A method for predicting wildlife migration routes based on environmental factor data and monitoring site data

Abstract

The present application relates to wild animal monitoring data prediction and simulation technical field, specifically relates to a kind of wild animal migration path prediction method based on environmental factor data and monitoring point data, S1: the digital extraction of point and path information system is carried out;S2: obtain digital orthophoto map DOM and preprocessing.S3: environmental factor data is sorted out;S4: environmental factor data and wild animal group appear and disappear point and its migration path data integration form wild animal group migration database;S5: using the way of constructing resource selection probability-resource oriented path algorithm (RSF-RGPA), migration path simulation model is established;S6: based on the model obtained in S5 analysis point, constantly iteration, to the end, form path;The present application is integrated by systematically environmental factor data and wild animal appear and disappear point data, the present application uses the resource selection probability-resource oriented path algorithm (RSF-RGPA) constructed, greatly improve the accuracy of migration path prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wild animal monitoring data prediction and simulation, and particularly relates to a wild animal migration path prediction method based on environmental factor data and monitoring point data. BACKGROUND

[0002] Wild animal migration route prediction has important ecological protection, species protection, ecological research, agricultural and resource management, and climate change research significance and application value. It plays an important role in promoting sustainable development, protecting natural ecosystems and human well-being.

[0003] Existing animal migration route prediction is often carried out by researchers such as scientists based on experience combined with some monitoring tools, such as:

[0004] 1. GPS tracking: By carrying GPS devices on animals, the position information of animals can be tracked and recorded in real time. These devices can transmit data through satellites to obtain detailed animal migration routes and behavior data.

[0005] 2. Satellite remote sensing: Use satellite remote sensing technology to obtain ground information such as vegetation type and ocean temperature. By analyzing satellite images and remote sensing data, the environmental preferences and route selection of animal migration can be revealed.

[0006] 3. Biological markers: By implanting or binding various biological marker devices such as rings, wireless tags and chips on animals, the position information of animals can be tracked and recorded in real time. These marker devices can provide valuable migration route and behavior data. This method is very dependent on the experience level of researchers, which leads to large differences in the accuracy of the paths predicted by different personnel and cannot be unified. SUMMARY

[0007] The purpose of the present application is to provide a wild animal migration path prediction method based on environmental factor data and monitoring point data.

[0008] In order to achieve the above technical purpose and achieve the above technical effect, the present application is realized by the following technical scheme:

[0009] A wild animal migration path prediction method based on environmental factor data and monitoring point data, comprising the following steps:

[0010] S1: The collected wild animal occurrence point and migration path information in the region are subjected to point and path information system digitization extraction work;

[0011] S2: Obtain the digital orthographic image map DOM with an initial resolution of 2m obtained by the China Resource Satellite Application Center, and perform remote sensing image preprocessing.

[0012] S3: Collate environmental factor data in remote sensing images through visual interpretation, including but not limited to terrain, population, NDVI, road network, food source distribution; establish a high-precision basic data set;

[0013] S4: Integrate the environmental factor data obtained in S3 and the wild animal occurrence point data and migration path data to form a wild animal migration database;

[0014] S5: Based on the database obtained in S4, a resource selection probability-resource oriented path algorithm is used to establish a migration path simulation model;

[0015] S6: Based on the model obtained in S5, the probability of the existence of the animal at each location under the influence of each environmental factor is analyzed according to the actual migration point of the wild animal, the probability of the six points around each point is determined, and the score value is obtained by combining the probability and the migration cost. The point with high score value is determined as the next point, and the iteration is continuously carried out until the end point is reached to form the path.

[0016] Further, the step S1 specifically comprises the following steps:

[0017] S1.1: Determine the migration behavior characteristics of the species by clearly defining the target animal species and the environmental factors;

[0018] S1.2: Select reliable data sources, including but not limited to national monitoring databases, scientific literature, and satellite tracking data;

[0019] S1.3: Convert the data into a unified format, including GeoJSON or Shapefile, to ensure compatibility of the data on the GIS platform;

[0020] S1.4: Extract the point information of the collected records, and through analysis and strict screening and verification of the point, abnormal points are excluded; the position accuracy of the collected data is corrected according to the known geographical markers to eliminate measurement errors;

[0021] S1.5: Based on the confirmed point information, integrate data from different time periods and sources to form a coherent historical data set, and record the data source and time stamp.

[0022] Further, the step S2 specifically comprises the following steps:

[0023] S2.1: Obtain the initial resolution of 2m digital orthographic image DOM obtained by China Resource Satellite Application Center, and pre-process the data by Resource 3 remote sensing image. First, data preparation, download the original image data package containing panchromatic (PAN) image and multispectral (MS) image from Resource 3 satellite. Resource 3 satellite data usually includes stereo imaging image, which is suitable for DEM extraction. Then, the image is radiometrically corrected to reduce atmospheric influence and adjust the brightness and contrast of the image to ensure the accuracy of the spectral information of the image. Geometric correction is performed on the image using known ground control points (GCP) and satellite orbit data to correct the spatial offset of the image and align it with the geographic coordinate system. In order to improve the spatial resolution of the image, PAN (panchromatic image, high resolution) and MS (multispectral image, low resolution) are fused. Using IHS (hue-saturation-brightness) fusion or other methods, high-resolution color images are generated. The image is orthorectified using DEM to eliminate image distortion caused by terrain undulations. After orthorectification, the results are saved as digital orthographic image (DOM), usually in GeoTIFF format. Resource 3 satellite has stereo imaging capability, and the front and rear images obtained can be used for stereo image matching. Use photogrammetry software (such as ERDAS IMAGINE or PCI Geomatica). Image stereo matching generates a disparity map, which is a displacement difference map between two images. Using the disparity map, calculate the height data to generate a preliminary DEM. The software will calculate the height value of each pixel according to the disparity between the stereo images using photogrammetry principles. The final accuracy of the digital orthographic image DOM of the region is 2m and the digital elevation model (DEM) is 5m, which shows the terrain distribution in the study area. Through further operation of DEM data, DEM is smoothed to remove noise and outliers to improve accuracy. Use known ground elevation data (such as GPS measurement data) to verify and correct the accuracy of DEM. Manually edit DEM to repair areas with obvious errors to ensure the accuracy of DEM data. Output DEM. Save the final corrected DEM data in a common format (such as GeoTIFF) for use in GIS software, and obtain the terrain undulation data;

[0024] S2.2: Verify the acquisition time, sensor type and resolution information of the image to ensure the freshness and applicability of the data.

