Forest Carbon Stock Estimation Method Based on Artificial Intelligence and Multimodal Remote Sensing Data

Abstract

The present invention discloses a method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data, which relates to the technical field of forest carbon storage estimation. The present invention integrates optical satellite images, lidar data, and hyperspectral image remote sensing data. Through multi-source data fusion, the comprehensiveness and estimation accuracy of the data are improved, and it can more accurately reflect the forest coverage, vertical structure, and biodiversity, overcoming the limitations of a single data source. The entire forest area is divided into multiple sub-regions, and a carbon storage estimation model is separately constructed based on the forest type of each sub-region. The method of estimating area by area refines the carbon storage estimation of different regions, accurately reflects the characteristics of different forest types, and introduces a dynamic carbon storage prediction model based on differential equations. Combining historical and real-time remote sensing data, it simulates the change of carbon storage in the time dimension and captures the impact of factors such as forest growth, climate change, natural disasters, and logging on carbon storage.

Description

Technical Field

[0001] The present invention relates to the technical field of forest carbon storage estimation, in particular to a forest carbon storage estimation method based on artificial intelligence and multimodal remote sensing data. Background Art

[0002] In the research of the global carbon cycle and response to climate change, forest carbon storage, as one of the key indicators of this research, the remote sensing technology applied provides the main data support for large-scale forest carbon storage estimation. Different remote sensing technologies can capture forest information from different perspectives, such as optical satellite images, lidar LiDAR.

[0003] Most of the current forest carbon storage estimation methods use one of the above-mentioned remote sensing technologies as the information capture means. Among them, optical satellite images are widely used to estimate the forest cover area, but it is not sensitive enough to information such as tree height and crown structure. Although LiDAR can provide detailed three-dimensional forest structure data, it can only cover part of the area and is restricted by weather conditions. The single data source collection leads to insufficient accuracy of the estimation results. In addition, there are significant differences in carbon storage and biomass distribution among different types of forests, such as tropical rainforests, coniferous forests, and broad-leaved forests. To ensure the accuracy of carbon storage estimation, accurate classification of forest types is required. However, the current forest classification methods still use manual division or predefined classification, with low classification efficiency and strong subjectivity.

[0004] To address the problem of single data source, most of the existing estimation methods adopt the method of fusing multiple data sources, such as linear regression and principal component analysis. However, such methods require a linear relationship between different data sources and cannot effectively handle the non-linear associations between data. Moreover, there are significant differences in the data formats, temporal resolutions, and spatial resolutions of different remote sensing technologies, making it difficult to effectively coordinate the differences during data fusion and easily resulting in data mismatch. In terms of forest type classification, most of the existing estimation methods classify forest types through remote sensing technology and regional empirical models, but they rely too much on optical images. The optical images obtained by remote sensing technology cannot guarantee clarity, making it difficult to conduct fine-grained classification of forest types. Some vegetation types are very similar in optical images, and insufficient classification accuracy will directly affect the zoning results of carbon storage estimation. In addition, the accuracy of regional empirical models is not good when applied to new geographical regions or climate conditions, and the models need to be re-optimized, making cross-regional application very difficult. Therefore, there is an urgent need for a forest carbon storage estimation method based on artificial intelligence and multimodal remote sensing data to solve such problems. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data to solve the problems in the existing solutions that principal component analysis and regression model fusion can only capture the linear relationships between data sources, ignoring the non-linear characteristics of forest structure and dynamic changes. The more complex the remote sensing data, the greater the limitations of such methods. And the experience-based forest classification methods are difficult to adapt to the forest types and climate conditions in different regions.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] The present invention provides a method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data, which includes:

[0009] Step S1, multi-source remote sensing data collection, collecting remote sensing data, including forest coverage information, vertical structure information, spectral feature information, and surface and canopy information;

[0010] Step S2, remote sensing data feature extraction, performing remote sensing data fusion, and extracting features related to carbon storage estimation and forest classification. The related features include: forest structure, tree species type, canopy height, and vegetation index;

[0011] Step S3, forest type classification, based on the features extracted in Step S2, using a classification algorithm to classify the forest types;

[0012] Step S4, constructing a carbon storage model, constructing a carbon storage estimation model for different types of forests, and based on the division of forest types, combining the carbon storage estimation models in different regions to estimate the forest carbon storage area by area;

[0013] Step S5, predicting dynamic carbon storage changes, using historical and real-time remote sensing data to construct a prediction model, and combining the forest types and the carbon storage estimation results in Step S4 to analyze the dynamic changes of carbon storage.

