Mangrove forest carbon sink monitoring and metering method based on unmanned aerial vehicle, radar and AI technology

Abstract

The invention relates to the technical field of ecological environment protection, in particular to a mangrove forest carbon sink monitoring and metering method based on an unmanned aerial vehicle, a radar and an AI technology, and the method comprises the steps: 1, obtaining the laser radar data of a mangrove forest through a laser radar carried by the unmanned aerial vehicle; step 2, acquiring elevation data in a mangrove forest vegetation layer area, and obtaining point cloud data after topographic error correction; 3, obtaining a multispectral image of the mangrove forest, and obtaining a multispectral data matrix; 4, identifying forest growth data features, constructing a mangrove forest carbon sink prediction model, and predicting the mangrove forest carbon sink amount; step 5, marking the image region with the NDVI value higher than a preset NDVI threshold value as a blade over-dense region; and according to the area of the overdense leaf region and the multispectral data matrix, calculating a light depression factor by using a photosynthetic depression factor formula, determining the carbon sink deviation of the overdense leaf region by using a regional carbon sink deviation formula, and obtaining a real carbon sink value of the mangrove forest according to a carbon sink calculated value obtained by prediction.

Description

Technical Field

[0001] This invention relates to the field of ecological and environmental protection technology, and in particular to a method for monitoring and measuring mangrove carbon sequestration based on drones, radar, and AI technologies. Background Technology

[0002] Mangroves, as important ecosystems, not only provide habitats for organisms but also play a crucial role in the global carbon cycle. Due to their strong carbon sequestration capacity, mangroves are considered one of the key ecosystems for mitigating climate change. However, with the increasing impact of climate change and human activities, the area of ​​mangroves is decreasing year by year. Therefore, accurately assessing and monitoring mangrove carbon sinks is of great significance for environmental protection, carbon emission management, and policy formulation. Currently, mangrove growth characteristics, such as tree height, diameter at breast height (DBH), and crown width, are key parameters for carbon sink estimation. However, current technologies struggle to accurately obtain this information, especially in high-density mangrove areas. Extracting precise growth characteristics from complex point cloud data remains a challenge. Furthermore, the carbon storage of mangrove ecosystems is influenced by various factors, such as vegetation density and leaf condition. Overly dense foliage can cause photosynthetic inhibition, thus affecting the accurate estimation of carbon sinks. However, existing technologies struggle to comprehensively consider these factors, leading to biases in carbon sink prediction results. Multispectral remote sensing technology provides effective data support for vegetation cover and leaf growth, but its data processing and analysis also face many challenges. Multispectral images are affected by noise, image distortion, and illumination variations, and image quality and processing effectiveness are highly dependent on environmental conditions during acquisition. Furthermore, existing multispectral image processing techniques still face technical difficulties in band alignment, image standardization, and feature extraction, affecting the accuracy of multispectral data in mangrove carbon sequestration assessment. Although deep learning technology has made significant progress in remote sensing data processing in recent years, existing carbon sequestration prediction models still face challenges in data fusion when processing multi-source data. How to effectively fuse lidar point cloud data, multispectral remote sensing image data, and other growth data, and train a deep learning model to achieve accurate carbon sequestration prediction, remains an urgent problem to be solved. Summary of the Invention

[0003] This invention addresses the problems existing in the prior art by providing a method for monitoring and measuring mangrove carbon sequestration based on drones, radar, and AI technologies, mainly including: Step 1: Acquire lidar data of mangrove forests using a drone equipped with lidar, acquire remote sensing image data using satellite remote sensing technology, classify and filter lidar point cloud data, identify the ground layer and vegetation layer of the forest, and label the ground layer and vegetation layer with signal category labels respectively. Step 2: Based on the vegetation layer data obtained in Step 1, obtain the elevation data within the mangrove vegetation layer area, and construct a mangrove point cloud data correction model by combining the point cloud data, elevation data, and signal category labels to correct the obtained point cloud data and obtain the point cloud data after terrain error correction. Step 3: Using a drone equipped with a multispectral imaging device, acquire multispectral images of mangroves, and perform noise reduction, SIFT key feature point extraction on the multispectral images. Align the images of different bands through affine transformation to form a multispectral data matrix. After standardizing the data of each band, obtain the multispectral data matrix. Step 4: Based on the point cloud data corrected in Step 2, identify the forest growth data characteristics. Based on the forest growth data characteristics, the remote sensing image data obtained in Step 1, and the multispectral data matrix obtained in Step 4, use a recurrent neural network to train the model and construct a mangrove carbon sink prediction model to predict the amount of mangrove carbon sink. Step 5: Based on the NDVI values ​​obtained in Step 1, image areas with NDVI values ​​higher than the preset NDVI threshold are marked as overly dense leaf areas. The area of ​​the overly dense leaf area is determined by calculating the number of pixels within it. The light suppression factor is calculated using the photosynthetic suppression factor formula based on the area of ​​the overly dense leaf area and the multispectral data matrix. The carbon sink deviation of the overly dense leaf area is determined using the regional carbon sink deviation formula based on the calculated light suppression factor. The true carbon sink value of the mangrove is obtained based on the carbon sink deviation of the overly dense leaf area and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in Step 4.