[0025] S2.3: Use radiation correction model to reduce atmospheric influence, considering the influence factors of aerosol, cloud and water vapor in the correction process.

[0026] S2.4: Splice and crop multiple adjacent images to ensure that the resulting image covers the entire study area.

[0027] Furthermore, step S3 specifically includes the following steps:

[0028] S3.1: Environmental factor data related to wildlife migration path analysis were collected through multiple channels, including but not limited to topography, vegetation cover, human activities, water resources, and road network factors. Important environmental factors were identified, including vegetation cover type, human activities, water resources, and road network. To ensure high data accuracy and spatial consistency, remote sensing image processing, digital elevation model (DEM) calculation, and other spatial analysis methods were used to extract and process multiple key environmental factors. Expert knowledge and references were used to determine the influence weight of each factor.

[0029] S3.2: Environmental factor data reflects the comprehensive ecological environment affecting the migration of wildlife populations. Based on publicly available data from the Open StreetMap platform, combined with visual interpretation and field observation, the distance to water sources was obtained and further corrected to ensure the accuracy and spatial consistency of water source data.

[0030] S3.3: Obtain food source data for wild animals from the Plant Science Data Center of the Chinese Academy of Sciences, including but not limited to corn, bamboo forests, orchards, and sugarcane. Through field investigation and DOM reading and identification, and through visual interpretation and correction, obtain the final food source data with a resolution of 2m.

[0031] S3.4: Obtain plantation data for this region within 1km from the 1km grid land use data dataset of China, including land type data for tea gardens and coffee gardens. Through field investigation and DOM reading and identification, and through visual interpretation and correction, obtain the final plantation data with a resolution of 2m.

[0032] S3.5: Obtain grassland and shrub data from the national 1m land use data, and obtain grassland and shrub data with a final resolution of 2m for use;

[0033] S3.6: Obtain natural mixed forest data from the national 1m land use data, identify it through field investigation and DOM reading, and correct it through visual interpretation to obtain the final natural mixed forest data with a resolution of 2m.

[0034] S3.7: Rubber forest data were obtained from the 1km grid land use data set in China. Through field investigation and DOM reading and identification, and through visual interpretation and correction, the final resolution of the rubber forest data was obtained to be 2m.

[0035] S3.8: Use a classification algorithm to classify images, verify the accuracy of the classification results, and perform manual correction;

[0036] S3.9: Convert the interpretation results into vector format to facilitate integration with other geographic data, and check the resolution and accuracy of the data to ensure that it meets research needs;

[0037] S3.10: Use field survey data to verify the interpretation results, make necessary corrections, establish metadata descriptions of factor data, and record the data acquisition, processing, and verification process.

[0038] Furthermore, step S4 specifically includes the following steps:

[0039] S4.1: Design the database structure, including tables, fields, and indexes, and determine the spatial data format and the storage method for non-spatial data.

[0040] S4.2: Import environmental factor data and animal movement data into the database, and verify the integrity and consistency of the data through SQL queries and spatial analysis tools;

[0041] S4.3: Spatial connectivity links different datasets to determine the topological relationships of spatial data and ensure spatial consistency of data.

[0042] Furthermore, step S5 specifically includes the following steps:

[0043] S5.1: Load data on locations visited and not visited by wild animals;

[0044] S5.2: Merge data from two locations;

[0045] S5.3: Extract environmental factor data from environmental raster files;

[0046] S5.4: Standardize the environmental factor data to obtain standardized continuous variables;

[0047] S5.5: Define the environment independent variable x and select the dependent variable y;

[0048] S5.6: Fit the logistic regression model to it. For the simulation of wildlife migration routes, the known migration points are taken as "used" points and the rarely existing points are taken as "unused" points. Based on the standardized environmental factors, the impact of various environmental factors is judged by logistic regression.

[0049] A log-linear model was used to quantify the attractiveness and hindering effects of each factor:

[0050] w(x) = pred_prob = exp(β1x) 1i +β2x 2i +...+β n x ni (1)

[0051] Where x represents the resource selection weight. ni β is the value of the i-th environmental factor at the n-th location. n represents the corresponding regression coefficient.

[0052] w(x): Represents the resource selection function; pred_prob: Represents the resource selection probability;

[0053] exp: represents the natural exponential function, i.e., e^(-1 / 2) x , where e is the base of the natural logarithm, approximately equal to 2.71828;

[0054] β1, β2, ..., β n These are model parameters, which are coefficients to be estimated, representing the degree of influence of different environmental variables or covariates on the probability of resource selection.

[0055] x 1i x 2i ...x ni : These are covariates, which are environmental characteristics that influence an animal's choice of a specific resource unit, including vegetation type, terrain slope, and distance from water source.

[0056] This function takes the form of an exponential model, mapping a linear combination of covariates to a probability. In resource selection studies, it is typically assumed that the probability of selecting a resource unit is exponentially related to the values ​​of the covariates. If the coefficient of a covariate is positive, then an increase in that covariate will increase the probability of the resource unit being selected; if the coefficient is negative, then the opposite is true.

[0057] S5.7: Output the resource selection probability formula and model validation results. During the fitting process, it is necessary to ensure that the model can fully explain the variation in the data and avoid overfitting. Overfitting refers to a situation where the model performs well on the training data but poorly on new data. To avoid this, techniques such as cross-validation can be used to evaluate the model's generalization ability.

[0058] After fitting, the goodness of fit of the model needs to be evaluated. This can be achieved by calculating statistical indicators such as the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). These indicators comprehensively consider the model's complexity and goodness of fit, and can help select the best model.

