Forest biomass estimation method and system, terminal and storage medium

Abstract

The invention relates to the field of remote sensing technology and ecological models, and discloses a forest biomass estimation method and system, a terminal and a storage medium, and the method comprises the steps: firstly, generating a high-precision initial biomass distribution map through fusing a ground sample plot, an unmanned plane laser radar and a point-line-plane multi-scale modeling technology of satellite-borne remote sensing; then, circularly executing the following steps in a dynamic data assimilation framework: predicting a prior state of biomass based on a forest growth digital twinborn model; acquiring multi-source remote sensing observation data; and through a data assimilation algorithm, model prediction and remote sensing observation are fused, a prior state is corrected, a more accurate posterior state is obtained, and a final biomass distribution diagram is generated. According to the invention, through bidirectional driving of the model and data, dynamic continuous monitoring of forest biomass is realized, an active sampling closed loop based on uncertainty can be formed optionally, and estimation precision, timeliness and reliability are significantly improved.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing technology and ecological modeling technology, and in particular to a method, system, terminal and storage medium for estimating forest biomass. Background Technology

[0002] Forests are the main body of terrestrial ecosystems and play a central role in the global carbon cycle and climate change. Accurate estimation of their aboveground biomass (AGB) is crucial for carbon storage assessment, forest resource management, and ecosystem health monitoring.

[0003] Currently, the estimation of aboveground biomass in large-area forests mainly relies on remote sensing technology, particularly methods that combine limited ground plot data with extensive spaceborne remote sensing imagery to construct statistical regression models. However, this traditional method based on "static snapshots" has inherent limitations. It typically only generates a "base map" of biomass stock for a specific year, making it difficult to reflect the dynamic processes of growth, death, and disturbance in forest ecosystems over time. This results in poor timeliness of the estimation results, failing to meet the needs of dynamic monitoring.

[0004] Furthermore, the accuracy of such methods is largely constrained by the quality of the initial modeling. Due to the high cost and sparse number of ground sample plots, remote sensing estimation models built upon them often face the challenge of insufficient true data samples. The resulting errors persist throughout the modeling process and are difficult to quantify and control. In addition, the biomass products generated by existing technologies are usually definite estimates, but they neglect the spatial assessment of the uncertainty of the estimation results. This makes it impossible for users to know which areas in the map have reliable estimates and which areas may have large deviations, thus limiting their application value in subsequent decision support.

[0005] Therefore, how to overcome the bottleneck of static estimation, achieve high-precision, dynamic and continuous tracking of forest biomass, and effectively manage its uncertainty is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a method, system, terminal and storage medium for estimating forest biomass, which solves the problems of low accuracy, difficulty in dynamic updating and inability to effectively control estimation uncertainty in existing large-area forest biomass estimation methods.

[0007] To achieve the above objectives, the present invention provides the following technical solution: The first aspect of this invention provides a method for estimating forest biomass, the method comprising the following steps: Step 1: Obtain the initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment; Step 2: Based on the aforementioned digital twin model of forest growth, perform time evolution on the state vector to obtain the prior state vector for the next time step; Step 3: Obtain remote sensing observation data of the target area at the next time step; Step 4: Based on the remote sensing observation data, the prior state vector is corrected using a data assimilation algorithm to obtain the posterior state vector for the next time step; Step 5: Generate the aboveground biomass distribution map for the next time step based on the posterior state vector.

[0008] In an optional embodiment, step four, the process of correcting the prior state vector using a data assimilation algorithm, is specifically implemented by constructing an observation operator. The observation operator maps the prior state vector to prior remote sensing observations. By calculating the observation residual between the actually acquired remote sensing observation data and the prior remote sensing observations, and updating the prior state vector based on this observation residual, the posterior state vector can be obtained. Specifically, this update process can be represented by the following formula: Where, x k Let be the posterior state vector. Let K be the prior state vector. k Let z be the Kalman gain matrix. k The remote sensing observation data, In an optional embodiment, to address potential sudden disturbances in the forest ecosystem such as fires and logging, the method of this invention further includes continuous monitoring of the observation residuals. When the observation residuals numerically exceed a preset statistical threshold, the system triggers a disturbance event detection program. After identifying the disturbance event, the program can perform a forced reset operation on the posterior state vector of the area according to preset event rules, so that the forest growth digital twin model can reflect the impact of the disturbance event.

