A large and medium-sized protection of mammalian diversity hot spot distribution area prediction method and system

Abstract

The present application relates to the technical field of biodiversity protection and ecological modeling, and particularly relates to a method and system for predicting hot spot distribution of large and medium-sized protected mammal diversity. According to the mammal diversity data obtained in the target area, spatio-temporal matching and standardization processing are performed to construct a comprehensive data set. By setting a variance inflation factor and through expert scoring method, core feature variables are selected from the comprehensive data set. Based on random forest and MaxEnt model as the basic prediction layer, the core feature variables are predicted, and the outputs of the random forest and MaxEnt model are respectively introduced into the meta learner model to obtain the initial distribution probability of each species. Habitat connectivity index and time series dynamic factor are constructed to generate a species diversity distribution matrix. A hot spot analysis algorithm is used to identify and output the hot spot area of each species. Through integration of multi-source data and dynamic correction mechanism, the accuracy and timeliness of the hot spot prediction are improved.

Description

Technical Field

[0001] This invention relates to the fields of biodiversity conservation technology and ecological modeling technology, specifically to a method and system for predicting the distribution hotspots of large and medium-sized protected mammal diversity. Background Technology

[0002] Large and medium-sized mammals, as key species in ecosystems, play an irreplaceable role in maintaining ecological balance. However, due to climate change and human activities, their habitats are becoming increasingly fragmented, and their populations are declining continuously. Accurately identifying biodiversity hotspots has become a core prerequisite for developing conservation strategies. Currently, species distribution models (SDMs) are the main tool for predicting hotspots, but traditional methods have significant limitations: First, the data sources are singular, relying heavily on species distribution points and climate variables, neglecting key biogeographical factors such as interspecific interactions and habitat connectivity; second, the models lack adaptability, failing to fully consider the wide range and strong spatiotemporal migration characteristics of large and medium-sized mammals; and third, the prediction results are static, making it difficult to reflect seasonal fluctuations and long-term trends, resulting in a lack of dynamic support for conservation decisions.

[0003] In the prior art, for example, the invention patent with patent publication number CN116912686A, entitled "A Method, System and Medium for Identifying Biodiversity Conservation Hotspots", discloses a method for identifying biodiversity hotspots based on spatiotemporal remote sensing indices and human footprint indices, but it does not incorporate species-specific characteristics and biological interactions, thus limiting the accuracy of prediction. Those skilled in the art have also used climate change scenarios to optimize prediction results, but have not addressed the impact of habitat fragmentation on large and medium-sized mammals. While existing research has demonstrated that integrating expert knowledge and interspecific interactions can improve model performance, this approach focuses on marine organisms and cannot be directly adapted to the ecological characteristics of large and medium-sized terrestrial mammals.

[0004] Therefore, there is an urgent need to develop a targeted and multi-dimensional integrated prediction method to meet the conservation needs of large and medium-sized terrestrial mammals and improve the accuracy of predicting the distribution hotspots of large and medium-sized protected mammal diversity. Summary of the Invention

[0005] The purpose of this invention is to design a method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals, thereby improving the accuracy of such predictions, in response to the technical deficiencies of existing solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals includes the following steps: S1. Based on the mammal diversity data obtained in the target area, perform spatiotemporal matching and standardization to construct a comprehensive dataset; Mammal diversity data includes species-based data, environmental variable data, human activity data, and expert knowledge data; S2. By setting the variance inflation factor and using the expert scoring method, core feature variables are selected from the comprehensive dataset after preprocessing in step S1. S3. Based on random forest and MaxEnt model as the basic prediction layer, predict the core feature variables respectively. Import the output of random forest and MaxEnt model into the meta learner model to obtain the initial distribution probability of each species. S4. Construct habitat connectivity index and time series dynamic factor, and correct the initial distribution probability of each species obtained in step S3 based on habitat connectivity index and time series dynamic factor to generate species diversity distribution matrix. S5. Based on the species diversity distribution matrix, a hotspot analysis algorithm is used to identify and output the hotspot areas of each species. The obtained hotspot areas are verified by field survey data and ROC-AUC index.