[0059] S5.8: Construct the Resource-Guided Path Algorithm (RGPA), a new algorithm developed for simulating wildlife migration paths. It is based on the resource selection function algorithm. The algorithm's logic is as follows: based on the resource selection function probability and the current location of the wildlife, analyze the surrounding areas gradually moving towards the migration endpoint; calculate the path cost of wildlife migration at each step, and consider the influence of positive attraction factors and negative obstacle factors, thereby simulating the actual migration path of wildlife.

[0060] S5.9: When constructing RGPA, the influence of spatial scale also needs to be considered. Different spatial scales (such as 1m*1m and 30m*30m) will have a significant impact on the simulation results. To determine the optimal spatial scale, we tried using different sizes of grid points to divide the study area, and simulated routes using RGPA respectively. Then, we compared the performance of these models in terms of prediction accuracy and interpretability, and selected the best-performing grid size for calculation. The grid size is not restricted, but a grid size suitable for the wildlife being applied is selected (if the wildlife being applied is small in size or has a small range of movement, a smaller grid size is selected). Here, we choose 30m*30m as the optimal scale for the final pathfinding grid.

[0061] S5.10: Calculate the probability value of each grid point being selected on a 30m*30m grid using the RSF model. These probability values ​​will be used as the basis for subsequent path generation.

[0062] S5.11: Determine the starting point (P) start ) and endpoint (P end The specific location of each point in the grid system is identified by its ID, and path planning is performed for these two points.

[0063] S5.12: Construct a hexagonal honeycomb structure of neighboring points, i.e., search for and obtain the 6 neighboring points of the current point. Determine if the neighboring points are empty; if yes, end the algorithm; otherwise, continue the algorithm. Starting from the initial point (P)... start ) is the current point (P) current Consider its six adjacent grid points as neighboring points (P). neighbors )

[0064]

[0065]

[0066] P current : Current point; P neighbors Neighboring points

[0067] d1: Distance between the current point and its nearest neighbors;

[0068] S5.13: Score the 6 neighboring points: resource selection probability × migration cost. The score is calculated based on the usage probability pred_prob of each neighboring point, calculated using the RSF model, and combined with the directional information of these points relative to the destination (i.e., which direction is closer to the destination).

[0069]

[0070] score neighbor =pred_prob neighbor ×d(6)

[0071] Prioritize selecting neighboring points with higher scores (higher probability and closer to the finish line) as the next candidate point.

[0072] Where: d2: distance from the destination, i.e., migration cost; d: migration cost coefficient.

[0073] pred_prob neighbor Resource selection probability for each neighboring point

[0074] score neighbor : The score value of each neighboring point;

[0075] S5.14: Iteration and Termination Condition. Select the nearest neighbor P with the highest score. neighbor As the next current point P current Repeat the above search process.

[0076] P next =ChooseNext(P current ,P neighbor1 ,P neighbor2 ,...)(7)

[0077] Until the current point is the destination P end Or it is impossible to find a next point P that satisfies the conditions. next The search terminates when the destination is reached; if the destination is successfully reached or the position is close enough to the destination, the algorithm ends and outputs the sequence of path points stored in the path list.

[0078] L = LineString(P start ,P next1 ,P next2 ,...P end (8)

[0079] S5.15: The generated path may contain some unnecessary loops or redundant points. Optimize the path through smoothing processes (such as linear interpolation, Bézier curves, etc.) to make it more natural and efficient. Simultaneously, further analyze and evaluate the path according to actual needs, such as calculating the total path length and the proportion of different habitat types traversed.

[0080] Furthermore, step S6 specifically includes the following steps:

[0081] S6.1: Set the starting point and potential ending point, initialize the current point as the starting point, set the initial path, and define the starting conditions;

[0082] S6.2: Obtain the six neighboring points of the current location of the wildlife herd and determine whether the six neighboring points are empty. If the neighboring points are empty, end the algorithm; if the neighboring points are not empty, analyze the surrounding areas as they gradually move towards the migration endpoint, use Monte Carlo simulation to analyze the environmental adaptability probability of each location, implement neighborhood analysis on the GIS platform, and calculate the migration probability of the six surrounding locations.

[0083] S6.3: Initially, algorithms such as A*, Dijkstra's algorithm, or circuit theory were used to find the optimal path. However, these methods were ineffective. For example, circuit theory algorithms failed to achieve the required accuracy for long-distance biological migration. The A* algorithm is mainly suitable for path planning in static road networks, requiring replanning after environmental changes. Furthermore, its computational efficiency is low because it cannot effectively utilize previously calculated information. It also has limited applicability in high-dimensional spaces due to the large computational burden. In complex maps, the paths searched by the A* algorithm may generate redundant turning points, leading to frequent turning of the mobile robot and increased losses. Additionally, the global optimality of the A* algorithm is limited. Due to the constraint of the 8-neighbor search node strategy, the planned path may not satisfy global optimality, reducing pathfinding efficiency. Therefore, the RSF algorithm was chosen, and an RGPA algorithm innovation was made based on it. This innovation abandoned the 8-neighbor search node structure and created a honeycomb structure with 6 neighbor search nodes, ensuring that the distance from the current point to its 6 neighboring points is equal. Considering the resource selection probability and the overall migration cost, a neighboring point score is performed to select the optimal neighboring point. The most important thing is to ensure that the resource selection probability is high and that the point is close to the destination. The path cost of migration is calculated at each step, and the effects of positive attraction factors and negative obstacle factors are considered.

[0084] S6.4: Add the optimal point to the path, update the current point to the optimal point, determine whether the current point is the destination, if the current point is not the destination, get the 6 neighboring points around the current point, and repeat the operations from S6.2 to S6.4; if the current point is the destination, use the path point set to create a path. During the iteration process, record the conditions and results of each path selection, and save the path.

[0085] S6.4: This simulates the actual migration paths of wildlife populations, verifies the ecological rationality of the generated paths, compares them with actual observation data, and corrects the path prediction results through expert review and further on-site investigation.