[0004] Further, step 1 includes acquiring lidar point cloud data of the mangrove area based on the settings of the lidar sensor on the UAV, and acquiring remote sensing image data using satellite remote sensing technology. The lidar point cloud data includes the three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest cover area, NDVI, and biomass. Based on the obtained raw point cloud data, noise removal is performed using high and low threshold filtering, voxel grid filtering, or statistical outlier filtering algorithms. Based on the filtered point cloud data, the height relationship of each point relative to surrounding points is determined, and points below a set threshold are classified as ground points, and points above a set threshold are classified as vegetation points, thus identifying the ground layer and vegetation layer of the mangrove.

[0005] Further, step 2 includes acquiring elevation data within the mangrove vegetation layer area, including the slope and altitude of the area; spatially interpolating the elevation data within the mangrove vegetation layer area using the Kriging method or inverse distance weighting method to obtain the elevation value of each pixel location; and training a model using the PointNet algorithm based on the acquired point cloud data, signal category labels, elevation data, and corrected point cloud data to construct a mangrove point cloud data correction model, correcting the acquired point cloud data to obtain point cloud data after terrain error correction, with signal category labels including ground layer and vegetation layer.

[0006] Further, step 3 includes acquiring multispectral images of mangroves using a drone equipped with a multispectral imaging device; removing noise from the multispectral images using a median filter; detecting and matching SIFT key feature points in images of different bands using the SIFT feature point matching algorithm; aligning images of different bands using affine transformation; integrating the reflectance data of each pixel in different bands into a vector to form a multispectral data matrix with a size of n×m, where n is the number of pixels and m is the number of bands, and each row of the multispectral data matrix represents the reflectance data of a pixel in all bands; standardizing the data of each band by subtracting the mean from the reflectance data of each band and then dividing by the standard deviation to obtain a standardized multispectral data matrix.

[0007] Further, step 4 includes identifying forest growth data characteristics based on the corrected lidar data of mangroves. These forest growth data characteristics include tree species distribution, tree height, diameter at breast height (DBH), crown width, and age structure. Based on the forest growth data characteristics, remote sensing image data, and multispectral data matrix, a recurrent neural network is used to train the model and construct a mangrove carbon sink prediction model to predict mangrove carbon sinks.

[0008] Further, step 5 includes calculating NDVI values ​​using multispectral images and generating an NDVI distribution image of the mangrove vegetation cover area. The NDVI distribution image is mapped to a grayscale image. Based on a preset NDVI threshold as a segmentation criterion, the grayscale image is binarized, and image areas with NDVI values ​​higher than the preset NDVI threshold are marked as overly dense leaf areas. The Canny edge detection algorithm is used to extract the regional contours of the overly dense leaf areas. The area of ​​the overly dense leaf areas is determined by calculating the number of pixels within the overly dense leaf areas. Based on the area of ​​the overly dense leaf areas and the multispectral data matrix, the photosynthetic suppression factor formula is used. To determine the photosynthetic suppression factor in areas of excessive leaf density, among which, m is the number of pixels in the region, and m is the number of bands. Let the normalized reflectance of the k-th pixel in the j-th band be denoted as . , representing the average value of the j-th band in region i; based on the photosynthetic suppression factor in areas with excessively dense leaves, the regional carbon sink deviation formula is used. Determine the carbon sink deviation in areas of excessively dense leaves. ,in, The carbon sequestration reduction factor per unit area was obtained by fitting historical data. This represents the area of ​​the overly dense leaf region. It is an exponential amplification of the total deviation based on the number of pixels and shading density within the region. It is a fine-tuning coefficient obtained by fitting historical data; based on the carbon sink deviation in the dense leaf area and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in step 4, the true carbon sink value of the mangrove is obtained.

