Remote sensing monitoring method and system for interaction of biochemical components of phyllostachys pubescens canopy under stress of phyllostachys pubescens poison moths

Abstract

The invention relates to a remote sensing monitoring method and system for interaction of biochemical components of a phyllostachys pubescens forest canopy under stress of phyllostachys pubescens, and belongs to the fields of forest science, geography, ecology and remote sensing science and technology. According to the method, a dynamic monitoring and evolution mechanism analysis system of the moso bamboo forest canopy biochemical components is constructed by fusing a multi-temporal remote sensing image inversion technology, a structural equation model SEM and an interpretable machine learning algorithm. According to the method, multi-temporal remote sensing image inversion, SEM, Stacking ensemble learning and interpretable machine learning are combined, so that dynamic monitoring and evolution mechanism analysis of the biochemical components of the moso bamboo forest canopy are realized.

Description

Technical Field

[0001] This invention belongs to the fields of forestry, geography, ecology and remote sensing science and technology, and relates to a remote sensing monitoring method and system for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth. Background Technology

[0002] Moso bamboo forests are an important ecological and economic resource in subtropical regions, but they are often attacked by pests such as the bamboo tussock moth (Pantanaphyllostachysae Chao), which leads to damage to the canopy structure and disorder of biochemical components, seriously affecting the health of bamboo forests. my country has 7.56 million hectares of bamboo forests, accounting for 3.31% of the total forest area; among them, moso bamboo forests cover 5.2776 million hectares, accounting for 69.78% of the total bamboo forest area. The bamboo tussock moth is listed as a Class III harmful forest pest, and it has periodic outbreaks in many provinces, becoming an important threat to the development of the bamboo industry. Existing research lacks in-depth analysis of the interaction of multiple components and their dynamic evolution mechanism, and traditional methods are difficult to quantify the synergistic effects of pest stress and seasonal changes on biochemical components, resulting in insufficient precision of pest control strategies. The main bottlenecks are reflected in the following three aspects: (1) Insufficient monitoring efficiency and scale connection: manual patrols and ground surveys are inefficient and cannot meet the needs of large-scale dynamic monitoring. Although remote sensing technology can partially replace manual labor, existing research is still insufficient in the collaborative application of multi-scale data from the ground, air, and space. Furthermore, the inversion of single biochemical components is easily affected by the heterogeneity of canopy structure, soil background reflection, and lighting conditions, resulting in large errors. It is also unable to analyze the synergistic changes of key components such as carotenoids, equivalent water thickness, and dry matter, as well as their response to pests. (2) Limitations of physical model adaptability: The parameterization process of mainstream physical models (such as PROSAIL) depends on the structure and biochemical characteristics of specific vegetation. As a special type of grass, the canopy structure of moso bamboo forest differs from the default assumptions of the model, leading to fluctuations in inversion accuracy. Existing models do not adequately analyze the interaction mechanism of biochemical components at the canopy scale, and mostly remain at the leaf scale. They lack systematic modeling and are difficult to quantify resource allocation strategies under pest stress. For example, the response pattern of chlorophyll-water interaction under different levels of pests is still unclear, and the trade-off between dry matter accumulation and photosynthetic pigments has not formed a unified theoretical framework; (3) Difficulty in analyzing dynamic evolution mechanisms: Although machine learning models have been used for pest prediction, their "black box" characteristics make it difficult to analyze the driving mechanism, and they have not effectively integrated the influence of seasonal meteorological fluctuations on the evolution path of biochemical component interactions, which limits the ecological guidance value of the models. Therefore, it is urgent to develop a technical system that can integrate multi-temporal remote sensing inversion, interaction quantification and evolution mechanism analysis, so as to fully reveal the dynamic adaptation pattern of biochemical components in the bamboo forest canopy under pest stress and provide a scientific basis for pest control and ecological management. Summary of the Invention

[0003] The purpose of this invention is to provide a remote sensing monitoring method and system for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth. By combining multi-temporal remote sensing image inversion, SEM, Stacking ensemble learning and interpretable machine learning, dynamic monitoring and evolution mechanism analysis of biochemical components in the canopy of moso bamboo forests can be achieved.