[0086] The beneficial effects of this invention are:

[0087] By systematically integrating environmental factor data with wildlife location data, this invention significantly improves the accuracy of migration path prediction using a constructed Resource Selection Probability-Resource Directed Path Algorithm (RSF-RGPA). The RSF model quantifies the effects of environmental factors through logistic regression, enabling the identification and assessment of the impact of various environmental variables (such as vegetation type, topography, roads, and human activities) on animal migration behavior. This scientific quantification method avoids the limitations of traditional methods relying on human experience and judgment. Through maximum likelihood estimation, the model can adaptively adjust to different species and environmental conditions, ensuring the scientific validity and consistency of the prediction results.

[0088] Using digital orthophoto maps (DOMs) with an initial resolution of 2 meters, and employing a series of preprocessing techniques (including geometric correction, radiometric correction, and image stitching) to ensure high-precision image data, forms the basis for path prediction. Spatial database technologies (such as PostGIS or SpatiaLite) enable efficient integration and management of environmental factors and animal movement data, ensuring data integrity and consistency. This high-precision data not only improves the reliability of model input but also provides detailed spatial contextual support for analyzing wildlife behavior in different environments.

[0089] The RSF-RGPA model effectively enhances its robustness by flexibly adapting to different environmental and species characteristics. Through logistic regression, the probability of animal behavior under given environmental conditions can be quantified, identifying key influencing factors and their relative importance. This adaptive capability makes the model less susceptible to changes in a single factor, thus maintaining stable performance across different geographical and ecological contexts and providing reliable migration route predictions.

[0090] In the path generation process, combining resource selection probability and neighborhood analysis, and considering migration costs, algorithms such as A*, Dijkstra's algorithm, or circuit theory were initially used to find the optimal path. However, due to unsatisfactory pathfinding results, these algorithms were abandoned, and the RSF algorithm was chosen instead. Based on this algorithm, the RGPA algorithm was innovated, abandoning the structure of 8 neighboring search nodes and innovating a honeycomb structure of 6 neighboring search nodes, ensuring that the distance from the current point to its 6 neighboring points is equal. This method intelligently selects the optimal path point by analyzing the probability and cost of multiple potential path points, ensuring not only the shortest or optimal path distance but also ecological rationality. Furthermore, through comparison and verification with actual observation data and expert review, the generated paths are confirmed to conform to biological logic and can effectively guide ecological research and species conservation. In this way, path prediction is not only mathematically sound but also has practical application value in ecology and conservation practice, providing a scientific basis for conservation decisions.

[0091] Xishuangbanna and Pu'er regions are major migration areas for wild animals. Existing ecological environment research and wildlife monitoring systems are relatively mature, and data on wildlife migration routes and related environmental factors are abundant and reliable. This invention utilizes existing ecological monitoring data and completed field surveys, including data on food sources, water sources, and wildlife tracking. Furthermore, by processing remote sensing imagery from the ZY-3 satellite, high-precision 2-meter resolution DOM data was obtained, providing accurate topographic and environmental information and improving road network surveys in Yunnan Province. These findings provide a solid foundation for constructing a high-precision migration database.

[0092] By constructing a high-precision simulation model of wildlife migration routes and visualizing the 3D model, the research findings will provide a scientific basis for reducing human-wildlife conflict and optimizing ecological protection strategies. The model's dynamic prediction capabilities and visualization will not only help managers better understand wildlife herd migration behavior but also provide strong support for formulating regional ecological protection policies. Furthermore, the introduction of digital twins will drive technological innovation in the field of ecological protection, possessing broad application prospects and academic value.

[0093] This technology still has many shortcomings. While the current path simulation accuracy can be maintained within the hundred-meter range, further improving it to the ten-meter level remains a challenge. This difficulty may stem from the relatively small size of the designated experimental area, leading to strong spatial correlations. Furthermore, the environments in which biological path simulations take place are typically complex and variable, which also increases the difficulty of the simulations.

[0094] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0095] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0096] Figure 1 This is a schematic diagram of the overall algorithm of the present invention;

[0097] Figure 2 This is a schematic diagram of (partial) environmental factor data for the study area of ​​this invention;

[0098] Figure 3 This is a schematic diagram of the simulated path in the experimental area of ​​this invention. Detailed Implementation

[0099] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0100] Example 1

[0101] This embodiment describes a method for predicting wildlife migration routes based on environmental factor data and monitoring site data. The wildlife targeted in this embodiment is the Asian elephant, and the method includes the following steps:

[0102] S1: In order to explore the activity patterns of Asian elephants in depth, the location and migration path information of Asian elephants in this area will be systematically extracted digitally.

[0103] S2: Obtain the digital orthophoto map (DOM) with an initial resolution of 2m obtained from the China Center for Resources Satellite Data and Application, and perform remote sensing image preprocessing.

[0104] S3: Visually interpret and organize environmental factor data, including topography, population, NDVI, road network, and food source distribution, in remote sensing images to establish a high-precision basic dataset;

[0105] S4: Integrate the environmental factor data obtained in S3 with the location data of Asian elephants and their migration routes to form an Asian elephant migration database, ensuring the spatial accuracy and completeness of the data;

[0106] S5: Based on data collection and processing, a migration path simulation model is established by constructing the Resource Selection Function-Resource-Guided Path Algorithm (RSF-RGPA).

[0107] S6: Based on the model obtained from S5, the probability of the presence of the Asian elephant at each location under the influence of various environmental factors is analyzed according to the actual migration points of the Asian elephant. The probability of the six surrounding points of each location is determined. Combining the probability and migration cost, a score value is obtained. The location with the higher score value is determined as the next location. This process is iterated until the destination is reached, thus forming a path.

[0108] In this embodiment, step S1 specifically includes the following steps:

[0109] S1.1: By clearly identifying the target animal species, the Asian elephant, and defining the border area between Pu'er and Xishuangbanna as the study area, specifically the Simao-Jinghong-Jiangcheng border zone, the study aims to explore the activity patterns of the Asian elephant population in this area. Through literature review and expert consultation, the study shows that the migration and habitat selection of Asian elephants at the border of Pu'er and Xishuangbanna are influenced by a variety of environmental factors, including the distribution of food resources, topographic features, and mineral requirements, which constitute their unique ecological behavior and migration patterns. The study also identifies the migratory behavior characteristics of this species.

