Grassland harmful insect suitable area monitoring method, device and system and storage medium

Abstract

The application discloses a grassland harmful insect suitable area monitoring method and device, system and storage medium, and comprises the following steps: acquiring sample point data containing longitude and latitude, ecological environment data and locust occurrence conditions; performing index screening on the sample point data to obtain key environmental factors affecting the growth and development of harmful insects; constructing and training a grassland harmful insect suitable area monitoring composite model based on an XGB-GWLR hybrid model according to the sample point data and the key environmental factors; extracting corresponding grid data pixel values by using field investigation points; predicting the harmful insect occurrence probability of each pixel by using the trained grassland harmful insect suitable area monitoring composite model; generating a spatial distribution grid map according to the prediction result, and dividing the risk levels according to the probability values. The technical scheme disclosed by the application can realize accurate extraction of a potential occurrence area of grassland harmful insects.

Description

Technical Field

[0001] This invention belongs to the field of pest and disease monitoring technology, specifically relating to a method, device, system, and storage medium for monitoring suitable habitats of harmful insects in grasslands. Background Technology

[0002] With global warming and excessive human exploitation of grassland resources, grassland landscape patterns have undergone significant changes. The original habitats of harmful grassland insects have shifted, while new habitats have emerged. The timing, location, and severity of harmful insect outbreaks are highly uncertain, increasing the difficulty of grassland pest control. Locusts, as a major pest in grasslands, pose a serious threat to grassland ecosystems and the stable development of the livestock economy. Therefore, identifying the main environmental factors affecting the growth and development of grassland locusts, elucidating their potential outbreak mechanisms, and constructing a monitoring and early warning model for suitable grassland locust habitats are prerequisites for effectively achieving early warning and preventing large-scale locust outbreaks. However, the current monitoring indicator system for suitable grassland locust habitats is chaotic, with the determination of indicator weight coefficients relying excessively on expert experience; habitat monitoring is mainly achieved through hierarchical analysis and niche models; and the spatial heterogeneity of indicator weight coefficients is often ignored in model construction. Furthermore, previous models have weak generalization ability and are difficult to extrapolate. Therefore, it is urgent to construct a complete method for monitoring and early warning of suitable locust habitats, specifically including indicator formulation, screening, determination of key indicator weight coefficients, and suitable habitat model extraction. This provides a reference and technical support for the formulation of early monitoring and prevention measures for grassland locust plagues. Summary of the Invention

[0003] To address the problems existing in the prior art, this invention provides a method, device, system, and storage medium for monitoring suitable habitats of harmful insects in grasslands. By synergistically coupling the high prediction accuracy of the XGBoost model with the excellent spatial nonstationarity modeling capability of the GWLR model, the spatial heterogeneity of the weights of each influencing factor is fully quantified while ensuring overall accuracy. This provides a solid scientific basis and effective technical support for the accurate monitoring and early warning of suitable habitats for locusts in grasslands.

[0004] To achieve the above objectives, the present invention provides the following solution: A method for monitoring suitable habitats of harmful insects in grasslands, comprising: Acquire sample point data including latitude and longitude, ecological environment data, and locust occurrence status; among which, key environmental factors include: soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, and aspect. By screening the data from the sample points, the key environmental factors affecting the growth and development of harmful insects were identified. Based on sample point data and key environmental factors, a composite model for monitoring suitable habitats of harmful insects in grasslands based on the XGB-GWLR hybrid model was constructed and trained. The corresponding raster data pixel values ​​are extracted from field survey points; the occurrence probability of harmful insects in each pixel is predicted using a trained composite model for monitoring suitable habitats of harmful insects in grasslands; the prediction results are used to generate a spatial distribution raster map, and risk levels are classified according to the probability values.

[0005] As a preferred approach, the index screening for sample point data includes: multicollinearity test, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values.

[0006] Preferably, a composite monitoring model for suitable habitats of harmful grassland insects is constructed based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: in, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. α represents the probability of occurrence of the Asian caricature locust based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models.

