Forest canopy carbon assimilation estimation method based on water-carbon coupling theory

Abstract

The invention belongs to the field of ecology and environmental science, and provides a forest canopy carbon assimilation estimation method based on a water-carbon coupling theory. Comprising the steps of forest remote sensing image and radar point cloud data acquisition, single tree crown segmentation, forest tree data extraction, monitoring equipment layout graph generation, sensor acquisition data acquisition, original canopy carbon assimilation rate estimation data calculation and forest canopy carbon assimilation estimation result generation. According to the method, the monitoring equipment layout diagram is generated by utilizing the equipment layout scheme generation model, so that the spatial layout of the monitoring equipment is optimized, and the reliability of the acquired data is improved; through the forest canopy carbon assimilation estimation optimization model, the influence on the forest remote sensing image is considered, errors caused by a conventional linear fitting method are avoided, and the generalization of the method is improved.

Description

Technical Field

[0001] The present invention relates to the fields of ecology and environmental science, and particularly to a method for estimating forest canopy carbon assimilation based on the theory of water-carbon coupling. Background Art

[0002] Forests are key ecosystems that regulate the global climate, and their carbon cycling process directly affects atmospheric carbon dioxide concentration and climate change trends. Through photosynthesis, forests absorb carbon dioxide and convert it into organic matter for storage, which is crucial for mitigating the greenhouse effect. The canopy is the main interface for gas exchange between the forest and the outside world, and its carbon dioxide and water vapor fluxes determine the carbon sequestration capacity and water use efficiency of the forest. Accurately estimating these fluxes is essential for understanding the carbon cycle and predicting ecological responses.

[0003] However, traditional methods have problems such as high cost and complex operation, making it difficult to meet the needs of long-term monitoring; in addition, traditional tree monitoring requires manual actual measurement, which is time-consuming, laborious, and has low efficiency; at the same time, since only dominant tree species are monitored, ignoring the influence of minority tree species, the estimation result will have a large error from the true value. Due to the dynamic changes in the forest, using conventional linear fitting will result in poor fitting accuracy. Summary of the Invention

[0004] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a method for estimating forest canopy carbon assimilation based on the theory of water-carbon coupling, which solves the problems of large errors in conventional linear fitting methods and unreasonable layout of monitoring equipment.

[0005] To achieve the above purpose, the present invention provides the following solution: A method for estimating forest canopy carbon assimilation based on the theory of water-carbon coupling, comprising: Using a drone equipped with a camera and a LiDAR radar to collect remote sensing data of the target monitoring forest, obtaining forest remote sensing images and radar point cloud data; Performing single-tree crown segmentation on the forest remote sensing images to obtain single-tree segmentation images; Extracting crown information from each single-tree sub-image in the single-tree segmentation images, and extracting the height information corresponding to each single-tree sub-image according to the radar point cloud data, obtaining forest tree data; Inputting the forest tree data into a pre-trained equipment layout scheme generation model to obtain monitoring equipment layout data, and mapping the monitoring equipment layout data to the forest remote sensing images to obtain a monitoring equipment layout map; Arranging monitoring sensors in the target monitoring forest according to the monitoring equipment layout map, and using the monitoring sensors to collect data to obtain sensor collection data; Calculate the data collected by the sensor using the pre-constructed canopy carbon assimilation rate estimation formula to obtain the original canopy carbon assimilation rate estimation data; Input the data collected by the sensor, the original canopy carbon assimilation rate estimation data, and the forest remote sensing image into the forest canopy carbon assimilation estimation optimization model for data processing to obtain the forest canopy carbon assimilation estimation result.

