Grassland looper habitat suitability distribution extraction method and system

Abstract

This invention provides a method and system for extracting the habitat suitability distribution of grassland caterpillars, comprising: acquiring multi-source data; constructing a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages; identifying key habitat factors in the multi-dimensional habitat factor system; constructing spatially structured samples based on the species-accessible areas of grassland caterpillars to obtain a structured sample set; constructing an integrated small model based on key habitat factors and the structured sample set; the integrated small model is obtained by weighted integration of multiple sub-models, each sub-model being constructed from at least one key habitat factor; evaluating and optimizing the integrated small model and decision threshold to determine the optimal model and optimal threshold; and using the optimal model and optimal threshold to extract and classify the habitat suitability of the test area, generating a grassland caterpillar habitat suitability distribution map, thereby enhancing the stability of the model in extrapolation prediction and improving the accuracy of prediction.

Description

Technical Field

[0001] This invention relates to the field of grassland caterpillar control technology, and in particular to a method and system for extracting habitat suitability distribution of grassland caterpillars. Background Technology

[0002] Grassland caterpillars are major pests of high-altitude and cold grasslands, and accurate monitoring and prediction of their occurrence range and damage are crucial for grassland ecological protection and pest control. Existing methods for analyzing the habitat suitability of grassland caterpillars are mostly based on traditional niche models or single machine learning models, predicting their distribution by screening environmental factors related to the occurrence points of grassland caterpillars.

[0003] However, these methods rely heavily on the statistical correlation between environmental factors when building models, which leads to insufficient stability of the models when extrapolating and predicting, and they are prone to incorrectly classifying areas that are not actually accessible as suitable habitats, resulting in over-prediction problems. Summary of the Invention

[0004] This invention provides a method and system for extracting the habitat suitability distribution of grassland caterpillars, which solves the problem that the existing models have insufficient stability when extrapolating and predicting, and are prone to incorrectly classifying unreachable areas as suitable habitats, resulting in over-prediction.

[0005] This invention provides a method for extracting habitat suitability distribution of grassland caterpillars, comprising: Acquire multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages. Key habitat factors in the multi-dimensional habitat factor system were identified, and spatially structured samples were constructed based on the species reachable areas of grassland caterpillars to obtain a structured sample set. Based on the key habitat factors and the structured sample set, an ensemble small model is constructed. The ensemble small model is obtained by weighted ensemble of multiple sub-models, and each sub-model is constructed by combining at least one key habitat factor. Evaluate and optimize the ensemble small model and decision threshold to determine the optimal model and optimal threshold; Using the optimal model and optimal threshold, habitat suitability of the test area is extracted and classified to generate a habitat suitability distribution map of grassland caterpillars.

[0006] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars is provided, wherein a multi-dimensional habitat factor system is constructed based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages, including: The key life cycle stages of the grassland caterpillar were determined, including the overwintering stage and the larval activity stage. For the overwintering stage, based on the ecological response of overwintering larvae to environmental factors, the first habitat factor related to overwintering survival rate was selected; For the aforementioned larval activity stages, based on the ecological response of the larvae to food sources, hydrothermal conditions, and accumulated temperature during growth, secondary habitat factors related to larval growth and development are selected. The first habitat factor and the second habitat factor are integrated to construct a multi-dimensional habitat factor system.

[0007] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars is provided, wherein determining the key habitat factors in the multi-dimensional habitat factor system includes: Identify factor pairs exhibiting collinearity within the multidimensional habitat factor system; To assess the contribution of each habitat factor in the multi-dimensional habitat factor system to the habitat suitability of grassland caterpillars; The factors with relatively large contributions in the factor pair are identified as key habitat factors.

[0008] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars is provided, wherein the method involves constructing spatially structured samples based on the species reachable areas of grassland caterpillars to obtain a structured sample set, including: Based on the potential dispersal ability of grassland caterpillars, the areas that the species can reach are delineated; Within the reachable area of ​​the species, the samples are spatially structured based on environmental heterogeneity and spatial distribution characteristics to obtain a structured sample set in terms of environmental gradient and spatial distribution.

[0009] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars, wherein delineating the species' reachable area based on the potential dispersal ability of grassland caterpillars includes: Establish a buffer zone centered on a known point of existence, and define the area covered by the buffer zone as the reachable region of the species; and / or, Based on landscape resistance surfaces, the dispersal cost of grassland caterpillars in different land surface types is simulated, and the reachable range of the minimum cost path is extracted as the species' reachable area; and / or, A convex hull polygon is generated based on known distribution points, and the convex hull polygon and its extended region are used as the reachable region of the species; and / or, Based on the boundaries of ecological geographical zoning or vegetation zoning, the ecological units where the historical distribution of grassland caterpillars is located are taken as the reachable areas of the species.

[0010] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars is provided, wherein the samples are spatially structured based on environmental heterogeneity and spatial distribution characteristics within the species' reachable area to obtain a structured sample set in terms of environmental gradient and spatial distribution, including: The reachable area of ​​the species is divided into environmental spatial partitions to obtain multiple spatial partitions with different environmental gradients; Within each spatial partition, background points are sampled centered on the existing point to obtain a background sample set; The samples in the background sample set are used as absent points and combined with the existent points in each spatial partition to construct a training sample set with a balanced number of existent and absent points. The training sample sets of each spatial partition are integrated to obtain a structured sample set in terms of environmental gradient and spatial distribution.

[0011] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars is provided, wherein the construction of an integrated small model includes: Based on the key habitat factors, multiple sub-models are constructed, each of which is composed of at least one key habitat factor, and different sub-models adopt different factor combination methods; Cross-validation was used to evaluate the predictive performance of each sub-model, and weights were assigned to each sub-model based on the predictive performance metrics. The multiple sub-models are weighted and integrated according to the weights to obtain an integrated small model.

[0012] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars, wherein the evaluation and optimization of the ensemble small model and decision threshold, and the determination of the optimal model and optimal threshold, includes: Multiple threshold partitioning methods are used to determine the corresponding candidate decision thresholds; For each candidate decision threshold, evaluate the ensemble small model on multiple evaluation metrics at the corresponding candidate decision threshold; By comparing the evaluation index values ​​under all candidate decision thresholds, the candidate decision threshold with the best overall performance is selected as the optimal decision threshold, and the corresponding ensemble small model is selected as the optimal model.

[0013] According to the present invention, a method for extracting the habitat suitability distribution of grassland caterpillars includes extracting and classifying the habitat suitability of the area to be tested to generate a grassland caterpillar habitat suitability distribution map, comprising: The multi-source data of the area to be tested are input into the optimal model to generate an initial habitat suitability probability map; Based on the species reachable area of ​​grassland caterpillars, spatial constraints are applied to the initial habitat suitability probability map to obtain a constrained habitat suitability probability map. According to the preset grading criteria, the constrained habitat suitability probability map is divided into multiple suitability levels to generate a habitat suitability grading map. The habitat suitability grading map is output as a habitat suitability distribution map of grassland caterpillars.