[0007] This invention also provides a monitoring device for suitable habitats of harmful insects in grasslands, comprising: The first processing module is used to acquire sample point data including latitude and longitude, ecological environment data, and locust occurrence status; among which, key environmental factors include: soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, and aspect. The second processing module is used to screen the sample point data for indicators to obtain the key environmental factors that affect the growth and development of harmful insects. The third processing module is used to construct and train a composite model for monitoring suitable habitats of harmful insects in grasslands based on sample point data and key environmental factors. The fourth processing module is used to extract the corresponding raster data pixel values ​​from field survey points; use a trained composite model for monitoring suitable habitats of harmful insects in grasslands to predict the probability of harmful insect occurrence for each pixel; generate a spatial distribution raster map from the prediction results; and classify the risk level according to the probability values.

[0008] As a preferred option, the second processing module performs index screening on the sample point data, including: multicollinearity test, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values.

[0009] Preferably, the third processing module constructs a composite monitoring model for suitable habitats of harmful grassland insects based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: in, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. α represents the probability of occurrence of the Asian caricature locust based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models.

[0010] The present invention also provides a grassland harmful insect habitat monitoring system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a grassland harmful insect habitat monitoring method when run by the processor.

[0011] The present invention also provides a storage medium storing a computer program, which executes a method for monitoring suitable habitats of harmful insects in grasslands when running.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: A balance between accuracy and interpretability: Through a hybrid model architecture, XGBoost's powerful nonlinear prediction capabilities are retained to ensure accuracy, while GWLR provides a clear explanation of spatial heterogeneity, enabling the answer to the key question of "which environmental factor plays a dominant role in which region". Strong robustness: The introduction of adaptive regularization guided by the global model effectively prevents the overfitting problem of GWLR when there are few local samples or poor data quality, and improves the generalization ability of the model. High practical value: This method can directly output a spatially intuitive risk distribution map, which makes it easier for agricultural management departments to accurately locate high-risk areas, rationally allocate prevention and control resources, and realize the transformation from "general prevention" to "precise prevention". Attached Figure Description

[0013] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a method for monitoring suitable habitats of harmful insects in grasslands, as described in an embodiment of the present invention. Figure 2 Correlation between ecological and environmental factors; Figure 3 The relative influence and importance of continuous variables on grasshopper occurrence. Detailed Implementation

[0015] 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.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Example 1 like Figure 1 As shown, this invention provides a method for monitoring suitable habitats of harmful insects in grasslands. It utilizes XGBoost to obtain global feature importance to guide the regularization process of a local GWLR model, and weights and fuses the prediction results of both models to achieve complementary advantages. The specific steps of the method are as follows: S1: Data Preprocessing and Feature Engineering Read sample point data containing latitude and longitude, ecological environment data (such as soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, aspect, etc.) and locust occurrence status (binary variable, 0 / 1). Label and encode the categorical variables, and remove missing values ​​to ensure data quality.

[0018] S2: Environmental Factor Screening and Optimization After data preprocessing, to ensure the optimal set of environmental factors for the input model, avoid multicollinearity, and identify core driving factors, this invention employs a multi-stage, multi-method fusion index selection strategy. Specifically, this includes: multicollinearity testing, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values.

[0019] S3: XGBoost Model Training and Global Feature Weight Calculation The hyperparameters of the XGBoost model are optimized using grid search cross-validation, and the optimal model is selected using the area under the curve (AUC) as the evaluation metric.

[0020] The optimal XGBoost model was trained using the full dataset, and its feature importance scores were extracted, normalized, and used as the global weights for each environmental factor.

