, a small amount of field-measured spectral data, and GPS positioning and mapping data. Among them, the hydrological factor data are the time series of lake water area, water level, and water volume changes; the water quality factors are chlorophyll a concentration, transparency, and suspended solids concentration; and the aquatic ecological factors are the spatial distribution of potential fish spawning grounds and habitats. Based on machine learning intelligent algorithms, using one or more different permutations and combinations of the hydrological factors, water quality factors and aquatic ecological factors as input, the correlation between the estimated fishery resources and the measured fishery resources under different combinations of lake environmental factors is simulated, and typical combinations of lake environmental factors that affect fishery resources are screened out. Remote sensing models were constructed by using combinations of typical lake environmental factors from different dimensions to estimate fishery resources.

[0007] As a further improvement of the present invention, the time series of the lake's water area and water level are extracted based on satellite remote sensing images and lidar altimetry satellite data, respectively. Chlorophyll a concentration, transparency, and suspended matter concentration were obtained by inversion based on a small amount of field-measured spectral data from Landsat OLI and Sentinel-2 MSI satellite remote sensing images. The NDVI index was calculated using preprocessed Landsat satellite imagery, and the spatial distribution of potential fish spawning grounds and habitats was extracted by combining the digital elevation model (DEM) data of the water area with the time series of the water area.

[0008] As a further improvement of the present invention, a remote sensing model is constructed by combining the selected typical lake environmental factor combinations with different machine learning intelligent algorithms, and the remote sensing model is selected with the goal of optimal model performance. Based on robustness testing using simulated iterations, the optimal number of iterations for the model is determined.

[0009] As a further improvement of the present invention, a multi-index joint analysis method is adopted to obtain the accuracy of the model method, and the remote sensing model is selected with the goal of optimal model performance; the indices used include correlation coefficient, root mean square error, absolute error, and Nash coefficient.

[0010] As a further improvement of the present invention, the NDWI water index is calculated based on Landsat series / Sentinel-2 remote sensing images respectively. The Edge Otsu threshold segmentation algorithm is used to automatically obtain the water area time series based on the two remote sensing data. Then, the water area results obtained from the two remote sensing data are fused into a continuous area time series.

[0011] As a further improvement to the present invention, the method of automatically obtaining the water area time series using the Edge Otsu threshold segmentation algorithm is as follows: The grayscale range of the remote sensing image after NDWI calculation is set to [0, T]. The threshold is incremented sequentially with a fixed step size of 1. The optimal threshold is obtained when the inter-class variance reaches its maximum and is used as the Otsu threshold.

[0012] As a further improvement of the present invention, a buffer is set one pixel outward from the water-land boundary range determined by NDWI. A histogram is constructed using the NDWI values ​​within the buffer edge, and used as input to the Edge Otsu algorithm. A segmentation threshold is automatically determined from the histogram sampled from the bimodal / unimodal regions representing water and non-water. This segmentation threshold maximizes the inter-class variance, and the image can be divided into binary images of water and non-water using this segmentation threshold.

[0013] As a further improvement of the present invention, the water level time series is reconstructed by fusing water level values ​​obtained from hydrological stations and lidar altimetry satellites; The time series of water volume changes is estimated based on the reservoir capacity curve constructed using area-water level.

[0014] As a further improvement of the present invention, the method for obtaining the water quality factor is as follows: Collect measured concentration data of the water quality factors to be tested and remote sensing optical characteristic information obtained by satellite remote sensing; By combining the band response functions of Landsat-8 OLI and Sentinel-2 MSI, hyperspectral reflectance is transformed into band reflectance, and the spectral reconstruction equation is optimized with the measured spectrum to reconstruct the sensitive bands for inverting lake water quality. A water quality factor inversion model was constructed by combining the aforementioned sensitive bands and measured concentration data; Transformation coefficients were established for the water quality factor inversion results from Landsat-8 OLI and Sentinel-2 MSI, and the error scale and spatial resolution were standardized. Then, an OLI-MSI virtual constellation was constructed to obtain the spatiotemporal variation sequence of lake water quality factors. Compared with existing semi-analytical inversion models and empirical models, the advantage of the inversion method in this invention lies in reconstructing the optimal inversion bands to construct the inversion model, rather than fitting the spectrum with existing band relationships or combining bands. This improvement can compensate for the lack of available bands for inverting water quality factors in high-resolution remote sensing data (e.g., Landsat-8 OLI and Sentinel-2 MSI data), thereby improving the spatial resolution of the inversion results.