[0014] Secondly, the present invention provides a grassland caterpillar habitat suitability distribution extraction system, comprising: The acquisition module is used to acquire multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages. The determination module is used to determine the key habitat factors in the multi-dimensional habitat factor system, and to construct spatially structured samples based on the species reachable area of ​​grassland caterpillars to obtain a structured sample set. A construction module is used to construct an integrated small model based on the key habitat factors and the structured sample set. The integrated small model is obtained by weighted integration of multiple sub-models, and the sub-model is constructed by combining at least one key habitat factor. The evaluation module is used to evaluate and optimize the integrated small model and decision threshold, and determine the optimal model and optimal threshold. The extraction module is used to extract and classify the habitat suitability of the test area using the optimal model and the optimal threshold, and generate a habitat suitability distribution map of grassland caterpillars.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the grassland caterpillar habitat suitability distribution extraction method as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the grassland caterpillar habitat suitability distribution extraction method as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the grassland caterpillar habitat suitability distribution extraction method as described above.

[0018] This invention provides a method and system for extracting habitat suitability distribution of grassland caterpillars. By constructing a multi-dimensional habitat factor system based on the ecological response relationships of grassland caterpillars to habitat factors at different life stages, the model input possesses a clear ecological orientation. Simultaneously, spatially structured sample construction based on the species' reachable areas strictly limits the model training and prediction range to the areas theoretically accessible to grassland caterpillars, effectively suppressing the blind expansion of suitability zones caused by traditional models that ignore species dispersal limitations. Through evaluation, optimization, and integration of small models and decision thresholds, the optimal model is determined, maintaining high sensitivity while possessing good specificity. This makes the spatial boundary of habitat suitability prediction more consistent with the actual ecological distribution range of the species, effectively improving the overall prediction accuracy and stability of the model. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the method for extracting the habitat suitability distribution of grassland caterpillars provided by the present invention. Figure 2 This is a schematic diagram illustrating the principle of the grassland caterpillar habitat suitability distribution extraction method provided by the present invention. Figure 3 This is a schematic diagram of the distribution characteristics of basic grassland caterpillar data provided by the present invention; Figure 4 This is a schematic diagram of habitat suitability distribution in 2022 provided by the present invention; Figure 5 This is a schematic diagram of habitat suitability distribution in 2023 provided by the present invention; Figure 6 This is a schematic diagram of habitat suitability distribution in 2024 provided by the present invention; Figure 7 This is a schematic diagram illustrating the correlation and weighting of rainfall and land surface temperature factors provided by the present invention; Figure 8 This is a schematic diagram illustrating the correlation and weighting of the number of rainy days and the radiation intensity factor provided by this invention; Figure 9 This is a schematic diagram of the correlation and weight of the accumulated temperature factor provided by the present invention; Figure 10 This is a correlation diagram of meteorological factors provided by the present invention; Figure 11 This is a schematic diagram of the terrain factors provided by the present invention; Figure 12 This is a schematic diagram of vegetation factors provided by the present invention; Figure 13 This is a schematic diagram of soil factors provided by the present invention; Figure 14 This is a schematic diagram of the grassland caterpillar habitat suitability distribution extraction system provided by the present invention; Figure 15 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Figure 1 This is a flowchart illustrating the method for extracting the habitat suitability distribution of grassland caterpillars provided by the present invention. Figure 2 This is a schematic diagram illustrating the principle of the grassland caterpillar habitat suitability distribution extraction method provided in this embodiment.

[0023] like Figure 1 As shown in this embodiment, a method for extracting the habitat suitability distribution of grassland caterpillars includes: 101. Obtain multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages.

[0024] Specifically, ground survey data of grassland caterpillars from 2022 to 2024 were obtained, along with multi-source data from meteorology, remote sensing, and plant protection. Relying on the big data platform (Google Earth Engine, GEE), GIS and multi-source data fusion technologies were used to quantitatively extract environmental factors such as temperature, precipitation, light intensity, vegetation type, soil moisture content, altitude, and slope aspect. A bilinear interpolation sampling method was used to unify the resolution of all data.

[0025] Based on the core biological characteristics of the grassland caterpillar: it is a pest unique to alpine meadows, with one generation per year. It overwinters as a first-instar larva under cover such as grass roots, soil clods, and cow dung. Its complete life cycle includes four stages: egg, larva, pupa, and adult. The second instar larvae last the longest, 6-7 months, while the other instars are approximately 15-20 days. Overwintering larvae begin to become active in April-May of the following year, pupating and emerging as adults in July-August, with concentrated hatching in September-October. As a poikilothermic animal, its development speed, survival rate, and individual growth are highly sensitive to temperature changes. The larvae's black hairs reduce ultraviolet radiation, and the red warning coloration on their heads helps them evade predators. Adults exhibit sexual dimorphism, which improves energy utilization. They are mainly concentrated in alpine meadows at altitudes of 3000-5000 meters, preferring sunny slopes and areas with good sunlight. Based on the above life history and ecological response patterns, the overwintering stage and the larval activity stage were identified as the key observation stages. Dynamic and static factors were distinguished, and the thresholds of precipitation-related factors were defined with reference to precipitation levels. A multi-dimensional habitat factor system including meteorology, vegetation, soil, and topography was constructed, as shown in Table 1.

[0026] Table 1

[0027] A support vector machine (SVM) model of overwintering habitat suitability for grassland caterpillars was constructed simultaneously, and overwintering habitat condition indicators were extracted and incorporated into the habitat factor system for consideration.

[0028] The GEE platform ensures efficient processing of multi-source data, and bilinear interpolation with uniform resolution guarantees data consistency and comparability. A multi-dimensional habitat factor system built upon the real ecological characteristics of grassland caterpillars gives the model input a clear ecological orientation, avoiding reliance solely on statistical correlation for factor selection.

[0029] 102. Identify key habitat factors in the multi-dimensional habitat factor system, construct spatially structured samples based on the species reachable areas of grassland caterpillars, and obtain a structured sample set.

[0030] Specifically, mathematical statistical methods such as Pearson correlation coefficient and multiple linear regression were used to analyze the multidimensional habitat factor system, identify factors with collinearity and evaluate the contribution of each factor to the habitat suitability of grassland caterpillars, screen key habitat factors according to established principles, and integrate overwintering habitat condition indicators with key habitat factors.

[0031] Based on the dispersal characteristics of grassland caterpillars, a 100km buffer zone was used to define their reachable area, and spatially structured sample processing was carried out within this area. A regular grid was established in a multidimensional environmental variable space, dividing the environmental gradient into five control grid scales. One point was randomly selected from each grid cell to reduce sample density bias. The optimal partitioning scale was determined through spatial autocorrelation of the three-dimensional environmental space, environmental similarity, and data volume differences to ensure that each partition is representative in spatial structure. Since species presence records are usually obtained from unsystematic collections, there is significant geographical sampling bias. A background thickening method was used for background point sampling. Simultaneously, a 1:1 presence-pseudo-absence dataset was generated based on the number of presence points to clarify that missing records do not necessarily represent true absence. Finally, a structured sample set balanced in both environmental gradient and spatial distribution was obtained, such as... Figure 3 The figure shows the distribution characteristics of basic grassland caterpillar data. During the modeling process, a distribution model constraint strategy for rare ecological species was also coupled in to perform preliminary corrections on the sample and factor systems, thus mitigating the risk of over-prediction in advance.

[0032] Screening key habitat factors effectively eliminates the interference of factor collinearity on the model and improves the effectiveness of model input. Based on the spatial structuring of samples in the species-accessible areas, it solves the problems of geographical sampling bias, uneven sample density, and sample imbalance caused by the unsystematic collection of grassland caterpillar records. It keeps the spatial structure, environmental gradient, and information content of training samples in balance, provides a high-quality sample foundation for model training, and enhances the generalization ability of the model.