[0014] Furthermore, perform data format conversion and standardization processing on the remote sensing data to remove noise and outliers;

[0015] Here, multi-source refers to multi-source data including optical satellite images, lidar data, and hyperspectral images.

[0016] Furthermore, the remote sensing data in Step S1 includes:

[0017] Optical satellite images: Lidar data: Hyperspectral images: Among them, H and W respectively represent the height and width of the image, and C1, C2, and C3 respectively represent the number of channels of the optical, LiDAR, and hyperspectral images;

[0018] Spatial features are extracted using a convolutional neural network. For optical satellite images, the convolutional operation is expressed as:

[0019] F optical = σ(W conv1 * X optical + b conv1 ),

[0020] The feature extraction of LiDAR data and hyperspectral data are respectively:

[0021] F LiDAR = σ(W conv2 * X LiDAR + b conv2 ),

[0022] F hyperspectral = σ(W conv3 * X hyperspectral + b conv3 ),

[0023] Among them, are the feature maps extracted from optical, LiDAR, and hyperspectral images respectively. D1, D2, and D3 respectively represent the number of channels of each feature map of the optical image, LiDAR data, and hyperspectral image after feature extraction. H'×W' represents the spatial size of the feature map after the convolutional operation. W conv1 , W conv2 , W conv3 is the convolutional kernel weight matrix, b conv1 , b conv2 , b conv3 are their respective bias terms. * represents the convolutional operation, and σ(·) is the ReLU activation function.

[0024] Furthermore, in step S2, remote sensing data fusion is performed:

[0025] The features of each modality are fused and dimensionally reduced using an autoencoder. The features of each modality are concatenated and expressed as:

[0026]

[0027] Then, dimensionality reduction processing is performed using an autoencoder to extract fused features:

[0028] F encoded = σ(W encoder · F concat + b encoder ), where F encoded ∈ R H′×W′×D represents the fused features after dimensionality reduction, W encoder represents the weight matrix of the autoencoder, b encoderdenotes the bias term of the autoencoder, D denotes the dimensionality of the features after dimensionality reduction, and σ(·) is the ReLU activation function;

[0029] Introduce the attention mechanism and calculate the attention weights:

[0030] where α i,j denotes the attention weight at feature positions i, j, q i and k j denote the query vector and the key vector respectively, which are obtained from the feature map through linear transformation, N denotes the total number of features, denotes the transpose operation;

[0031] Use the attention weights to perform weighted summation on the features:

[0032] where v j denotes the value vector of the feature, and F attention denotes the fused feature after being weighted by the attention mechanism.

[0033] Furthermore, in step S3, according to the features extracted after fusion, the forest is divided into different types including coniferous forest, broad-leaved forest, and tropical rainforest, and according to the classification results, the different types of forests are regionally divided.

[0034] Furthermore, the method for classifying forest types in step S3 is as follows:

[0035] The fused feature in step S2 is F attention ∈R H′×W′×D , convert this feature map into a vector, and perform global average pooling on each channel:

[0036] where F attention (i, j) denotes the feature value of the feature map at position (i, j), and F pooled ∈R D denotes the pooled feature vector, retaining the global information of each channel and serving as the input to the classifier, H′, W′ denote the spatial dimensions of the feature map, and D denotes the number of channels of the feature map;

[0037] Use a support vector machine for classification and construct the optimization objective:

[0038] where w denotes the normal vector of the classification hyperplane, b denotes the bias term of the classification hyperplane, and y i denotes the class label of sample i, taking values of {-1, +1}, |w| 2 is the regularization term, denotes the transpose operation;

[0039] The classification decision function of SVM is as follows: Among them, is the predicted category, that is, the forest type, and sign(·) represents the sign function.