[0021] S4: Build and train the XGB-GWLR hybrid model Specifically, it includes: ① Data Input and XGBoost Global Learning: The model receives sample data with geographic labels. XGBoost is trained first as a global model, and its core output is the global feature importance weights (W). XGBj This weight quantifies the global average impact of each environmental factor (such as soil type and vegetation index) on locust occurrence. ② Generation of adaptive regularized weights: This involves transforming the global weights generated by XGBoost into prior knowledge that guides the local model. This is a key decision point. For continuous environmental factors (such as elevation), the importance weights of XGBoost are used directly. ). For categorized environmental factors (such as soil type), their weights are attenuated ( This move aims to balance the model complexity introduced by categorical variables. ③ GWLR Local Modeling and Spatial Weight Calculation: In parallel, the GWLR model begins to work. It first calculates the spatial weights (wi) between sample points using a Gaussian kernel function, based on the first law of geography (distance decay effect). ij This ensures that neighboring points have a greater impact on the local model. ④ Local Parameter Estimation: GWLR fits a local logistic regression model for each geographic location. The innovation here lies in the regularization term introduced into its objective function, the penalty strength of which is determined by the adaptive regularization weights generated in the second step (…). XGBoost considers important features to be more flexible in the local model, while less important features are subject to stronger constraints, effectively preventing overfitting and improving model robustness. ⑤ Predictive fusion: The predicted probabilities from the GWLR output, taking into account spatial heterogeneity, are fused together. ) and the global prediction probability output by XGBoost ( The model is fused using a weighted average. The weight coefficient α is set to 0.7, which means that the model focuses more on interpretable GWLR results. At the same time, XGBoost is used for global correction, and finally a spatial distribution map of locust occurrence risk with high accuracy and high interpretability is generated.

[0022] S5: Regional-scale prediction and mapping of locust habitats The corresponding raster data pixel values ​​are extracted from field survey points; a trained composite model for monitoring suitable habitats of grassland pests is used to predict the probability of pest occurrence in each pixel; the prediction results are used to generate a spatial distribution raster map, and risk levels are classified according to the probability values: unsuitable habitat: P<0.5, low suitability: 0.5≤P≤0.6, medium suitability: 0.6≤P≤0.75, high suitability: 0.75≤P≤0.9, and extremely suitable habitat: P>0.9, thus achieving visualized early warning. Example 2 This invention uses the Asian grasshopper (Oedaleus asiaticus) as the monitoring and early warning target in the grasslands of Inner Mongolia. This insect species is a dominant grasshopper species in the northern grasslands of China, exhibiting migratory behavior and posing a significant threat to grassland ecology and animal husbandry. The monitoring method for suitable habitats of harmful grassland insects in this invention includes: Step 1: Data Sources and Processing 1.1 Data Collection from Locust Field Survey Historical field survey data of the Asian carp (Hemiberlesia asiatica) were collected. The data from each year's survey were processed, removing data with poor quality or missing information. Only one survey point was retained within a one-kilometer radius. Locust density data were recoded: a density greater than or equal to 1 was assigned a value of 1, indicating a locust infestation; a density of 0 was assigned a value of 0, indicating no infestation. When there were too few or no survey points with a locust density of 0, the same number of spurious infestation points were randomly generated based on the spatial distribution of historical locust field survey data, and coded as 0. This generated the Asian carp field survey dataset.

[0023] 1.2 Ecological and Environmental Data Processing Based on the physiological characteristics, habits, and developmental cycle of the Asian pit viper, and referring to existing literature, a preliminary indicator system influencing the growth and development of the Asian pit viper was proposed, including 64 indicators related to vegetation, soil, climate, and topography. Ecological and environmental data were preprocessed to generate raster data with the same row and column numbers, coordinates, and projection.

[0024] Step 2: Data Preprocessing and Feature Selection 2.1 Data Preprocessing Overlay analysis was performed on field survey data of the Asian carrion locust and a raster dataset of ecological and environmental factors to extract the pixel values ​​corresponding to each survey point. Missing data points were directly deleted to ensure data accuracy. Categorical variables, such as soil type and grassland type, were coded and uniformly converted into numerical variables.

[0025] 2.2 Multi-stage environmental factor screening To reduce the redundancy of environmental factors, improve computational efficiency, and ensure that the ecological and environmental factors input to the model are closely related to the growth and development of the Asian grasshopper, a three-step screening method was used to select core factors: (1) Variance inflation factor analysis Objective: To eliminate environmental factors with severe multicollinearity and ensure the stability and interpretability of model estimation.