[0006] Preferably, perform single-tree crown segmentation on the forest remote sensing image to obtain a single-tree segmentation image, including: Convert the radar point cloud data into a canopy height model, and use the watershed segmentation algorithm to segment the canopy height model to obtain an initial crown contour candidate area; Perform adaptive histogram equalization on the forest remote sensing image to obtain a canopy-enhanced image; Perform pixel-level registration on the initial crown contour candidate area and the canopy-enhanced image through affine transformation according to the position and attitude information of the unmanned aerial vehicle to obtain a registered image; Use the U-Net network to extract, fuse, and decode spectral features, structural features, and boundary perception features of the registered image to obtain a single-tree crown probability map and an instance mask; Count the mask number and average probability of the pixels to be analyzed within a preset neighborhood range according to the single-tree crown probability map and the instance mask. When the mask number and the average probability are greater than the preset standard, determine the area where the pixels to be analyzed are located as a candidate area; Calculate the volume centroid of each candidate area according to the canopy height model, and use the Delaunay triangulation algorithm to construct a triangular network based on the volume centroid to obtain a Delaunay triangular network; Determine the two candidate areas corresponding to the two endpoints of each edge in the Delaunay triangular network as a pair of overlapping tree crowns, and determine the overlapping area of the pair of overlapping tree crowns; Extract the cutting line of the overlapping area using the height gradient; Calculate the point cloud density of each pixel within a preset range of the cutting line. If the difference in the point cloud density of adjacent pixel points is greater than the density difference threshold, determine the adjacent pixel points as boundary points; Calculate the distances from the boundary points to the two volume centroids respectively, and classify the boundary points into the candidate area corresponding to the volume centroid with the minimum distance to obtain the image segmentation result; Match the geometric feature constraint conditions for each tile in the image segmentation result using a preset tree species prior knowledge base to obtain the single-tree segmentation image with determined tree species types.

[0007] Preferably, the forest tree data includes: tree ID, tree species type, crown area, crown width, tree height, crown base height, crown volume, spectral characteristics, and location coordinates.

[0008] Preferably, the training process of the equipment layout scheme generation model includes: Collecting forest tree data of the target research forest area, and constructing a virtual forest model according to the forest tree data by using a 3D modeling tool; Performing parameter simulation on the forest tree data of the virtual forest model by adjusting environmental variables to obtain forest simulation data; Inputting the forest simulation data into a preset graph neural network for layout scheme generation to obtain intermediate equipment layout data; Evaluating the intermediate equipment layout data by using a preset layout scheme evaluation formula to obtain a scheme evaluation value; Optimizing the parameters of the graph neural network by using the scheme evaluation value to obtain the equipment layout scheme generation model after iteration is completed.

[0009] Preferably, the expression of the canopy carbon assimilation rate estimation formula is: ; where is the estimation result of the canopy carbon assimilation rate; is the transpiration rate per unit leaf area; is the water density; is the water vapor gas constant; represents temperature data; is the forest leaf area index; is the vapor pressure deficit; is the ambient carbon dioxide concentration; is the stable carbon isotope resolution of plant leaves.

[0010] Preferably, the training process of the forest canopy carbon assimilation estimation optimization model includes: Arranging the monitoring sensors in the target area according to the monitoring equipment layout map, and using the monitoring sensors to obtain the sensor acquisition data of the target area; Using the drone to collect multi-spectral images of the target area according to the fixed flight altitude and route planning; Obtaining the overall canopy carbon assimilation data of the target area by using the eddy covariance method; Collecting and calculating the sensor acquisition data according to the monitoring equipment layout map to obtain the original canopy carbon assimilation rate estimation data; Performing geometric correction and radiometric calibration on the multi-spectral images, and removing outliers and aligning timestamps on the sensor acquisition data; The ResNet18 network is pre-trained using the ImageNet image set, and the convolutional layer of the ResNet18 network is used to extract features from the multi-spectral image to obtain an image feature vector; The data collected by the sensor is standardized to obtain standard sensor data; The image feature vector, the standard sensor data, and the original canopy carbon assimilation rate estimation data are concatenated to obtain a fused feature vector; The fused feature vector is input into a multi-layer fully connected network for prediction to obtain overall canopy carbon assimilation prediction data; The mean square error between the overall canopy carbon assimilation data and the overall canopy carbon assimilation prediction data is calculated to obtain a loss value, and the loss value is used to iterate the network parameters of the multi-layer fully connected network to obtain the optimized forest canopy carbon assimilation estimation model after iteration.

[0011] Preferably, the monitoring sensor includes: a heat dissipation probe measurement device.