[0110] S1.2: Select reliable data sources, such as national monitoring databases, scientific literature, satellite tracking data, etc.

[0111] S1.3: Convert the data into a uniform format, such as GeoJSON or Shapefile, to ensure data compatibility on GIS platforms.

[0112] S1.4: The location information of the collected records was extracted, and a total of 245 Asian elephant sighting locations were extracted. Then, through analysis and strict screening and verification of the locations, 233 Asian elephant sighting locations were finally retained. Known geographical features were used to correct the location accuracy of the collected data to eliminate measurement errors.

[0113] S1.5: Based on the confirmed location information, data from different time periods and sources are integrated to form a coherent historical dataset, and the data source and timestamp are recorded. In this embodiment, 26 major migration routes of Asian elephants are identified and mapped. These routes clearly show the activity trajectory of Asian elephants in the study area.

[0114] In this embodiment, step S2 specifically includes the following steps:

[0115] S2.1: The acquisition primarily involved obtaining a DOM (Digital Orthophoto Map) with an initial resolution of 2m from the China Resources Satellite Application Center. This DOM was preprocessed using ZY-3 remote sensing imagery to produce a final DOM with a resolution of 2m and a 5-meter Digital Elevation Model (DEM) for the study area, showcasing the topographic distribution within the area. Further calculations on the DEM data yielded topographic relief data, reflecting the magnitude of topographic changes at different locations within the study area, which helps in understanding the impact of topographic changes on Asian elephant migration routes.

[0116] S2.2: Verify the image acquisition time, sensor type, resolution, and other information to ensure the freshness and applicability of the data.

[0117] S2.3: Use radiation correction models (such as FLAASH) to reduce atmospheric effects, taking into account the influence of aerosols, clouds and water vapor during the correction process.

[0118] S2.4: Stitch and crop multiple adjacent images to ensure that the resulting image covers the entire study area.

[0119] In this embodiment, step S3 specifically includes the following steps:

[0120] S3.1: Environmental factor data related to the analysis of Asian elephant migration routes were collected through multiple channels, covering factors such as topography, vegetation cover, human activities, water resources, and road networks. Important environmental factors, such as vegetation cover type, human activities, water resources, and road networks, were identified. To ensure high data accuracy and spatial consistency, this embodiment used remote sensing image processing, digital elevation model (DEM) calculation, and other spatial analysis methods to extract and process multiple key environmental factors, and used expert knowledge and references to determine the influence weight of each factor.

[0121] S3.2: Environmental factor data mainly includes topographic relief, slope, aspect, vegetation cover (NDVI), distance to water sources, distribution of food sources such as orchards and bamboo forests, population density, and road network distribution, aiming to reflect the comprehensive ecological environment affecting elephant migration. Water sources provide crucial survival conditions for Asian elephants, especially during the dry season or during migration, when they preferentially choose areas close to water sources. The existence of road networks may hinder or guide the migration path of the elephant herd; in particular, the density of roads and traffic flow affect the activity range of the elephant herd. Based on publicly available data from the Open Street Map platform (2016), combined with visual interpretation and field observation, the distance to water sources was obtained and further corrected to ensure the accuracy and spatial consistency of the water source data.

[0122] S3.3: Obtained food source data of wild animals published in the Flora of China in 2022 by the Plant Science Data Center of the Chinese Academy of Sciences, including maize, bamboo forests, orchards, sugarcane, etc., and obtained the final food source data with a resolution of 2m through field investigation and DOM reading and identification, and through visual interpretation and correction.

[0123] S3.4: Obtain plantation data for this region within 1km from the 2020 China 1km grid land use data set, including land types such as tea gardens and coffee plantations. Through field investigation and DOM reading and identification, and through visual interpretation and correction, obtain the final plantation data with a resolution of 2m.

[0124] S3.5: Obtain grassland and shrub data from the national 1m land use data (SinoLC-1) to obtain grassland and shrub data with a final resolution of 2m for use.

[0125] S3.6: Obtain natural mixed forest data from the national 1m land use data (SinoLC-1), identify them through field surveys and DOM reading, and correct them through visual interpretation to obtain the final natural mixed forest data with a resolution of 2m.

[0126] S3.7: Rubber forest data were obtained from the 2020 China 1km grid land use data set. Through field investigation and DOM reading and identification, and through visual interpretation and correction, the final resolution of the rubber forest data was obtained to be used.

[0127] S3.8: Use classification algorithms (such as maximum likelihood method, support vector machine) to classify images, verify the accuracy of the classification results, and perform manual correction.

[0128] S3.9: Convert the interpretation results into a vector format for easy integration with other geographic data. Check the resolution and accuracy of the data to ensure it meets research needs.

[0129] S3.10: Use field survey data to verify the interpretation results, make necessary corrections, establish metadata descriptions of factor data, and record the data acquisition, processing, and verification process.

[0130] In this embodiment, step S4 specifically includes the following steps:

[0131] S4.1: Design the database structure, including tables, fields, and indexes, determine the spatial data format (such as PostGIS or SpatiaLite), and the storage method for non-spatial data.

[0132] S4.2: Import environmental factor data and animal movement data into the database, and verify the integrity and consistency of the data through SQL queries and spatial analysis tools.

[0133] S4.3: Spatial joins are used to link different datasets, determine the topological relationships of the spatial data, and ensure the spatial consistency of the data.

[0134] In this embodiment, step S5 specifically includes the following steps:

[0135] S5.1: Load data on Asian elephant traversal and non-traversal points;

[0136] S5.2: Merge data from two locations;

[0137] S5.3: Extract environmental factor data from environmental raster files;

[0138] S5.4: Standardize the environmental factor data to obtain standardized continuous variables;

[0139] S5.5: Define the environment independent variable x and select the dependent variable y;

[0140] S5.6: Fitting a logistic regression model to the model. For the simulation of Asian elephant migration routes, since we are concerned with the probability of Asian elephants choosing habitats under different environmental conditions, GLM is a commonly used choice. In particular, the logistic regression model is suitable for handling binary response variables (i.e., "used" and "unused" resource units) and can estimate model parameters through maximum likelihood estimation. Known migration points are used as "used" points, and rarely existing points are used as "unused" points. Based on standardized environmental factors (such as water sources, food sources, roads, distribution of human activities, etc.), logistic regression is used to determine the impact of various environmental factors.

