Canopy leaf fuel load inversion method and system based on coupled radiative transfer model

Abstract

This invention discloses a method and system for inverting canopy leaf combustible load based on a coupled radiative transfer model, belonging to the field of remote sensing inversion technology. The method includes: constructing a two-layer coupled radiative transfer model to characterize the vertically heterogeneous canopy spectral response through a nested approach; performing parameter sensitivity analysis and ecological rule constraint preprocessing to screen core parameters and construct ecological constraints based on the synergistic laws of vegetation parameters; constructing a physically autonomous simulation lookup table based on the core parameters and ecological constraints; and performing pixel-level backward inversion using the lookup table based on sensitive vegetation indices to obtain an estimated canopy leaf combustible load. This invention constructs a vertically heterogeneous forward modeling framework by nesting and coupling the PROSAIL and PROGeoSail models, introduces ecological rule constraints to compress the simulation parameter space, and combines the KDTree search algorithm to achieve efficient pixel-level inversion. This effectively solves the problems of understory background interference and pathological multiple solutions in traditional inversion, achieving physically interpretable FFL inversion at a high resolution of 10m.

Description

Technical Field

[0001] This invention relates to the field of remote sensing inversion technology, and in particular to a method and system for inverting the combustible load of canopy leaves based on a coupled radiative transfer model. Background Technology

[0002] Forest fires are a crucial ecological process affecting atmospheric and biosphere carbon and nutrient cycles. With the intensification of global climate change, the frequency and intensity of forest fires are constantly increasing, posing significant ecological risks. Canopy fire leaf (FFL), defined as the dry mass of canopy leaves per unit area of ​​land surface, is a core indicator for determining crown fire behavior, fire spread rate, and fire hazard level. Accurate monitoring of the spatiotemporal distribution of forest canopy FFL is essential for forest fire risk assessment, early warning system construction, and the formulation of disaster mitigation strategies.

[0003] In recent years, multi-source remote sensing technology has become the main means of large-scale fuel load monitoring. Compared with statistical regression models that rely on a large number of ground-based measured samples, physics-driven remote sensing (RTM) has been widely used in vegetation biochemical parameter inversion due to its clear physical meaning and strong spatiotemporal mobility. RTM simulates the absorption and scattering process of solar radiation within leaves and canopy through mathematical equations, establishing a causal relationship between vegetation traits and spectral reflectance. In practical applications, localized underscores (LUTs), as the mainstream technology for RTM inversion, effectively solve the problem of large computational costs in inverting complex physical equations by matching and searching through pre-generated simulation databases. However, existing inversion frameworks still face several significant bottlenecks when dealing with fuel load (FFL) monitoring: First, radiative transfer models are insufficient in representing vertically heterogeneous canopies. Traditional RTMs (such as the commonly used PROSAIL model) typically assume that the canopy is a horizontally homogeneous and continuous "large leaf" structure, neglecting the vertical structure of the forest canopy and the contribution of understory vegetation to the overall reflectance. At a 10m resolution scale, a pixel often contains a complex combination of "canopy-forest gap-underground background," and a single-level physical model cannot accurately isolate background noise, leading to serious estimation biases for parameters closely related to biomass, such as FFL.

[0004] Secondly, the pathological problems in the inversion process remain severe. When using LUTs for backward inversion, the phenomenon of "same spectrum, different objects" occurs—different combinations of vegetation parameters (such as a high leaf area index with low dry matter content versus a low leaf area index with high dry matter content) can produce extremely similar spectral responses. Without effective prior knowledge constraints, the inversion method often gets trapped in local optima, generating unrealistic parameter combinations that conform to physical equations but violate plant growth patterns.