[0040] Furthermore, the method of classifying forest types in step S3 also includes:

[0041] Define the loss function as: Among them, L represents the cross-entropy loss value, and y i,c represents the true label of sample i, which is represented by one-hot encoding. represents the predicted probability that classifier belongs to category c for sample i, and the parameters of the classifier are updated using the gradient descent method: Among them, θ represents the parameters of the classifier, including w and b, and η represents the learning rate, which controls the update step size. represents the gradient of the loss function L with respect to the parameter θ. After training is completed, input the features F of new test data pooled , and use the trained classifier to predict the forest type

[0042] In step S3, use the features F extracted in step S2 pooled , vectorize the features through global average pooling, and then use a support vector machine to build a classification model to classify the forest type. During the training process, use the cross-entropy loss function and optimize the parameters of the classifier through gradient descent.

[0043] Furthermore, the method of constructing the carbon storage estimation model in step S4 is as follows:

[0044] Design a carbon storage estimation model for each forest type T, and the carbon storage is expressed as:

[0045] C T =β T ·B T , where C T is the total carbon storage of forest type T, β T is the carbon conversion coefficient, which represents the conversion ratio between forest biomass and carbon storage, and B T is the biomass of forest type T;

[0046] The biomass B of the forest T The estimation formula is: Among them, a T , b T , c T are empirical coefficients related to forest type T, obtained through field measurement, H T is the average tree height of the trees in forest type T, and D T is the average breast diameter of the trees in forest type T;

[0047] The biomass estimation formula is adjusted according to the tree species characteristics of different forest types, a T , b T , c T and is dependent on the forest type;

[0048] After the forest type is divided, the entire forest area is divided into multiple sub-areas according to different types. Assuming that the total forest area is divided into N sub-areas, the forest type of each sub-area i is T i , and its carbon storage C i The estimation formula is:

[0049] Among them, C i is the carbon storage of the i-th sub-area, is the carbon conversion coefficient related to the forest type T i to which the sub-area belongs, is the biomass per unit area of the forest type T i to which the sub-area belongs,

[0050] A i is the area of the sub-area;

[0051] Sum up the carbon storage of all sub-areas: Among them, C total is the total carbon storage of the entire forest area, C i is the carbon storage of the i-th sub-area; for different types of forests, a carbon storage estimation model is constructed based on the biomass estimation formula. By calculating the biomass and carbon conversion coefficient of the forest type area by area, combined with the area of the region, the carbon storage is estimated area by area, and then by accumulating the carbon storage of all sub-areas, the total carbon storage of the entire forest area is obtained.

[0052] Furthermore, in step S5, the method for predicting the dynamic change of carbon storage is:

[0053] Use a first-order differential equation to model the change of carbon storage C i (t) at time t:

[0054] Among them, C i (t) is the carbon storage of the i-th sub-area at time t, is the carbon storage growth rate of the i-th area, L i (t) is the carbon loss of the i-th area at time t, represents the change rate of the carbon storage of the i-th area at time t;

[0055] The growth rate of carbon storage Depending on the forest type and the biomass growth model, the relationship between the growth rate of carbon storage and the growth of forest biomass is expressed as: where r max is the maximum growth rate of this forest type, B i (t) is the biomass of the i-th region at time t, B max is the maximum biomass of the i-th region, that is, the maximum forest biomass value that this region can reach. When the biomass approaches the maximum value, the growth rate tends to zero.

[0056] Furthermore, in step S5, the method for predicting the dynamic change of carbon storage also includes:

[0057] Construct a biomass growth model to describe the growth of biomass:

[0058] where B i (t) is the biomass of the i-th region at time t, B max is the maximum biomass of this region, k is the growth rate parameter, t0 is the initial moment of biomass growth, and m is the parameter affecting the curve shape;

[0059] Construct a carbon loss model to describe the loss of carbon:

[0060] L i (t) = L0·e -λ·t , where L i (t) is the carbon loss of the i-th region at time t, L0 is the initial value of carbon loss, and λ is the loss attenuation rate, indicating the decreasing speed of carbon loss over time;

[0061] Combine the biomass growth model and the carbon loss model to represent the change of carbon storage at time t:

[0062] where C i (0) represents the initial carbon storage of the i-th region, represents the net carbon storage growth, is the cumulative change of carbon storage from time 0 to t;

[0063] The carbon storage of the entire forest area is the sum of the carbon storages of each sub-region. For N sub-regions, the change of the total carbon storage at time t is: where C total (t) is the carbon storage of the entire forest area at time t, C i (t) is the carbon storage of the i-th region at time t;

[0064] The dynamic change of the total carbon storage of the entire forest area over time is obtained by estimating each region one by one.