[0026] Method: Calculate the variance inflation factor (VIF) for each environmental factor. VIF quantifies the extent to which the variance of its regression coefficient increases due to the correlation between independent variables. The formula is: VIF = 1 / (1-R²), where R² is the coefficient of determination obtained by regressing this factor against all other factors.

[0027] Screening criteria: Stepwise elimination method is used.

[0028] VIF<5: This indicates weak collinearity, and the factor is retained.

[0029] 5 ≤ VIF ≤ 10: This indicates moderate collinearity, which needs to be carefully judged in conjunction with other methods.

[0030] A VIF value greater than 10 indicates severe multicollinearity, and this factor should be removed. Factors with lower VIF values ​​are typically retained to eliminate redundant information.

[0031] Variance inflation factor analysis was performed on 64 variables to remove environmental factors that had a weak impact on the occurrence of Asian trolley locusts.

[0032] (2) Correlation analysis of variables Objective: To further analyze the linear correlation between pairs of variables and avoid introducing highly repetitive information.

[0033] Method: Calculate the Pearson correlation coefficient (r) or Spearman rank correlation coefficient between each pair of the remaining environmental factors (for non-normally distributed data).

[0034] Screening criteria: When |r|>0.8, the two variables are considered highly correlated. In this case, the results of the XGBoost-Shap analysis are combined to remove factors with lower SHAP importance, retaining the core factors that contribute more to locust occurrence prediction, thereby simplifying the model structure (e.g., ...). Figure 2 (As shown).

[0035] (3) Identification of core driving factors based on XGBoost and SHAP values Objective: To screen out the core environmental drivers that have the most significant impact on locust outbreaks from the perspectives of predictive ability and contribution.

[0036] Methods: An initial XGBoost model is trained using residual environmental factors. The SHAP value for each sample is calculated. SHAP is a unified interpretation method based on game theory, which can accurately reflect the marginal contribution of each feature in the prediction of each sample.

[0037] The global SHAP importance ranking is obtained by calculating the average absolute value of the SHAP values ​​for each feature. This ranking is more consistent and reliable than the feature importance ranking built into XGBoost (such as gain).

[0038] Selection criteria: Based on SHAP importance ranking, the top-ranked core factors are retained. This ensures that the variable set input to the XGB-GWLR hybrid model contains key information while being as concise as possible, improving the model's efficiency and generalization ability (e.g., Figure 3 (As shown).

[0039] After the above three-step screening, a set of core environmental driving factors with low redundancy, good representativeness and strong predictive ability was obtained. Finally, 28 environmental factors were retained as model input variables.

[0040] Step 3: Construction, training, and validation of the XGB-GWLR hybrid model 3.1 XGBoost Global Model Training and Weight Calculation The aforementioned 28 environmental factors and locust field survey data (including spurious occurrence points) were input into the XGBoost model. The binarized data of locust occurrence points and spurious occurrence points were used as dependent variables, and the 28 environmental factors were used as independent variables to train the XGBoost model (Equation 1). The optimal hyperparameter combination of the XGBoost classifier was determined through GridSearchCV optimization. Based on the trained model, the feature importance scores of each environmental factor were extracted (Equation 2). The importance scores were then normalized (Equation 3). (1) (2) (3) In the formula, L XGB This represents the total loss function of the XGBoost model. n and K These represent the number of variables and the number of trees in the model, respectively. y i Location i The values ​​of the two variables correspond to the actual observed values ​​of what happened in the field. The function represents the predicted probability of a grasshopper outbreak. Depend on express, f k The model representing the k-th tree. Ω(f k ) Indicates the first k The regularization term for the trees (Formula 4). I j This represents the feature importance score of the j-th environmental factor. t Indicates the first k A split node in a tree, v(t) It is the feature index used when node t splits. h(v(t)=j) Represents an indicator function. This represents the gain after a node splits. Indicates the first j Normalized weight vector of environmental factors. (4) In the formula, The penalty coefficient representing the number of leaf nodes, T Represents the number of leaf nodes in the tree. λ Represents the regularization coefficient ( λ =1), Indicates the first j The weights of each environmental factor.