[0015] As a further improvement of the present invention, considering that fish will choose areas with a certain water depth and aquatic vegetation distribution for spawning and habitat, the distribution of potential spawning grounds and habitats can be obtained by overlaying DEM and NDVI spatial data.

[0016] As a further improvement of the present invention, in view of the differences in spawning and habitat time of different fish species, a mask cut of the water area corresponding to the spawning and habitat time of different fish species with a certain depth (DEM threshold) and aquatic vegetation distribution (NDVI threshold) can be selected to obtain the spatial distribution of potential spawning grounds and habitats of different fish species.

[0017] As a further improvement of the present invention, the small amount of field-measured spectral data includes chlorophyll a and irradiance information of suspended matter, which can increase the accuracy of remote sensing inversion of water quality factors.

[0018] As a further improvement of the present invention, the GPS positioning and mapping data includes the distribution of aquatic vegetation and the location information of the maximum boundary of the water area, which can increase the accuracy of remote sensing inversion of hydrological factors and aquatic ecological factors.

[0019] As a further improvement of the present invention, the average value of simulation results from ≥1000 iterations is preferably used as the final estimation result.

[0020] This invention has the following advantages: (1) The method for quantitatively estimating lake fishery resources based on multidimensional environmental factors proposed in this invention does not rely on a large amount of field measurement data, but can be completed mainly based on publicly available remote sensing data; (2) The algorithm of this invention is simple to implement and can quickly build models; (3) This invention can be applied to the spatial feature simulation of the entire lake area without considering the morphology and geomorphological features of the lake itself, and can be promoted to other regional lakes. Attached Figure Description

[0021] Figure 1 This is a sample area diagram provided in the example of the present invention.

[0022] Figure 2 This is a flowchart of the model construction process of the present invention.

[0023] Figure 3 This is the monthly average result of time-series remote sensing reconstruction of the lake water area in this invention example.

[0024] Figure 4 This is the remote sensing inversion result of lake water quality factors in an example of the present invention.

[0025] Figure 5 This is the remote sensing inversion result of the distribution of potential spawning grounds for fish in an example of the present invention.

[0026] Figure 6 These are simulation results under different combinations of environmental factors in the examples of this invention.

[0027] Figure 7 This is the model accuracy verification result in the example of this invention.

[0028] Figure 8 These are the robustness test results of the model in the example of this invention. Detailed Implementation

[0029] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0030] The embodiments of this application take Poyang Lake as the study area. Poyang Lake is the largest freshwater lake in my country and is connected to the main stream of the Yangtze River. Due to the frequent exchange of matter and energy between the river and the lake, Poyang Lake has abundant freshwater fishery resources, making it a representative embodiment of this invention.

[0031] like Figure 2 The diagram shown is a flowchart of an embodiment, which includes the following steps: Step 1: Remote Sensing Extraction of Hydrological Factors from Lakes in the Example. First, the NDWI water index was calculated using Landsat series / Sentinel-2 remote sensing image data, and the maximum water area was constrained using GPS-assisted mapping of the maximum water body boundary information. Then, the monthly water area of ​​Poyang Lake was extracted using the Edge Otsu threshold segmentation algorithm. Figure 3 Then, the water level time series of Poyang Lake is reconstructed by integrating the water level values ​​obtained from hydrological stations and lidar altimetry satellites; finally, the water volume change time series of Poyang Lake is estimated by referencing the reservoir capacity curve constructed from the area and water level.

[0032] Step 2: Remote Sensing Inversion of Lake Water Quality Factors (Example). The purpose of this step is to invert the water quality of the entire lake area by utilizing the relationship between the spectral characteristics of the lake environment and the concentration of water quality parameters.

[0033] Water quality factors were retrieved using Landsat-8 OLI and Sentinel-2 MSI remote sensing image data.