[0005] In summary, the core challenge in improving the accuracy of FFL remote sensing monitoring is how to construct a coupled physical model that can characterize vertical heterogeneous structures and combine ecological principles to constrain the LUT sampling space in order to reduce the interference of multiple solutions in LUT inversion. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method and system for inverting canopy leaf combustible load based on a coupled radiative transfer model. This method solves the problem that traditional models cannot remove background interference from the forest understory by constructing a two-layer nested coupled radiative transfer model. By introducing ecological rule constraints and a sensitive vegetation index lookup table, it solves the problem of pathological multiple solutions in the inversion process and achieves high-precision FFL inversion.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The method for inverting canopy leaf combustible load based on a coupled radiative transfer model provided by this invention includes the following steps: Step 1: Construct a two-layer radiative transfer coupling model driven by physics, and build a two-layer radiative transfer framework; the two-layer radiative transfer coupling model characterizes the spectral response of the vertical heterogeneous canopy through a nested approach; Step 2: Perform parameter sensitivity analysis and ecological rule constraint preprocessing to screen the core parameters affecting the simulated reflectance, and construct ecological constraints based on the synergistic relationship between vegetation parameters; Step 3: Based on the core parameters and the ecological constraints, construct a simulation lookup table for physical autonomy; use the ecological constraint loss function from Step 2 to calculate and filter the weights of the randomly generated parameter combinations, retain high-weight sample points for forward simulation, and resample the sensor band response function to construct the physical lookup table; Step 4: Based on the sensitive vegetation index, perform pixel-level backward inversion using the simulated lookup table to obtain the estimated canopy leaf combustible load. The pixel-level backward inversion uses the KDTree spatial indexing algorithm to search for multiple candidate samples in the simulated lookup table that are closest to the Euclidean distance of the pixel feature vector, and calculates the estimated canopy leaf combustible load using the inverse distance weighting method.

[0008] Furthermore, in step one, the two-layer radiative transfer coupling model is specifically as follows: The PROSAIL model was used to simulate the response of understory vegetation. The simulated hemispherical directional reflectance of the understory was used as a background input and nested into the PROGeoSail model to simulate the total pixel reflectance response, including leaf scattering, canopy geometric shading and underlying surface reflection.

[0009] Furthermore, in step two, the parameter sensitivity analysis employs the Sobol global sensitivity analysis method, selecting core parameters by calculating the overall order sensitivity index of each parameter; the ecological constraints are implemented through the following ecological constraint loss function: in, Representative parameters and The joint weights, These are parameters extracted from a vegetation database. and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

[0010] Furthermore, in step three, the construction of the simulation lookup table for physical autonomy specifically includes: calculating the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, arranging them in descending order of total weight, selecting and retaining the top 10% of samples with the highest weights, using the selected samples for forward simulation, and resampling for the sensor band response function to obtain the lookup table.

[0011] Furthermore, in step four, the sensitive vegetation index includes at least one of the following: Normalized Infrared Index NDII7, NDII6, Water Stress Index MSI, Global Vegetation Moisture Index GVMI, and Global Environmental Monitoring Index GEMI, which are used to construct the feature vector.

[0012] Furthermore, the calculation formula for the reciprocal distance weighted method is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. To minimize the amount and avoid division by zero, This represents the weights before normalization. This represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

[0013] The canopy blade combustible load inversion system based on a coupled radiative transfer model provided by this invention includes: The model building module is used to construct a physics-driven two-layer radiative transfer coupling model, which characterizes the spectral response of the vertical heterogeneous canopy through a nested approach. The preprocessing module is used to perform parameter sensitivity analysis and ecological rule constraint preprocessing, screen the core parameters that affect the simulated reflectance, and construct ecological constraints based on the synergistic laws among vegetation parameters. The lookup table construction module is used to construct a simulated lookup table of physical autonomy based on the core parameters and the ecological constraints; The inversion module is used to perform pixel-level backward inversion based on the sensitive vegetation index and the simulated lookup table to obtain an estimated canopy leaf combustible load.

[0014] Furthermore, the preprocessing module includes: The sensitivity analysis unit is used to calculate the overall order sensitivity index of each parameter using the Sobol global sensitivity analysis method, and to screen core parameters. Ecological constraint units are used to extract the collaborative patterns between parameters based on vegetation databases and to construct ecological constraint loss functions. in, Representative parameters and The joint weights, It is a parameter and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

[0015] Furthermore, the lookup table construction module includes: The parameter sampling and filtering unit is used to calculate the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, sort them in descending order of total weight, and filter and retain the top 10% of samples with the highest weight. The forward simulation and resampling unit is used to perform forward simulation using the filtered samples and to resample the sensor band response function to generate a lookup table.

[0016] Furthermore, the inversion module includes: A feature construction unit is used to construct a feature vector of a pixel based on a sensitive vegetation index, wherein the sensitive vegetation index includes at least one of the Normalized Infrared Index NDII7, NDII6, Water Stress Index MSI, Global Vegetation Moisture Index GVMI, and Global Environmental Monitoring Index GEMI. The search unit is used to search the simulated lookup table for multiple candidate samples that are closest to the Euclidean distance of the cell feature vector using the KDTree spatial indexing algorithm. The weighted inversion unit is used to estimate the combustible load of canopy blades using the inverse distance weighting method. The calculation formula for the inverse distance weighting method in the weighted inversion unit is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. Extremely small amount, This represents the weights before normalization. This represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