[0033] 103. Based on key habitat factors and structured sample sets, construct an integrated small model. The integrated small model is obtained by weighted integration of multiple sub-models. Each sub-model is constructed by combining at least one key habitat factor.

[0034] Specifically, in modeling rare species, the number of explanatory variables often far exceeds the number of valid existence records. This makes the model prone to high fitting during training, but its accuracy decreases in extrapolation or independent region prediction. Too many predictors can cause the model to capture noise in the samples rather than ecological signals, thereby weakening generalization ability and limiting applicability to new environments. To avoid this structural problem, this application adopts a small model ensemble strategy. By constructing a large number of sub-models composed of a small number of predictors, and using cross-validation to evaluate their performance, these sub-models are then integrated into the final model using performance weights. This strategy uses all possible bivariate models as the core, reducing the risk of overfitting through a small and simple model framework, while ensuring that ecological information of environmental factors is not overlooked.

[0035] Small model ensembles are particularly effective for rare species, small sample sizes, and hard-to-detect species. They maintain a stable structure even when the number of predictors far exceeds the number of existing records, allowing the model to perform well in independent region predictions. The performance advantage of small model ensembles over traditional models becomes more significant as the sample size decreases, indicating that this method is particularly important in data-scarce scenarios. This embodiment selects an ensemble of small models based on artificial neural networks to construct a habitat suitability model.

[0036] 104. Evaluate and optimize the integrated small model and decision threshold, and determine the optimal model and optimal threshold.

[0037] Specifically, as shown in Table 2, six different threshold classification methods were used to determine corresponding candidate decision thresholds for the ensemble small model. Then, for each candidate decision threshold, as shown in Table 3, seven evaluation indicators were used to comprehensively evaluate the model's predictive performance, quantifying the model's ability to detect suitable habitats, distinguish unsuitable habitats, and its variability under different thresholds. Finally, the evaluation indicator values ​​under all candidate decision thresholds were compared, as shown in Tables 4, 5, and 6, and the candidate decision threshold with the best overall performance was selected as the optimal decision threshold. The corresponding ensemble small model is the optimal model.

[0038] Table 2 Threshold Classification Methods for 6 Categories Threshold method meaning equal_sens_spec Thresholds where sensitivity equals specificity max_sens_spec The threshold that maximizes the sum of sensitivity and specificity max_jaccard The highest threshold of the Jaccard index max_sorensen The highest threshold of the Sorensen index max_fpb The highest threshold for FPB (Presence-Background F-metric) sensitivity Threshold based on specified sensitivity Table 3 Seven evaluation indicators Column names meaning threshold Threshold method thr_value Threshold value TPR_mean Average True Positive Rate TPR_sd True positive rate standard deviation TNR_mean Average True Negative Rate TNR_sd True negative rate standard deviation W_TPR_TNR_mean Weighted average TPR and TNR Table 4. Model Evaluation Results in 2022 threshold thr TPR_mean TPR_sd TNR_mean TNR_sd W_TPR_TNR_mean equal_sens_sence 0.694 0.822 0.0942 0.809 0.0849 0.813 max_fpb 0.613 0.973 0.0309 0.647 0.29 0.749 max_jaccard 0.613 0.973 0.0309 0.647 0.29 0.749 max_sens_spec 0.673 0.816 0.207 0.892 0.141 0.872 max_sorensen 0.613 0.973 0.0309 0.647 0.29 0.749 sensitivity 0.694 0.813 0.00884 0.774 0.229 0.79 Table 5. Model Evaluation Results in 2023 threshold thr TPR_mean TPR_sd TNR_mean TNR_sd W_TPR_TNR_mean equal_sens_sence 0.669 0.785 0.0627 0.795 0.0795 0.792 max_fpb 0.626 0.982 0.0213 0.547 0.224 0.686 max_jaccard 0.626 0.982 0.0213 0.547 0.224 0.686 max_sens_spec 0.653 0.872 0.0866 0.817 0.0919 0.84 max_sorensen 0.626 0.982 0.0213 0.547 0.224 0.686 sensitivity 0.663 0.81 0.0116 0.792 0.126 0.798 Table 6. Model Evaluation Results in 2024 threshold thr TPR_mean TPR_sd TNR_mean TNR_sd W_TPR_TNR_mean equal_sens_sence 0.320 0.670 0.106 0.670 0.090 0.675 max_fpb 0.308 0.920 0.075 0.578 0.146 0.810 max_jaccard 0.308 0.920 0.075 0.578 0.146 0.810 max_sens_spec 0.308 0.920 0.075 0.578 0.146 0.810 max_sorensen 0.308 0.920 0.075 0.578 0.146 0.810 sensitivity 0.313 0.818 0.013 0.636 0.119 0.763 The multi-threshold, multi-index evaluation system comprehensively measures the model's classification ability and stability, avoiding the one-sidedness of a single evaluation method. Continuous evaluation from 2022 to 2024 verified the model's cross-time stability. By comparing the selected optimal model and optimal threshold, the best balance between sensitivity and specificity of the model was ensured, improving the accuracy and reliability of the model's predictions.

[0039] 105. Using the optimal model and optimal threshold, the habitat suitability of the test area is extracted and classified to generate a habitat suitability distribution map of grassland caterpillars.

[0040] Specifically, the multi-source data of the area to be tested is preprocessed using bilinear interpolation to achieve uniform resolution and factor extraction. This data is then input into the optimal model and combined with the optimal threshold to generate an initial habitat suitability probability map. Figure 4 , Figure 5 and Figure 6 This paper presents the initial habitat suitability distribution results for grassland caterpillars from 2022 to 2024. A posterior method is then used to spatially constrain the initial results. This constraint is inserted after modeling is complete, without interfering with the earlier modeling process, and only removes suitability results outside the species' reachable areas. Finally, the constrained habitat suitability probability map is divided into multiple levels according to preset criteria of unsuitable, low-suitable, moderately suitable, and highly suitable, generating the final grassland caterpillar habitat suitability distribution map. It can also output suitability classification results under six threshold classification methods.

[0041] Preprocessing of the data for the test area ensures the consistency between the input data and the model training data, avoiding prediction bias caused by data differences. The spatial constraints of the posterior method effectively reduce the model's over-prediction of the potential distribution of grassland caterpillars without leading to higher omission errors. This makes the spatial boundary of habitat suitability more consistent with the actual ecological distribution range of grassland caterpillars. Habitat suitability classification makes the distribution map more intuitive and easy to understand, providing accurate spatial data support for grassland caterpillar pest monitoring and early warning.

[0042] Furthermore, based on the above embodiments, this embodiment constructs a multi-dimensional habitat factor system based on the ecological response relationships of grassland caterpillars to habitat factors at different life stages. This includes: determining the key life stages of grassland caterpillars, which include the overwintering stage and the larval activity stage; for the overwintering stage, selecting the first habitat factor related to overwintering survival rate based on the ecological response relationships of overwintering larvae to environmental factors; for the larval activity stage, selecting the second habitat factor related to larval growth and development based on the ecological response relationships of larvae to food sources, hydrothermal conditions, and accumulated temperature during the larval stage; and integrating the first and second habitat factors to construct a multi-dimensional habitat factor system.