[0141] A log-linear model (as shown in the formula below) is used to quantify the attractiveness and hindering effects of each factor:

[0142] w(x) = pred_prob = exp(β1x) 1i +β2x 2i +...+β n x ni (1)

[0143] Where x represents the resource selection weight. ni β is the value of the i-th environmental factor at the n-th location. n represents the corresponding regression coefficient.

[0144] w(x): Represents the resource selection function; pred_prob: Represents the resource selection probability;

[0145] exp: represents the natural exponential function, i.e., e^(-1 / 2) x, where e is the base of the natural logarithm, approximately equal to 2.71828;

[0146] β1, β2, ..., β n These are model parameters, which are coefficients to be estimated, representing the degree of influence of different environmental variables (or covariates) on the probability of resource selection.

[0147] x 1i x 2i ...x ni : These are covariates, which are environmental characteristics that influence an animal's choice of a specific resource unit, such as vegetation type, terrain slope, and distance from water source.

[0148] This function takes the form of an exponential model, mapping a linear combination of covariates to a probability. In resource selection studies, it is typically assumed that the probability of selecting a resource unit is exponentially related to the values ​​of the covariates. If the coefficient of a covariate is positive, then an increase in that covariate will increase the probability of the resource unit being selected; if the coefficient is negative, then the opposite is true.

[0149] S5.7: Output the resource selection probability formula and model validation results. During the fitting process, it is necessary to ensure that the model can fully explain the variation in the data and avoid overfitting. Overfitting refers to a situation where the model performs well on the training data but poorly on new data. To avoid this, techniques such as cross-validation can be used to evaluate the model's generalization ability.

[0150] After fitting, the goodness of fit of the model needs to be evaluated. This can be achieved by calculating statistical indicators such as the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). These indicators comprehensively consider the model's complexity and goodness of fit, and can help select the best model.

[0151] S5.8: Construct the Resource-Guided Path Algorithm (RGPA), a new algorithm developed for simulating Asian elephant migration paths. It is based on the resource selection function algorithm. The algorithm's logic is as follows: based on the resource selection function probability and the current location of the Asian elephant, analyze the surrounding areas gradually moving towards the migration endpoint; calculate the path cost of the Asian elephant's migration at each step, and consider the influence of positive attraction factors and negative obstacle factors, thereby simulating the actual migration path of wild animals.

[0152] S5.9: When constructing RGPA, the influence of spatial scale also needs to be considered. Different spatial scales (such as 1m*1m and 30m*30m) will have a significant impact on the simulation results. To determine the optimal spatial scale, we tried using different sizes of grid points to divide the study area, and simulated routes using RGPA respectively. Then, we compared the performance of these models in terms of prediction accuracy and interpretability, and selected the best-performing grid size for calculation. The grid size is not restricted, but a grid size suitable for the wildlife being applied is selected (if the wildlife being applied is small in size or has a small range of movement, a smaller grid size is selected). Here, we choose 30m*30m as the optimal scale for the final pathfinding grid.

[0153] S5.10: Calculate the probability value of each grid point being selected on a 30m*30m grid using the RSF model. These probability values ​​will be used as the basis for subsequent path generation.

[0154] S5.11: Determine the starting point (P) start ) and endpoint (P end The specific location of each point in the grid system is identified by its ID, and path planning is performed for these two points.

[0155] S5.12: Search and obtain the 6 neighboring points of the current point. Check if the neighboring points are empty; if so, end the algorithm; otherwise, continue the algorithm. Starting from the initial point (P) start ) is the current point (P) current Consider its six adjacent grid points as neighboring points (P). neighbors )

[0156]

[0157] P current : Current point; P neighbors Neighboring points

[0158] d1: Distance between the current point and its nearest neighbors;

[0159] S5.13: Score the 6 neighboring points: resource selection probability × migration cost. The score is calculated based on the usage probability pred_prob of each neighboring point, calculated using the RSF model, and combined with the directional information of these points relative to the destination (i.e., which direction is closer to the destination).

[0160]

[0161] score neighbor =pred_prob neighbor ×d(6)

[0162] Prioritize selecting neighboring points with higher scores (higher probability and closer to the finish line) as the next candidate point.

[0163] Where: d2: distance from the destination, i.e., migration cost; d: migration cost coefficient.

[0164] pred_prob neighbor Resource selection probability for each neighboring point

[0165] score neighbor : The score value of each neighboring point;

[0166] S5.14: Iteration and Termination Condition. Select the nearest neighbor P with the highest score. neighbor As the next current point P current Repeat the above search process.

[0167] P next =ChooseNext(P current ,P neighbor1 ,P neighbor2 (7)

[0168] Until the current point is the destination P end Or it is impossible to find a next point P that satisfies the conditions. next The search terminates when the destination is reached; if the destination is successfully reached or the position is close enough to the destination, the algorithm ends and outputs the sequence of path points stored in the path list.

[0169] L = LineString(P start ,P next1 ,P next2 ,…P end (8)

[0170] S5.15: The generated path may contain some unnecessary loops or redundant points. Optimize the path through smoothing processes (such as linear interpolation, Bézier curves, etc.) to make it more natural and efficient. Simultaneously, further analyze and evaluate the path according to actual needs, such as calculating the total path length and the proportion of different habitat types traversed.

[0171] In this embodiment, step S6 specifically includes the following steps:

[0172] S6.1: Set the starting point and potential ending point, initialize the current point as the starting point, set the initial path, and define the starting conditions;

[0173] S6.2: Obtain the six neighboring points of the current location of the elephant herd and determine whether the six neighboring points are empty. If the neighboring points are empty, end the algorithm; if the neighboring points are not empty, analyze the surrounding areas as they gradually move towards the migration endpoint, use Monte Carlo simulation to analyze the environmental adaptability probability of each location, implement neighborhood analysis on the GIS platform, and calculate the migration probability of the six surrounding locations.