[0004] To achieve the above objectives, the technical solution of the present invention is: a remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forest under the stress of the bamboo tussock moth, which integrates multi-temporal remote sensing image inversion technology, structural equation modeling (SEM), and interpretable machine learning algorithms to construct a dynamic monitoring and evolution mechanism analysis system for biochemical components in the canopy of moso bamboo forest.

[0005] Furthermore, the method includes the following steps:

[0006] (1) Acquire multi-temporal multispectral remote sensing images and preprocess them to generate representative image datasets for spring, summer and autumn;

[0007] (2) Based on the extreme gradient boosting XGBoost classification model, combined with the Normalized Difference Vegetation Index (NDVI), Bamboo Index (BI), Vegetation Health and Stress Index (SI), and Red-edge Normalized Difference Vegetation Index (NDVI), the model was developed. re1 The spatial distribution information of moso bamboo forests was extracted using vegetation indices.

[0008] (3) Using the leaf area index (LAI), vegetation greenness index (GI), moisture index (MI), and novel characteristic spectral index (CSI), the pest level of bamboo tussock moth was classified by the XGBoost classification model.

[0009] (4) Based on the improved P-PROSAIL radiative transfer model, the PROSAIL model was optimized by introducing pest stress parameters, and the leaf area index (LAI), chlorophyll (CCC), carotenoids (CCCa), equivalent water thickness (CEWT), and dry matter (CDMC) of the bamboo forest canopy were simulated and inverted by combining the lookup table method and Gaussian noise.

[0010] (5) The significance of the interaction was verified by Scheiger-Ray-Hare nonparametric analysis of variance. The interaction effect of seasonal factors, pest severity and biochemical components was quantified by structural equation modeling (SEM), and a seasonal interaction path model was constructed.

[0011] (6) Define the Normalized Pest Interaction Index (NPII), construct a spatiotemporal evolution model based on Stacking ensemble learning, select the optimal base learner combination through genetic algorithm, and combine SHAP interpretable machine learning to analyze the key driving features of NPII.

[0012] Furthermore, in step (1), the preprocessing methods include atmospheric correction, terrain correction, cloud masking, and seasonal synthesis.

[0013] Furthermore, in step (4), the improved P-PROSAIL model screens highly sensitive parameters through Sobol global sensitivity analysis, dynamically adjusts the parameter ranges of CCC, CEWT, CDMC and LAI according to the pest level, and optimizes spectral matching using the minimum absolute error cost function.

[0014] Furthermore, in step (5), SEM evaluates the model fitting effect using absolute fit index, relative fit index, and reduction index.

[0015] Furthermore, in step (5), seasonal factors include precipitation, temperature, and wind speed.

[0016] Furthermore, in step (6), the NPII formula is:

[0017]

[0018] PII stands for Biochemical Component Interaction Index. max and PII min This represents the extreme value of the sample.

[0019] Furthermore, in step (6), the spring and summer models use Random Forest (RF), Lightweight Gradient Boosting (LightGBM), and Ridge Regression (Ridge), while the summer and autumn models use XGBoost, LightGBM, and K-Nearest Neighbors (KNN).

[0020] Furthermore, the Stacking ensemble learning spring and summer model is suitable for the peak vegetation growth period and captures nonlinear relationships through RF, while the summer and autumn model is suitable for the stress response sensitive period and improves classification accuracy through XGBoost. The difference between the two base learner combinations is designed based on seasonal spectral feature variations.

[0021] This invention also provides a remote sensing monitoring system for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth, comprising:

[0022] Data acquisition module: used to receive multi-temporal multispectral remote sensing images and meteorological and topographic data;

[0023] Preprocessing module: Performs image preprocessing, including atmospheric correction, terrain correction, and cloud masking;

[0024] Information extraction module: Extracts the distribution and pest levels of moso bamboo forests based on the XGBoost classification model;

[0025] Parameter inversion module: Run the P-PROSAIL radiative transfer model to invert leaf area index (LAI), canopy chlorophyll (CCC), canopy carotenoids (CCCa), canopy equivalent water thickness (CEWT), and canopy dry matter (CDMC) of bamboo forest.