[0009] In an optional embodiment, to optimize the allocation of observation resources, the data assimilation algorithm outputs the posterior state vector along with its corresponding error covariance matrix. The diagonal elements of this matrix represent the uncertainty of the aboveground biomass estimation results at each spatial location. Based on this error covariance matrix, this invention generates a spatial distribution map of the uncertainty in aboveground biomass estimation. By analyzing this map, the regions with the highest uncertainty in the model estimation results, i.e., uncertainty hotspots, can be identified, and active sampling tasks can be generated for these regions, such as instructing drones to collect higher-precision data.

[0010] In an optional embodiment, after performing the active sampling task, UAV lidar data for the uncertain hotspot area is acquired. This high-precision data not only participates in the data assimilation process as new remote sensing observation data, but is also used to locally optimize or recalibrate the observation operators. Through this feedback mechanism, the model's performance in specific areas can be improved in a targeted manner, thereby reducing future estimation uncertainties.

[0011] In an optional embodiment, to obtain a high-precision initial aboveground biomass distribution map, step one is specifically implemented as follows: First, obtain measured aboveground biomass values ​​from a small number of ground sample plots within the target area, along with UAV lidar data covering these plots, and construct a high-precision first regression model based on these two data. Then, apply this first regression model to a wider range of UAV lidar data to generate a large number of high-precision aboveground biomass samples. Finally, using these high-precision samples as training data, combined with spaceborne remote sensing data covering the entire target area, construct a second regression model with a wider applicability, and apply this model to generate the initial aboveground biomass distribution map.

[0012] A second aspect of the present invention provides a forest biomass estimation system, the system comprising: The initialization module is used to obtain the initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment. The prediction module is used to perform time evolution on the state vector based on the forest growth digital twin model to obtain the prior state vector at the next moment. The observation module is used to acquire remote sensing observation data of the target area at the next time step; An assimilation module is used to correct the prior state vector based on the remote sensing observation data using a data assimilation algorithm to obtain the posterior state vector at the next time step. The generation module is used to generate the aboveground biomass distribution map at the next time step based on the posterior state vector.

[0013] In an optional embodiment, the assimilation module is further configured to output the error covariance matrix of the posterior state vector; the system further includes a decision module, which is configured to generate a spatial distribution map of uncertainty in aboveground biomass estimation based on the error covariance matrix, identify uncertainty hotspots based on the map, and finally generate an active sampling task for the uncertainty hotspots.

[0014] A third aspect of the present invention provides a terminal comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method as described in any embodiment of the first aspect of the present invention.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any embodiment of the first aspect of the present invention.

[0016] In summary, the present invention has at least one of the following beneficial technical effects: 1. This invention constructs a digital twin model of forest growth and introduces a data assimilation algorithm. This invention combines a process model with physical mechanisms with actual remote sensing observations and uses observation residuals to continuously correct the model state. This makes the estimation results not only fit the observation data, but also follow the inherent laws of forest growth, thereby improving the tracking ability and estimation accuracy of dynamic changes in forest aboveground biomass.

[0017] 2. By establishing a disturbance event detection and response mechanism based on observation residuals, this invention can promptly identify sudden events such as fires and logging, and reset the model state, ensuring the timeliness of the digital twin model and its ability to represent the real situation after drastic environmental changes.

[0018] 3. By introducing an active sampling decision-making mechanism based on uncertainty quantification, this invention can accurately allocate high-cost observation resources to the most uncertain regions of the model and use the acquired data to back-optimize the observation operators, forming a closed loop of cognition and observation, which improves the targeting of data acquisition and the convergence efficiency of the overall estimation framework. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a structural block diagram of the forest biomass estimation system of the present invention; Figure 3 This is a schematic diagram of the scale expansion modeling process of the present invention; Figure 4 This is a schematic diagram of the data assimilation loop of the present invention; Figure 5 This is a schematic diagram of the active sampling closed loop based on uncertainty in this invention.