[0034] (1) Using the lake in the example as the test area, remote sensing optical characteristics information and chlorophyll a concentration, transparency and suspended matter concentration information of the sample area were collected through field survey and satellite-ground synchronous experiment.

[0035] (2) By combining the band response functions of OLI and MSI, the hyperspectral reflectance is transformed into band reflectance, and the spectral reconstruction equation is optimized with the measured spectrum to reconstruct the sensitive band for inverting lake water quality.

[0036] (3) Based on the spectral bands reconstructed by OLI and MSI, the inversion model of lake water quality is revised using the measured concentration of water quality parameters.

[0037] (4) Establish conversion coefficients for the water quality factor inversion results of OLI and MSI, unify the error scale and spatial resolution, and then construct an OLI-MSI virtual constellation to obtain the spatiotemporal variation sequence of lake water quality factors. Compared with existing water quality factor remote sensing inversion technologies, this implementation scheme improves the spatiotemporal resolution of the inversion results. Figure 4 ).

[0038] Step 3: Remote Sensing Implementation Example - Aquatic Ecological Factors of Lakes. The dominant fish species in Poyang Lake are carp and crucian carp. Their spawning grounds are distributed along aquatic grasslands and in shallow water areas with a depth of no more than 0.5 meters, while their habitats are generally distributed in the depth range of 2-4 meters (carp) and 3-4 meters (crucian carp).

[0039] First, NDVI was calculated using Landsat 8 series remote sensing imagery. The NDVI threshold was determined by GPS-based measurements of aquatic vegetation distribution areas, and the areas within the NDVI threshold range were extracted as the aquatic vegetation distribution areas. Second, the water areas corresponding to the spawning season (March-August) were selected as masked cropping areas for the aquatic vegetation distribution areas and areas with DEM values ​​below 0.5. These two cropped areas were then overlaid to represent the potential spawning grounds for carp and crucian carp. Furthermore, the potential habitat distribution areas for crucian carp and carp were extracted by overlaying DEM data at different depth thresholds onto water areas corresponding to the non-breeding months of September to February of the following year. Compared to existing technologies, this method independently extracts ecological factors for different fish species and different habitat characteristics, improving the accuracy and diversity of ecological factor extraction. Figure 5 ).

[0040] Step 4: Employ various machine learning algorithms (Extreme Gradient Boosting Tree, Deep Neural Network (DNN), Random Forest (RF)) to construct a model for estimating fishery resources based on remote sensing monitoring of multi-dimensional lake environmental factors. Simulation experiments on fishery resources are conducted within the constructed model, setting different combinations of lake environmental factors. Figure 6 The simulation scenarios include those involving lake environmental hydrological factors, lake environmental water quality factors, lake environmental aquatic ecological factors, lake environmental hydrological factors combined with water quality factors, lake environmental hydrological factors combined with aquatic ecological factors, lake environmental water quality factors combined with aquatic ecological factors, and lake environmental hydrological factors combined with water quality and aquatic ecological factors. These scenarios are used to simulate the impact of different dimensions of lake environmental factors on fishery resources and determine the optimal combination scenario as the three-dimensional combination of hydrological factors combined with water quality and aquatic ecological factors.

[0041] Step 5: Select the optimal machine learning intelligent algorithm ( Figure 7 Four simulation accuracy verification metrics—R, RMSE, MAE, and NSE—were selected. The simulation performance of three machine learning intelligent algorithms—XGBoost, DNN, and RF—involved in modeling was compared and verified against measured values. Based on the accuracy verification results (…),… Figure 7 (b) The overall accuracy of the model built using the XGBoost machine learning intelligent algorithm selected in this embodiment is as high as 97%, which meets the modeling conditions of this invention (accuracy > 90%). Therefore, the XGBoost machine learning intelligent algorithm is ultimately determined as the optimal algorithm for remote sensing modeling in this embodiment.

[0042] Step Six: Robustness Testing of the Model. For the remote sensing model built using the XGBoost machine learning algorithm, tests with different maximum iteration counts are conducted to find the maximum number of iterations required for robust model performance. In this embodiment, the average of robust simulation results after 1000 iterations is preferred as the final estimation result. Figure 8 ).