[0007] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, in step S1, the basic species data includes distribution point data of large protected mammals, distribution point data of medium-sized protected mammals, species protection level data, population size data, and GPS tracking data within the target area. The distribution point data of large and medium-sized protected mammals integrates citizen scientific observation data, infrared camera monitoring data, and specimen collection data.

[0008] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, in step S1, the environmental variable data includes climate variable data, topographic variable data, vegetation variable data and hydrological variable data within the target area; Climate variable data include annual average temperature, precipitation, and seasonal fluctuation data for the target region during the statistical period; Topographic variable data include target area elevation data, slope data, topographic roughness data, and topographic abundance index data; The vegetation variable data consisted of NDVI time-series data and vegetation type classification data obtained within the target area; Hydrological variables include water source distribution data and soil moisture data within the target area.

[0009] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, in step S1, human activity data includes human footprint index, road density, proportion of construction land, protected area boundary data, and land use type transfer data within the target area; Human activity data were obtained through standardized data analysis within the target area using a 1km × 1km grid. Standardized data is generated through remote sensing image interpretation and GIS spatial analysis within the target area.

[0010] As a further improvement to the method for predicting the distribution hotspots of biodiversity of large and medium-sized protected mammals in this application, in step S2, the core characteristic variables include species characteristic variables, environmental adaptation variables, human disturbance variables, and interspecific interaction variables.

[0011] As a further improvement to the method for predicting the distribution hotspots of biodiversity of large and medium-sized protected mammals proposed in this application, interspecific interaction variables are obtained through the calculation of a species interaction matrix. The matrix elements in the species interaction matrix include predator-prey relationship strength coefficient, symbiotic relationship strength coefficient and competition relationship strength coefficient. The matrix elements are determined based on bibliometric analysis and IUCN expert evaluation.

[0012] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, in step S4, the habitat connectivity index is obtained by calculating the patch connectivity index and effective mesh size of the target area using landscape ecology software.

[0013] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, the time series dynamic factor is generated based on a trained LSTM neural network model. The LSTM neural network is trained using time series data of species activity intensity in the target area within a set calculation time as training samples.

[0014] As a further improvement to the method for predicting the distribution hotspots of large and medium-sized protected mammal diversity in this application, in step S5, the hotspot analysis algorithm uses the Getis-Ord Gi statistic to determine grid cells with Z values ​​≥ 1.96 as hotspot areas; The validation process includes model accuracy validation and result authenticity validation. Model accuracy validation uses the ROC-AUC metric, while result authenticity validation combines measured data from infrared cameras with on-site expert evaluation.

[0015] This application also relates to a system for predicting the distribution hotspots of large and medium-sized protected mammal diversity, including: Data acquisition module: used to collect basic species data, environmental variable data, human activity data, and expert knowledge data; Data processing module: used to perform spatiotemporal matching, missing value imputation, and standardization on the data output by the data acquisition module; Feature selection module: Based on variance inflation factor and expert scoring method, the core feature variables of the data output by the data processing module are selected; Model building module: Integrates random forest, MaxEnt model and meta learner model to build predictive models and predict core feature variables; Dynamic correction module: Integrates the prediction results from the habitat connectivity index and the time series dynamic factor correction model construction module, and outputs a species diversity distribution matrix; Identification and verification module: Identifies hotspots based on the species diversity distribution matrix using the Getis-Ord Gi algorithm and completes accuracy verification.

[0016] The above technical solution produces the following technical effects: This application provides a method for predicting the distribution hotspots of large and medium-sized protected mammal diversity. By integrating multi-source data and a dynamic correction mechanism, the accuracy and timeliness of hotspot prediction are improved. Specifically, the technical effects of this application's solution include the following: 1) Higher prediction accuracy: By integrating interspecific interactions and habitat connectivity factors, the problem of traditional models ignoring core biogeographical elements is solved. According to experimental verification, the ROC-AUC value is 12% to 18% higher than that of the single MaxEnt model, and the hotspot identification accuracy is ≥85%.

[0017] 2) Enhanced spatiotemporal adaptability: The introduction of LSTM time series models and spatial refinement processing with 1km grids can accurately capture the seasonal migration and habitat fragmentation response of large and medium-sized mammals, adapting to their wide range of activities.