[0174] S6.3: The RSF algorithm was chosen, and RGPA was innovated upon it. The original structure of 8 neighboring search nodes was abandoned, and a new honeycomb structure of 6 neighboring search nodes was created, ensuring that the distance from the current point to its 6 neighboring points is equal. Neighboring points are scored considering resource selection probability and overall migration cost to select the optimal neighboring point. Most importantly, the resource selection probability is high, and the point is close to the destination. The path cost of elephant herd migration is calculated at each step, taking into account the influence of positive attraction factors and negative obstacle factors.

[0175] S6.4: Add the optimal point to the path, update the current position of the elephant herd to the optimal point, determine whether the current point is the destination, if the current point is not the destination, obtain the 6 neighboring points around the current point, and repeat the operations from S6.2 to S6.4; if the current point is the destination, use the path point set to create a path. During the iteration process, record the conditions and results of each path selection, and save the path.

[0176] S6.4: This simulates the actual migration path of the elephant herd, verifies the ecological rationality of the generated path, compares it with actual observation data, and corrects the path prediction results through expert review and further on-site investigation.

[0177] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1.A method for predicting a migration path of a wild animal based on environmental factor data and monitoring point data, the method comprising: obtaining environmental factor data and monitoring point data; and predicting a migration path of a wild animal based on the environmental factor data and the monitoring point data. It comprises the following steps: S1: The collected wild animal occurrence point and migration path information in the region is subjected to digital extraction of point and path information system; S2: Obtain the digital orthographic image (DOM) with an initial resolution of 2 m obtained by the China Resource Satellite Application Center, and perform remote sensing image preprocessing; S3: The environmental factor data in the remote sensing image is sorted out by visual interpretation, including but not limited to terrain, population, NDVI, road network, and food source distribution; and a high-precision basic data set is established; S4: The environmental factor data obtained in S3 and the wild animal occurrence point and migration path data are integrated to form a wild animal migration database; S5: Based on the database obtained in S4, a resource selection probability-resource oriented path algorithm is used to establish a migration path simulation model; S6: Based on the model obtained in S5, the probability of the existence of the animal at each location under the influence of various environmental factors is analyzed according to the actual migration point of the wild animal, the probability of the six points around each point is determined, and the scoring value is obtained by combining the probability and the migration cost, the point with high scoring value is determined as the next point, and the iteration is continuously performed until the end point is reached, thereby forming the path; The step S5 specifically comprises the following steps: S5.1: Load the wild animal passing point and non-passing point data; S5.2: Merge the two point data; S5.3: Extract environmental factor data from the environmental grid file; S5.4: Standardize the environmental factor data to obtain standardized continuous variables; S5.5: Define the environmental independent variable x and the selection dependent variable y; S5.6: Fit the logistic regression model, for the simulation of wild animal migration path, the known migration point is used as the "used" point, the point rarely existing is used as the "unused" point, and the influence of various environmental factors is judged based on the standardized multiple environmental factors through logistic regression; S5.7: Output the resource selection probability formula and the model test result, in the fitting process, it is necessary to ensure that the model can fully explain the variation in the data and there is no overfitting; cross-validation is used to evaluate the generalization ability of the model; After fitting, the goodness of fit of the model needs to be evaluated; this is realized by calculating the Akaike information criterion and the Bayesian information criterion statistical indicators; S5.8: Resource oriented path algorithm construction, RGPA algorithm is a new algorithm constructed for wild animal migration path simulation; it is formed on the basis of resource selection function algorithm; based on the resource selection function probability and the current point of the wild animal, the movement towards the migration end point is analyzed step by step; the path cost of the wild animal migration is calculated at each step, and the influence of positive attraction factors and negative hindering factors is considered, thereby simulating the actual migration path of the wild animal; S5.9: Try to divide the study area using different sizes of grid points, and simulate the route using RGPA respectively, and then compare the performance of these models in prediction accuracy and interpretability, select the best grid size calculation, the selected grid size is not limited, select the grid size suitable for the applied wildlife, here we choose 30m*30m as the final pathfinding grid best scale; S5.10: Calculate the probability value of each grid point being selected on the 30m*30m grid using the RSF model, which will be used as the basis for subsequent path generation; S5.11: Determine the starting point P start and the end point P end The specific location in the grid system is identified by ID, and path planning is performed for the two points; S5.12: Build a neighboring point honeycomb structure, that is, search for the surrounding 6 neighboring points of the current point; judge whether the neighboring points are empty, yes, end the algorithm running, no, continue the algorithm running; S5.13: Score the surrounding 6 neighboring points: resource selection probability x migration cost; According to the use probability of each neighboring point calculated by the RSF model, and combining the direction information of these points relative to the terminal point, score; preferentially select the neighboring point with high score as the next candidate point; S5.14: Iteration with termination condition; select the neighboring point with the highest score P neighbor As the next current point P current , repeat the search process described above until the current point is the end point P end or the search is terminated when no next point meeting the conditions can be found P next ; if the end point or a location sufficiently close to the end point is successfully reached, the algorithm ends and outputs the sequence of path points stored in the path list; S5.15: The generated path may contain some unnecessary turns or redundant points, which can be optimized by smoothing processing to make the path more natural and efficient; At the same time, according to the actual demand, further analyze and evaluate the path, including calculating the total length of the path and the proportion of passing through different habitat types. 