[0026] Interactive Analysis Module: Constructing structural equation modeling (SEM) to quantify the interactions of biochemical components;

[0027] Evolution Simulation Module: Through Stacking ensemble learning and SHAP interpretable machines, it outputs spatiotemporal evolution mechanisms.

[0028] Compared with existing technologies, the present invention has the following advantages: by combining multi-temporal remote sensing image inversion, SEM, Stacking ensemble learning and interpretable machine learning, the present invention realizes dynamic monitoring and evolution mechanism analysis of biochemical components in the canopy of moso bamboo forests. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the method implementation of an embodiment of the present invention.

[0030] Figure 2 This is a remote sensing thematic map of the bamboo forest in the study area during spring, summer and autumn in an embodiment of the present invention.

[0031] Figure 3 This is a diagram of the biochemical composition of the bamboo forest canopy retrieved by the P-PROSAIL model in an embodiment of the present invention.

[0032] Figure 4 This is a SEM path coefficient diagram in an embodiment of the present invention.

[0033] Figure 5 This is a contribution map of key driving features analyzed by SHAP in an embodiment of the present invention.

[0034] Figure 6 This is a diagram showing the NPII spatiotemporal evolution simulation results in an embodiment of the present invention. Detailed Implementation

[0035] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0036] This invention provides a remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth. It integrates multi-temporal remote sensing image inversion technology, structural equation modeling (SEM), and interpretable machine learning algorithms to construct a dynamic monitoring and evolution mechanism analysis system for biochemical components in the moso bamboo forest canopy. The method includes the following steps:

[0037] (1) Acquire multi-temporal multispectral remote sensing images and preprocess them to generate representative image datasets for spring, summer and autumn;

[0038] (2) Based on the extreme gradient boosting XGBoost classification model, combined with the Normalized Difference Vegetation Index (NDVI), Bamboo Index (BI), Vegetation Health and Stress Index (SI), and Red-edge Normalized Difference Vegetation Index (NDVI), the model was developed. re1The spatial distribution information of moso bamboo forests was extracted using vegetation indices.

[0039] (3) Using the leaf area index (LAI), vegetation greenness index (GI), moisture index (MI), and novel characteristic spectral index (CSI), the pest level of bamboo tussock moth was classified by the XGBoost classification model.

[0040] (4) Based on the improved P-PROSAIL radiative transfer model, the PROSAIL model was optimized by introducing pest stress parameters, and the leaf area index (LAI), chlorophyll (CCC), carotenoids (CCCa), equivalent water thickness (CEWT), and dry matter (CDMC) of the bamboo forest canopy were simulated and inverted by combining the lookup table method and Gaussian noise.

[0041] (5) The significance of the interaction was verified by Scheiger-Ray-Hare nonparametric analysis of variance. The interaction effect of seasonal factors, pest severity and biochemical components was quantified by structural equation modeling (SEM), and a seasonal interaction path model was constructed.

[0042] (6) Define the Normalized Pest Interaction Index (NPII), construct a spatiotemporal evolution model based on Stacking ensemble learning, select the optimal base learner combination through genetic algorithm, and combine SHAP interpretable machine learning to analyze the key driving features of NPII.

[0043] This invention also provides a remote sensing monitoring system for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth, comprising:

[0044] Data acquisition module: used to receive multi-temporal multispectral remote sensing images and meteorological and topographic data;

[0045] Preprocessing module: Performs image preprocessing, including atmospheric correction, terrain correction, and cloud masking;

[0046] Information extraction module: Extracts the distribution and pest levels of moso bamboo forests based on the XGBoost classification model;

[0047] Parameter inversion module: Run the P-PROSAIL radiative transfer model to invert leaf area index (LAI), canopy chlorophyll (CCC), canopy carotenoids (CCCa), canopy equivalent water thickness (CEWT), and canopy dry matter (CDMC) of bamboo forest.