[0017] The beneficial effects of this invention are as follows: This invention provides a method and system for inverting canopy leaf combustible load based on a coupled radiative transfer model, belonging to the field of remote sensing inversion technology. The method includes: constructing a two-layer coupled radiative transfer model to characterize the vertically heterogeneous canopy spectral response through a nested approach; performing parameter sensitivity analysis and ecological rule constraint preprocessing to screen core parameters and construct ecological constraints based on the synergistic laws of vegetation parameters; constructing a physically autonomous simulation lookup table based on the core parameters and ecological constraints; and performing pixel-level backward inversion using the lookup table based on sensitive vegetation indices to obtain an estimated canopy leaf combustible load. This invention constructs a vertically heterogeneous forward modeling framework by nesting and coupling the PROSAIL and PROGeoSail models, introduces ecological rule constraints to compress the simulation parameter space, and combines the KDTree search algorithm to achieve efficient pixel-level inversion. It effectively solves the problems of understory background interference and pathological multiple solutions in traditional inversion, achieving physically interpretable FFL inversion at a high resolution of 10m. Compared with existing technologies, this invention has the following beneficial effects: Improve inversion accuracy: By constructing a two-layer nested coupled radiative transfer model, the spectral contribution of the forest background is explicitly considered, effectively removing background noise and making the FFL inversion results more consistent with the real ground scene.

[0018] Mitigating pathological problems: An ecological constraint loss function based on vegetation statistical laws is introduced, which eliminates parameter combinations that do not conform to the logic of natural growth and significantly reduces the interference of multiple solutions in the inversion process.

[0019] Improve inversion efficiency: The KDTree spatial indexing algorithm is used to perform fast searches in the lookup table, replacing the traditional linear search, which greatly improves the matching efficiency in the high-dimensional feature space and makes pixel-level FFL inversion of 10m resolution satellite imagery possible.

[0020] Enhanced method universality: This invention is based on a physical model and does not rely on a large number of ground-based measured samples in a specific region. It has strong spatiotemporal mobility and can be applied to FFL monitoring in different geographical regions and forest types.

[0021] The above and other objects, advantages, and features of the present invention will be more fully set forth and demonstrated through the following detailed description of specific embodiments in conjunction with the accompanying drawings. Those skilled in the art, upon referring to the following detailed description and the accompanying drawings, will be able to better understand and realize the above advantages of the present invention. Other objects, features, and advantages of the present invention will become clearer after being described in detail in the detailed description section in conjunction with the accompanying drawings. Attached Figure Description

[0022] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following drawings are provided for illustration.

[0023] Figure 1 This is a flowchart illustrating the method. Figure 2 A bar chart showing the Sobol total sensitivity index of each parameter of the coupled radiative transfer model to the simulated Sentinel-2 satellite band reflectivity. Figure 3 The fitting relationship and scatter plot of EWT and DMC based on samples from the publicly available vegetation leaf databases LOPEX1993 and ANGERS2003. Figure 4 The fitting relationship and scatter plot of EWT and N; Figure 5 The fitting relationship and scatter plot of DMC and N; Figure 6 A bar chart showing the correlation between candidate vegetation indices and simulated FFL; Figure 7 RGB color satellite imagery of Baigongyan Park in the test area; Figure 8 This is a thematic map showing the inversion of FFL in the test area. The unit of FFL in the map is g / m. 2 Due to cloud cover, there are some blank areas in the image; Figure 9 This is a scatter plot of the measured and inverted FFL values ​​at the sampling points in the test area. Detailed Implementation

[0024] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0025] Example 1 like Figure 1 As shown in this embodiment, the method for inverting canopy blade combustible load based on a coupled radiative transfer model includes the following steps: Step 1: Construct a two-layer radiative transfer coupling model driven by physics, and build a two-layer radiative transfer framework; the two-layer radiative transfer coupling model characterizes the spectral response of the vertical heterogeneous canopy through a nested approach; Step 2: Perform parameter sensitivity analysis and ecological rule constraint preprocessing to screen the core parameters affecting the simulated reflectance, and construct ecological constraints based on the synergistic relationship between vegetation parameters; Step 3: Based on the core parameters and the ecological constraints, construct a simulation lookup table for physical autonomy; use the ecological constraint loss function from Step 2 to calculate and filter the weights of the randomly generated parameter combinations, retain high-weight sample points for forward simulation, and resample the sensor band response function to construct the physical lookup table; Step 4: Based on the sensitive vegetation index, perform pixel-level backward inversion using the simulated lookup table to obtain the estimated canopy leaf combustible load. The pixel-level backward inversion uses the KDTree spatial indexing algorithm to search for multiple candidate samples in the simulated lookup table that are closest to the Euclidean distance of the pixel feature vector, and calculates the estimated canopy leaf combustible load using the inverse distance weighting method.