[0043] Specifically, considering the growth and development patterns of grassland caterpillars and the environmental characteristics of high-altitude regions, their key life cycle stages are identified as the overwintering stage and the larval activity stage. The overwintering stage is the most critical survival bottleneck in their life cycle. The first instar larva is the only overwintering stage, and its successful overwintering directly determines the population size the following year. Overwintering larvae can only begin to move at temperatures above 7°C, and are most active at 13-16°C. Extreme low temperatures or fluctuating spring temperatures can cause mass mortality. The larval activity stage is the core stage of their feeding and growth. This stage has the most significant requirements for environmental factors such as food, water, heat, and accumulated temperature. Moreover, larvae consume a large number of forage grasses during the larval stage, and their feeding habits highly overlap with those of the main livestock in high-altitude grasslands. The habitat conditions of these two stages directly affect the occurrence and spread of grassland caterpillars.

[0044] Based on the ecological response of overwintering larvae to environmental factors, average temperature, average snow depth, altitude, slope, aspect, and soil type were selected as the primary habitat factors related to overwintering survival rate from stable topographic factors and key climatic factors. Among these, average temperature determines the metabolism and activity of overwintering larvae; snow depth provides insulation and reduces drastic fluctuations in surface temperature; topographic factors alter the local temperature environment by regulating light and heat; soil type affects the larvae's cold resistance and water balance; and microenvironments such as dry grass, rocks, and surface crevices provide physical barriers for overwintering larvae, including insulation, wind protection, and avoidance of predators.

[0045] Based on the ecological response of larvae to food sources, hydrothermal conditions, and accumulated temperature during the larval stage, four major categories of factors—meteorology, vegetation, soil, and topography—were selected as secondary habitat factors related to larval growth and development. Specific information on these factors is shown in Table 1. The definitions, thresholds, and time scales of each factor are also clearly defined. Meteorological factors include average precipitation intensity, number of days with heavy rain (rainfall ≥25mm, refer to GB / T 28592-2012), extreme rainfall, number of rainy days (rainfall ≥0.1mm, refer to GB / T 28592-2012), average temperature, effective accumulated temperature (starting from April 1st), and cumulative solar radiation within 12 hours, all on a monthly timescale. Furthermore, precipitation has a significant stage-specific impact on larval activity. Low temperatures and high precipitation during the emergence period delay larval activity, while continuous overcast and rainy weather during the breeding period inhibits adult mating. Prolonged waterlogging can lead to larval death or promote their dispersal through runoff.

[0046] Vegetation factors include vegetation type groups representing vegetation function and NDVI (Normalized Mean Vegetation Index), measured on a monthly scale. Vegetation is the main food source for grassland caterpillars. Statistical analysis of NDVI variation curves in key high-altitude areas shows that the period with NDVI > 0.4 is their critical growth season, from June to September.

[0047] Soil factors include soil moisture content (0~100%, monthly scale), soil type (static factor, soil category), soil organic matter (0~15cm), total nitrogen content, and soil pH. Increased soil moisture will increase the risk of pupa mortality, reduce the emergence rate, and cause egg decay.

[0048] Topographic factors include slope aspect (sunny / shady / partially sunny / partially shaded, static factor), slope (0~90°), and elevation (altitude). Elevation has a controlling effect on the distribution of grassland caterpillars, while slope has a slight controlling effect. The influence of sunny slopes is the most obvious.

[0049] All dynamic factors, including meteorology, vegetation, and soil moisture content, are measured on a monthly timescale, while static factors, including vegetation type groups, soil type, and topography, are measured on a fixed timescale, in accordance with the actual environmental change characteristics of each factor.

[0050] By comprehensively integrating the aforementioned primary and secondary habitat factors and combining them with multi-source data with unified resolution obtained through bilinear interpolation, a multi-dimensional habitat factor system that meets the monitoring requirements of grassland caterpillars is constructed. This system comprehensively covers all core environmental factors that affect the key life cycle stages of grassland caterpillars, and all factors have clear ecological significance.

[0051] By selecting habitat factors in stages, the habitat requirements of grassland caterpillars at different growth stages were precisely matched, avoiding the blind selection of factors. The time scale division of dynamic and static factors conforms to the actual environmental change patterns, and the inclusion of the stage-specific effects of factors such as precipitation and soil moisture makes the factor system more ecologically rational. The reference to GB / T 28592-2012 and the inclusion of all factors in Table 1 ensure the standardization and comprehensiveness of the factor system.

[0052] Furthermore, based on the above embodiments, this embodiment identifies key habitat factors in the multi-dimensional habitat factor system, including: identifying factor pairs that exhibit collinearity in the multi-dimensional habitat factor system; evaluating the contribution of each habitat factor in the multi-dimensional habitat factor system to the habitat suitability of grassland caterpillars; and determining the factor with a relatively large contribution in the factor pair as the key habitat factor.

[0053] Specifically, in the Python 3.9 analysis environment, the Pearson correlation coefficient is used to calculate the correlation coefficient between factors in the multidimensional habitat factor system. By analyzing the magnitude and significance of the correlation coefficient, factor pairs with collinearity are accurately identified. Figure 10 The correlation analysis results of meteorological factors visually demonstrate the correlations between factors such as temperature, precipitation, radiation, and accumulated temperature. Figure 11 , 12 Figures 1 and 13 show the correlation and distribution characteristics of topography, vegetation, and soil microclimate influencing factors, respectively. The core reason for eliminating collinearity factors is that collinear environmental variables reduce the model's prediction accuracy and interfere with the model's characterization of the grassland caterpillar's habitat response mechanism.

[0054] The contribution of each habitat factor was assessed using a multiple linear regression method. A linear equation was established between the habitat factors and the habitat characteristic indicators of the grassland caterpillar. The importance of each habitat factor was calculated through the equation, thereby objectively assessing the contribution of each factor to the habitat suitability of the grassland caterpillar. The expression of the linear equation is (1): (1) in, Indicates the first One habitat suitability index; Indicates the first The intercept term of each equation is a constant term; This represents the total number of key habitat factors, i.e., the number of independent variables included in the equation; , Representing the first Each habitat factor and its corresponding weight Represents the random error term. Figure 7 , Figure 8 and Figure 9 The weighted analysis results for each pair of meteorological factors show the differences in the contribution and interaction of different meteorological factors to grassland caterpillars in different months.

[0055] When screening key habitat factors, the following principles should be followed: a. The final variable factors should be the same each year. The correlation distribution calculated from the background correlation values ​​only represents the correlation characteristics at the location of occurrence; correlation may also exist for years and locations that were not observed.

[0056] b. For paired factors exhibiting collinearity, the more important one should be retained, while the less important one should be discarded. Factors with higher importance contribute more to the model and should be retained.

[0057] c. Factors that do not contribute significantly to the model should be removed.

[0058] The specific factors ultimately determined for model construction include: rainfall in April, June, and September (Rain4, rain6, rain9); number of rainy days in May and July (rd5, rd7); average temperature in April and August (LST4, LST8); average monthly radiation (ra8); accumulated temperature in July and September (GDD7, GDD9); soil moisture content in June (soilmostire6); soil pH (Phh2o); soil type (soil); average vegetation index in July and September (NDVI7, NDVI9); vegetation type; elevation; and slope. The extracted overwintering habitat condition index (ow) was also incorporated into these factors, forming the final set of input factors for the model.