[0009] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention effectively improves the measurement accuracy of mangrove carbon sequestration capacity by fusing lidar point cloud data and multispectral data matrices and employing recurrent neural networks to construct a mangrove carbon sequestration prediction model. It innovatively achieves the quantitative integration of the photosynthetic suppression effect caused by sub-light saturation within the mangrove canopy. Based on the calculated normalized difference vegetation index (NDVI) value, and using a preset NDVI threshold, the spatial range and area of ​​overly dense leaf regions are identified and determined. Using this region area and the multispectral data matrix, the photosynthetic suppression factor is calculated using the photosynthetic suppression factor formula. Combined with the regional carbon sequestration deviation formula, the carbon sequestration capacity loss caused by the photosynthetic suppression effect in overly dense leaf regions is accurately quantified, i.e., the carbon sequestration deviation. Finally, the preliminary carbon sequestration estimate output by the mangrove carbon sequestration prediction model is used to correct and compensate for this carbon sequestration deviation, obtaining a more accurate carbon sequestration value that better reflects the actual physiological and ecological response of mangroves. This solution effectively addresses the core problem of existing technologies that systematically overestimate carbon sequestration capacity due to neglecting the photosynthetic suppression effect caused by light competition within the canopy. It significantly improves the measurement errors caused by excessively dense canopies in mature forest areas, providing a more accurate data foundation for blue carbon accounting. Attached Figure Description

[0010] Figure 1 This is a flowchart of the steps of the present invention; Figure 2 The flowchart illustrates the steps involved in constructing a mangrove carbon sink prediction model for this invention. Figure 3 This is a flowchart illustrating the steps for calculating the true carbon sink value of mangroves according to the present invention. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0012] This invention takes a coastal mangrove wetland in Futian District, Shenzhen as an example. The mangrove area covers approximately 126.09 hectares and belongs to the South Asian subtropical marine monsoon climate zone, possessing characteristics of both estuary and bay. River-sea interaction, abundant fine-grained sediment deposition, and fertile water quality provide a favorable geomorphological environment for mangrove development. The mangrove communities in the area have a simple appearance, neat canopies, and a clustered distribution with a simple stand composition. Most are dominant mangrove communities composed of a single species or two or three species of mangrove plants, with the tallest trees reaching approximately 15 meters. Some artificially planted *Rhizophora stylosa* trees reach a height of approximately 7 meters. Shrubs growing under the mangroves are about 1.5 meters tall and often form dominant mangrove communities with other dominant mangrove plants. *Symplocos edulis* trees are distributed in clumps along the outer edge of the coastal communities, and are also scattered under the mangrove canopy, reaching a height of approximately 2 meters.

[0013] Please refer to Figure 1 This embodiment of a mangrove carbon sequestration monitoring and measurement method based on drones, radar, and AI technologies may specifically include: Step 1: Acquire lidar data of the mangrove forest using a drone equipped with lidar, and acquire remote sensing image data using satellite remote sensing technology. Classify and filter the lidar point cloud data to identify the forest ground layer and vegetation layer, and label the ground layer and vegetation layer with signal category tags respectively. Step 1 includes: acquiring lidar point cloud data of the mangrove area according to the settings of the lidar sensor on the drone, and acquiring remote sensing image data using satellite remote sensing technology. The lidar point cloud data includes the three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest cover area, NDVI, and biomass. Based on the obtained raw point cloud data, noise removal is performed using high and low threshold filtering, voxel grid filtering, or statistical outlier filtering algorithms. Based on the filtered point cloud data, determine the height relationship of each point relative to surrounding points, classifying points below a set threshold as ground points and points above a set threshold as vegetation points, thus identifying the mangrove ground layer and vegetation layer.