[0018] 3) More direct decision support: The output includes the spatial coordinates of hotspot areas and potential hotspot areas and their protection priorities (based on species protection levels), which can be directly used in practices such as protected area boundary demarcation and ecological corridor planning, improving the efficiency of conservation resource allocation by 3% to 5%. Attached Figure Description

[0019] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0020] 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.

[0021] To facilitate an accurate understanding of the solutions provided in the following embodiments of the present invention, the terms involved in the present invention are explained as follows before describing the technical solutions provided by the present invention: Interspecific interaction variables: These are characteristic variables that describe the relationships between species. They are achieved by constructing a species interaction matrix, whose elements include the strength of predation, symbiosis, and competition. The data are derived from the interaction strength coefficients determined by bibliometric analysis and IUCN expert evaluation. This is the key innovative dimension that distinguishes this application from traditional models.

[0022] Habitat connectivity index: an indicator used to quantify the connectivity between habitat patches. This application uses the patch connectivity index (PC) and effective mesh size (MESH) to calculate the integrity of migration corridors for large and medium-sized mammals using landscape ecology software (such as Fragstats). The formula incorporates patch area and distance attenuation coefficients.

[0023] Time-series dynamic factors: seasonal / annual correction factors generated based on LSTM neural network models, such as time-series data of species activity intensity over the past 5 years (e.g., monthly average activity frequency of infrared cameras) as training samples, used to capture the temporal heterogeneity of species migration (e.g., correction coefficient of 0.85 in winter and 1.12 in summer).

[0024] Species diversity distribution matrix: The core output data after dynamic correction, with 1km×1km grid as the unit. The rows of the matrix represent spatial grids, the columns represent species, and the elements are the corrected species distribution probabilities, which are used as the basic input for subsequent hotspot analysis.

[0025] Variance inflation factor (VIF): A multicollinearity test index used in the feature selection stage. The calculation formula is VIF = 1 / (1-R²), where R² is the goodness of fit of the variable to other variables. In this application, variables with VIF < 10 are retained to avoid redundant interference between features.

[0026] Random Forest Model: This is an ensemble learning algorithm. Specifically, this application constructs 500 decision trees (n_estimators=500), each trained based on randomly sampled samples and features. The final output is the average vote of all trees. This application sets a maximum depth of 10 (max_depth=10) and optimizes the parameters using 5-fold cross-validation to capture non-linear interactions of features.

[0027] MaxEnt model: A species distribution model based on the maximum entropy principle, which predicts potential distributions under environmental constraints by maximizing the entropy value of the species existence probability distribution. In this application, altitude, mean NDVI, and distance from water source are selected as key variables, with a regularization parameter of 1.0 to avoid overfitting. It is suitable for scenarios with limited distribution point data.

[0028] Meta-learner: The upper-level learner in the model ensemble stage, employing a stacking strategy: using the prediction results of Random Forest and MaxEnt as input features, it is trained through a logistic regression algorithm, outputting the initial distribution probability (0-1) of a single species. The ensemble process incorporates IUCN species distribution range data constraints to avoid over-prediction at geographical barriers.

[0029] Getis-Ord Gi Statistic: The core algorithm for hotspot analysis, it identifies high-value clustering regions by calculating the local spatial autocorrelation coefficient. The formula is: ; Where x j This is the grid cell diversity index. This is the spatial weight matrix. The global mean. The standard deviation is given. The calculation results are converted to Z-values, where Z ≥ 1.96 corresponds to hotspot areas at a 95% confidence level.

[0030] ROC-AUC: A validation metric for model accuracy. The ROC curve is plotted with the false positive rate (FPR) on the horizontal axis and the true positive rate (TPR) on the vertical axis. AUC is the area under the curve (values ​​from 0 to 1). This application requires a validation set AUC ≥ 0.85. In actual tests, the corrected AUC reached 0.94, reflecting the model's ability to distinguish between the presence and absence of species.