[0012] Preferably, the U-Net network is used to extract, fuse, and decode spectral features, structural features, and boundary-aware features from the registered image to obtain a single-tree crown probability map and an instance mask, including: The convolutional layer of the U-Net network is used to extract texture features, color features, and spectral depth features from the multi-spectral channel data of the registered image to obtain the spectral features; The pooling layer of the U-Net network is used to reduce the spatial resolution of the registered image to obtain the structural features; The Sobel operator is used to identify the crown edge of the registered image to obtain the boundary-aware features; The spectral features, the structural features, and the boundary-aware features are skip-connected to obtain cross-layer fusion features; The cross-layer fusion features are upsampled and feature maps are concatenated to obtain fused multi-scale information; Through the decoder, the Sigmoid activation function is used to predict the crown probability of the fused multi-scale information to obtain the single-tree crown probability map, and the Softmax activation function is used to assign instance labels to the fused multi-scale information to obtain the instance mask.

[0013] Preferably, the tree species prior knowledge base includes: crown morphology data, ratio data, and structural data.

[0014] Preferably, the environmental variables include: environmental temperature, precipitation pattern, solar radiation, atmospheric carbon dioxide concentration, altitude, slope, aspect, soil water content, nutrient concentration, wind speed, and crown competition intensity.

[0015] The present invention discloses the following technical effects: The present invention provides a method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory. By using the equipment layout plan generation model to generate the monitoring equipment layout map, the problem of unreasonable layout of monitoring equipment is solved, and the optimization of the spatial layout of monitoring equipment is realized; through the forest canopy carbon assimilation estimation optimization model, the problem of large errors in conventional linear fitting methods is solved, and the optimization of the original canopy carbon assimilation rate estimation data using sensor-collected data and forest remote sensing images is realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0017] Figure 1 It is a schematic diagram of the forest canopy carbon assimilation estimation process based on the water-carbon coupling theory provided by the embodiments of the present invention; Figure 2 It is a schematic diagram of the process for obtaining the single-tree segmentation image provided by the embodiments of the present invention; Figure 3 It is a schematic diagram of the training process of the equipment layout plan generation model provided by the embodiments of the present invention; Figure 4 It is a schematic diagram of the training process of the forest canopy carbon assimilation estimation optimization model provided by the embodiments of the present invention; Figure 5 It is a schematic diagram of the process for obtaining the single-tree crown probability map and instance mask provided by the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0019] The purpose of the present invention is to provide a method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory, and solve the problems of large errors in conventional linear fitting methods and unreasonable layout of monitoring equipment.

[0020] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Figure 1 As shown in the schematic diagram of the forest canopy carbon assimilation estimation process based on the water-carbon coupling theory provided by the embodiment of the present invention, Figure 1 the present invention provides a method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory, including: Step 100: Use a drone equipped with a camera and a LiDAR radar to collect remote sensing data of the target monitoring forest to obtain a forest remote sensing image and radar point cloud data; Step 200: Perform single-tree crown segmentation on the forest remote sensing image to obtain a single-tree segmentation image; Step 300: Extract crown information from each single-tree sub-image in the single-tree segmentation image, and extract the height information corresponding to each single-tree sub-image according to the radar point cloud data to obtain forest tree data; Step 400: Input the forest tree data into a pre-trained device layout scheme generation model to obtain monitoring device layout data, and map the monitoring device layout data to the forest remote sensing image to obtain a monitoring device layout map; Step 500: Arrange monitoring sensors in the target monitoring forest according to the monitoring device layout map, and use the monitoring sensors to collect data to obtain sensor collection data; Step 600: Calculate the sensor collection data using a pre-constructed canopy carbon assimilation rate estimation formula to obtain original canopy carbon assimilation rate estimation data; Step 700: Input the sensor collection data, the original canopy carbon assimilation rate estimation data, and the forest remote sensing image into a forest canopy carbon assimilation estimation optimization model for data processing to obtain a forest canopy carbon assimilation estimation result.