[0059] The Python 3.9 analysis environment ensures the feasibility and reproducibility of statistical analysis. Pearson correlation coefficients can accurately and quickly identify factor collinearity, and multiple linear regression can objectively quantify the contribution of each factor. These three screening principles fundamentally guarantee the quality of key habitat factors and eliminate the interference of invalid factors on the model. The factor analysis results provided intuitive visual support for the screening, and the final list of key habitat factors makes the model input more targeted. The factor set after incorporating overwintering habitat condition indicators better reflects the ecological characteristics of grassland caterpillars, further improving the accuracy of modeling.

[0060] Furthermore, based on the above embodiments, this embodiment constructs spatially structured samples based on the species reachable regions of grassland caterpillars to obtain a structured sample set, including: delineating the species reachable regions based on the potential dispersal ability of grassland caterpillars; and within the species reachable regions, performing spatial structuring processing on the samples based on environmental heterogeneity and spatial distribution characteristics to obtain a structured sample set in terms of environmental gradient and spatial distribution.

[0061] Specifically, considering the actual dispersal characteristics of grassland caterpillars, the geographical features of the plateau, and the actual distribution of pest outbreaks, grassland caterpillars damaged over 16 million mu (approximately 1.1 million hectares) in xx province in 2024, including 9.958 million mu (approximately 667,000 hectares) in xx province. The study delineates the species-reachable areas of grassland caterpillars, clarifies the spatial range for model training and prediction, avoids ineffective predictions by the model for areas that grassland caterpillars theoretically cannot spread to, and mitigates the risk of over-prediction from the spatial source.

[0062] The complete pre-modeling process consists of four steps: a. Establish a regular grid in the multidimensional environmental variable space, control the grid scale by dividing the environmental gradient into five categories, and randomly select a point in each grid cell to fundamentally reduce the bias caused by the difference in sample density.

[0063] b. Within the three-dimensional environment, the optimal partitioning scale is determined by considering three factors: spatial autocorrelation, environmental similarity, and data volume differences. The environmental space accessible to species is then partitioned according to this scale to ensure that each partition is representative in terms of spatial structure and environmental gradient, and to avoid homogenization / fragmentation of sample environmental characteristics caused by partitions that are too large or too small.

[0064] c. To address the geographical sampling bias caused by the unsystematic collection of grassland caterpillar records, a buffer zone is superimposed around the existing point, and background points are extracted in a concentrated manner around it. The background thickening method is used to obtain a background sample set, so that the spatial bias of the background samples is consistent with that of the existing point, effectively reducing the interference of uneven sampling on the model structure.

[0065] d. Considering that the missing records of grassland caterpillars cannot represent the true absence of habitat, a 1:1 presence-pseudo-absence dataset is generated in each spatial partition based on the number of presence points to keep the sample distribution of presence points and pseudo-absence points balanced, thus solving the model overfitting problem caused by sample imbalance.

[0066] During the structuring of the samples, a distribution model constraint strategy for rare ecological species is simultaneously coupled to correct the spatial distribution and factor matching of the samples, further enhancing their ecological rationality. Finally, the presence-pseudo-absence datasets within each spatial partition are integrated to form the final training sample set, which maintains equilibrium in both environmental gradient and spatial distribution; this is the structured sample set.

[0067] Furthermore, based on the above embodiments, this embodiment delineates the species reachable area based on the potential dispersal ability of the grassland caterpillar, including: establishing a buffer zone centered on a known location, and using the area covered by the buffer zone as the species reachable area; and / or, based on the landscape resistance surface, simulating the dispersal cost of the grassland caterpillar in different land surface types, and extracting the reachable range of the minimum cost path as the species reachable area; and / or, generating a convex hull polygon based on known distribution points, and using the convex hull polygon and its extended area as the species reachable area; and / or, based on the boundaries of ecological geographic zoning or vegetation zoning, using the ecological unit where the grassland caterpillar has historically distributed as the species reachable area.

[0068] Specifically, the buffer method. A 100km buffer zone is established centered on the known locations of grassland caterpillars, and the coverage area of ​​the buffer zone is taken as the reachable area of ​​the species. This method is simple to operate, computationally efficient, and can quickly define the basic reachable area of ​​grassland caterpillars.

[0069] The minimum cost path method simulates the diffusion cost of grassland caterpillars in different land surface types, such as grassland, mountains, water bodies, and bare rocks, based on landscape resistance surfaces. The reachable range of the minimum cost path is extracted as the species' reachable area. This method closely reflects the actual diffusion process of grassland caterpillars, fully considering the influence of diffusion resistance such as topography and vegetation, and the delineation results are more realistic.

[0070] The convex hull polygon method generates convex hull polygons based on the known distribution points of grassland caterpillars. The convex hull polygons and their extended regions are used as the reachable regions of the species. This method can comprehensively cover the existing distribution range of grassland caterpillars and is suitable for research scenarios where the distribution points are relatively concentrated.

[0071] Ecological zoning method. Based on the boundaries of ecological geographical zoning or vegetation zoning, the ecological unit where the grassland caterpillar has historically distributed is taken as the species' accessible area. This method combines the ecological environment characteristics of high-altitude and cold regions, which is more in line with the habitat ecological requirements of grassland caterpillars, and the delineation results have the highest ecological rationality.

[0072] Each of the four delineation methods has its own advantages and can be flexibly selected based on actual research data. They can also be used in combination to make the delineation of the species' reach area more accurate and more in line with the actual dispersal characteristics of grassland caterpillars and the environmental characteristics of the plateau. This reduces the model's over-prediction from the source and improves the stability of the model's extrapolation prediction.

[0073] Furthermore, based on the above embodiments, in this embodiment, within the species-accessible area, samples are spatially structured based on environmental heterogeneity and spatial distribution characteristics to obtain a structured sample set in terms of environmental gradient and spatial distribution. This includes: dividing the species-accessible area into environmental spatial partitions to obtain multiple spatial partitions with different environmental gradients; within each spatial partition, sampling background points centered on the presence points to obtain a background sample set; combining the samples in the background sample set as absent points with the presence points in each spatial partition to construct a training sample set with a balanced number of presence and absence points; and integrating the training sample sets of each spatial partition to obtain a structured sample set in terms of environmental gradient and spatial distribution.

[0074] Specifically, within a three-dimensional environmental space, by considering three core factors—spatial autocorrelation, environmental similarity, and data volume differences—an optimization program determines the optimal partitioning scale. Based on this scale, the accessible areas of species are spatially partitioned to obtain multiple spatial partitions with different environmental gradients. Alternatively, automatic environmental spatial partitioning based on K-means, hierarchical partitioning methods based on environmental heterogeneity gradients, and regional block strategies based on ecological region boundaries can all achieve equivalent partitioning results.

[0075] Within each spatial partition, a buffer zone is overlaid centered on the location of the grassland caterpillar, and background points are extracted around the buffer zone. A background thickening method is used to obtain a background sample set, ensuring that the spatial bias of the background samples is consistent with that of the location, effectively reducing the interference of uneven geographic sampling on the model structure. Other methods can also be used for background sampling, such as bias matching based on the target population background, spatially weighted background sampling based on kernel density estimation (KDE), a dual strategy combining spatial dilution and density-weighted background points, and background point selection methods based on biased raster, etc.

[0076] Samples in the background sample set cannot be directly used as absent points because the missing records of grassland caterpillars cannot represent true habitat absence. In practice, a 1:1 presence-pseudo-absence dataset is generated within each spatial partition based on the number of presence points. The background sample set is used as the basic data source, and pseudo-absence points are constructed through random selection, spatial matching, and other methods to maintain a balanced distribution of presence and pseudo-absence points, thus solving the model overfitting problem caused by imbalanced samples.