[0020] Figure 6 This is a schematic diagram of the structure of a terminal device used to implement the method of the present invention.

[0021] The system consists of: 100, Forest Biomass Estimation System; 10, Initialization Module; 20, Prediction Module; 30, Observation Module; 40, Assimilation Module; 50, Generation Module; and 60, Decision Module. Detailed Implementation

[0022] See attached document Figure 1 , Figure 1 This is a schematic flowchart of a forest biomass estimation method according to an embodiment of the present invention. The forest biomass estimation method provided by this embodiment of the present invention may include the following steps: S101, Obtain the initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment; S102, based on the digital twin model of forest growth, performs time evolution on the state vector to obtain the prior state vector at the next moment; S103, acquire remote sensing observation data of the target area at the next moment; S104, based on remote sensing observation data, the prior state vector is corrected through a data assimilation algorithm to obtain the posterior state vector at the next moment; S105, Based on the posterior state vector, generate the aboveground biomass distribution map for the next time step.

[0023] See attached document Figure 2 , Figure 2 This is a structural block diagram of a forest biomass estimation system according to an embodiment of the present invention. The forest biomass estimation system is configured to perform the forest biomass estimation method described above. The system can be deployed on a single high-performance computing terminal or in a distributed computing environment consisting of a data server, a computing server, and user terminals.

[0024] Specifically, the forest biomass estimation system includes: an initialization module 10, a prediction module 20, an observation module 30, an assimilation module 40, and a generation module 50.

[0025] Initialization module 10 is used to obtain an initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment. The function of this module corresponds to step S101 in the flowchart.

[0026] The prediction module 20 is connected to the output of the initialization module 10. This module is used to perform temporal evolution on the state vector based on the forest growth digital twin model to obtain the prior state vector for the next time step. The function of this module corresponds to step S102 in the flowchart.

[0027] The observation module 30 is used to acquire remote sensing observation data of the target area at the next time step. The function of this module corresponds to step S103 in the flowchart.

[0028] The assimilation module 40 has its inputs connected to the outputs of the prediction module 20 and the observation module 30, respectively. This module is used to correct the prior state vector based on remote sensing observation data using a data assimilation algorithm to obtain the posterior state vector for the next time step. The function of this module corresponds to step S104 in the flowchart.

[0029] The generation module 50 is connected to the output of the assimilation module 40. This module is used to generate a map of aboveground biomass distribution at the next time step based on the posterior state vector, and output or store this map as the final result. The function of this module corresponds to step S105 in the flowchart.

[0030] In a practical implementation, the forest biomass estimation system can be implemented by one or more computing devices. For example, the computing device includes a processor and a memory. The memory stores computer program instructions, and when the processor executes these instructions, it performs the steps described above. The initialization module 10, prediction module 20, observation module 30, assimilation module 40, and generation module 50 can be program units stored in memory, which are called and executed by the processor. These modules interact with each other via an internal data bus or a defined communication protocol to collaboratively estimate forest aboveground biomass.