[0026] In this embodiment, the two-layer radiation transmission coupling model in step one is specifically as follows: The PROSAIL model was used to simulate the response of understory vegetation. The simulated hemispherical directional reflectance of the understory was used as a background input and nested into the PROGeoSail model to simulate the total pixel reflectance response, including leaf scattering, canopy geometric shading and underlying surface reflection.

[0027] In this embodiment, in step two, the parameter sensitivity analysis adopts the Sobol global sensitivity analysis method, and the core parameters are screened by calculating the total order sensitivity index of each parameter; the ecological constraints are implemented through the following ecological constraint loss function: in, Representative parameters and The joint weights, These are parameters extracted from a vegetation database. and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

[0028] In this embodiment, step three, the construction of the simulation lookup table for physical autonomy specifically includes: calculating the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, arranging them in descending order of total weight, selecting and retaining the top 10% of samples with the highest weight, using the selected samples for forward simulation, and resampling for the sensor band response function to obtain the lookup table.

[0029] In this embodiment, in step four, the sensitive vegetation index includes at least one of the Normalized Infrared Index NDII7, NDII6, Water Stress Index MSI, Global Vegetation Moisture Index GVMI, and Global Environmental Monitoring Index GEMI, which is used to construct the feature vector.

[0030] In this embodiment, the KDTree algorithm searches for 40 nearest neighbor samples.

[0031] The calculation formula for the distance reciprocal weighted method described in this embodiment is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. To minimize the amount and avoid division by zero, This represents the weights before normalization, and N represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

[0032] The canopy blade combustible load inversion system based on the coupled radiative transfer model provided in this embodiment includes: The model building module is used to construct a physics-driven two-layer radiative transfer coupling model, which characterizes the spectral response of the vertical heterogeneous canopy through a nested approach. The preprocessing module is used to perform parameter sensitivity analysis and ecological rule constraint preprocessing, screen the core parameters that affect the simulated reflectance, and construct ecological constraints based on the synergistic laws among vegetation parameters. The lookup table construction module is used to construct a simulated lookup table of physical autonomy based on the core parameters and the ecological constraints; The inversion module is used to perform pixel-level backward inversion based on the sensitive vegetation index and the simulated lookup table to obtain an estimated canopy leaf combustible load.

[0033] The model building module described in this embodiment includes: The understory simulation unit is used to simulate the response of understory vegetation using the PROSAIL model and output the hemispherical directional reflectance of the understory. The upper canopy coupling unit is used to nest the hemispherical directional reflectance of the understory as a background input into the PROGeoSail model to simulate the total pixel reflectance response, including leaf scattering, canopy geometric occlusion, and underlying surface reflection.

[0034] The preprocessing module described in this embodiment includes: The sensitivity analysis unit is used to calculate the overall order sensitivity index of each parameter using the Sobol global sensitivity analysis method, and to screen core parameters. Ecological constraint units are used to extract the collaborative patterns between parameters based on vegetation databases and to construct ecological constraint loss functions. in, Representative parameters and The joint weights, It is a parameter and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

[0035] The lookup table construction module described in this embodiment includes: The parameter sampling and filtering unit is used to calculate the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, sort them in descending order of total weight, and filter and retain the top 10% of samples with the highest weight. The forward simulation and resampling unit is used to perform forward simulation using the filtered samples and to resample the sensor band response function to generate a lookup table.

[0036] The inversion module described in this embodiment includes: A feature construction unit is used to construct a feature vector of a pixel based on a sensitive vegetation index, wherein the sensitive vegetation index includes at least one of the Normalized Infrared Index NDII7, NDII6, Water Stress Index MSI, Global Vegetation Moisture Index GVMI, and Global Environmental Monitoring Index GEMI. The search unit is used to search the simulated lookup table for multiple candidate samples that are closest to the Euclidean distance of the cell feature vector using the KDTree spatial indexing algorithm. The weighted inversion unit is used to calculate the estimated combustible load of canopy blades using the inverse distance weighting method.

[0037] In this embodiment, the KDTree algorithm searches for 40 nearest neighbor samples.