[0022] Referring to Figure 2 , performing single-tree crown segmentation on the forest remote sensing image to obtain a single-tree segmentation image includes: Step 201: Convert the radar point cloud data into a canopy height model, and use the watershed segmentation algorithm to segment the canopy height model to obtain an initial crown contour candidate area; Step 202: Perform adaptive histogram equalization on the forest remote sensing image to obtain a canopy enhanced image; Step 203: Perform pixel-level registration on the initial crown contour candidate area and the canopy enhanced image through affine transformation according to the position and attitude information of the drone to obtain a registered image; Step 204: Use the U-Net network to extract, fuse, and decode spectral features, structural features, and boundary-aware features from the registered image to obtain the single-tree crown probability map and instance mask; Step 205: According to the single-tree crown probability map and the instance mask, count the mask number and average probability of the pixels to be analyzed within a preset neighborhood range. When the mask number and the average probability are greater than the preset standard, determine the area where the pixels to be analyzed are located as the candidate area; Step 206: Calculate the volume centroid of each candidate area according to the canopy height model, and use the Delaunay triangulation algorithm to construct a triangular network based on the volume centroid to obtain the Delaunay triangular network; Step 207: Determine the two candidate areas corresponding to the two endpoints of each edge in the Delaunay triangular network as a pair of overlapping crowns, and determine the overlapping area of the pair of overlapping crowns; Step 208: Use the height gradient to extract the cutting line of the overlapping area; Step 209: Calculate the point cloud density of each pixel within a preset range of the cutting line. If the difference in the point cloud density of adjacent pixels is greater than the density difference threshold, determine the adjacent pixel as a boundary point; Step 210: Calculate the distances from the boundary points to the two volume centroids respectively, and classify the boundary points into the candidate area corresponding to the volume centroid with the minimum distance to obtain the image segmentation result; Step 211: Use the preset tree species prior knowledge base to match the geometric feature constraint conditions for each tile in the image segmentation result to obtain the single-tree segmentation image with determined tree species types.

[0023] Preferably, the forest tree data includes: tree ID, tree species type, crown area, crown width, tree height, crown base height, canopy volume, spectral features, and position coordinates.

[0024] Reference Figure 3 The training process of the device layout scheme generation model includes: Step 401: Collect forest tree data of the target research forest area, and use a 3D modeling tool to construct a virtual forest model based on the forest tree data; Step 402: Simulate the parameters of the forest tree data of the virtual forest model by adjusting the environmental variables to obtain forest simulation data; Step 403: Input the forest simulation data into a preset graph neural network for layout scheme generation to obtain intermediate device layout data; Step 404: Evaluate the intermediate device layout data using a preset layout scheme evaluation formula to obtain a scheme evaluation value; Step 405: Use the scheme evaluation value to optimize the parameters of the graph neural network to obtain the device layout scheme generation model after iteration.

[0025] Further, the expression of the canopy carbon assimilation rate estimation formula is: ; where is the estimation result of the canopy carbon assimilation rate; is the transpiration rate per unit leaf area; is the water density; is the water vapor gas constant; represents temperature data; is the forest leaf area index; is the vapor pressure deficit; is the ambient carbon dioxide concentration; is the stable carbon isotope resolution of plant leaves.

[0026] Reference Figure 4 , the training process of the forest canopy carbon assimilation estimation optimization model includes: Step 701: Arrange the monitoring sensors in the target area according to the monitoring device layout map, and use the monitoring sensors to obtain the sensor acquisition data of the target area; Step 702: Use the drone to collect multi-spectral images of the target area according to the fixed flight altitude and flight path planning; Step 703: Use the eddy covariance method to obtain the overall canopy carbon assimilation data of the target area; Step 704: Collect and calculate the sensor acquisition data according to the monitoring device layout map to obtain the original canopy carbon assimilation rate estimation data; Step 705: Perform geometric correction and radiometric calibration on the multi-spectral images, and perform outlier rejection and timestamp alignment on the sensor acquisition data; Step 706: Pre-train the ResNet18 network using the ImageNet image set, and use the convolutional layer of the ResNet18 network to extract features from the multi-spectral images to obtain image feature vectors; Step 707: Standardize the sensor acquisition data to obtain standard sensor data; Step 708: Concatenate the image feature vectors, the standard sensor data, and the original canopy carbon assimilation rate estimation data to obtain a fused feature vector; Step 709: Input the fused feature vector into a multi-layer fully connected network for prediction to obtain the overall canopy carbon assimilation prediction data; Step 710: Calculate the mean square error between the overall canopy carbon assimilation data and the overall canopy carbon assimilation prediction data to obtain a loss value, and use the loss value to iterate the network parameters of the multi-layer fully connected network to obtain the optimized forest canopy carbon assimilation estimation model after iteration.

[0027] Preferably, the monitoring sensor includes: a heat dissipation probe measurement device.