[0031] Delphi method: an expert knowledge data collection method that uses anonymous scoring (1-5 points) from 10-15 mammalian ecology experts, and after multiple rounds of feedback to converge opinions, transforms qualitative information such as species habitat preferences and interspecific interactions into standardized indices.

[0032] DBSCAN algorithm: An outlier removal method in the data cleaning stage. Based on the principle of density clustering, it identifies and removes outlier distribution points (such as abnormally high altitude records) that exceed the known distribution range of a species by setting a neighborhood radius and a minimum number of samples.

[0033] Stacking ensemble strategy: A two-level framework for model ensemble. The first level (base model layer) trains random forest and MaxEnt. The second level (meta learner layer) uses the output of the base model as features and trains the final predictor through logistic regression, which improves the AUC by 0.03-0.05 compared to a single model.

[0034] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0035] Example 1 like Figure 1 As shown, the purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for predicting the distribution hotspots of large and medium-sized protected mammal diversity. By integrating multi-source data and a dynamic correction mechanism, the accuracy and timeliness of hotspot prediction are improved, providing a scientific basis for conservation decisions. Specifically, this application provides a method for predicting the distribution hotspots of large and medium-sized protected mammal diversity, including the following steps: S1. Based on the mammal diversity data obtained in the target area, perform spatiotemporal matching and standardization to construct a comprehensive dataset; Mammal diversity data includes species-based data, environmental variable data, human activity data, and expert knowledge data; S2. By setting the variance inflation factor and using the expert scoring method, core feature variables are selected from the comprehensive dataset after preprocessing in step S1. S3. Based on random forest and MaxEnt model as the basic prediction layer, predict the core feature variables respectively. Import the output of random forest and MaxEnt model into the meta learner model to obtain the initial distribution probability of each species. S4. Construct habitat connectivity index and time series dynamic factor, and correct the initial distribution probability of each species obtained in step S3 based on habitat connectivity index and time series dynamic factor to generate species diversity distribution matrix. S5. Based on the species diversity distribution matrix, a hotspot analysis algorithm is used to identify and output the hotspot areas of each species. The obtained hotspot areas are verified by field survey data and ROC-AUC index.

[0036] Furthermore, in step S3, the core feature variables obtained in the "multidimensional feature variable screening" stage are used as inputs. These core variables include species characteristic variables (such as conservation level and active home range area), environmental adaptation variables (such as annual mean temperature and NDVI mean), human disturbance variables (such as human footprint index and road density), and interspecific interaction variables (such as predator / competition relationship intensity). These are key influencing factors retained after quantitative screening using variance inflation factor (VIF) and expert verification. The basic prediction layer is constructed in parallel using two models. This application preferably uses two classic species distribution models (SDM) to construct the basic prediction layer: the random forest model and the MaxEnt model. Furthermore, the outputs of the two basic models are integrated using the "meta-learner" method. Specifically, a stacking ensemble strategy is adopted: the prediction results of the random forest and MaxEnt are used as new input features and input into the meta-learner model (logistic regression algorithm is used in the document). Through training, the "initial distribution probability" of a single species (value 0-1, representing the probability of the target species appearing in the grid cell) is obtained. Specifically, step S3 ultimately generates the "initial distribution probability," which is the potential distribution probability value of the species before dynamic correction, providing basic data for subsequent correction by introducing habitat connectivity index and time series dynamic factors (step S4).

[0037] Specifically, the data presented in step S1 includes: Basic species data: Distribution point data of large and medium-sized protected mammals (such as leopards and takins) in the target area are obtained through the Global Biodiversity Information Platform (GBIF) and the National Key Protected Wildlife Monitoring Network. Combined with population activity data obtained from infrared camera monitoring (≥1 year) and migration data tracked by GPS collars, the protection level (critically endangered / endangered / vulnerable) and population size information in the IUCN Red List are supplemented.

[0038] Environmental variable data: Climate data such as annual average temperature and precipitation in the target area were obtained from the WorldClim database (spatial resolution 30s). Topographic variables such as elevation and slope were calculated using a digital elevation model (DEM). Vegetation growth status information was extracted using NDVI time series data from the MODIS satellite (time resolution 16 days). Soil moisture and distance from water source data were generated by combining soil database and hydrological monitoring station data.