Claims

1. A method for quantitatively estimating lake fishery resources based on multidimensional environmental factors, characterized in that, include: Data on hydrological, water quality, and aquatic ecological factors of the lake environment were obtained using satellite data, limited field-measured spectral data, and GPS positioning and mapping data, including: Based on satellite remote sensing imagery and lidar altimetry satellite data, the time series of lake area, water level, and water volume changes were extracted as hydrological factor data. Chlorophyll a concentration, transparency, and suspended matter concentration were retrieved based on Landsat-8 OLI and Sentinel-2 MSI remote sensing images and a small amount of field-measured spectral data, and used as water quality factor data. The NDVI index was calculated using preprocessed Landsat satellite imagery. The NDVI threshold was determined by measuring the distribution of aquatic vegetation using GPS positioning data. The area within the NDVI threshold range was extracted as the aquatic vegetation distribution area. Considering the differences in spawning and habitat times for different fish species, the water area corresponding to the spawning and habitat times of different fish species was masked and cropped into aquatic vegetation distribution areas and areas with DEM values ​​below the DEM threshold. The two cropped areas were then superimposed as the spatial distribution of potential spawning grounds and habitats for the corresponding fish species, serving as aquatic ecological factor data. Based on machine learning intelligent algorithms, using one or more different permutations and combinations of the hydrological factors, water quality factors and aquatic ecological factors as input, the correlation between the estimated fishery resources and the measured fishery resources under different combinations of lake environmental factors is simulated, and typical combinations of lake environmental factors that affect fishery resources are screened out. Remote sensing models were constructed by using combinations of typical lake environmental factors from different dimensions to estimate fishery resources.

2. The method according to claim 1, characterized in that, Remote sensing models were constructed by combining typical lake environmental factors with different machine learning intelligent algorithms, and the models with the best performance were selected. Based on robustness testing using simulated iterations, the optimal number of iterations for the model is determined.

3. The method according to claim 2, characterized in that, A multi-indicator joint analysis method was used to obtain the accuracy of the model method, and the remote sensing model was selected with the goal of optimizing the model performance. The indicators used included correlation coefficient, root mean square error, mean absolute error, and Nash coefficient.

4. The method according to claim 1, characterized in that, The NDWI water index was calculated based on Landsat series / Sentinel-2 satellite remote sensing data. The Edge Otsu threshold segmentation algorithm was used to automatically extract the water area time series of the two remote sensing data. The water area time series extraction results of the two remote sensing data were then merged into a continuous water area time series.

5. The method according to claim 4, characterized in that, The method for automatically obtaining the time series of water area using the Edge Otsu threshold segmentation algorithm is as follows: The grayscale range of the remote sensing image after NDWI calculation is set to [0, T]. The threshold is incremented sequentially with a fixed step size of 1. When the inter-class variance reaches its maximum, the optimal threshold is obtained as the NDWI segmentation threshold, and finally, a binary image of water and non-water is generated.

6. The method according to claim 1, characterized in that, The water level time series is reconstructed by fusing water level values ​​obtained from hydrological stations and lidar altimetry satellites. The time series of water volume changes is estimated based on the reservoir capacity curve constructed using area-water level.

7. The method according to claim 1, characterized in that, The water quality factors are obtained in the following way: Collect measured concentration data of the water quality factors to be tested and remote sensing optical characteristic information obtained by satellite remote sensing; By combining the band response functions of Landsat-8 OLI and Sentinel-2 MSI, hyperspectral reflectance is transformed into band reflectance, and the spectral reconstruction equation is optimized with the measured spectrum to reconstruct the sensitive bands for inverting lake water quality. A water quality factor inversion model was constructed by combining the aforementioned sensitive bands and measured concentration data; Transformation coefficients were established for the water quality factor inversion results of Landsat-8 OLI and Sentinel-2 MSI, and the error scale and spatial resolution were unified. Then, the OLI-MSI virtual constellation was constructed to obtain the spatiotemporal variation sequence of lake water quality factors.

8. The method according to claim 1, characterized in that, The limited amount of field-measured spectral data includes chlorophyll a and irradiance information of suspended matter; the GPS positioning and mapping data includes the distribution of aquatic vegetation and the location information of the maximum boundary of the water area.

Images