[0065] The beneficial effects of the present invention are as follows:

[0066] In the present invention, optical satellite images, lidar data, and hyperspectral image remote sensing data are integrated, making full use of the complementarity of different data sources. The optical images provide spatial resolution information, the lidar data captures the vertical structure of the forest, and the hyperspectral images reflect the spectral characteristics of different forest types. Through multi-source data fusion, the comprehensiveness and estimation accuracy of the data are greatly improved, and it can more accurately reflect the forest cover, vertical structure, and biodiversity, overcoming the limitations of a single data source.

[0067] In the present invention, an attention mechanism is introduced, which can adaptively focus on the key features related to carbon storage estimation. It not only improves the accuracy of feature extraction but also, through data dimensionality reduction, avoids the interference of redundant information on the results, further enhancing the estimation accuracy.

[0068] In the present invention, the entire forest area is divided into multiple sub-regions, and a carbon storage estimation model is separately constructed based on the forest type of each sub-region. The method of estimating area by area refines the carbon storage estimation of different regions, accurately reflecting the characteristics of different forest types. For different types of forests such as coniferous forests, broad-leaved forests, and tropical rainforests, specific biomass models and carbon conversion coefficients are respectively constructed, improving the accuracy and regional adaptability of carbon storage estimation.

[0069] In the present invention, a dynamic carbon storage prediction model based on differential equations is introduced. Combining historical and real-time remote sensing data, it simulates the change of carbon storage in the time dimension, establishes a logarithmic relationship between the carbon storage growth rate and forest biomass, captures the impacts of factors such as forest growth, climate change, natural disasters, and deforestation on carbon storage, and can predict future change trends. Brief Description of the Drawings

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0071] Figure 1 It is a schematic flow chart of the forest carbon storage estimation method of the present invention. Detailed Embodiments

[0072] To make the above objects, features, and advantages of the present invention more obvious and understandable, the detailed embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0073] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, the present invention may be practiced in other ways than those specifically described herein, and those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0074] Secondly, as used herein, "one embodiment" or "an embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other.

[0075] Example 1, referring to Figure 1 , this example provides a method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data, including the following steps:

[0076] Step S1, multi-source remote sensing data collection, collect remote sensing data, including forest coverage information, vertical structure information, spectral feature information, and surface and canopy information;

[0077] Perform data format conversion and standardization processing on the remote sensing data to remove noise and outliers;

[0078] Here, multi-source refers to multi-source data including optical satellite images, lidar data, and hyperspectral images.

[0079] The remote sensing data in step S1 includes:

[0080] Optical satellite images: Lidar data: Hyperspectral images: Where H and W respectively represent the height and width of the image, and C1, C2, and C3 respectively represent the number of channels of the optical, LiDAR, and hyperspectral images;

[0081] Use a convolutional neural network to extract spatial features. For optical satellite images, the convolutional operation is expressed as:

[0082] F optical = σ(W conv1 *X optical +b conv1 ),

[0083] The feature extraction of LiDAR data and hyperspectral data are respectively:

[0084] F LiDAR = σ(W conv2 ·X LiDAR +b conv2 ),

[0085] F hyperspectral = σ(W conv3 * X hyperspectral + b conv3 ),

[0086] wherein, are the feature maps extracted from optical, LiDAR, and hyperspectral images respectively. D1, D2, and D3 represent the number of channels of each feature map of the optical image, LiDAR data, and hyperspectral image after feature extraction. H'×W' represents the spatial size of the feature map after convolution operation. W conv1 , W conv2 , W conv3 is the convolutional kernel weight matrix, b conv1 , b conv2 , b conv3 is the respective bias term, * represents the convolution operation, and σ(·) is the ReLU activation function;

[0087] Specifically, the introduction of multi-source remote sensing data provides spatial, structural, and spectral information for the model. Spatial features are extracted through a convolutional neural network, combined with an autoencoder for dimensionality reduction and fusion processing, improving the utilization efficiency of multi-modal information, avoiding information insufficiency or bias brought by a single data source, and introducing an attention mechanism enabling the model to better highlight key features related to carbon storage and enhancing the understanding ability of complex forest structures.