[0038] The calculation formula for the distance reciprocal weighting method in the weighted inversion unit described in this embodiment is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. Extremely small amount, This represents the weights before normalization. This represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

[0039] Example 2 This embodiment further illustrates the method with specific illustrations and implementation processes. It addresses the shortcomings of existing high-resolution canopy leaf combustible load (FFL) inversion techniques, such as the scarcity of ground samples, pathological issues caused by observation-model bias, and the inability of conventional radiative transfer models to characterize vertically heterogeneous canopy structures. By constructing a two-layer forward modeling framework with a clearly defined physical mechanism, combined with ecological rule constraints and sensitive vegetation index screening, it achieves accurate 10m resolution FFL inversion with high physical interpretability without requiring a large number of ground samples. The high-resolution canopy leaf combustible load inversion method based on a coupled radiative transfer model provided in this embodiment includes the following steps: Step 1: Prepare to construct a physically driven two-layer coupled radiative transfer model. First, use the PROSAIL model to simulate the understory reflectance response, including understory vegetation and soil. Input parameters include: leaf structure parameter N, chlorophyll content (Cab), carotenoid content (Car), dry matter content (DMC), and equivalent water thickness (EWT). Second, nest these parameters as dynamic background radiation inputs into the PROGeoSail canopy model. This two-layer framework, through coupling, explicitly considers the significant contribution of the understory background to the total reflectance of pixels, thus characterizing the radiative transfer and spectral response in a vertically heterogeneous canopy. To distinguish it from traditional methods, commonly used linear mixture formulas are listed below: in, , , , representing the reflectance of the canopy, shadow, and bright background in a given wavelength band, respectively; C, S, and B are the corresponding area fractions, respectively. The total reflectance of a pixel is represented as a combination of reflectances of different components. In this embodiment, the simulated reflectance of the lower-layer model is used as background input into the upper-layer model to simulate the total reflectance response of the pixel, including leaf scattering, canopy geometric shading, and underlying surface reflection. This coupling method internally considers the interaction between the canopy and the underlying surface in a unified manner through the radiative transfer equation, thus eliminating the need to consider each component of the total reflectance separately and abandoning the traditional linear weighted summation method shown in formula (1).

[0040] Step 2: Parameter sensitivity analysis and introduction of ecological rule constraints.

[0041] The Sobol global sensitivity analysis method is used to quantitatively evaluate the contribution of each input parameter in the coupled radiative transfer model to different spectral bands. This method decomposes the total variance of the model into the sum of partial variances of each order: in For the total variance, , , For each order of partial variance, For the variable dimension, this embodiment focuses on examining the total order sensitivity index of each parameter, which is calculated using the following formula: in, The overall sensitivity index. The target parameters currently being evaluated Indicates the difference between the target parameter and the target parameter. The set of all other input parameters besides This means that all other parameters are fixed. Without changing the objective parameter Output caused by change The variance. By introducing the above total order sensitivity index. This method can effectively identify which parameters have a decisive influence on the Sentinel-2 band signal in complex vertical heterogeneous canopies. Analysis of key bands of Sentinel-2, such as red and near-infrared light (see...) Figure 2 Based on the consideration of focusing on the analysis of canopy characteristics, this embodiment identifies the core driving parameters affecting simulated reflectance as: LAI, ccov, DMC, Cab, EWT, and N. Based on these results, the aforementioned highly sensitive parameters are set as active variables in the lookup table, while parameters with lower sensitivity, such as carotenoid content, are set as constants based on experience to effectively reduce the ambiguity of model inversion.

[0042] After identifying the core variables, this embodiment introduces a constraint mechanism based on ecological statistical laws to avoid unrealistic parameter combinations during the inversion process. Utilizing the LOPEX1993 and ANGERS2003 vegetation leaf databases, statistical analysis is used to extract the co-evolutionary patterns among parameters. Specifically, based on the distribution of vegetation parameters in the databases, this embodiment determines a reasonable parameter distribution range and step size when constructing the LUT. Simultaneously, the parameter variables are sampled uniformly or normally to eliminate extreme combinations in the sampling space that do not conform to natural growth logic, thereby pre-narrowing the parameter search boundary before constructing the lookup table.

[0043] Step 3: Construct a physically self-consistent simulation lookup table.

[0044] The value ranges and distribution characteristics of key parameters were defined, and forward simulation was performed using the aforementioned coupled model. Subsequently, the co-development patterns between parameters extracted from the vegetation database were used to further filter the simulation samples using a loss function, ultimately generating a LUT containing 6920 samples. The simulated continuous spectrum was resampled based on the band response function of the Sentinel-2 satellite, and the reflectance of each band was calculated. in, For the simulated Sentinel-2 satellite band reflectivity, The spectral reflectance is simulated by the coupled radiative transfer model. This is the spectral response function for the corresponding band of the Sentinel-2 satellite.