[0014] Specifically, a UAV equipped with a RIEGL VUX-1LR lidar system was used at a flight altitude of 150 meters to acquire raw point cloud data with a point density of 150 points / square meter. Remote sensing images of the mangrove region were acquired using the Landsat 8 satellite remote sensing platform, collecting image data in different bands. Geometric correction was performed on the raw satellite images to ensure consistency between the image coordinate system and the lidar point cloud data coordinate system. Based on the satellite remote sensing images, NDVI values ​​were calculated using near-infrared and red band remote sensing image data, with an NDVI threshold set between 0.2 and 0.6. Areas with high vegetation cover were selected, forest boundaries were extracted, and the forest cover area of ​​the mangrove region was determined. Based on the relationship between NDVI and biomass, a vegetation biomass inversion model, such as a spectral feature regression model, was used in conjunction with the remote sensing image data to generate the biomass and biomass distribution map of the entire mangrove region. To address high-altitude and low-altitude noise in the original point cloud data, such as bird and tidal reflections, a three-stage filtering process was first applied to remove outliers with elevations above 60 meters and below the chart datum of -2 meters. Next, voxel grid filtering was used to merge multiple points within a 0.1m x 0.1m x 0.1m cube into a single point, reducing redundant data. Finally, statistical outlier filtering was performed, calculating the standard deviation of elevation for each point within a 0.5m radius of its 15 nearest neighbors, removing outliers deviating more than 1.5 standard deviations from the mean. After these three stages of filtering, the total number of point clouds was reduced from 5 million to 4.2 million valid points. Based on this, a height thresholding method was used to separate the ground from vegetation. By setting 0.3 meters as the classification threshold, points with an elevation 0.3 meters lower than the surrounding terrain were classified as ground points, including mudflats and tidal channels, while points with abrupt elevation changes exceeding 0.3 meters were classified as vegetation points, including mangrove prop roots and canopies. Ultimately, the three-dimensional structure of the Kandelia candel community with an average canopy height of 8.2 meters was successfully identified.

[0015] Step 2: Based on the vegetation layer data obtained in Step 1, acquire elevation data within the mangrove vegetation layer area. Combine this with point cloud data, elevation data, and signal category labels to construct a mangrove point cloud data correction model. Correct the acquired point cloud data to obtain point cloud data after terrain error correction. Step 2 includes: acquiring elevation data within the mangrove vegetation layer area, including the area's slope and altitude; performing spatial interpolation on the elevation data within the mangrove vegetation layer area using Kriging or inverse distance weighting to obtain the elevation value for each pixel; training the model using the PointNet algorithm based on the acquired point cloud data, signal category labels, elevation data, and corrected point cloud data to construct a mangrove point cloud data correction model. Correct the acquired point cloud data to obtain point cloud data after terrain error correction. Signal category labels include ground layer and vegetation layer.

[0016] Specifically, to avoid systematic elevation deviations in the sloping areas caused by the influence of tidal gully topography on the original vegetation point cloud acquired by lidar, and to ensure the accuracy of subsequent quantification of photosynthetic suppression effects, the elevation characteristics of this area were first extracted: slope gradient 0.8°-7.5°, elevation range 1.8-4.3 meters. The terrain surface was reconstructed using Kriging interpolation with a range parameter of 18 meters / Kriging value of 0.15, generating a 1-meter resolution elevation grid. The interpolated elevations, the original point cloud, and manually labeled vegetation / ground layers were input into the PointNet model for training, focusing on learning the elevation error patterns caused by laser incident angle distortion at the tidal gully edges. After correction, the elevation deviation in the Avicennia marina prop root cluster area decreased from 0.25 meters to 0.03 meters; for example, the canopy point cloud elevation at a certain location was restored from 3.62 meters before correction to the actual 3.35 meters. This topographic correction directly affects the accuracy of canopy structure analysis. Without correction, canopy height in steep slope areas will be overestimated, leading to misjudgments of NDVI thresholds for areas with excessively dense leaves, such as identifying sloping canopies as vertically closed areas. Furthermore, errors in prop root height will distort the calculation of canopy thickness parameters in the light suppression factor formula, ultimately affecting the accuracy of carbon sink bias. This step lays the foundation for subsequent elimination of topographic interference in canopy photosynthetic efficiency assessment.

[0017] Step 3: Acquire multispectral images of the mangrove forest using a drone equipped with a multispectral imaging device. Denoise the multispectral images, extract SIFT key feature points, and align the images across different bands using affine transformation to form a multispectral data matrix. Standardize the data for each band to obtain the final multispectral data matrix. Specifically, Step 3 includes: acquiring multispectral images of the mangrove forest using a drone equipped with a multispectral imaging device; removing noise from the multispectral images using a median filter; detecting and matching SIFT key feature points in different bands using the SIFT feature point matching algorithm; aligning the images across different bands using affine transformation; integrating the reflectance data of each pixel across different bands into a vector to form a multispectral data matrix of size n×m, where n is the number of pixels and m is the number of bands; and standardizing the data for each band by subtracting the mean from the reflectance data and then dividing by the standard deviation to obtain a standardized multispectral data matrix.