[0048] Interactive Analysis Module: Constructing structural equation modeling (SEM) to quantify the interactions of biochemical components;

[0049] Evolution Simulation Module: Through Stacking ensemble learning and SHAP interpretable machines, it outputs spatiotemporal evolution mechanisms.

[0050] The following is a detailed implementation process of the present invention.

[0051] This invention is based on Sentinel-2A remote sensing image data of the study area in spring, summer, and autumn. Figure 1 The data is not limited to the provided embodiments; the following are specific implementations of the present invention.

[0052] (1) Data preprocessing: Atmospheric correction and band synthesis were performed on Sentinel-2A imagery using SNAP software, cloud interference was removed using the Fmask algorithm, and terrain correction was performed using the SCS+C model to generate remote sensing datasets for spring, summer, and autumn. Figure 1 ).

[0053] (2) Biochemical Component Retrieval: Based on the P-PROSAIL model, the One-variable at-a-time (OAT) approach and Sobol sensitivity analysis were used sequentially to conduct local and global sensitivity analyses on various PROSAIL parameters and simulated reflectance. Based on the Sobol global sensitivity analysis, highly sensitive parameters (CCC, CEWT, CDMC, LAI) were selected, and the parameter ranges were dynamically adjusted in conjunction with the pest severity level. A lookup table was constructed and Gaussian noise was added. The minimum absolute error cost function was used to optimize the matching, and parameters such as LAI and CCC were retrieved. Figure 2 ).

[0054]

[0055] In the formula, ρ s (λ i ) represents the simulated reflectance after resampling, and λ max , λ min These represent the maximum and minimum values ​​for the wavelength range across each band. Let ρ(λ) be the spectral response function of Sentinel-2. i () represents the canopy spectral reflectance simulated by P-PROSAIL.

[0056]

[0057] In the formula: To simulate the spectrum after adding Gaussian noise to the lookup table, G x To add Gaussian white noise, G x ~N(0,ρ s (λ)×4%).

[0058] (3) Interaction analysis: Redundant variables were removed using the KMO test (value > 0.7) and Bartlett's test of sphericity (p < 0.001), ultimately including precipitation, temperature, wind speed, pest severity, and four biochemical components. The ADF estimation method was used to process non-normal data, and path coefficients were calculated through Bootstrap resampling (200 times). The SEM model was optimized based on model extension and constraint methods. Figure 3 ), χ 2 The model fit metrics, including / df, RMSEA, AGFI, NFI, TLI, CFI, IFI, AIC, and ECVI, all meet the requirements.

[0059] (4) Evolutionary simulation and analysis: Based on SEM path coefficients, the Normalized Interaction Index (NPII) of Biochemical Components in the Canopy of Bambusa multiplex under Bambusa tussock moth stress is defined, and the specific formula is as follows:

[0060] PII = DE + IE

[0061] In the formula: DE represents the direct effect of the stress of the bamboo tussock moth on the changes in the biochemical components of the bamboo forest, which characterizes the direct effect of the bamboo tussock moth on the characteristics of each biochemical component in the canopy layer; IE represents the interaction effect between components, which reflects the contribution of the interaction between different biochemical component variables to the overall system.

[0062]

[0063] PII stands for Biochemical Component Interaction Index. max and PII min This represents the extreme value of the sample.

[0064] Input the Stacking ensemble model, R for the spring / summer and summer / autumn models. 2 The values ​​were 0.6783 and 0.7962, respectively; based on SHAP interpretable machine learning, feature contribution was calculated, generating a global importance ranking and local dependency graph, revealing chlorophyll and carotenoids as key driving features. Figure 4 ), and quantify the nonlinear effects of meteorological factors (such as precipitation and wind speed) on NPII. Figure 5 ).

[0065] Figure 6 This is a diagram showing the NPII spatiotemporal evolution simulation results in an embodiment of the present invention.

