GEDI AGBD data correction and quality improvement method based on three-dimensional structure information

Abstract

The invention discloses a GEDIAGBD data correction and quality improvement method based on three-dimensional structure information, and belongs to the field of remote sensing science and technology, and the method comprises the steps: respectively obtaining L2A canopy height data, L2B canopy coverage information and L4A above-ground biomass density (AGBD) data of GEDI; preprocessing is carried out to extract effective canopy height, total coverage, vegetation area index, vegetation area volume density and vertical structure information thereof; taking overground biomass data measured by an actual sample plot as a reference, performing biomass inversion of an airborne area, and performing spatial position matching with the L4A data to obtain spatially overlapped reference AGBD data; the method comprises the following steps: selecting data with better quality in L4A data as a training sample, taking three-dimensional structure information as a characteristic variable, carrying out model training by using an improved different-speed growth equation, carrying out nonlinear parameter fitting by using a GBM machine learning algorithm, comparing different model precisions, obtaining an optimal training model, carrying out data correction, and obtaining final L4AAGBD data.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing science and technology, and in particular relates to a method for GEDI AGBD data correction and quality improvement based on three-dimensional structural information. Background Technology

[0002] Forest aboveground biomass is an important parameter characterizing the structure and function of terrestrial ecosystems. Accurately understanding forest canopy structure and efficiently estimating forest aboveground biomass are key requirements for deepening carbon cycling and promoting scientific forest management. It is also a necessary support for achieving the "dual carbon targets" and implementing "carbon measurement," and an important issue that has received widespread attention and urgent research from forestry departments and related scholars.

[0003] Traditional methods for measuring canopy height in the field are time-consuming and labor-intensive, while advancements in remote sensing technology offer a relatively low-cost approach for large-scale coverage. However, optical remote sensing primarily captures horizontal forest structure information and is prone to saturation in dense forest areas. LiDAR, due to its penetrating power, has an absolute advantage in acquiring canopy and structural information. The emergence of the Global Ecosystem Dynamics Investigation (GEDI) space-based lidar system not only breaks the limitation of optical remote sensing in acquiring vegetation vertical structure information but also solves the problem of airborne lidar's inability to cover large areas. GEDI provides a new data foundation and research approach for acquiring forest vertical structure and aboveground biomass, especially with the public release of aboveground biomass density (AGBD) products at both the spot and grid scales. However, the relatively coarse resolution, the limitations of plot-scale AGBD data, and the difficulty in aligning it with spot-scale AGBD data make the quality evaluation and data calibration of GEDI L4A products challenging. Furthermore, due to data limitations, GEDI products lack effective validation in several regions, including East Asia, resulting in significant uncertainty in the application of data in these areas. Furthermore, the GEDI AGBD product uses only the relative height of the vegetation canopy as input, which can easily lead to insufficient interpretability of biomass results, thus limiting the further application of the data. Although some scholars have conducted assessments of AGBD data quality, corrected geolocation errors, and calibrated and improved the data, the uncertainty of aboveground biomass data remains high compared to the data quality of canopy height. For example, in the natural forests of Laos, the absolute deviation of the current GEDI L4A AGBD product ranges from -54.24 to 106.23 Mg / ha across different forest types. In some areas of India, GEDI AGBD estimates are severely mismatched with reference AGBD maps, and in Japan, the data uncertainty exceeds 62%. These studies further emphasize the uncertainty of the data and the necessity for data quality assessment and improvement.

[0004] To address the low accuracy and uncertainty of AGBD data, studies have proposed methods for improvement, including stratified calibration of GEDI AGBD using least-squares linear regression with local data and auxiliary variables, local calibration using generalized linear models and random forest models, and direct calibration via linear fitting models. These studies have effectively improved data quality. However, most studies utilize forest attribute information and attribute variables for quality improvement and local refinement. While some studies considered vegetation area indices or vegetation cover in GEDI L2B data, these were based on overall horizontal values ​​and did not consider the vertical structure of the forest canopy and the corresponding three-dimensional information of its horizontal profile. Furthermore, most of their models were constructed by taking the square root of the canopy height index and then performing linear regression. Additionally, some studies have used GEDI waveform simulation, using waveform errors or waveforms as model input variables, combined with geostatistical interpolation, to correct and estimate GEDI AGBD data. While interpolation-based methods may be superior to calibration based on airborne lidar data, they are less sensitive to errors. Meanwhile, similar to the "same object, different spectrum" and "same spectrum, different object" characteristics of optical images, forest structures under different AGBDs may produce highly similar waveforms, while different waveforms may exist under the same AGBD. This "same wave, different structure, different waveform" characteristic will contribute certain uncertainties to AGBD prediction, especially posing challenges to prediction methods that use GEDI waveforms as input.