[0041] Potential suitable habitats for the Asian caricature locust in the Inner Mongolian grasslands, determined using the global weight coefficients of the XGBoost model. Specific model details: (5) In the formula, σ represents the predicted probability output of the XGBoost model, and σ represents the sigmoid function. It is responsible for mapping the original output to the [0,1] interval. f k (x i ) Representing variables x i In the kThe weights on each tree.

[0042] 3.2 Adaptive Regularized XGB-GWLR Hybrid Model Training During the hybrid model training phase, the great circle distance between sample points is first calculated based on the Haversine formula, and a Gaussian kernel function is used to construct the spatial weight matrix. The optimal bandwidth is determined using the AICc criterion. Subsequently, the model is constructed, and the global feature importance weights output by the XGBoost model (…) are used to… WXGB j ) is transformed into adaptive regularized weights in a geographic weighted logistic regression (GWLR) model. ), where continuous variables (such as precipitation) directly use the original weights ( j ), while categorical variables (such as soil type) are reduced in size. j The regularization strength coefficient λ=5 was set. During model training, iterative weighted least squares fitting was used for point-by-point fitting. The model formula is referenced in (6-9). (6) (7) (8) (9) In the formula, wij (Values ​​range from 0 to 1) Represents the site j The observed values ​​for the estimated station i parameter βi Influence weight; dij Point i With point j The distance between them b Represents bandwidth. ui,vi and uj, vj They represent the sites respectively. i and sites j Latitude and longitude coordinates. pjGWLR The probability predicted by the adaptively regularized XGB-GWLR mixture model; σ This represents the Sigmoid function. n and m Representing the total number of variables and the first variable, respectively. m One variable, β0(uj,vj) The intercept term represents spatial variation. βm(uj,vj) Indicates the first m The spatial variation regression coefficients of the variables xjm It is a website j First m The values ​​of the variables. Indicates site iThe optimal set of coefficients for all variables, including the intercept term, at a given point, i.e. ( T ), yj These are the binary variable observations for station j. λ Let λ be the regularization intensity hyperparameter (set to λ=5 in this model). Indicates the first m Adaptive penalty weights for each variable (categorical variable: Continuous variables: ), β2 im Indicates site i First m The square of the regression coefficients of each variable.

[0043] A composite monitoring model for suitable habitats of harmful grassland insects is constructed based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: (10) In the formula, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. This represents the probability of Asian caricature occurrence based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models, which is set to 0.7 here.

[0044] The final model showed good performance on the training set, with AUC=0.813, prediction accuracy of 0.738, and pseudo R² of 0.252, indicating that the model has high explanatory power and robustness.

[0045] Step 4: Suitable habitat prediction mapping and early warning application Taking the extraction of potential suitable habitats in 2022 as an example, 28 environmental factors were input into the model to obtain the suitability index, suitability grade, and spatial distribution map of the Asian small locust in the Inner Mongolia grassland in 2022.

[0046] Example 3: This invention also provides a monitoring device for suitable habitats of harmful insects in grasslands, comprising: The first processing module is used to acquire sample point data including latitude and longitude, ecological environment data, and locust occurrence status; among which, key environmental factors include: soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, and aspect. The second processing module is used to screen the sample point data for indicators to obtain the key environmental factors that affect the growth and development of harmful insects. The third processing module is used to construct and train a composite model for monitoring suitable habitats of harmful insects in grasslands based on sample point data and key environmental factors. The fourth processing module is used to extract the corresponding raster data pixel values ​​from field survey points; use a trained composite model for monitoring suitable habitats of harmful insects in grasslands to predict the probability of harmful insect occurrence for each pixel; generate a spatial distribution raster map from the prediction results; and classify the risk level according to the probability values.

[0047] As one embodiment of the present invention, the second processing module performs index screening on the sample point data, including: multicollinearity test, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values.