[0028] Reference Figure 5 , use a U-Net network to extract, fuse and decode spectral features, structural features, and boundary perception features from the registered image to obtain a single-tree crown probability map and an instance mask, including: Step 20401: Use the convolutional layer of the U-Net network to extract texture features, color features, and spectral depth features from the multi-spectral channel data of the registered image to obtain the spectral features; Step 20402: Use the pooling layer of the U-Net network to reduce the spatial resolution of the registered image to obtain the structural features; Step 20403: Use the Sobel operator to identify the crown edge of the registered image to obtain the boundary perception features; Step 20404: Perform skip connections on the spectral features, the structural features, and the boundary perception features to obtain cross-layer fusion features; Step 20405: Perform upsampling and feature map splicing on the cross-layer fusion features to obtain fused multi-scale information; Step 20406: Use the Sigmoid activation function through the decoder to perform crown probability prediction on the fused multi-scale information to obtain the single-tree crown probability map, and use the Softmax activation function to perform instance label assignment on the fused multi-scale information to obtain the instance mask.

[0029] Optionally, the tree species prior knowledge base includes: crown morphology data, proportion data, and structural data.

[0030] Preferably, the environmental variables include: environmental temperature, precipitation pattern, solar radiation, atmospheric carbon dioxide concentration, altitude, slope, aspect, soil water content, nutrient concentration, wind speed, and crown competition intensity.

[0031] Specifically, install a heat dissipation probe measurement system to observe trunk sap flow and obtain trunk sap flow data. Obtain real-time meteorological data in the forest area, such as atmospheric temperature (T), rainfall (P), relative humidity (RH), total radiation (TBQ), photosynthetically active radiation (PAR), and vapor pressure deficit (VPD), where VPD is obtained through the following empirical formula: In the formula, represents the vapor pressure deficit at temperature T; a, b, and c are all empirical parameters, a is 0.611, b is 17.502, and c is 240.97.

[0032] Furthermore, calculate the trunk sap flow density based on the Granier model and the empirical formula. The trunk sap flow density can be expressed as: In the formula, Js is the sap flow density, is the instantaneous temperature difference between the upper and lower probes, is the maximum temperature difference within a day; Preferably, based on the sap flow density, the method of converting whole-tree transpiration to canopy transpiration. The whole-tree transpiration ( ) is calculated as follows: In the formula, As is the sapwood area of the trees within this diameter class.

[0033] The canopy transpiration of a certain tree species ( ) is: In the formula, is the average sap flow density of the trees in diameter class i of the stand, is the sum of the sapwood areas of the trees of class i.

[0034] The transpiration rate per unit leaf area of the stand ( ) is: In the formula, LAI is the forest leaf area index, is the total occupied area.

[0035] Furthermore, calculate the canopy stomatal conductance ( ) based on the simplified formula of Kostner, which can be expressed as: In the formula, Gv is the water vapor gas constant, ρ is the density of water, and D is the vapor pressure deficit.

[0036] Preferably, leaf samples at different canopy heights and air samples near the canopy are collected during the plant growth season and the non-growth season respectively for measuring the stable carbon isotope compositions of the leaves and the air. , The measurement of the value is based on the PDB (Pee Dee Belemnite) standard and can be expressed as: represents the sample permil deviation from the standard sample; represents the carbon ratio of the standard substance PDB.

[0037] The calculation method for the stable carbon isotope resolution Δ of plant leaves is: where: is the carbon isotope ratio of atmospheric carbon dioxide.

[0038] Furthermore, for the construction of the water-carbon coupling model at the leaf scale, according to Farquhar's construction of the water-carbon coupling model for C3 plants, it can be expressed by the following formula: where, is the canopy carbon dioxide absorption rate, is the stomatal conductance of CO2, is the stomatal conductance of water, a is the discrimination rate difference of 13 C / 12 C during the diffusion process of the gas under the stomata, 13 C / 12 C during the carboxylation process, represents the intercellular carbon dioxide concentration, represents the ambient carbon dioxide concentration.

[0039] Specifically, for the construction of the water-carbon coupling model at the canopy scale, continuous forest canopy transpiration can be obtained by monitoring the sap flow in the trunk, and the continuous average canopy stomatal conductance can be calculated. The estimation of the canopy carbon assimilation rate can be achieved through the following formula: Specifically, a drone platform with multi-payload mounting capabilities is selected. The payload interface needs to be compatible with the synchronous operation of the camera and the LiDAR radar and support timestamp synchronization. The camera uses a multispectral camera or a high-resolution visible light camera; the LiDAR radar selects a medium-range 3D lidar with a scanning frequency of not less than 200 kHz, and the vertical field of view is consistent with that of the camera.