[0077] The presence-pseudo-absence datasets within each spatial partition are integrated to form the final training sample set. This sample set maintains a balance in both environmental gradient and spatial distribution, thus becoming a structured sample set that can be directly used to build integrated small models.

[0078] Furthermore, based on the above embodiments, this embodiment constructs an integrated small model, including: constructing multiple sub-models based on key habitat factors, each sub-model being composed of at least one key habitat factor, and different sub-models using different factor combination methods; using cross-validation to evaluate the predictive performance of each sub-model, and assigning weights to each sub-model according to the predictive performance index; and weighting and integrating the multiple sub-models according to the weights to obtain the integrated small model.

[0079] Specifically, based on key habitat factors integrated with overwintering habitat condition indicators, and using all possible bivariate combinations as the core, multiple sub-models based on artificial neural networks (ANNs) were constructed. Different sub-models adopted different bivariate factor combinations. The reason for choosing the bivariate model as the core of the sub-models is that the small and simple model framework can effectively reduce the risk of overfitting, while ensuring that no ecological information of environmental factors is missed. The reason for choosing artificial neural networks as the basis of the sub-models is that they have better fitting ability and prediction accuracy under rare species and small sample conditions.

[0080] Spatial block cross-validation was employed to comprehensively and objectively evaluate the predictive performance of each sub-model based on pre-defined evaluation metrics such as true positive rate, true negative rate, and accuracy. This quantifies the predictive ability and stability of each sub-model for the habitat suitability of grassland caterpillars. The core reason for choosing spatial block cross-validation over ordinary cross-validation is that it can effectively test the extrapolation robustness of the model, better reflect the spatial heterogeneity of the Qinghai-Tibet Plateau, and avoid model overfitting caused by ordinary cross-validation.

[0081] Based on the prediction performance metrics obtained from the spatial block cross-validation, each sub-model is assigned a corresponding weight. The sub-model with better prediction performance and stronger stability is assigned a higher weight, thereby reflecting the prediction value of different sub-models and making the weight allocation more objective and scientific.

[0082] According to the assigned weights, all bivariate sub-models based on artificial neural networks are weighted and integrated to obtain the ensemble small model of this invention. The ensemble small model can also be implemented using several equivalent alternatives: ensemble small models using Random Forest (RF) or Gradient Boosting Tree (GBM); bivariate small models using MaxEnt or GLM; small model combinations using Bayesian Additive Regression Tree (BART); and using Model Averaging instead of ESM, all of which achieve equivalent ensemble results.

[0083] The bivariate artificial neural network sub-model is perfectly suited to the current research status of grassland caterpillars with small samples, multiple factors, and rare species. It reduces the risk of overfitting while preserving the ecological information of environmental factors. Spatial block cross-validation ensures the objectivity and accuracy of the sub-model's performance evaluation and effectively verifies the model's extrapolation robustness.

[0084] Furthermore, based on the above embodiments, this embodiment evaluates and optimizes the ensemble small model and decision threshold to determine the optimal model and optimal threshold, including: using multiple threshold division methods to determine the corresponding candidate decision thresholds respectively; for each candidate decision threshold, evaluating multiple evaluation indicators of the ensemble small model under the corresponding candidate decision threshold; comparing the evaluation indicator values ​​under all candidate decision thresholds, selecting the candidate decision threshold with the best overall performance as the optimal decision threshold, and using the corresponding ensemble small model as the optimal model.

[0085] Specifically, as shown in Table 2, six different threshold classification methods are used to determine corresponding candidate decision thresholds for the ensemble small model. Different threshold methods provide different decision boundaries to adapt to different pest monitoring needs. The core meaning of the six methods is as follows: a.equal_sens_spec: The threshold where sensitivity and specificity are equal, emphasizing a balance between positive and negative sample recognition capabilities.

[0086] b.max_sens_spec: The threshold that maximizes the sum of sensitivity and specificity, aiming for the best overall performance of both.

[0087] c.max_jaccard: The highest threshold for the Jaccard index, which improves classification consistency by maximizing similarity.

[0088] d.max_sorensen: The highest threshold for the Sorensen index, similar to the Jaccard index, focusing on classification similarity.

[0089] e.max_fpb: The highest threshold for FPB (Presence-Background F-index), suitable for presence-background data, which enhances prediction accuracy by optimizing the F-index.

[0090] f.sensitivity: Based on a specified sensitivity threshold, it allows the corresponding classification limits to be deduced at a set recall level.

[0091] As shown in Table 3, for each candidate decision threshold, seven evaluation metrics are used to comprehensively and quantitatively assess the predictive performance of the ensemble small model. The core meanings of the seven metrics are as follows: a.threshold: Threshold method; b.thr_value: Threshold value; c. TPR_mean: The average true positive rate, reflecting the model's ability to detect suitable habitats; d.TPR_sd: Standard deviation of true positive rate, reflecting the degree of fluctuation in the model's detection of suitable habitats; e.TNR_mean: The average true negative rate, reflecting the model's ability to distinguish unsuitable habitats; f.TNR_sd: True negative rate standard deviation, reflecting the degree of fluctuation in the model's ability to distinguish unsuitable habitats; g.W_TPR_TNR_mean: The weighted average of the true positive rate and the true negative rate, which strikes a balance between sensitivity and specificity and comprehensively measures the model performance.

[0092] The models for 2022, 2023, and 2024 were evaluated separately, and the specific evaluation values ​​for each year are shown in Tables 4, 5, and 6.

[0093] By comparing the values ​​of the seven evaluation indicators under all candidate decision thresholds, and comprehensively considering the model's detection ability, discrimination ability, stability, and overall balance, the candidate decision threshold with the best overall performance is selected as the optimal decision threshold, and the corresponding ensemble small model is the optimal model.

[0094] The evaluation results from 2022 to 2024 show that the overall predictive performance remained at a high level, but there were still some fluctuations between different years and different threshold methods. The model generally exhibited a high true positive rate (TPR) in each year, indicating that its ability to identify positive samples remained stable. However, the true negative rate (TNR) was significantly affected by the threshold method, and the accuracy of identifying negative samples fluctuated between years, reflecting that there is still room for optimization in the model's false positive control. Among all thresholds, `max_sens_spec`, which maximizes the sum of sensitivity and specificity, consistently showed the best overall performance. In 2022, this method achieved a comprehensive index of 0.872, giving it an advantage among various thresholds. In 2023, it maintained a leading position at 0.840. Although the overall performance declined slightly in 2024, this method still possessed acceptable specificity while maintaining high sensitivity, and remained among the preferred thresholds. In contrast, max_jaccard, max_sorensen, and max_fpb all showed extremely high true positive rates over the three years, but with relatively low true negative rates, indicating a greater tendency to capture positive samples and increase the risk of false positives. equal_sens_spec and sensitivity showed relatively balanced performance, achieving a reasonable balance between sensitivity and specificity in different years. Particularly in 2024, when the model tended to be conservative, the sensitivity method achieved the best overall performance that year. A comprehensive comparison of the three years' results shows that the model structure is stable and the recognition ability is reliable. The threshold strategy is the key factor affecting the differences in model performance. Among them, max_sens_spec is the most outstanding in terms of overall balance, robustness, and consistency of performance across years, and therefore can be used as the core threshold selection scheme for model application and optimization.