[0045] Step 4: Backward inversion based on sensitive vegetation indices.

[0046] A vegetation index that is highly sensitive to FFL and has strong anti-interference capabilities is selected as the feature space. The feature vectors of the Sentinel-2 image pixels to be inverted are input into the inversion framework. The KDTree algorithm is used to search the lookup table for the k closest candidate samples to the feature vectors in Euclidean distance, and the final FFL estimate is calculated using a distance-weighted scheme. in and These represent the eigenvectors calculated from satellite data and the eigenvectors simulated in the LUT, respectively. That is, the Euclidean distance between two eigenvectors. This represents the weights before normalization, and N represents the set of matched LUT entries. For a single simulated FFL in the LUT, This is the final inversion result.

[0047] Example 3 The coupled radiative transfer model for inverting forest canopy leaf combustible load provided in this embodiment will be further explained below with reference to specific embodiments and accompanying drawings: Physical Model Nesting Coupling and Sensitivity Analysis This embodiment first constructs a two-layer radiative transfer coupling framework that considers the contribution of understory vegetation (see...). Figure 1 In specific implementation, the PROSPECT5 model is used to simulate leaf-level spectra, which are then used as input to the SAIL model to characterize the understory vegetation response. Subsequently, the simulated hemispherical directional reflectance of the understory is used as background input and nested into the PROGeoSail canopy model to achieve a refined characterization of the upper forest canopy spectrum. To determine the core driving variables for FFL inversion, this embodiment uses the Sobol global sensitivity analysis method to calculate the contribution rates of multiple parameters, including leaf structure parameter N, chlorophyll content Cab, equivalent water thickness EWT, dry matter content DMC, leaf area index LAI, and canopy cover ccov, to the simulated band reflectance. Figure 2 As shown, Figure 2 The bar chart shows the Sobol total sensitivity index of each parameter of the coupled radiative transfer model to the reflectivity of the simulated Sentinel-2 satellite bands. The four sub-charts show the sensitivity ranking of each parameter of the coupled radiative transfer model to the reflectivity of the simulated Sentinel-2 satellite bands in green, red, near-infrared, and shortwave infrared bands.

[0048] The results show that LAI, ccov, and DMC are the dominant factors affecting the near-infrared and short-wave infrared bands, and the combination of the three directly determines the physical quantity value of FFL. Based on this analysis, this embodiment selects the above-mentioned sensitive parameters as the state variables for constructing the lookup table, and removes redundant parameters that contribute little to the spectral response to reduce the complexity of the forward model.

[0049] Ecological rule constraints and physical lookup table construction In constructing the physical lookup table, to make the forward simulation results more closely resemble the physiological characteristics of real forests, this embodiment introduces ecological rule constraints based on existing vegetation datasets (such as a global leaf trait database or specific forest plot observation data) to optimize the physical self-consistency of parameter combinations. First, by statistically analyzing existing vegetation datasets, the synergistic relationships between key biochemical parameters corresponding to different vegetation types are extracted. In this embodiment, a one-dimensional linear fitting relationship is established between each pair of leaf structure parameter N, leaf dry matter content (DMC), and equivalent water thickness (EWT). Second, in the initially generated large-scale random parameter combination space, this invention sets an ecological constraint loss function to measure the degree to which each randomly generated parameter combination deviates from the aforementioned statistical regularity. Loss Function The calculation formula is as follows: in, Represents the joint weights of x and y (such as N and DMC). It is a fitting equation between two variables, such as Figure 3 , Figure 4 , Figure 5 As shown, based on samples from the LOPEX1993 and ANGERS2003 vegetation databases, the top three parameters with the highest correlation ranking are selected to plot scatter plots of the samples and to perform linear fitting equations, which are used to construct ecological rule constraints. Figure 3 The fitting relationship and scatter plot of EWT and DMC based on samples from the publicly available vegetation leaf databases LOPEX1993 and ANGERS2003. Figure 4 The fitting relationship and scatter plot of EWT versus N, Figure 5 The fitted relationship between DMC and N is shown in the scatter plot, where r is the correlation coefficient and k is a control factor; a higher k value generates a more constrained combination of variables. Subsequently, the total weight is calculated for each parameter combination based on the loss function; the smaller the loss, the higher the weight. All generated simulated samples are sorted in descending order of weight, and the top 10% with the highest weights are selected and retained as the final lookup table. This constraint mechanism eliminates anomalous parameter combinations that, while conforming to the physical range, do not conform to the ecological growth logic, significantly mitigating the ill-conditioned problems of the inversion.