[0088] Step S2, remote sensing data feature extraction, perform remote sensing data fusion, and extract features related to carbon storage estimation and forest classification. The related features include: forest structure, tree species type, canopy height, and vegetation index;

[0089] Perform remote sensing data fusion in step S2:

[0090] Fuse and reduce the dimensionality of the features of each modality using an autoencoder. Concatenate the features of each modality and represent them as:

[0091]

[0092] Then perform dimensionality reduction processing using an autoencoder to extract fused features:

[0093] F encoded = σ(W encoder · F concat + b encoder ), where F encoded ∈ R H′×W′×D represents the fused features after dimensionality reduction, W encoder represents the weight matrix of the autoencoder, and b encoder represents the bias term of the autoencoder. D represents the dimensionality of the features after dimensionality reduction, and σ(·) is the ReLU activation function;

[0094] Introduce the attention mechanism and calculate the attention weights:

[0095] Among them, α i,j represents the attention weight of feature positions i, j, q i and k j respectively represent the query vector and the key vector, which are obtained from the feature map through linear transformation. N represents the total number of features, represents the transpose operation;

[0096] Use the attention weights to perform weighted summation on the features:

[0097] Among them, v j represents the value vector of the feature, and F attention represents the fused feature weighted by the attention mechanism;

[0098] Specifically, based on the extracted multi-modal features, use the support vector machine (SVM) for classification, and retain the information in the feature vectorization process through global average pooling. The SVM combines the gradient descent method to optimize the parameters of the classifier, so that the classifier can still effectively handle the complexity and diversity of different forest types when dealing with different types of forests. Through forest type classification, the forest area is divided more precisely.

[0099] Step S3, forest type classification, based on the features extracted in step S2, use a classification algorithm to classify the forest type;

[0100] In step S3, according to the features extracted after fusion, the forest is divided into different types including coniferous forest, broad-leaved forest, and tropical rainforest. According to the classification results, the areas of different types of forests are divided;

[0101] The way to perform forest type classification in step S3 is:

[0102] The fused feature in step S2 is F attention ∈R H′×W′×D , convert this feature map into a vector, and perform global average pooling on each channel:

[0103] Among them, F attention (i, j) represents the feature value of the feature map at position (i, j), and F pooled ∈R D represents the pooled feature vector, retaining the global information of each channel and serving as the input of the classifier. H′, W′ represent the spatial dimensions of the feature map, and D represents the number of channels of the feature map;

[0104] Classification is performed using a support vector machine, and an optimization objective is constructed:

[0105] Among them, w represents the normal vector of the classification hyperplane, b represents the bias term of the classification hyperplane, and y i represents the class label of sample i, taking values in {-1, +1}, and |w| 2 is the regularization term, represents the transpose operation;

[0106] The classification decision function of the SVM is: Among them, is the predicted class, that is, the forest type, and sign(·) represents the sign function;

[0107] The method of classifying forest types in step S3 also includes:

[0108] The loss function is defined as: Among them, L represents the cross-entropy loss value, and y i,c represents the true label of sample i, represented using one-hot encoding, represents the predicted probability that the classifier assigns sample i to class c, and the parameters of the classifier are updated using the gradient descent method: Among them, θ represents the parameters of the classifier, including w and b, η represents the learning rate, which controls the update step size, represents the gradient of the loss function L with respect to the parameter θ. After training is completed, the new test data feature F pooled is input, and the trained classifier is used to predict the forest type

[0109] In step S3, using the feature F extracted in step S2 pooled , the feature is vectorized through global average pooling, and then a classification model is constructed using a support vector machine to classify the forest type. During the training process, a cross-entropy loss function is used, and the parameters of the classifier are optimized through gradient descent;

[0110] Specifically, the carbon storage estimation model estimates the carbon storage for different forest types. Through sub-region division, differential modeling is carried out according to the ecological characteristics, carbon absorption capacity, and tree growth laws of different types of forests, and the spatial accuracy of carbon storage estimation is effectively improved by estimating region by region, solving the problem of large-scale and diverse forest ecosystems that are difficult to handle by traditional estimation methods;

[0111] Step S4, construct a carbon storage model, construct a carbon storage estimation model for different types of forests, and based on the division of forest types, combine the carbon storage estimation models of different regions to estimate the carbon storage of forests region by region;

[0112] In step S4, the carbon storage estimation model is constructed as follows:

[0113] For each forest type T, a carbon storage estimation model is designed, and the carbon storage is expressed as:

[0114] C T = β T ·B T where C T is the total carbon storage of forest type T, β T is the carbon conversion coefficient, representing the conversion ratio between forest biomass and carbon storage, and B T is the biomass of forest type T;

[0115] The biomass B of the forest T is estimated by the formula: where a T , b T , c T are empirical coefficients related to forest type T, obtained through field measurement, H T is the average tree height of the trees in forest type T, and D T is the average breast diameter of the trees in forest type T;

[0116] The biomass estimation formula is adjusted according to the tree species characteristics of different forest types, and a T , b T , c T are dependent on the forest type;

[0117] After the forest type is divided, the entire forest area is divided into multiple sub-areas according to different types. Assuming that the total forest area is divided into N sub-areas, and the forest type of each sub-area i is T i , its carbon storage C i is estimated by the formula:

[0118] where C i is the carbon storage of the i-th sub-area, is the carbon conversion coefficient related to the forest type T i to which the sub-area belongs, is the biomass per unit area of the forest type T i to which the sub-area belongs,

[0119] A i is the area of the sub-area;

[0120] The carbon storage of all sub-areas is summed: where C total is the total carbon storage of the entire forest area, and C iis the carbon storage of the i-th sub-region; for different types of forests, a carbon storage estimation model is constructed based on the biomass estimation formula. By calculating the biomass and carbon conversion coefficient of the forest type zone by zone, combined with the regional area, the carbon storage is estimated zone by zone, and then by accumulating the carbon storage of all sub-regions, the total carbon storage of the entire forest area is obtained;

[0121] Step S5, predicting the dynamic change of carbon storage, using historical and real-time remote sensing data, constructing a prediction model, and analyzing the dynamic change of carbon storage in combination with the forest type and the carbon storage estimation result of step S4;

[0122] In step S5, the method for predicting the dynamic change of carbon storage is as follows:

[0123] Use a first-order differential equation to model the change of carbon storage C i (t) over time t:

[0124] where C i (t) is the carbon storage of the i-th sub-region at time t, is the carbon storage growth rate of the i-th region, and L i (t) is the carbon loss of the i-th region at time t, represents the change rate of the carbon storage of the i-th region over time t;

[0125] The growth rate of carbon storage depends on the forest type and the biomass growth model. The relationship between the growth rate of carbon storage and the growth of forest biomass is expressed as: where r max is the maximum growth rate of this forest type, B i (t) is the biomass of the i-th region at time t, and B max is the maximum biomass of the i-th region, that is, the maximum forest biomass value that this region can reach. When the biomass approaches the maximum value, the growth rate tends to zero;

[0126] In step S5, the method for predicting the dynamic change of carbon storage also includes:

[0127] Construct a biomass growth model to describe the growth of biomass:

[0128] where B i (t) is the biomass of the i-th region at time t, and B max is the maximum biomass of this region, k is the growth rate parameter, t0 is the initial moment of biomass growth, and m is the parameter affecting the curve shape;

[0129] Construct a carbon loss model to describe the loss of carbon:

[0130] L i L(t)=L0·e -λ·t , where L i (t) is the carbon loss in the i-th region at time t, L0 is the initial value of carbon loss, and λ is the loss attenuation rate, indicating the decreasing rate of carbon loss over time;

[0131] Combining the biomass growth model and the carbon loss model, the change in carbon storage at time t is represented as:

[0132] where C i (0) represents the initial carbon storage in the i-th region, represents the net carbon storage growth, is the cumulative change in carbon storage from time 0 to t;

[0133] The carbon storage of the entire forest area is the sum of the carbon storages of each sub-region. For N sub-regions, the change in total carbon storage at time t is: where C total (t) is the carbon storage of the entire forest area at time t, and C i (t) is the carbon storage in the i-th region at time t;

[0134] The dynamic change of the total carbon storage of the entire forest area over time is obtained by estimating region by region;

[0135] Specifically, a first-order differential equation is used to model the change in carbon storage over time. Combining historical and real-time remote sensing data, the growth and loss processes of forest carbon storage are dynamically captured. The relationship between the biomass growth rate and the maximum forest biomass is introduced, and the impact of natural disasters, logging, and degradation factors on carbon storage is expressed through an exponential decay model of carbon loss, enabling the carbon storage prediction to have the ability to respond to complex natural processes and more accurately reflect the temporal change of carbon storage.