[0095] Furthermore, based on the above embodiments, this embodiment extracts and classifies the habitat suitability of the test area to generate a grassland caterpillar habitat suitability distribution map, including: inputting multi-source data of the test area into the optimal model to generate an initial habitat suitability probability map; spatially constraining the initial habitat suitability probability map based on the species reachable area of ​​the grassland caterpillar to obtain a constrained habitat suitability probability map; dividing the constrained habitat suitability probability map into multiple suitability levels according to a preset classification standard to generate a habitat suitability classification map; and outputting the habitat suitability classification map as the grassland caterpillar habitat suitability distribution map.

[0096] Specifically, the multi-source data of the test area undergoes standardized preprocessing, including data collection, factor extraction, and bilinear interpolation to unify resolution, ensuring that the format, resolution, and factor type of the test area data are consistent with those of the model training data. The preprocessed standardized data is then input into the optimal model, and combined with the optimal decision threshold, an initial habitat suitability probability map of the test area is generated. This probability map quantifies the habitat suitability of grassland caterpillars at each point within the test area using numerical values ​​of 0-1. Figure 4 , Figure 5 and Figure 6 The images show the probability distribution of initial habitat suitability for alpine meadow caterpillars in 2022, 2023, and 2024, respectively.

[0097] A posterior approach is employed to spatially constrain the initial habitat suitability probability map based on the defined reachable areas of grassland caterpillars. The core characteristic of the posterior approach is that the spatial constraints are inserted after modeling is complete, without interfering with the earlier modeling process. It only removes habitat suitability results outside the reachable areas of the species. This approach neither alters the model's fitting results nor over-predicts the potential distribution of grassland caterpillars, while avoiding higher omission errors. This is the core advantage of the posterior approach compared to traditional spatial constraint methods.

[0098] Based on a pre-defined grading standard of unsuitable, slightly suitable, moderately suitable, and highly suitable, the spatially constrained habitat suitability probability map is divided into multiple suitability levels. Specific grading thresholds can be adjusted according to the actual needs of pest monitoring. For example, 0-0.2 is unsuitable, 0.2-0.4 is slightly suitable, 0.4-0.6 is moderately suitable, and 0.6-1 is highly suitable. Habitat suitability grading makes the probability results, originally expressed numerically, more intuitive and easier to interpret, adapting to the actual needs of grassroots pest monitoring and control.

[0099] The resulting spatial distribution map, generated from the classified habitat suitability, is the final habitat suitability distribution map for grassland caterpillars. Simultaneously, the suitability classification results under six threshold classification methods can be used as auxiliary map outputs, providing multi-dimensional reference for the formulation of pest control strategies. For example, highly suitable areas require focused monitoring, low-suitability areas require regular inspections, and unsuitable areas can have reduced monitoring efforts.

[0100] The method of the present invention has the following advantages: (1) Advantage 1: Reduced over-prediction of potential distribution. This application constructs a variable system with clear ecological significance and adopts methods such as accessible area limitation, environmental spatial zoning, and background thickening to enable the model to fit the response mechanism of grassland caterpillar habitat based on real ecological relationships, rather than relying on simple statistical correlation to select variables. Compared with the prior art, the model of this application has stronger ecological constraints in the process of suitable area identification, avoiding the expansion of suitable areas caused by the neglect of non-climate limiting factors in traditional models, making the spatial boundary of habitat suitability prediction more consistent with the actual ecological distribution range of species, thereby significantly reducing the over-prediction of potential distribution.

[0101] (2) Advantage 2: Effectively enhances the generalization ability of the grassland caterpillar habitat suitability model. This application uses techniques such as spatial partitioning optimization, sample structure balancing, and pseudo-absence construction regularization to ensure that the training data of the model maintains a stable structure in terms of spatial autocorrelation, environmental gradient, and sample quantity, thereby effectively enhancing the generalization ability of the model. Compared with traditional niche models, which are prone to model divergence or accuracy decline in extrapolation prediction, the model of this application can still maintain high prediction stability under new regions, future climate scenarios, and unknown environmental conditions. Multi-year evaluation results show that the model using the method of this application can obtain high TPR and TNR under different threshold systems. In particular, the max_sens_spec index has consistently maintained the best performance, indicating that this application has significant technical advantages in improving the reliability of model extrapolation.

[0102] (3) Advantage 3: The overall prediction accuracy and stability are significantly better than traditional single modeling methods. This application adopts an extreme small model ensemble method based on artificial neural networks, which enables the model to still have high performance output under small sample conditions, breaking through the limitation of traditional models that are prone to overfitting and resulting in a decline in generalization ability when the sample size is insufficient. Through cross-validation weighted ensemble of a large number of bivariate small models, the model can simultaneously retain the contribution of all key predictive factors and suppress the fitting bias caused by noise variables. Compared with the prior art, this application significantly improves the structural robustness and parameter interpretability of the model, making the performance of grassland caterpillar suitability prediction more consistent across years, improving the balance between sensitivity and specificity of the model, and the overall prediction accuracy and stability are significantly better than traditional single modeling methods.

[0103] The grassland caterpillar habitat suitability distribution extraction system provided by the present invention is described below. The grassland caterpillar habitat suitability distribution extraction system described below can be referred to in correspondence with the grassland caterpillar habitat suitability distribution extraction method described above.

[0104] Figure 14 This is a schematic diagram of the grassland caterpillar habitat suitability distribution extraction system provided by the present invention. Figure 14 As shown in this embodiment, a grassland caterpillar habitat suitability distribution extraction system is provided, comprising: The acquisition module 1901 is used to acquire multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages. The determination module 1902 is used to determine the key habitat factors in the multi-dimensional habitat factor system, and to construct spatially structured samples based on the species reachable area of ​​grassland caterpillars to obtain a structured sample set. Module 1903 is used to build ensemble small models based on key habitat factors and structured sample sets; Evaluation module 1904 is used to evaluate the optimized integrated small model and decision threshold, and to determine the optimal model and optimal threshold. Extraction module 1905 is used to extract and classify the habitat suitability of the test area using the optimal model and optimal threshold, and generate a distribution map of grassland caterpillar habitat suitability.

[0105] Figure 15 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0106] like Figure 15As shown, the electronic device may include: a processor 2010, a communication interface 2020, a memory 2030, and a communication bus 2040, wherein the processor 2010, the communication interface 2020, and the memory 2030 communicate with each other through the communication bus 2040. The processor 2010 can call logical instructions in the memory 2030 to execute a method for extracting the habitat suitability distribution of grassland caterpillars. This method includes: acquiring multi-source data; constructing a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages; determining key habitat factors in the multi-dimensional habitat factor system; constructing spatially structured samples based on the species reachable areas of grassland caterpillars to obtain a structured sample set; constructing an integrated small model based on the key habitat factors and the structured sample set; evaluating and optimizing the integrated small model and decision threshold to determine the optimal model and optimal threshold; and using the optimal model and optimal threshold to extract and classify the habitat suitability of the test area to generate a grassland caterpillar habitat suitability distribution map.

[0107] Furthermore, the logical instructions in the aforementioned memory 2030 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the grassland caterpillar habitat suitability distribution extraction method provided by the above methods. This method includes: acquiring multi-source data; constructing a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages; determining key habitat factors in the multi-dimensional habitat factor system; constructing spatially structured samples based on the species reachable areas of grassland caterpillars to obtain a structured sample set; constructing an integrated small model based on the key habitat factors and the structured sample set; evaluating and optimizing the integrated small model and decision threshold to determine the optimal model and optimal threshold; and using the optimal model and optimal threshold to extract and classify the habitat suitability of the test area to generate a grassland caterpillar habitat suitability distribution map.