[0050] Building upon this, this embodiment further investigates the screening of FFL-sensitive features. First, nine candidate vegetation indices were pre-selected, such as... Figure 6 As shown, Figure 6 A bar chart showing the correlation between candidate vegetation indices and simulated FFL is provided. Correlation analysis was performed on nine candidate vegetation indices and simulated FFL, and the top five vegetation indices with the highest correlations—NDII7, NDII6, MSI, GVMI, and GEMI—were selected as feature combinations for constructing the lookup table.

[0051] Subsequently, using the filtered lookup table samples, the correlation coefficient R² between each vegetation index and the simulated FFL was calculated. Based on the correlation analysis results, the top five vegetation indices—NDII7, NDII6, MSI, GVMI, and GEMI—were selected as matching features for the lookup table, and feature vectors were constructed. This feature selection strategy based on physical simulation data ensures that the selected indices have extremely high physical sensitivity and anti-interference ability for FFL, laying a characteristic foundation for subsequent FFL inversion.

[0052] like Figure 7 , 8 As shown in Figure 9, the user-estimated FFL is compared with the actual measured value. Figure 7 This is an RGB color satellite image of Baigongyan Park, the test area. Figure 8 This is a thematic map showing the inversion of FFL in the test area. The unit of FFL in the map is g / m. 2 Due to cloud cover, there are some blank areas in the image. Figure 9 A scatter plot of the measured and inverted FFL values ​​at the sampling points in the test area is shown. The field sampling points used for the test were located in Baigongyan Park, Chengdu, Sichuan Province, with a total of 9 valid sampling points. After pairing with the inversion results, a scatter plot of the measured and inverted values ​​was plotted, and the validation index, the coefficient of determination R, was calculated. 2 =0.17, RMSE=157.84, Accuracy=46.6%. Subsequently, a thematic map of FFL distribution was generated. Overall, the spatial distribution of FFL and vegetation showed a high degree of similarity. High values ​​were mainly concentrated in densely canopied areas, while low values ​​appeared in open grasslands, roads, water areas, etc.

[0053] FFL Inversion Based on KDTree Search: For processing actual remote sensing imagery, this embodiment first performs L2A-level atmospheric and topographic correction on Sentinel-2 imagery to obtain 10m resolution ground reflectance data. During the inversion implementation phase, to overcome the low efficiency and susceptibility to local optima of conventional lookup tables, the KDTree spatial indexing algorithm is used to organize the lookup table. The observed vegetation index vector of the pixel to be inverted is used as the query input, and the k=40 closest candidate simulation samples with the closest Euclidean distance are quickly retrieved from the lookup table. To achieve continuous and stable quantitative estimation, this embodiment performs distance-inverse weighting on the retrieved candidate set. This weighting strategy effectively smooths out random errors between physical simulation and actual observation. The final output FFL spatial distribution map maintains 10m high-resolution spatial detail while its value range is highly consistent with the forest fire risk level assessment standard, significantly improving the physical interpretability and practical accuracy of forest fuel load monitoring.

[0054] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A method for inverting canopy leaf combustible load based on a coupled radiative transfer model, characterized in that, Includes the following steps: Step 1: Construct a two-layer radiative transfer coupling model driven by physics, and build a two-layer radiative transfer framework; the two-layer radiative transfer coupling model characterizes the spectral response of the vertical heterogeneous canopy through a nested approach; Step 2: Perform parameter sensitivity analysis and ecological rule constraint preprocessing to screen the core parameters affecting the simulated reflectance, and construct ecological constraints based on the synergistic relationship between vegetation parameters; Step 3: Based on the core parameters and the ecological constraints, construct a simulation lookup table for physical autonomy; use the ecological constraint loss function from Step 2 to calculate and filter the weights of the randomly generated parameter combinations, retain high-weight sample points for forward simulation, and resample the sensor band response function to construct the physical lookup table; Step 4: Based on the sensitive vegetation index, perform pixel-level backward inversion using the simulated lookup table to obtain the estimated canopy leaf combustible load. The pixel-level backward inversion uses the KDTree spatial indexing algorithm to search for multiple candidate samples in the simulated lookup table that are closest to the Euclidean distance of the pixel feature vector, and calculates the estimated canopy leaf combustible load using the inverse distance weighting method.