[0039] Human activity data: Based on Landsat 8 remote sensing imagery, land use types within the target area from 2000 to 2024 were interpreted, and indicators such as road density and construction land ratio were calculated. Human footprint index (HFI) data was used to generate a human disturbance intensity layer by combining protected area boundary vector data.

[0040] Expert knowledge data: The Delphi method is used to collect opinions from 10-15 mammalian ecology experts to obtain qualitative data such as species habitat preferences and interspecific interactions, which are then converted into standardized indices.

[0041] Furthermore, in the data processing steps described above, the spatiotemporal matching method unifies all data to the WGS84 coordinate system and resamples it into 1km×1km grid cells to ensure consistency in time scale (data from the past 10 years). The data cleaning stage employs local weighted regression to fill in missing values ​​in the climate data and uses the DBSCAN algorithm to remove outliers from the distribution point data (such as records exceeding the known distribution range of a species).

[0042] Finally, numerical variables were standardized using the Z-score standardization formula: x'=(x-μ) / σ, where μ is the mean and σ is the standard deviation. Categorical variables (such as vegetation type) were transformed using independent coding.

[0043] Furthermore, in step S2, this application mainly achieves multidimensional feature variable screening through feature initial selection and feature optimization. The feature initial selection step extracts initial feature variables from the preprocessed data, including species characteristic variables (conservation level, active home range area), environmental adaptation variables (annual mean temperature, mean NDVI, terrain roughness, etc.), human disturbance variables (human footprint index, road density), and interspecific interaction variables (predator / competition / symbiotic relationship strength).

[0044] In the feature optimization stage, the Pearson correlation coefficient between variables is calculated. Variables with a rank correlation coefficient ≥ 0.7 are redundantly removed using the variance inflation factor (VIF) (variables with VIF < 10 are retained). Furthermore, ecological experts score the importance of the selected variables (1-5 points), retaining variables with an average score ≥ 3.5 as core feature variables, forming the final feature set.

[0045] Furthermore, in step S4, this application uses Fragstats software to calculate the patch connectivity index (PC) and effective mesh size (MESH) of each grid cell, and constructs a connectivity correction matrix, the formula of which is: ; in, Let i be the patch area of ​​grid cells i and j. This represents the distance decay coefficient between the two. The connectivity index is coupled with the initial probability distribution, and the weighting coefficients are determined through linear regression. k The patch area of ​​a grid cell, d k This refers to the distance attenuation coefficient between grid cells. Patch area reflects the size attribute of the habitat, while the distance attenuation coefficient quantifies the impact of spatial distance between patches on connectivity; both are used in the calculation of the habitat connectivity index. Therefore, based on the past 5 years of time-series data on species activity intensity (monthly average activity frequency monitored by infrared cameras), an LSTM neural network model is constructed to predict changes in activity intensity in different seasons, generating a time-dynamic correction factor (values ​​ranging from 0.8 to 1.2). The connectivity-corrected results are then adjusted a second time to obtain a species diversity distribution matrix (rows represent grid cells, columns represent species, and elements are corrected distribution probabilities).

[0046] Furthermore, this application sums the species diversity distribution matrix by grid cells to obtain a comprehensive diversity index. The Z-value is calculated using ArcGIS's hotspot analysis tool (Getis-Ord Gi). Grids with Z ≥ 1.96 are defined as hotspot areas (high diversity areas), and those with 1.65 ≤ Z < 1.96 are considered potential hotspot areas. 20% of the field monitoring data is used as the test set, and ROC-AUC values ​​are calculated. The model is considered valid when AUC ≥ 0.85. 30-50 infrared camera monitoring points are deployed within the predicted hotspot areas and monitored continuously for 6 months. If the relative error between the measured species richness and the predicted value is ≤ 15%, and the area contains ≥ 2 critically endangered / endangered species, the hotspot area is confirmed as valid.