[0031] In one specific embodiment, the forest biomass estimation method of the present invention first performs step S101, namely, obtaining an initial aboveground biomass distribution map of the target area. (Refer to the appendix...) Figure 3 The figure schematically illustrates the "point-line-area" scale expansion modeling process used in an embodiment of the present invention to generate an initial aboveground biomass distribution map. The implementation process may include: establishing several ground sample plots within the target area; calculating the measured aboveground biomass value A at the sample plot scale by measuring each tree and combining it with an allometric growth equation applicable to the region. plot The allometric growth equation takes the form of: Where N is the total number of trees in the sample plot, i is the index of a single tree, and f s D is a function corresponding to tree species s. i and H i These are the diameter at breast height (DBH) and tree height of a single tree. Simultaneously, point cloud data of UAV-LiDAR covering the ground sample plot and part of the extended area were acquired, and the point cloud data were preprocessed such as denoising, classification, and normalization to extract structural feature variables such as point cloud height quantiles and density. lidar Based on the measured value A of the ground sample plot. plot and the corresponding point cloud feature variable v lidar Construct a first regression model The model was applied to all areas covered by UAV-LiDAR data to generate high-precision ground biomass samples. Subsequently, using the ground biomass values ​​of these high-precision samples as the dependent variable, and feature variables v extracted from spaceborne remote sensing data (such as Sentinel-2 optical imagery and Sentinel-1 synthetic aperture radar imagery) at the corresponding spatial locations, were used as the basis for the analysis. sat Construct a second regression model with as the independent variable. Finally, the second regression model was applied to spaceborne remote sensing data covering the entire target area to generate a spatial distribution map of ground biomass A(x,y,t0) at the initial time t0. This distribution map serves as the state vector of the forest growth digital twin model at the initial time.

[0032] See attached document Figure 4 The figure details the core process of the data assimilation loop in this embodiment of the invention. This loop mainly consists of steps S102 and S104. Next, step S102 is executed, where the state vector undergoes temporal evolution based on the forest growth digital twin model. This step is implemented through a process model (state transition operator) to predict the state change from the previous time k-1 to the current time k. The state transition equation for this process can be expressed as: in, Let x be the prior state vector at time k; k -1 is the corrected posterior state vector at time k-1; It is a nonlinear state transition operator, which internally encapsulates a set of functions describing biophysical processes such as forest growth, respiration, and shedding; u k The data vector representing meteorological driving forces at time k includes temperature, precipitation, and illumination; w k-1 The process noise vector is used to characterize the uncertainty of the process model itself. It follows a Gaussian distribution with zero mean and covariance matrix Q.

[0033] Subsequently, step S103 is executed to acquire remote sensing observation data of the target area at the current time k. This data can be a combination of one or more data sources, such as spaceborne optical sensors and synthetic aperture radar sensors. Next, step S104 is executed to correct the prior state vector based on the remote sensing observation data using a data assimilation algorithm. To establish a connection between the model state and the remote sensing observations, this step first constructs an observation operator. The function of this observation operator is to convert the state vector x inside the model. k Mapped to the remote sensing observation space, the relationship can be expressed as: Among them, z kLet k be the actual remote sensing observation vector at time k; This is a nonlinear observation operator, which can take the form of a radiative transfer model or a machine learning model trained offline; v k The observation noise vector is used to characterize the uncertainty of remote sensing observations and observation operators. It follows a Gaussian distribution with zero mean and covariance matrix R.

[0034] In this embodiment, the data assimilation algorithm employs ensemble Kalman filtering (EnKF). This algorithm uses a set to represent the probability distribution of the states. First, the observation residual is calculated, which is the difference between the actual remote sensing observation and the remote sensing observation predicted based on the prior state vector. Then, based on the observation residual, each member in the set is corrected using the following update equation to obtain the posterior state vector: Where, x k,i and These are the posterior and prior states of the i-th member in the set, respectively; K k It is the Kalman gain, calculated from the covariance of the set: Here, It is the cross-covariance matrix between state predictions and observation predictions. It is the covariance matrix of the observation and prediction.

[0035] In another embodiment of this method, the observation residuals generated during the data assimilation process are... Continuous monitoring is conducted. When the norm of the observed residual sequence of a spatial cell exceeds the value determined by its statistical covariance matrix at multiple consecutive time steps... When a predefined threshold is reached, a disturbance event detection program is triggered. This program identifies the type of disturbance event (such as fire, pest infestation, or human logging) based on the characteristics of the residuals (such as numerical value and direction of change), and performs a forced reset operation on the posterior state vector of the unit according to predefined rules to ensure that the state of the forest growth digital twin model is consistent with the abrupt changes in the real world.