[0040] Furthermore, obtain the boundary range, terrain undulation, and general vegetation distribution of the target forest through an online map to delimit the flight area; set the flight altitude: determined according to the sensor performance and the tree crown height. Usually, the effective scanning height of the LiDAR is 50 - 150 m (to avoid signal attenuation), and maintain a fixed altitude cruise during the flight to avoid altitude differences. Plan the flight route: adopt a "zigzag" flight route, with the overlap rate set such that the forward overlap is not less than 80% and the side overlap is not less than 70%, and the flight speed is about 8 m / s.

[0041] Preferably, collect multispectral images (including RGB, red edge, and near-infrared bands) of the target area according to the fixed flight altitude and route plan, ensure that a single collection covers the entire research area (such as 1 km × 1 km), record the collection timestamp (accurate to seconds), and the position and attitude data of the drone for subsequent geometric correction. Uniformly arrange 5 - 10 monitoring sample points in the research area to collect the trunk sap flow rate, and record the temperature, relative humidity, carbon dioxide concentration, light intensity in real time, and measure the leaf area index.

[0042] Furthermore, based on the POS data and ground control points, perform image registration through ENVI / ArcGIS to eliminate the distortion caused by the flight attitude of the drone; convert the pixel values to actual reflectance to generate a radiometrically corrected remote sensing image.

[0043] Preferably, what the network outputs are the single-tree crown probability map and the instance mask. These data are pixel-level segmentation results, which are essentially a preliminary judgment of the crown area in the image. And the subsequent processing is exactly carried out based on these preliminary segmentation results. It can be said that the output of the neural network provides the basic data for the subsequent post-processing, and the post-processing process is to further optimize and correct these data.

[0044] Furthermore, use the single-tree crown probability map and the instance mask output by the neural network to carry out the identification of overlapping crowns; analyze the probability map and the instance mask to find the areas where crown overlap may exist. These areas usually show blurred boundaries between adjacent masks or probability values that are intertwined within a certain range; based on the three-dimensional spatial distribution information of the LiDAR point cloud, calculate the canopy volume centroid of each candidate area (i.e., the possible single-tree crown area). Then, construct a Delaunay triangulation network to describe the spatial relationship between trees, providing a geometric basis for the subsequent separation of overlapping crowns.

[0045] Specifically, let the probability map of individual tree crowns be P(x, y), where (x, y) represents the image pixel coordinates, and the value range is 0 ≤ P(x, y) ≤ 1. The closer the value is to 1, the higher the probability that the pixel belongs to the tree crown; the instance mask is M(x, y), and different integers represent different tree crown instances. To determine the overlapping area, first define a probability threshold and a 3×3 neighborhood N(x, y) centered at (x, y). For each pixel (x, y), calculate the number n(x, y) of different instance masks in its neighborhood and the average probability . When n(x, y) > 1 and > , the area where the pixel is located is considered the overlapping area.

[0046] Furthermore, the graph structure modeling of the adjacency relationship. Through the triangular mesh constructed by the centroid set, the spatial relationship of trees is transformed into non-crossing triangular connections. The vertices (centroids) of any two adjacent triangles represent spatially adjacent tree crowns; the tree crown pairs (i, j) corresponding to the edges of the triangular mesh are potential overlapping pairs, and the subsequent separation operation processes the intersection area of these adjacent tree crowns.

[0047] Preferably, the generation of the device layout scheme needs to consider the relevance of the spatial position of trees, tree species characteristics, and canopy structure. The graph neural network (GNN) can model forest tree data as a graph structure, which can effectively capture the spatial dependence and structural characteristics between trees and is suitable for dealing with device layout problems with spatial relevance. The node features include: tree ID, tree species type, canopy area, crown width, tree height, canopy volume, position coordinates; the edge features include: Euclidean distance between two trees, canopy overlap degree, tree species correlation (ecological relevance from the prior knowledge base); the graph structure of the graph neural network adopts an undirected graph structure.