[0047] Furthermore, this application also relates to a system for predicting the distribution hotspots of large and medium-sized protected mammal diversity, comprising: Data acquisition module: used to collect basic species data, environmental variable data, human activity data, and expert knowledge data; Data processing module: used to perform spatiotemporal matching, missing value imputation, and standardization on the data output by the data acquisition module; Feature selection module: Based on variance inflation factor and expert scoring method, the core feature variables of the data output by the data processing module are selected; Model building module: Integrates random forest, MaxEnt model and meta learner model to build predictive models and predict core feature variables; Dynamic correction module: Integrates the prediction results from the habitat connectivity index and the time series dynamic factor correction model construction module, and outputs a species diversity distribution matrix; Identification and verification module: Identifies hotspots based on the species diversity distribution matrix using the Getis-Ord Gi algorithm and completes accuracy verification.

[0048] Example 2 The following example, using the prediction of biodiversity hotspots for large and medium-sized protected mammals in the Kazila Mountain area of ​​western Sichuan as an example, details the implementation process of this invention: 1. Study Area and Data Preparation The target area is the core region of western Sichuan (29.98°N-30.14°N, 100.58°E-100.80°E), covering an area of ​​approximately 400 km². The collected data includes: Species data: Distribution points (more than 200) of 8 key protected mammal species, including sambar deer, leopard, and tufted deer, and infrared camera monitoring data (2023-2025).

[0049] Environmental data: 19 bioclimatic variables from WorldClim, 30m resolution DEM data, MODISNDVI data (2019-2024), and soil moisture data (Soil Science Database, Chinese Academy of Sciences).

[0050] Human activity data: Land use map interpreted from Landsat 8 imagery in 2024, road vector data (National Geographic Information Public Service Platform), and Human Footprint Index (2023 version).

[0051] 2. Data Preprocessing and Feature Selection All data were resampled into a 1km×1km grid. Locally weighted regression was used to fill in missing values ​​in the climate data, and the DBSCAN algorithm was used to remove outliers (retaining 1080 valid points). Initially, 32 feature variables were extracted. After screening by variance inflation factor (VIF<10) and expert scoring (by 12 experts), 15 core feature variables were ultimately retained, including annual mean temperature, mean NDVI, terrain roughness, human footprint index, and interspecies competition intensity.

[0052] 3. Model Construction and Calibration 3.1 Basic Model Training: The random forest model was set with n_estimators=500 and max_depth=10; the MaxEnt model selected altitude, mean NDVI, and distance from water source as key variables, with a regularization parameter of 1.0, and cross-validation AUCs of 0.89 and 0.86, respectively.

[0053] 3.2 Integration and Correction: By integrating the basic model through logistic regression and incorporating IUCN-leopard distribution range data constraints, the initial predicted AUC was improved to 0.91. The habitat connectivity index (mean PC 0.62) and seasonal dynamic factors (winter correction coefficient 0.85, summer 1.12) were calculated, and the final corrected AUC reached 0.94.

[0054] 4. Hotspot identification and verification 4.1 Hotspot Identification: A comprehensive diversity index matrix was obtained. Getis-Ord Gi* analysis showed that the hotspot areas were mainly concentrated in the core area of ​​western Sichuan (Z value 2.3-3.1), covering an area of ​​approximately 6200 km²; potential hotspot areas were distributed in the western section (Z value 1.7-2.2), covering an area of ​​approximately 400 km².

[0055] 4.2 Validation results: The test set ROC-AUC=0.94. The 40 infrared camera monitoring points deployed in the field showed that the relative error between the measured and predicted values ​​of species richness in the hotspot area was 11.2%, including key species such as tufted deer (near threatened) and leopard (endangered), and the validation was successful.

[0056] It is noteworthy that those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals, characterized in that, Includes the following steps: S1. Based on the mammal diversity data obtained in the target area, perform spatiotemporal matching and standardization to construct a comprehensive dataset; The mammalian diversity data includes basic species data, environmental variable data, human activity data, and expert knowledge data; S2. By setting the variance inflation factor and using the expert scoring method, the core feature variables are selected from the comprehensive dataset after preprocessing in step S1. S3. Based on random forest and MaxEnt model as the basic prediction layers, the core feature variables are predicted respectively. The outputs of random forest and MaxEnt model are respectively imported into the meta learner model to obtain the initial distribution probability of each species. S4. Construct a habitat connectivity index and a time-series dynamic factor, and correct the initial distribution probability of each species obtained in step S3 based on the habitat connectivity index and the time-series dynamic factor to generate a species diversity distribution matrix. S5. Based on the species diversity distribution matrix, a hotspot analysis algorithm is used to identify and output the hotspot areas of each species. The obtained hotspot areas are verified by field survey data and ROC-AUC index.