[0066] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth, characterized in that, By integrating multi-temporal remote sensing image inversion technology, structural equation modeling (SEM), and interpretable machine learning algorithms, a dynamic monitoring and evolution mechanism analysis system for biochemical components in the bamboo canopy was constructed.

2. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 1, is characterized in that... The method includes the following steps: (1) Acquire multi-temporal multispectral remote sensing images and preprocess them to generate representative image datasets for spring, summer and autumn; (2) Based on the extreme gradient boosting XGBoost classification model, combined with the Normalized Difference Vegetation Index (NDVI), Bamboo Index (BI), Vegetation Health and Stress Index (SI), and Red-edge Normalized Difference Vegetation Index (NDVI), the model was developed. re1 The spatial distribution information of moso bamboo forests was extracted using vegetation indices. (3) Using the leaf area index (LAI), vegetation greenness index (GI), moisture index (MI), and novel characteristic spectral index (CSI), the pest level of bamboo tussock moth was classified by the XGBoost classification model. (4) Based on the improved P-PROSAIL radiative transfer model, the PROSAIL model was optimized by introducing pest stress parameters, and the leaf area index (LAI), chlorophyll (CCC), carotenoids (CCCa), equivalent water thickness (CEWT), and dry matter (CDMC) of the bamboo forest canopy were simulated and inverted by combining the lookup table method and Gaussian noise. (5) The significance of the interaction was verified by Scheiger-Ray-Hare nonparametric analysis of variance. The interaction effect of seasonal factors, pest severity and biochemical components was quantified by structural equation modeling (SEM), and a seasonal interaction path model was constructed. (6) Define the Normalized Pest Interaction Index (NPII), construct a spatiotemporal evolution model based on Stacking ensemble learning, select the optimal base learner combination through genetic algorithm, and combine SHAP interpretable machine learning to analyze the key driving features of NPII.

3. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (1), the preprocessing methods include atmospheric correction, terrain correction, cloud masking, and seasonal synthesis.

4. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (4), the improved P-PROSAIL model screens highly sensitive parameters through Sobol global sensitivity analysis, dynamically adjusts the parameter ranges of CCC, CEWT, CDMC and LAI according to the pest level, and optimizes spectral matching using the minimum absolute error cost function.

5. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (5), SEM evaluates the model fit using absolute fit index, relative fit index, and reduction index.

6. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (5), seasonal factors include precipitation, temperature, and wind speed.

7. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (6), the NPII formula is: PII stands for Biochemical Component Interaction Index. max and PII min This represents the extreme value of the sample.

8. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... In step (6), the spring and summer models use Random Forest (RF), Lightweight Gradient Boosting (LightGBM), and Ridge Regression (Ridge), while the summer and autumn models use XGBoost, LightGBM, and K-Nearest Neighbors (KNN).

9. The remote sensing monitoring method for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the tussock moth, as described in claim 2, is characterized in that... The Stacking ensemble learning model for spring and summer is suitable for the peak vegetation growth period and captures nonlinear relationships through RF. The model for summer and autumn is suitable for the stress response sensitive period and improves classification accuracy through XGBoost. The difference between the two base learner combinations is designed based on seasonal spectral feature variations.

10. A remote sensing monitoring system for the interaction of biochemical components in the canopy of moso bamboo forests under the stress of the bamboo tussock moth, characterized in that, include: Data acquisition module: used to receive multi-temporal multispectral remote sensing images and meteorological and topographic data; Preprocessing module: Performs image preprocessing, including atmospheric correction, terrain correction, and cloud masking; Information extraction module: Extracts the distribution and pest levels of moso bamboo forests based on the XGBoost classification model; Parameter inversion module: Run the P-PROSAIL radiative transfer model to invert leaf area index (LAI), canopy chlorophyll (CCC), canopy carotenoids (CCCa), canopy equivalent water thickness (CEWT), and canopy dry matter (CDMC) of bamboo forest. Interactive Analysis Module: Constructing structural equation modeling (SEM) to quantify the interactions of biochemical components; Evolution Simulation Module: Through Stacking ensemble learning and SHAP interpretable machines, it outputs spatiotemporal evolution mechanisms.

Images