[0048] Specifically, this embodiment provides an evaluation formula for the layout scheme for the training process of the network model, and its expression is as follows: Among them, , , , are the uniformity weight, representativeness weight, diversity weight, and interference weight respectively; , , , They are the spatial uniformity index, tree species representativeness index, environmental heterogeneity diversity index, and equipment interference index respectively; is the coefficient of variation of the area; , are the standard deviation of the regional area and the average area respectively; , , are the tree species weight, the number of individuals of the k-th tree species, and the total number of individuals of the k-th tree species respectively; is the total number of tree species; is the proportion of the number of devices in the m-th environmental gradient interval; is the total number of environmental gradient intervals; is the number of pairs of devices with a distance less than the minimum reasonable spacing; is the total number of devices.

[0049] The beneficial effects of the present invention are as follows: By using the device layout scheme generation model to generate the monitoring device layout map, the present invention optimizes the spatial layout of the monitoring devices and improves the reliability of the collected data; through the forest canopy carbon assimilation estimation optimization model, the present invention realizes the consideration of the influence of forest remote sensing images, avoids the errors brought by the conventional linear fitting method, and improves the generalization of the method.

[0050] Each embodiment in this specification is described in a progressive manner. The key points of each embodiment are the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0051] Specific examples are used in this article to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for estimating forest canopy carbon assimilation based on the theory of water-carbon coupling, characterized in that, Including: Using a drone equipped with a camera and a LiDAR radar to collect remote sensing data of the target monitoring forest, obtaining forest remote sensing images and radar point cloud data; Performing single-tree crown segmentation on the forest remote sensing images to obtain single-tree segmentation images; Extracting crown information from each single-tree sub-image in the single-tree segmentation images, and extracting the height information corresponding to each single-tree sub-image according to the radar point cloud data to obtain forest tree data; Inputting the forest tree data into a pre-trained device layout plan generation model to obtain monitoring device layout data, and mapping the monitoring device layout data to the forest remote sensing images to obtain a monitoring device layout map; Arranging monitoring sensors in the target monitoring forest according to the monitoring device layout map, and using the monitoring sensors to collect data to obtain sensor collection data; Calculating the original canopy carbon assimilation rate estimation data by using a pre-constructed canopy carbon assimilation rate estimation formula for the sensor collection data; Inputting the sensor collection data, the original canopy carbon assimilation rate estimation data, and the forest remote sensing images into a forest canopy carbon assimilation estimation optimization model for data processing to obtain a forest canopy carbon assimilation estimation result.

2. The method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, wherein Performing single-tree crown segmentation on the forest remote sensing images to obtain single-tree segmentation images, including: Converting the radar point cloud data into a canopy height model, and using a watershed segmentation algorithm to segment the canopy height model to obtain an initial canopy contour candidate area; Performing adaptive histogram equalization on the forest remote sensing images to obtain a canopy enhanced image; Performing pixel-level registration on the initial canopy contour candidate area and the canopy enhanced image through affine transformation according to the position and attitude information of the drone to obtain a registered image; Using a U-Net network to extract, fuse, and decode spectral features, structural features, and boundary perception features of the registered image to obtain a single-tree crown probability map and an instance mask; Counting the mask number and average probability of the pixels to be analyzed within a preset neighborhood range according to the single-tree crown probability map and the instance mask. When the mask number and the average probability are greater than a preset standard, determining the area where the pixels to be analyzed are located as a candidate area; Calculating the volume centroid of each candidate area according to the canopy height model, and using the Delaunay triangulation algorithm to construct a triangular network based on the volume centroid to obtain a Delaunay triangular network; Determining the two candidate areas corresponding to the two endpoints of each edge in the Delaunay triangular network as a group of overlapping crown pairs, and determining the overlapping area of the overlapping crown pairs; Extracting the cutting line of the overlapping area by using a height gradient; Calculating the point cloud density of each pixel within a preset range of the cutting line. If the difference in the point cloud density of adjacent pixel points is greater than a density difference threshold, determining the adjacent pixel points as boundary points; Calculating the distances between the boundary points and the two volume centroids respectively, and classifying the boundary points into the candidate area corresponding to the volume centroid with the minimum distance to obtain an image segmentation result; Using a preset prior knowledge base of tree species to match geometric feature constraint conditions for each tile in the image segmentation result, the single-tree segmentation image with determined tree species types is obtained.