[0018] Specifically, the drone was equipped with a RedEdge-MX multispectral camera, whose wavelengths are: blue 475nm, green 560nm, red 668nm, and red edge 717nm / 842nm, to acquire multispectral images of the Avicennia marina community. To address image noise caused by tidal reflection, a 5×5 pixel windowed mid-range filter was used for elimination, reducing abnormal pixel value fluctuations in a certain tidal channel area in the 717nm band from ±25 to ±8. Due to wind effects during high tide causing inter-band shifts, the SIFT algorithm was used to match feature points. Approximately 1500 key points were extracted from a single image, revealing an average displacement of 1.8 pixels in the red edge band relative to the blue band, approximately 0.27 meters above the ground. Accurate alignment was achieved through an affine transformation matrix, with a residual of <0.3 pixels. After alignment, the reflectance of each pixel across five bands is integrated into a vector, forming a multispectral data matrix with a resolution of 1024×768 pixels, consisting of 786432 rows × 5 columns. The data in the k-th row and 3rd column represents the reflectance of the k-th pixel in the red band. To eliminate the influence of differences in illumination conditions, each band is standardized. The mean reflectance of all pixels in the blue band is calculated to be 0.21 and the standard deviation is 0.07. The original value of 0.35 is converted to (0.35-0.21) / 0.07=2.0. This standardized matrix directly serves the NDVI threshold analysis of the 717nm / 842nm band in step 5, calculating the reflectance parameter input in the NDVI and light suppression factor formulas to ensure that the subsequent quantification of photosynthetic suppression effect is not affected by atmospheric scattering.

[0019] Please refer to Figure 2 Step 4: Based on the point cloud data corrected in Step 2, identify forest growth data characteristics. Using these characteristics, the remote sensing image data obtained in Step 1, and the multispectral data matrix obtained in Step 3, train a recurrent neural network to construct a mangrove carbon sink prediction model and predict mangrove carbon sink amounts. Step 4 includes: identifying forest growth data characteristics based on the corrected mangrove lidar data. These characteristics include tree species distribution, tree height, diameter at breast height (DBH), crown width, and age structure. Based on these characteristics, the remote sensing image data, and the multispectral data matrix, train a recurrent neural network to construct a mangrove carbon sink prediction model and predict mangrove carbon sink amounts.

[0020] Specifically, based on the point cloud data corrected for terrain errors using the PointNet model in step 2, the individual trees in the mangrove area were first segmented and their structural features extracted. The DBSCAN algorithm, based on density clustering, was used to cluster the point cloud, setting the spatial neighborhood radius ε to 0.45 meters and the minimum neighbor number MinPts to 25 to eliminate excessive connections between support roots. The segmentation results showed that 1265 individual trees were identified in the Avicennia marina region and 842 individual trees were identified in the Kandelia candel region. Based on the individual tree point cloud data, various forest growth data features were calculated by extracting the elevation difference between the highest and lowest ground points of each tree. For example, the height of the Kandelia candel sample was 8.34 meters, and the height of the Avicennia marina sample was 6.97 meters. A point cloud slice at a height of 1.3 meters above the ground was taken, and a circular profile was fitted to calculate the diameter, for example, the diameter at breast height (DBH) of the Kandelia candel was 0.21 meters, and that of the Avicennia marina was 0.17 meters. Projecting the canopy points onto a horizontal plane, the average major and minor axes of the minimum circumscribed ellipse were calculated. The average canopy width of *Kandelia candel* was 3.62 meters, and that of *Avicennia marina* was 2.95 meters. Based on species-specific growth curve models, the age distribution of individual trees was estimated. *Kandelia candel* trees were concentrated in the 12-18 year age range, while *Avicennia marina* trees were concentrated in the 9-14 year age range. Species distribution information was obtained by combining the red-edge band features from the Landsat 8 remote sensing imagery in step 1 with the multispectral matrix reflectance data from step 3, using random forest pre-classification to obtain spatial distribution maps showing *Kandelia candel* accounting for 54.8%, *Avicennia marina* for 41.6%, and *Rhizophora stylosa* for 3.6%. Extracted forest growth feature vectors, including tree species, tree height, diameter at breast height (DBH), crown width, tree age, vegetation index (NDVI) from remote sensing imagery, and 5-band reflectance values ​​from the multispectral matrix, were input into a recurrent neural network (RNN) for model training to construct a mangrove carbon sink prediction model. The RNN employed a three-layer structure: the input layer had a feature dimension of 15, containing growth features, spectral features, and vegetation index; the hidden layers had 64 and 32 nodes respectively; ReLU was used as the activation function; and the output layer contained the predicted carbon sink value. The training and validation set ratio was 7:3, the Adam optimizer was selected, and the initial learning rate was 0.001. After 1000 iterations, the validation set loss converged to 0.018. On the test dataset, the model achieved an R² of 0.94 (p > 0.05), passing the unbiasedness test. Based on the constructed mangrove carbon sink prediction model, the average monthly carbon sink of mangroves was estimated to be 255.3 kgC.