2.The method of claim 1, wherein the method further comprises: determining a migration path of the wild animal based on the environmental factor data and the monitoring point data. The step S2 specifically comprises the following steps: S2.1: Obtain the digital orthographic image map DOM with an initial resolution of 2m obtained by the China Resource Satellite Application Center, and perform data preprocessing on the Resource No. 3 remote sensing image. First, data preparation is performed, and the original image data package containing panchromatic PAN image and multispectral MS image is downloaded from the Resource No. 3 satellite; The data of the Resource No. 3 satellite includes stereo imaging image, which is suitable for DEM extraction; Then, the image is radiometrically corrected to reduce the atmospheric influence and adjust the brightness and contrast of the image to ensure the accuracy of the spectral information of the image; and the image is geometrically corrected using known ground control points GCP and satellite orbit data to correct the spatial offset of the image to align it with the geographic coordinate system; Fuse PAN and MS using IHS fusion or other methods to generate high-resolution color images; Use DEM to orthorectify the image to eliminate image distortion caused by terrain undulations; After orthorectification, save the result as a digital orthographic image DOM in GeoTIFF format; The Resource No. 3 satellite has stereo imaging capability, and the forward-looking and rear-view images obtained are stereo image matched; Use photogrammetry software to; Image stereo matching generates a parallax map, which is a displacement difference map between two images; Use the parallax map to calculate the elevation data to generate a preliminary DEM; The software calculates the elevation value of each pixel based on the parallax between the stereo images using photogrammetry principles; And get the final accuracy of 2m of the area of digital orthographic image DOM and 5 meters of digital elevation model DEM, show the terrain distribution in the study area; Through further operation of DEM data, smooth processing of DEM, remove noise and outliers, to improve the accuracy; Using known ground elevation data to verify and correct the accuracy of DEM; Manual editing of DEM, repair the area with obvious error, ensure the accuracy of DEM data; Output DEM, save the final corrected DEM data as GeoTIFF format, so as to use in GIS software, get the terrain relief data; S2.2: Verify the acquisition time, sensor type and resolution information of the image to ensure the freshness and applicability of the data; S2.3: Use the radiation correction model to reduce the influence of atmosphere, and consider the influence factors of aerosol, cloud and water vapor in the correction process; S2.4: Splice and crop multiple adjacent images to ensure that the formed image covers the entire study area. 3.The method of claim 1, wherein the method further comprises: determining a migration path of the wild animal based on the environmental factor data and the monitoring point data. The step S3 specifically comprises the following steps: S3.1: Collect environmental factor data related to wild animal migration path analysis through multiple channels, including but not limited to terrain, vegetation cover, human activity, water source, road network factors; Determine important environmental factors, including vegetation cover type, human activity, water source, road network, in order to ensure the high accuracy and spatial consistency of data, use remote sensing image processing, digital elevation model calculation and other spatial analysis methods to extract and process multiple key environmental factors, use expert knowledge and reference literature to determine the influence weight of each factor; S3.2: Environmental factor data reflects the comprehensive ecological environment affecting the migration of wild animals, based on the public data of Open Street Map platform, combined with visual interpretation reading identification and field observation, the distance of water source is obtained, and further correction is carried out to ensure the accuracy and spatial consistency of water source data; S3.3: Obtain wild animal food source data from China Plant Science Data Center of Chinese Academy of Sciences, including but not limited to corn, bamboo forest, orchard, sugarcane, through field exploration and DOM reading identification, and through visual interpretation correction, obtain the final resolution of 2m of food source data; S3.4: Obtain 1km planting garden data of the region from China 1km grid land use degree data set, including tea garden, coffee garden land data, through field exploration and DOM reading identification, and through visual interpretation correction, obtain the final resolution of 2m of planting garden data; S3.5: Obtain grassland and shrub data from national 1m land use data, obtain the final resolution of 2m of grassland and shrub data; S3.6: Obtain natural mixed forest data from national 1m land use data, through field exploration and DOM reading identification, and through visual interpretation correction, obtain the final resolution of 2m of natural mixed forest data; S3.7: Obtain rubber plantation data from the China 1km grid land use extent dataset, identify and correct through field investigation and DOM reading, and obtain the final resolution of 2m rubber plantation data used; S3.8: Use classification algorithms for image classification, verify the accuracy of the classification results, and perform manual correction; S3.9: Convert the interpretation results to vector format to facilitate integration with other geographic data, check the resolution and accuracy of the data to ensure that it meets the research needs; S3.10: Verify the interpretation results using field survey data and make necessary corrections, establish metadata description of factor data, and record data acquisition, processing and verification process. 4.The method of claim 1, wherein the method further comprises: determining a migration path of the wild animal based on the environmental factor data and the monitoring point data. The step S4 specifically includes the following steps: S4.1: Design the structure of the database, including tables, fields and indexes, determine the spatial data format, and the storage method of non-spatial data; S4.2: Import environmental factor data and animal movement data into the database, verify the integrity and consistency of the data through SQL query and spatial analysis tools; S4.3: Associate different data sets through spatial connection, determine the topological relationship of spatial data, and ensure the spatial consistency of data. 5.The method of claim 1, wherein the method further comprises: determining a migration path of the wild animal based on the environmental factor data and the monitoring point data. The step S6 specifically includes the following steps: S6.1: Set the starting point and potential end point, initialize the current point as the starting point, set the initial path, and define the starting conditions; S6.2: Obtain the 6 adjacent points around the current point, judge whether the 6 adjacent points are empty; If the adjacent points are empty, end the algorithm; If the adjacent points are not empty, analyze the movement towards the migration end point, use Monte Carlo simulation to analyze the environmental adaptability probability of each point, realize neighborhood analysis on the GIS platform, and calculate the migration probability of the 6 surrounding points; S6.3: Use RSF algorithm to find the best path, including 6 adjacent search nodes in honeycomb structure, to ensure that the distance from the current point to the 6 adjacent points is equal; Consider resource selection probability and comprehensive migration cost to score adjacent points, select the optimal adjacent point to ensure large resource selection probability and proximity to the end point; Calculate the migration path cost at each step, and consider the influence of positive attraction factors and negative hindering factors; S6.4: Add the optimal point to the path, update the current point to the optimal point, judge whether the current point is the end point, if the current point is not the end point, obtain the 6 adjacent points around the current point, repeat the operation of S6.2 to S6.4; If the current point is the end point, use the path point set to create the path, record the conditions and results of each path selection in the iteration process, and save the path; S6.4: Thus simulate the actual migration path, test the ecological rationality of the generated path, compare with the actual observation data, and correct the path prediction results through expert review and further field investigation.

Images