[0136] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for estimating forest carbon stocks based on artificial intelligence and multi-modal remote sensing data, characterized in that: Including, Step S1, multi-source remote sensing data acquisition, collecting remote sensing data, including forest coverage information, vertical structure information, spectral feature information, and surface and canopy information; Step S2, remote sensing data feature extraction, performing remote sensing data fusion, and extracting features related to carbon stock estimation and forest classification. The related features include: forest structure, tree species type, canopy height, and vegetation index; Step S3, forest type classification, based on the features extracted in Step S2, using a classification algorithm to classify forest types; Step S4, constructing a carbon stock model, constructing a carbon stock estimation model for different types of forests, and based on the division of forest types, combining carbon stock estimation models in different regions to estimate the carbon stock of forests area by area; Step S5, dynamic carbon stock change prediction, using historical and real-time remote sensing data, constructing a prediction model, and combining forest types and the carbon stock estimation results in Step S4 to analyze the dynamic changes of carbon stock; In Step S2, remote sensing data fusion is performed: the features of each modality are fused and dimension-reduced using an autoencoder, and the features of each modality are concatenated and expressed as: Then, dimension reduction processing is performed using an autoencoder to extract fused features: F encoded = σ(W encoder · F concat + b encoder ), where F encoded ∈ R H'×W'×D represents the fused feature after dimensionality reduction, W encodeR represents the weight matrix of the autoencoder, b encoder represents the bias term of the autoencoder, D represents the dimensionality of the feature after dimensionality reduction, and σ(·) is the ReLU activation function; Introduce an attention mechanism to calculate attention weights: Among them, α i,j represents the attention weight of the feature positions i, j, q i and k j respectively represent the query vector and the key vector, which are obtained from the feature map by linear transformation. N represents the total number of features, and T represents the transpose operation; Use the attention weights to perform weighted summation on the features: Among them, v j represents the value vector of the feature, and F attention represents the fused feature weighted by the attention mechanism; The method for constructing the carbon stock estimation model in Step S4 is: Design a carbon stock estimation model for each forest type T, and the carbon stock is expressed as: C T = β T ·B T , where C T is the total carbon storage of forest type T, β T is the carbon conversion coefficient, representing the conversion ratio between forest biomass and carbon storage, and B T is the biomass of forest type T; The biomass B of the forest T The estimation formula is as follows: where a T , b T , c T are empirical coefficients related to the forest type T, obtained through on-site measurement, H T is the average tree height of the trees in the forest type T, and D T is the average diameter at breast height of the trees in the forest type T; After the forest type is divided, the entire forest area is divided into multiple sub-areas according to different types. Assuming that the total forest area is divided into N sub-areas, the forest type of each sub-area i is T i , and its carbon storage C i The estimation formula is as follows: Among them, C i is the carbon storage of the i-th sub-region, is the carbon conversion coefficient related to the forest type T i to which the sub-region belongs, is the biomass per unit area of the forest type T i to which the sub-region belongs, and A i is the area of the sub-region; Sum the carbon stocks of all sub-regions: where C total is the total carbon stock of the entire forest area, and C i is the carbon stock of the i-th sub-region; In Step S5, the method for performing dynamic carbon stock change prediction is: Modeling the carbon storage C with a first-order differential equation to represent the change of i (t) over time t: Among them, C i (t) is the carbon storage of the i-th sub-region at time t, and r Ti is the carbon storage growth rate of the i-th region, and L i (t) is the carbon loss of the i-th region at time t, represents the change rate of the carbon storage of the i-th region over time t; Growth rate of carbon storage Depending on the forest type and biomass growth model, the relationship between the growth rate of carbon storage and the growth of forest biomass is expressed as: where r max The maximum growth rate of this forest type, B i (t) is the biomass of the i-th region at time t, B max The maximum biomass of the i-th region, that is, the maximum forest biomass value that this region can reach.

2. The forest carbon stock estimation method based on artificial intelligence and multi-modal remote sensing data according to claim 1, characterized in that Perform data format conversion, standardization processing, and remove noise and outliers on the remote sensing data; Here, multi-source refers to multi-source data including optical satellite images, LiDAR data, and hyperspectral images.