3. The method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, characterized in that The forest tree data includes: tree ID, tree species type, crown area, crown width, tree height, crown base height, crown volume, spectral characteristics, and position coordinates.

4. A method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, characterized in that The training process of the device layout scheme generation model includes: Collecting forest tree data of the target research forest area and constructing a virtual forest model using a 3D modeling tool based on the forest tree data; Performing parameter simulation on the forest tree data of the virtual forest model by adjusting environmental variables to obtain forest simulation data; Inputting the forest simulation data into a preset graph neural network for layout scheme generation to obtain intermediate device layout data; Evaluating the intermediate device layout data using a preset layout scheme evaluation formula to obtain a scheme evaluation value; Using the scheme evaluation value to optimize the parameters of the graph neural network to obtain the iteratively completed device layout scheme generation model.

5. A method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, characterized in that, The expression of the canopy carbon assimilation rate estimation formula is: ； Among them, is the estimation result of canopy carbon assimilation rate; is the transpiration rate per unit leaf area; is the water density; is the water vapor gas constant; represents temperature data; is the forest leaf area index; is the vapor pressure deficit; is the ambient carbon dioxide concentration; is the stable carbon isotope resolution of plant leaves.

6. The method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, wherein The training process of the forest canopy carbon assimilation estimation optimization model includes: Arranging the monitoring sensors in the target area according to the monitoring device layout map and using the monitoring sensors to obtain sensor acquisition data of the target area; Using the drone to collect multi-spectral images of the target area according to a fixed flight altitude and flight path planning; Using the eddy covariance method to obtain the overall canopy carbon assimilation data of the target area; Collecting and calculating the sensor acquisition data according to the monitoring device layout map to obtain the original canopy carbon assimilation rate estimation data; Performing geometric correction and radiometric calibration on the multi-spectral images, and removing outliers and aligning timestamps for the sensor acquisition data; Pre-training the ResNet18 network using the ImageNet image set and using the convolutional layer of the ResNet18 network to extract feature vectors of the multi-spectral images; Performing standardization processing on the sensor acquisition data to obtain standard sensor data; Performing splicing processing on the image feature vectors, the standard sensor data, and the original canopy carbon assimilation rate estimation data to obtain a fused feature vector; Inputting the fused feature vector into a multi-layer fully connected network for prediction to obtain overall canopy carbon assimilation prediction data; Calculating the mean square error between the overall canopy carbon assimilation data and the overall canopy carbon assimilation prediction data to obtain a loss value, and using the loss value to iterate the network parameters of the multi-layer fully connected network to obtain the iteratively completed forest canopy carbon assimilation estimation optimization model.

7. A method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 1, characterized in that The monitoring sensor includes: a thermal diffusion probe measurement device.

8. The method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 2, wherein Using a U-Net network to extract, fuse, and decode spectral features, structural features, and boundary-aware features of the registered image to obtain a single-tree crown probability map and an instance mask, including: Extract texture features, color features, and spectral depth features from the multi-spectral channel data of the registered image using the convolutional layer of the U-Net network to obtain the spectral features; Reduce the spatial resolution of the registered image using the pooling layer of the U-Net network to obtain the structural features; Identify the canopy edge of the registered image using the Sobel operator to obtain the boundary perception features; Perform skip connections on the spectral features, the structural features, and the boundary perception features to obtain cross-layer fusion features; Perform upsampling and feature map stitching on the cross-layer fusion features to obtain fused multi-scale information; Use the Sigmoid activation function through the decoder to perform canopy probability prediction on the fused multi-scale information to obtain the single-tree canopy probability map, and use the Softmax activation function to perform instance label assignment on the fused multi-scale information to obtain the instance mask.

9. The method for estimating forest canopy carbon assimilation based on the water-carbon coupling theory according to claim 2, wherein The tree species prior knowledge base includes: canopy morphology data, ratio data, and structural data.

10. A method for estimating forest canopy carbon assimilation based on the theory of water-carbon coupling according to claim 4, characterized in that The environmental variables include: environmental temperature, precipitation pattern, solar radiation, atmospheric carbon dioxide concentration, altitude, slope, aspect, soil water content, nutrient concentration, wind speed, and canopy competition intensity.

Images