[0021] Please refer to Figure 3Step 5: Calculate the NDVI value based on the multispectral data matrix obtained in Step 3. Mark the image areas with NDVI values ​​higher than the preset NDVI threshold as overly dense leaf areas. Determine the area of ​​the overly dense leaf areas by calculating the number of pixels within them. Calculate the light suppression factor using the photosynthetic suppression factor formula based on the area of ​​the overly dense leaf areas and the multispectral data matrix. Determine the carbon sink deviation of the overly dense leaf areas using the regional carbon sink deviation formula based on the calculated light suppression factor. Obtain the true carbon sink value of the mangroves based on the carbon sink deviation of the overly dense leaf areas and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in Step 4.

[0022] Step 5 includes calculating NDVI values ​​using multispectral imagery and generating an NDVI distribution image of the mangrove vegetation cover. The NDVI distribution image is then mapped to a grayscale image. Based on a preset NDVI threshold as a segmentation criterion, the grayscale image is binarized, and image regions with NDVI values ​​higher than the preset NDVI threshold are marked as overly dense leaf regions. The Canny edge detection algorithm is used to extract the contours of the overly dense leaf regions. The area of ​​the overly dense leaf regions is determined by calculating the number of pixels within each region. Finally, based on the area of ​​the overly dense leaf regions and the multispectral data matrix, the photosynthetic suppression factor formula is used. To determine the photosynthetic suppression factor in areas of excessive leaf density, among which, m is the number of pixels in the region, and m is the number of bands. Let the normalized reflectance of the k-th pixel in the j-th band be denoted as . , representing the average value of the j-th band in region i; based on the photosynthetic suppression factor in areas with excessively dense leaves, the regional carbon sink deviation formula is used. Determine the carbon sink deviation in areas of excessively dense leaves. ,in, The carbon sequestration reduction factor per unit area was obtained by fitting historical data. This represents the area of ​​the overly dense leaf region. It is an exponential amplification of the total deviation based on the number of pixels and shading density within the region. It is a fine-tuning coefficient obtained by fitting historical data; based on the carbon sink deviation in the dense leaf area and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in step 4, the true carbon sink value of the mangrove is obtained.

[0023] Specifically, based on NDVI calculations and identification of excessively dense areas in the blades, and using the standardized multispectral data matrix generated in step 3, the reflectance of the target grid in the 842nm and 668nm bands is extracted from a 1024×768 pixel×5 band matrix. Reflectance at pixel k=42: .

[0024] Calculate NDVI: ; ; The preset NDVI threshold is 0.68, which is set based on the threshold of LAI > 4.0. Pixels with NDVI ≥ 0.68 are set to 1, and the rest to 0. Connectivity regions are counted, and 92 consecutive pixels exceeding the threshold, including pixel 57, are marked. The Canny algorithm parameters are used, with Gaussian kernel σ = 1.0, low threshold = 0.1, and high threshold = 0.3, outputting a closed polygon area of ​​185 ± 1.5 m², corresponding to 92 1m × 1m pixels.

[0025] Based on the area of ​​the overly dense leaf region and the multispectral data matrix, the photosynthetic suppression factor formula was used. The photosynthetic suppression factor LPSI in dense leaf regions was determined, among which... m is the number of pixels in the region, and m is the number of bands. Let the normalized reflectance of the k-th pixel in the j-th band be denoted as . , where represents the average value of the j-th band in region i.