[0036] In another embodiment of this method, the data assimilation algorithm calculates the error covariance matrix P while outputting the posterior state vector set. k The diagonal elements of this matrix quantify the uncertainty in the aboveground biomass estimation for each spatial unit. Based on this matrix, a spatial distribution map of uncertainty is generated. The system then identifies several regions with the highest uncertainty values ​​as uncertainty hotspots and automatically generates active sampling tasks for these regions, such as planning the flight path of a UAV lidar system. The acquired UAV lidar data is used, on the one hand, as new high-precision remote sensing data input to the assimilation module, and on the other hand, for the observation operators corresponding to the uncertainty hotspots. Perform parameter optimization or model structure recalibration to form a feedback loop that reduces model uncertainty. Finally, step S105 is executed to generate the aboveground biomass distribution map for the next time step based on the posterior state vector. Specifically, the mean of the posterior state set after assimilation correction is used as the optimal estimate of the aboveground biomass at that time step, thereby generating the final aboveground biomass distribution map A(x,y,t). k ), used for output, display or storage.

[0037] See attached document Figure 2 , Figure 2 This is a structural block diagram of a forest biomass estimation system 100 according to an embodiment of the present invention. The system 100 is configured to perform the forest biomass estimation method in any of the foregoing embodiments. In a specific embodiment, the system 100 may be implemented by one or more computers, servers, or cloud computing platforms.

[0038] As attached Figure 2 As shown, the forest biomass estimation system 100 includes: an initialization module 10, a prediction module 20, an observation module 30, an assimilation module 40, a generation module 50, and an optional decision module 60. These modules can be hardware, software, or firmware, or any combination thereof. In a software-based implementation, these modules are a set of computer-executable instructions stored in memory and executed by one or more processors.

[0039] The initialization module 10 is used to execute step S101 in the method. Specifically, this module is configured to receive ground sample plot survey data and multi-source remote sensing data, perform scale expansion modeling through built-in first and second regression models, and finally generate an initial aboveground biomass distribution map covering the entire target area, and output it as the initial state vector.

[0040] The input of the prediction module 20 is connected to the output of the initialization module 10, and is used to execute step S102 in the method. This module integrates a forest growth digital twin model, which calculates the prior state vector at the current time based on the state vector of the previous time step and the externally input meteorological driving data through the state transition operator F.

[0041] The observation module 30 is used to execute step S103 in the method. This module is configured to acquire remote sensing observation data for a specified time and area from an external data source (such as a remote sensing data archive or real-time data stream), and preprocess the data to form an observation vector z that can be assimilated. k .

[0042] See attached document Figure 5The figure illustrates an uncertainty-based active sampling closed-loop process in an optional embodiment of the present invention. The input of the assimilation module 40 is connected to the outputs of the prediction module 20 and the observation module 30, respectively, for executing step S104 in the method. This module internally implements a data assimilation algorithm, such as ensemble Kalman filtering (EnKF). It receives a prior state vector and a remote sensing observation vector, and uses a built-in observation operator... The calculation is performed, and the posterior state vector, corrected for the observed data, is finally output. In some embodiments, the assimilation module 40 is further configured to calculate and output the error covariance matrix P of the posterior state vector. k .

[0043] The input of the generation module 50 is connected to the output of the assimilation module 40 to execute step S105 in the method. This module receives the posterior state vector, extracts the aboveground biomass component from it, calculates its optimal estimate (such as the ensemble mean), and organizes it into spatial raster data to generate the final aboveground biomass distribution map. This distribution map can be stored in a storage device or visualized using a display device.

[0044] In an optional embodiment, system 100 further includes a decision module 60. The input of decision module 60 is connected to the output of assimilation module 40, and is used to receive the error covariance matrix P. k The decision module 60 is configured to generate an uncertainty spatial distribution map based on the matrix, identify uncertainty hotspots according to set criteria, and automatically generate active sampling tasks based on these areas. This task instruction can be output to an external UAV control system. Furthermore, the decision module 60 is also configured to receive new data acquired through active sampling and utilize this data to adjust the observation operators in the assimilation module 40. Update or optimize.