[0109] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the grassland caterpillar habitat suitability distribution extraction method provided by the above methods. The method includes: acquiring multi-source data; constructing a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages; determining key habitat factors in the multi-dimensional habitat factor system; constructing spatially structured samples based on the species reachable areas of grassland caterpillars to obtain a structured sample set; constructing an integrated small model based on the key habitat factors and the structured sample set; evaluating and optimizing the integrated small model and decision threshold to determine the optimal model and optimal threshold; and using the optimal model and optimal threshold to extract and classify the habitat suitability of the test area to generate a grassland caterpillar habitat suitability distribution map.

[0110] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0111] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0112] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting habitat suitability distribution of grassland caterpillars, characterized in that, include: Acquire multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages. Key habitat factors in the multi-dimensional habitat factor system were identified, and spatially structured samples were constructed based on the species reachable areas of grassland caterpillars to obtain a structured sample set. Based on the key habitat factors and the structured sample set, an ensemble small model is constructed. The ensemble small model is obtained by weighted ensemble of multiple sub-models, and each sub-model is constructed by combining at least one key habitat factor. Evaluate and optimize the ensemble small model and decision threshold to determine the optimal model and optimal threshold; Using the optimal model and optimal threshold, habitat suitability of the test area is extracted and classified to generate a habitat suitability distribution map of grassland caterpillars.

2. The method for extracting habitat suitability distribution of grassland caterpillars according to claim 1, characterized in that, The above describes the construction of a multi-dimensional habitat factor system based on the ecological response relationships of grassland caterpillars to habitat factors at different life stages, including: The key life cycle stages of the grassland caterpillar were determined, including the overwintering stage and the larval activity stage. For the overwintering stage, based on the ecological response of overwintering larvae to environmental factors, the first habitat factor related to overwintering survival rate was selected; For the aforementioned larval activity stages, based on the ecological response of the larvae to food sources, hydrothermal conditions, and accumulated temperature during growth, secondary habitat factors related to larval growth and development are selected. The first habitat factor and the second habitat factor are integrated to construct a multi-dimensional habitat factor system.

3. The method for extracting the habitat suitability distribution of grassland caterpillars according to claim 1, characterized in that, The determination of key habitat factors in the multi-dimensional habitat factor system includes: Identify factor pairs exhibiting collinearity within the multidimensional habitat factor system; To assess the contribution of each habitat factor in the multi-dimensional habitat factor system to the habitat suitability of grassland caterpillars; The factors with relatively large contributions in the factor pair are identified as key habitat factors.

4. The method for extracting the habitat suitability distribution of grassland caterpillars according to claim 1, characterized in that, The spatially structured sample construction based on the species reachable regions of grassland caterpillars yields a structured sample set, including: Based on the potential dispersal ability of grassland caterpillars, the areas that the species can reach are delineated; Within the reachable area of ​​the species, the samples are spatially structured based on environmental heterogeneity and spatial distribution characteristics to obtain a structured sample set in terms of environmental gradient and spatial distribution.

5. The method for extracting the habitat suitability distribution of grassland caterpillars according to claim 4, characterized in that, The delineation of species reachable areas based on the potential dispersal ability of grassland caterpillars includes: Establish a buffer zone centered on a known point of existence, and define the area covered by the buffer zone as the reachable region of the species; and / or, Based on landscape resistance surfaces, the dispersal cost of grassland caterpillars in different land surface types is simulated, and the reachable range of the minimum cost path is extracted as the species' reachable area; and / or, A convex hull polygon is generated based on known distribution points, and the convex hull polygon and its extended region are used as the reachable region of the species; and / or, Based on the boundaries of ecological geographical zoning or vegetation zoning, the ecological units where the historical distribution of grassland caterpillars is located are taken as the reachable areas of the species.

6. The method for extracting the habitat suitability distribution of grassland caterpillars according to claim 4, characterized in that, Within the species' reachable region, samples are spatially structured based on environmental heterogeneity and spatial distribution characteristics to obtain a structured sample set in terms of environmental gradient and spatial distribution, including: The reachable area of ​​the species is divided into environmental spatial partitions to obtain multiple spatial partitions with different environmental gradients; Within each spatial partition, background points are sampled centered on the existing point to obtain a background sample set; The samples in the background sample set are used as absent points and combined with the existent points in each spatial partition to construct a training sample set with a balanced number of existent and absent points. The training sample sets of each spatial partition are integrated to obtain a structured sample set in terms of environmental gradient and spatial distribution.

7. The method for extracting habitat suitability distribution of grassland caterpillars according to claim 1, characterized in that, The construction of the integrated small model includes: Based on the key habitat factors, multiple sub-models are constructed, each of which is composed of at least one key habitat factor, and different sub-models adopt different factor combination methods; Cross-validation was used to evaluate the predictive performance of each sub-model, and weights were assigned to each sub-model based on the predictive performance metrics. The multiple sub-models are weighted and integrated according to the weights to obtain an integrated small model.

8. The method for extracting the habitat suitability distribution of grassland caterpillars according to claim 1, characterized in that, The evaluation and optimization of the ensemble small model and decision threshold, to determine the optimal model and optimal threshold, includes: Multiple threshold partitioning methods are used to determine the corresponding candidate decision thresholds; For each candidate decision threshold, evaluate the ensemble small model on multiple evaluation metrics at the corresponding candidate decision threshold; By comparing the evaluation index values ​​under all candidate decision thresholds, the candidate decision threshold with the best overall performance is selected as the optimal decision threshold, and the corresponding ensemble small model is selected as the optimal model.

9. The method for extracting the habitat suitability distribution of grassland caterpillars according to any one of claims 1-8, characterized in that, The habitat suitability of the area to be tested is extracted and classified to generate a distribution map of grassland caterpillar habitat suitability, including: The multi-source data of the area to be tested are input into the optimal model to generate an initial habitat suitability probability map; Based on the species reachable area of ​​grassland caterpillars, spatial constraints are applied to the initial habitat suitability probability map to obtain a constrained habitat suitability probability map. According to the preset grading criteria, the constrained habitat suitability probability map is divided into multiple suitability levels to generate a habitat suitability grading map. The habitat suitability grading map is output as a habitat suitability distribution map of grassland caterpillars.

10. A system for extracting habitat suitability distribution of grassland caterpillars, characterized in that, include: The acquisition module is used to acquire multi-source data and construct a multi-dimensional habitat factor system based on the ecological response relationship of grassland caterpillars to habitat factors at different life stages. The determination module is used to determine the key habitat factors in the multi-dimensional habitat factor system, and to construct spatially structured samples based on the species reachable area of ​​grassland caterpillars to obtain a structured sample set. A construction module is used to construct an integrated small model based on the key habitat factors and the structured sample set. The integrated small model is obtained by weighted integration of multiple sub-models, and the sub-model is constructed by combining at least one key habitat factor. The evaluation module is used to evaluate and optimize the integrated small model and decision threshold, and determine the optimal model and optimal threshold. The extraction module is used to extract and classify the habitat suitability of the test area using the optimal model and the optimal threshold, and generate a habitat suitability distribution map of grassland caterpillars.

Images