3. The forest carbon stock estimation method based on artificial intelligence and multi-modal remote sensing data according to claim 2, characterized in that, The remote sensing data in Step S1 includes: Optical satellite imagery: LiDAR data: Hyperspectral imagery: Among them, H and W respectively represent the height and width of the imagery, and C1, C2, and C3 respectively represent the number of channels of the optical, LiDAR, and hyperspectral imagery; Use a convolutional neural network to extract spatial features. For optical satellite images, the convolutional operation is expressed as: F optical = σ(W conv1 * X optical + b conv1 ) The feature extraction of LiDAR data and hyperspectral data are respectively: F LiDAR = σ(W conv2 * X LiDAR + b conv2 ) F hyperspectral = σ(W conv3 * X hyperspectral + b conv3 ) Among them, are the feature maps extracted from optical, LiDAR, and hyperspectral images respectively. D1, D2, and D3 represent the number of channels of each feature map of the optical image, LiDAR data, and hyperspectral image after feature extraction. H'×W' represents the spatial size of the feature map after the convolution operation, and W conv1 , W conv2 , W conv3 is the convolutional kernel weight matrix, and b conv1 , b conv2 , b conv3 are their respective bias terms. * represents the convolution operation, and σ(·) is the ReLU activation function.

4. The forest carbon stock estimation method based on artificial intelligence and multi-modal remote sensing data according to claim 3, characterized in that, In Step S3, according to the features extracted after fusion, the forest is divided into different types including coniferous forests, broad-leaved forests, and tropical rainforests. According to the classification results, different types of forests are divided into regions.

5. The method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data according to claim 4, wherein The method for performing forest type classification in Step S3 is: The fused feature in step S2 is F attention ∈R H'×W'×D , map and transform this feature into a vector, and perform global average pooling on each channel: Among them, F attention (i,j) represents the eigenvalue of the feature map at position (i,j), and F pooled ∈R D represents the pooled feature vector, retaining the global information of each channel and serving as the input to the classifier. H' and W' represent the spatial dimensions of the feature map, and D represents the number of channels of the feature map; Use a support vector machine for classification and construct an optimization objective: Among them, w represents the normal vector of the classification hyperplane, b represents the bias term of the classification hyperplane, and y i represents the class label of sample i, taking values in {-1, +1}, |w| 2 is the regularization term, represents the transpose operation; The classification decision function of SVM is as follows: where is the predicted class, i.e., the forest type, and sign(·) represents the sign function.

6. The method for estimating forest carbon storage based on artificial intelligence and multimodal remote sensing data according to claim 5, characterized in that The method for performing forest type classification in Step S3 also includes: Define the loss function as: where L represents the cross-entropy loss value, y i,c represents the true label of sample i, represented by one-hot encoding, represents the predicted probability that classifier assigns sample i to class c, and update the classifier parameters using gradient descent method: where θ represents the parameters of the classifier, including w and b, η represents the learning rate, controlling the update step size, represents the gradient of the loss function L with respect to the parameter θ. After training is completed, input the feature F of the new test data pooled , and use the trained classifier to predict the forest type 7. The method for estimating forest carbon storage based on artificial intelligence and multi-modal remote sensing data according to claim 6, characterized in that, In Step S5, the method for performing dynamic carbon stock change prediction also includes: Construct a biomass growth model to describe the growth of biomass: where B i (t) is the biomass of the i-th region at time t, B max is the maximum biomass of this region, k is the growth rate parameter, t0 is the initial time of biomass growth, and m is the parameter affecting the curve shape; Construct a carbon loss model to describe the loss of carbon: L i L(t)=L0·e -λ·t , where L i (t) is the carbon loss in the i-th area at time t, L0 is the initial value of carbon loss, and λ is the loss attenuation rate, indicating the decreasing speed of carbon loss over time; Combine the biomass growth model and the carbon loss model to represent the change of carbon stock at time t: Among them, C i (0) represents the initial carbon storage of the i-th region, represents the net carbon storage growth, is the change in carbon storage accumulated from time 0 to t; The carbon storage of the entire forest area is the sum of the carbon storages of each sub-region. For N sub-regions, the change in the total carbon storage at time t is as follows: where C total (t) is the carbon storage of the entire forest area at time t, and C i (t) is the carbon storage of the i-th region at time t.

Images