[0026] Mean reflectance calculation: For the 92 pixels in the region, calculate the mean reflectance of each of the 5 bands according to the formula: .

[0027] Single-band standard deviation calculation: Normalized reflectance of pixel k=57: ; Deviation squared: ; Summation of regions: ; Standard deviation: ; LPSI Integrated Computing:

[0028] Carbon sink deviation Parameter acquisition: Based on flux tower data fitting, the average loss rate of the shaded area is 18.2%, α=0.18, and the Canny profile area is... For a surface area of ​​185 m², a canopy height model was generated from the lidar point cloud. The calculated surface roughness was 3.8, and γ = 0.003. The total number of pixels in the region was n_i = 92, with a tidal humidity correction term of δ = 0.015.

[0029] Logarithmic calculations: ; .

[0030] Carbon sink deviation integration: .

[0031] Real carbon sequestration value generation: The carbon sequestration output in step 4 is 255.3 kgC, and the isal density after unit conversion is... .

[0032] The actual carbon sink value = 255.3 - 10.715 = 244.585 kgC.

[0033] Unit conversion: .

[0034] Model validation: Six LI-6800 leaf chambers were arranged in a hexagonal grid in a 185m² area. The photosynthetic parameters of the lower leaves (2.5-3.0 meters from the canopy top) were measured, as shown in the table below: Table 1 Photosynthetic parameters

[0035] Accuracy comparison: Output of this solution: 1.322 kgC / m² → Deviation +0.38% Uncorrected model: 1.38 kgC / m² → Deviation +4.78% Traditional NDVI model: 1.41 kgC / m² → Deviation +7.06% Verification shows that by correcting and compensating for the carbon sequestration deviation in the initial carbon sequestration estimate output by the mangrove carbon sequestration prediction model, a more accurate carbon sequestration value that reflects the actual physiological and ecological response of mangroves can be obtained. This solution effectively addresses the core problem of existing technologies systematically overestimating carbon sequestration capacity due to neglecting the photosynthetic suppression effect caused by light competition within the canopy. It significantly improves the measurement error caused by excessively dense canopies in mature forest areas, providing a more accurate data foundation for blue carbon accounting.

[0036] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for monitoring and measuring mangrove carbon sequestration based on drones, radar, and AI technologies, characterized in that, The method includes: Step 1: Acquire lidar data of mangrove forests using a drone equipped with lidar, acquire remote sensing image data using satellite remote sensing technology, classify and filter lidar point cloud data, identify the ground layer and vegetation layer of the forest, and label the ground layer and vegetation layer with signal category labels respectively. Step 2: Based on the vegetation layer data obtained in Step 1, obtain the elevation data within the mangrove vegetation layer area, and construct a mangrove point cloud data correction model by combining the point cloud data, elevation data, and signal category labels to correct the obtained point cloud data and obtain the point cloud data after terrain error correction. Step 3: Using a drone equipped with a multispectral imaging device, acquire multispectral images of mangroves, and perform noise reduction, SIFT key feature point extraction on the multispectral images. Align the images of different bands through affine transformation to form a multispectral data matrix. After standardizing the data of each band, obtain the multispectral data matrix. Step 4: Based on the point cloud data corrected in Step 2, identify the forest growth data characteristics. Based on the forest growth data characteristics, the remote sensing image data obtained in Step 1, and the multispectral data matrix obtained in Step 3, use a recurrent neural network to train the model and construct a mangrove carbon sink prediction model to predict the amount of mangrove carbon sink. Step 5: Based on the NDVI values ​​obtained in Step 1, image areas with NDVI values ​​higher than the preset NDVI threshold are marked as overly dense leaf areas. The area of ​​the overly dense leaf area is determined by calculating the number of pixels within it. The light suppression factor is calculated using the photosynthetic suppression factor formula based on the area of ​​the overly dense leaf area and the multispectral data matrix. The carbon sink deviation of the overly dense leaf area is determined using the regional carbon sink deviation formula based on the calculated light suppression factor. The true carbon sink value of the mangrove is obtained based on the carbon sink deviation of the overly dense leaf area and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in Step 4.