[0048] As one embodiment of the present invention, the third processing module constructs a composite monitoring model for suitable habitats of harmful grassland insects based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: in, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. α represents the probability of occurrence of the Asian caricature locust based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models.

[0049] Example 4 The present invention also provides a grassland harmful insect habitat monitoring system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a grassland harmful insect habitat monitoring method when run by the processor.

[0050] Example 5 The present invention also provides a storage medium storing a computer program, which executes a method for monitoring suitable habitats of harmful insects in grasslands when running.

[0051] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for monitoring suitable habitats of harmful insects in grasslands, characterized in that, include: Acquire sample point data including latitude and longitude, ecological environment data, and locust occurrence status; among which, key environmental factors include: soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, and aspect. By screening the data from the sample points, the key environmental factors affecting the growth and development of harmful insects were identified. Based on sample point data and key environmental factors, a composite model for monitoring suitable habitats of harmful insects in grasslands based on the XGB-GWLR hybrid model was constructed and trained. The corresponding raster data pixel values ​​are extracted from field survey points; the occurrence probability of harmful insects in each pixel is predicted using a trained composite model for monitoring suitable habitats of harmful insects in grasslands; the prediction results are used to generate a spatial distribution raster map, and risk levels are classified according to the probability values. The index screening for sample point data includes: multicollinearity test, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values. A composite monitoring model for suitable habitats of harmful grassland insects is constructed based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: in, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. α represents the probability of occurrence of the Asian caricature locust based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models. During the hybrid model training phase, the great circle distance between sample points is first calculated based on the Haversine formula, and a Gaussian kernel function is used to construct the spatial weight matrix. The optimal bandwidth is determined using the AICc criterion. Subsequently, the model is constructed, and the global feature importance weights output by the XGBoost model are applied. WXGB j Adaptive regularization weights in a geographic weighted logistic regression (GWLR) model Continuous variables directly use their original weights. j Categorical variables, on the other hand, are reduced in size. j In model training, iterative weighted least squares method is used for point-by-point fitting.

2. A monitoring device for suitable habitats of harmful insects in grasslands, characterized in that, include: The first processing module is used to acquire sample point data including latitude and longitude, ecological environment data, and locust occurrence status; among which, key environmental factors include: soil type, soil moisture, surface temperature, vegetation type, vegetation index, altitude, slope, and aspect. The second processing module is used to screen the sample point data for indicators to obtain the key environmental factors that affect the growth and development of harmful insects. The third processing module is used to construct and train a composite model for monitoring suitable habitats of harmful insects in grasslands based on sample point data and key environmental factors. The fourth processing module is used to extract the corresponding raster data pixel values ​​from field survey points; use a trained composite model for monitoring suitable habitats of harmful insects in grasslands to predict the probability of harmful insect occurrence in each pixel; generate a spatial distribution raster map from the prediction results; and classify the risk level according to the probability values. The second processing module performs index screening on the sample point data, including: multicollinearity test, correlation analysis between variables, and identification of core driving factors based on XGBoost and SHAP values. The third processing module constructs a composite monitoring model for suitable habitats of harmful grassland insects based on the XGBoost global prediction model and the adaptive regularized XGB-GWLR prediction model. The composite monitoring model for suitable habitats of harmful grassland insects is as follows: in, This represents the suitability index obtained based on a composite model for monitoring suitable habitats of harmful insects in grasslands. This represents the occurrence probability of the Asian caricature locust based on the adaptive regularized XGB-GWLR prediction model. α represents the probability of occurrence of the Asian caricature locust based on the XGBoost global prediction model; α represents the hyperparameter used to determine the global coefficients of the global and local models.

3. A monitoring system for suitable habitats of harmful insects in grasslands, characterized in that, include: The system includes a memory and a processor, wherein the memory stores a computer program that is executed by the processor, and the computer program, when executed by the processor, performs the grassland pest habitat monitoring method as described in claim 1.

4. A storage medium, characterized in that, The storage medium stores a computer program, which executes the grassland pest habitat monitoring method as described in claim 1 when running.

Images