[0045] See attached document Figure 6 The figure illustrates a hardware structure diagram of a terminal device that can be used to implement the method of the present invention. The present invention also provides a terminal. This terminal can be a personal computer, workstation, server, or mobile computing device. Its internal structure includes at least one processor, memory, communication interface, and input / output devices. The memory stores a computer program containing instructions for implementing any of the foregoing embodiments of the method of the present invention. When the terminal is running, the processor reads and executes the computer program, thereby completing the estimation of forest aboveground biomass.

Claims

1. A method for estimating forest biomass, characterized in that, Includes the following steps: Step 1: Obtain the initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment; Step 2: Based on the aforementioned digital twin model of forest growth, perform time evolution on the state vector to obtain the prior state vector for the next time step; Step 3: Obtain remote sensing observation data of the target area at the next time step; Step 4: Based on the remote sensing observation data, the prior state vector is corrected using a data assimilation algorithm to obtain the posterior state vector for the next time step; Step 5: Generate the aboveground biomass distribution map for the next time step based on the posterior state vector.

2. The forest biomass estimation method according to claim 1, characterized in that, Step four specifically includes: Construct an observation operator to map the prior state vector to prior remote sensing observations; Calculate the observation residual between the remote sensing observation data and the prior remote sensing observation values; The prior state vector is updated based on the observed residuals to obtain the posterior state vector.

3. The forest biomass estimation method according to claim 1, characterized in that, The method further includes: Continuously monitor the observed residuals; When the observed residual exceeds a preset statistical threshold, disturbance event detection is triggered to identify disturbance events occurring within the target area; Based on the type of the disturbance event, the posterior state vector is forcibly reset to update the forest growth digital twin model.

4. The forest biomass estimation method according to claim 1, characterized in that, The data assimilation algorithm also outputs the error covariance matrix of the posterior state vector; The method further includes: Based on the error covariance matrix, a spatial distribution map of the uncertainty in the aboveground biomass estimation is generated; Based on the aforementioned uncertainty spatial distribution map, uncertainty hotspot areas were identified; Active sampling tasks are generated for the aforementioned uncertain hotspot areas.

5. The forest biomass estimation method according to claim 4, characterized in that, The active sampling task is to acquire UAV lidar data for the uncertain hotspot area; The method further includes: The observation operator is optimized or recalibrated using the acquired UAV lidar data.

6. The forest biomass estimation method according to claim 1, characterized in that, Step one includes: Obtain measured values ​​of aboveground biomass in ground sample plots within the target area, as well as UAV lidar data covering the ground sample plots; Based on the measured values ​​of aboveground biomass and the UAV lidar data, a first regression model is constructed; The first regression model was applied to a wider range of UAV lidar data to generate high-precision aboveground biomass samples. Based on the high-precision ground biomass sample and spaceborne remote sensing data covering the target area, a second regression model is constructed. The second regression model is applied to the spaceborne remote sensing data to generate the initial ground biomass distribution map.

7. A forest biomass estimation system, applied to a forest biomass estimation method according to any one of claims 1-6, characterized in that, include: The initialization module is used to obtain the initial aboveground biomass distribution map of the target area and use it as the state vector of the forest growth digital twin model at the initial moment. The prediction module is used to perform time evolution on the state vector based on the forest growth digital twin model to obtain the prior state vector at the next moment. The observation module is used to acquire remote sensing observation data of the target area at the next time step; An assimilation module is used to correct the prior state vector based on the remote sensing observation data using a data assimilation algorithm to obtain the posterior state vector at the next time step. The generation module is used to generate the aboveground biomass distribution map at the next time step based on the posterior state vector.

8. A forest biomass estimation system according to claim 7, characterized in that, The assimilation module is also used to output the error covariance matrix of the posterior state vector; The system also includes: The decision module is used to generate an uncertainty spatial distribution map of aboveground biomass estimation based on the error covariance matrix, identify uncertainty hotspot areas based on the uncertainty spatial distribution map, and generate active sampling tasks for the uncertainty hotspot areas.

9. A terminal, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-6.

Images