2. The method for inverting canopy leaf combustible load based on a coupled radiative transfer model as described in claim 1, characterized in that, In step one, the two-layer radiative transfer coupling model is specifically as follows: The PROSAIL model was used to simulate the response of understory vegetation. The simulated hemispherical directional reflectance of the understory was used as a background input and nested into the PROGeoSail model to simulate the total pixel reflectance response, including leaf scattering, canopy geometric shading and underlying surface reflection.

3. The method for inverting canopy blade combustible load based on a coupled radiative transfer model as described in claim 1, characterized in that, In step two, the parameter sensitivity analysis employs the Sobol global sensitivity analysis method, selecting core parameters by calculating the overall order sensitivity index of each parameter; the ecological constraints are implemented through the following ecological constraint loss function: in, Representative parameters and The joint weights, These are parameters extracted from a vegetation database. and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

4. The method for inverting canopy leaf combustible load based on a coupled radiative transfer model as described in claim 1, characterized in that, In step three, the construction of the simulation lookup table for physical autonomy specifically includes: calculating the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, arranging them in descending order of total weight, selecting and retaining the top 10% of samples with the highest weights, using the selected samples for forward simulation, and resampling for the sensor band response function to obtain the lookup table.

5. The method for inverting canopy blade combustible load based on a coupled radiative transfer model as described in claim 1, characterized in that, In step four, the sensitive vegetation indices include at least one of the Normalized Infrared Index (NDII7), NDII6, Water Stress Index (MSI), Global Vegetation Moisture Index (GVMI), and Global Environmental Monitoring Index (GEMI), which are used to construct the feature vector.

6. The method for inverting canopy leaf combustible load based on a coupled radiative transfer model as described in claim 1, characterized in that, The calculation formula for the inverse distance weighted method is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. To minimize the amount and avoid division by zero, Represents the weights before normalization. This represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

7. A canopy blade combustible load inversion system based on a coupled radiative transfer model, characterized in that, include: The model building module is used to construct a physics-driven two-layer radiative transfer coupling model, which characterizes the spectral response of the vertical heterogeneous canopy through a nested approach. The preprocessing module is used to perform parameter sensitivity analysis and ecological rule constraint preprocessing, screen the core parameters that affect the simulated reflectance, and construct ecological constraints based on the synergistic laws among vegetation parameters. The lookup table construction module is used to construct a simulated lookup table of physical autonomy based on the core parameters and the ecological constraints; The inversion module is used to perform pixel-level backward inversion based on the sensitive vegetation index and the simulated lookup table to obtain an estimated canopy leaf combustible load.

8. The canopy blade combustible load inversion system based on the coupled radiative transfer model as described in claim 7, characterized in that, The preprocessing module includes: The sensitivity analysis unit is used to calculate the overall order sensitivity index of each parameter using the Sobol global sensitivity analysis method, and to screen core parameters. Ecological constraint units are used to extract the collaborative patterns between parameters based on vegetation databases and to construct ecological constraint loss functions. in, Representative parameters and The joint weights, It is a parameter and The fitting equation between them The correlation coefficient, The parameter is a factor that controls the constraint strength. and This includes blade structural parameters N, dry matter content DMC, and equivalent water thickness EWT.

9. The canopy blade combustible load inversion system based on the coupled radiative transfer model as described in claim 7, characterized in that, The lookup table construction module includes: The parameter sampling and filtering unit is used to calculate the total weight of the randomly generated parameter combinations according to the ecological constraint loss function, sort them in descending order of total weight, and filter and retain the top 10% of samples with the highest weight. The forward simulation and resampling unit is used to perform forward simulation using the filtered samples and to resample the sensor band response function to generate a lookup table.

10. The canopy blade combustible load inversion system based on the coupled radiative transfer model as described in claim 7, characterized in that, The inversion module includes: A feature construction unit is used to construct a feature vector of a pixel based on a sensitive vegetation index, wherein the sensitive vegetation index includes at least one of the normalized infrared index NDII7, NDII6, water stress index MSI, global vegetation moisture index GVMI, and global environmental monitoring index GEMI. The search unit is used to search the simulated lookup table for multiple candidate samples that are closest to the Euclidean distance of the cell feature vector using the KDTree spatial indexing algorithm. The weighted inversion unit is used to estimate the combustible load of canopy blades using the inverse distance weighting method. The calculation formula for the inverse distance weighting method in the weighted inversion unit is as follows: in, and These represent the feature vectors calculated from satellite data and the feature vectors simulated in the lookup table, respectively. Extremely small amount Represents the weights before normalization. This represents the set of matched lookup table entries. To find the combustible load of a single simulated canopy leaf in the table, This is the final inversion result.

Images