2. The method for monitoring and measuring mangrove carbon sequestration based on UAV, radar, and AI technologies according to claim 1, characterized in that, Step 1 includes: acquiring lidar point cloud data of the mangrove area based on the settings of the lidar sensor on the UAV, and acquiring remote sensing image data using satellite remote sensing technology. The lidar point cloud data includes the three-dimensional coordinate information and signal reflection intensity of each point, and the remote sensing image data includes forest cover area, NDVI, and biomass. Based on the obtained raw point cloud data, noise removal is performed using high and low threshold filtering, voxel grid filtering, or statistical outlier filtering algorithms. Based on the filtered point cloud data, the height relationship of each point relative to surrounding points is determined, and points below a set threshold are classified as ground points, and points above a set threshold are classified as vegetation points, thus identifying the ground layer and vegetation layer of the mangrove.

3. The mangrove carbon sequestration monitoring and measurement method based on UAV, radar, and AI technologies according to claim 1, characterized in that, Step 2 includes: acquiring elevation data within the mangrove vegetation layer area, including the slope and altitude of the area; spatially interpolating the elevation data within the mangrove vegetation layer area using Kriging or inverse distance weighting to obtain the elevation value of each pixel location; training a model using the PointNet algorithm based on the acquired point cloud data, signal category labels, elevation data, and corrected point cloud data to construct a mangrove point cloud data correction model, correcting the acquired point cloud data to obtain point cloud data after terrain error correction, with signal category labels including ground layer and vegetation layer.

4. The method for monitoring and measuring mangrove carbon sequestration based on UAV, radar, and AI technologies according to claim 1, characterized in that, Step 3 includes: acquiring multispectral images of mangroves using a drone equipped with a multispectral imaging device; removing noise from the multispectral images using a median filter; detecting and matching SIFT key feature points in images of different bands using the SIFT feature point matching algorithm; aligning images of different bands using affine transformation; integrating the reflectance data of each pixel in different bands into a vector to form a multispectral data matrix with a size of n×m, where n is the number of pixels and m is the number of bands, and each row of the multispectral data matrix represents the reflectance data of a pixel in all bands; standardizing the data of each band by subtracting the mean from the reflectance data of each band and then dividing by the standard deviation to obtain a standardized multispectral data matrix.

5. The mangrove carbon sequestration monitoring and measurement method based on UAV, radar, and AI technologies according to claim 1, characterized in that: Step 4 includes: identifying forest growth data characteristics based on the corrected mangrove lidar data, including tree species distribution, tree height, diameter at breast height, crown width, and age structure; using a recurrent neural network to train a model based on the forest growth data characteristics, remote sensing image data, and multispectral data matrix, constructing a mangrove carbon sink prediction model, and predicting mangrove carbon sinks.

6. The method for monitoring and measuring mangrove carbon sequestration based on UAV, radar, and AI technologies according to claim 1, characterized in that, Step 5 includes generating an NDVI distribution image of the mangrove vegetation cover using the NDVI value obtained in step 1, mapping the NDVI distribution image to a grayscale image, performing binarization processing on the grayscale image based on a preset NDVI threshold as a segmentation standard, and marking image areas with NDVI values ​​higher than the preset NDVI threshold as areas with excessively dense leaves. The Canny edge detection algorithm is used to extract the region contour of the overly dense area of ​​the leaf. The area of ​​the overly dense area of ​​the leaf is determined by calculating the number of pixels in the overly dense area. Based on the area of ​​the overly dense leaf region and the multispectral data matrix, the photosynthetic suppression factor formula was used. To determine the photosynthetic suppression factor in areas of excessive leaf density, among which, m is the number of pixels in the region, and m is the number of bands. Let the normalized reflectance of the k-th pixel in the j-th band be denoted as . , representing the average value of the j-th band in region i; based on the photosynthetic suppression factor in areas with excessively dense leaves, the regional carbon sink deviation formula is used. Determine the carbon sink deviation in areas of excessively dense leaves. ,in, The carbon sequestration reduction factor per unit area was obtained by fitting historical data. This represents the area of ​​the overly dense leaf region. It is an exponential amplification of the total deviation based on the number of pixels and shading density within the region. It is a fine-tuning coefficient obtained by fitting historical data; based on the carbon sink deviation in the dense leaf area and the carbon sink calculation value predicted by the mangrove carbon sink prediction model constructed in step 4, the true carbon sink value of the mangrove is obtained.

Images