2. The method for predicting the distribution hotspots of large and medium-sized protected mammal diversity according to claim 1, characterized in that, In step S1, the basic species data includes distribution point data of large protected mammals, distribution point data of medium protected mammals, species protection level data, population size data, and GPS tracking data within the target area. The distribution point data of large protected mammals and the distribution point data of medium protected mammals integrate citizen scientific observation data, infrared camera monitoring data, and specimen collection data.

3. The method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals according to claim 1, characterized in that, In step S1, the environmental variable data includes climate variable data, topographic variable data, vegetation variable data, and hydrological variable data within the target area; The climate variable data used are the annual average temperature data, precipitation data and seasonal fluctuation data of the target area during the statistical period; The terrain variable data includes the target area's elevation data, slope data, terrain roughness data, and terrain abundance index data; The vegetation variable data consists of NDVI time-series data and vegetation type classification data obtained within the target area; The hydrological variables include water source distribution data and soil moisture data within the target area.

4. The method for predicting the distribution hotspots of large and medium-sized protected mammal diversity according to claim 1, characterized in that, In step S1, the human activity data includes human footprint index, road density, construction land ratio, protected area boundary data, and land use type transfer data within the target area; The human activity data was obtained through standardized data analysis within the target area using a grid scale of 1km × 1km. The standardized data is generated through remote sensing image interpretation and GIS spatial analysis within the target area.

5. The method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals according to claim 1, characterized in that, In step S2, the core characteristic variables include species characteristic variables, environmental adaptation variables, human disturbance variables, and interspecific interaction variables.

6. The method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals according to claim 5, characterized in that, The interspecific interaction variables were obtained by calculating the species interaction matrix, whose matrix elements include predator-prey relationship strength coefficient, symbiotic relationship strength coefficient, and competition relationship strength coefficient. The matrix elements were determined based on bibliometric analysis and IUCN expert evaluation.

7. The method for predicting the distribution hotspots of large and medium-sized protected mammal diversity according to claim 1, characterized in that, In step S4, the habitat connectivity index is obtained by calculating the patch connectivity index and effective mesh size of the target area using landscape ecology software.

8. The method for predicting the distribution hotspots of biodiversity in large and medium-sized protected mammals according to claim 1, characterized in that, The time-series dynamic factor is generated based on a trained LSTM neural network model, which is trained using time-series data of species activity intensity within the target area as training samples within a set calculation time period.

9. The method for predicting the distribution hotspots of large and medium-sized protected mammal diversity according to claim 1, characterized in that, In step S5, the hotspot analysis algorithm uses the Getis-Ord Gi statistic to identify grid cells with a Z value ≥ 1.96 as hotspot regions. The verification includes model accuracy verification and result authenticity verification. Model accuracy verification uses the ROC-AUC index, while result authenticity verification combines infrared camera measured data with expert on-site evaluation.

10. A prediction system for hotspot distribution areas of large and medium-sized protected mammal diversity, characterized in that, include: Data acquisition module: used to collect basic species data, environmental variable data, human activity data, and expert knowledge data; Data processing module: used to perform spatiotemporal matching, missing value imputation, and standardization on the data output by the data acquisition module; Feature filtering module: Based on variance inflation factor and expert scoring method, the core feature variables of the data output by the data processing module are filtered; Model building module: Integrates random forest, MaxEnt model and meta learner model to build a prediction model to predict the core feature variables; Dynamic correction module: Integrates habitat connectivity index and time series dynamic factors to correct the prediction results output by the model building module, and outputs a species diversity distribution matrix; Identification and verification module: Identifies hotspots based on the species diversity distribution matrix using the Getis-Ord Gi algorithm and completes accuracy verification.

Images