[0005] In summary, although the GEDI spaceborne lidar system provides a large-scale data foundation for forest aboveground biomass estimation, there are significant uncertainties. Even though existing studies have calibrated, standardized, and improved the quality of the data, they have not yet taken into account three-dimensional structural information, and the calibration coefficients in different regions are not applicable to other study areas.

[0006] To address the shortcomings of existing technologies, this invention provides a method for GEDIAGBD data correction and quality improvement based on three-dimensional structural information, aiming to solve the aforementioned problems. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for GEDIAGBD data correction and quality improvement based on three-dimensional structural information. This method can solve the problems of large uncertainty and insufficient interpretability of biomass data of existing spaceborne lidar GEDI and provide a methodological reference for biomass estimation of airborne lidar.

[0008] To achieve the above objectives, in a first aspect, the present invention provides a method for GEDIAGBD data correction and quality improvement based on three-dimensional structural information, the method comprising the following steps:

[0009] S1: Obtain aboveground biomass data of L4AAGBD, canopy height data of L2A, and canopy cover information data of GEDI in the study area, and perform data preprocessing.

[0010] S2: Extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD data, and perform preliminary data quality screening and geolocation error correction.

[0011] S3: Based on the aboveground biomass data measured in the actual sample plots, biomass inversion of the airborne area is performed, and spatial location matching is performed with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform quality evaluation.

[0012] S4: Select high-quality data from the L4AAGBD dataset as training samples, train the model using the improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model.

[0013] S5: Based on the optimized model in step S4, correct all L4AAGBD data in the study area to obtain corrected and improved L4AAGBD data.

[0014] In conjunction with the first aspect, the data preprocessing in step S1 includes:

[0015] The acquired L4A aboveground biomass data, L2A canopy height data, and L2B canopy cover information data were format converted and attribute information extracted. Furthermore, the L2A canopy height data, L2B canopy cover information data, and L4A biomass data were matched for location, and empty data and non-forest area data were deleted.

[0016] In conjunction with the first aspect, the specific process of step S2 is as follows:

[0017] Based on the L2A canopy height data, the corresponding relative canopy height data, canopy top height information, and elevation information were extracted; simultaneously, based on the L2B canopy cover information data, the total coverage, vegetation area index data, vegetation area volume density data, and their vertical structure information for the corresponding geographical location were extracted; and AGBD information was extracted from the L4A biomass data.

[0018] Based on S1, preliminary screening of the quality of L2A canopy height data, L2B canopy cover information data and L4A biomass data and correction of geolocation errors were carried out.

[0019] In conjunction with the first aspect, the specific process of step S4 is as follows:

[0020] S41: Based on the traditional allometric growth equation of "biomass-tree height-diameter at breast height" (6-1), calculate the aboveground biomass of the forest.

[0021]

[0022] Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; D represents diameter at breast height (DBH); and H represents canopy top height or tree height.

[0023] S42: From a three-dimensional perspective, the original biomass-tree height-diameter-thickness model is refined and updated to a three-dimensional structure model of biomass-tree height vertical structure-canopy cover, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area index, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area volume density.

[0024] S43: Taking the three-dimensional structure of biomass-tree height vertical structure-canopy cover as an example, the product between tree height and tree cover is proportional to the diameter at breast height, as shown in formula (6-2);

[0025]

[0026] Where D represents diameter at breast height (DBH); cover represents canopy coverage; h represents canopy height; H1 represents canopy base height; and H represents canopy top height.

[0027] S44: Substitute formula (6-2) into formula (6-1) to obtain formula (6-3);

[0028]

[0029] S45: The canopy cover is partitioned using GEDI L2B data partitioning intervals, and the partition at each i is defined using cover... i and h i The number of i is determined by the canopy top height and the profile interval. Therefore, formula (6-3) can be converted into the following formula (6-4), and the aboveground biomass AGB is divided by the area S to obtain the aboveground biomass density AGBD formula (6-5).

[0030]

[0031] Where dh represents the profile interval; h i Vertical subdivision information representing relative canopy height; S represents area;

[0032] S46: The parameters are standardized to obtain the final AGBD calculation model, as shown in formula (6-6);

[0033]

[0034] Where k0, k1, and k2 represent coefficients; H represents the height of the canopy top, and when i equals H / dh, h i =H.

[0035] In conjunction with the first aspect, step S4 can also employ the following process: From a volumetric perspective, the existing allometric growth equation based on biomass-tree height-canopy area / diameter is updated to derive the three-dimensional structural information relating biomass to canopy height and canopy coverage (or vegetation area index or vegetation area volume density), thus obtaining a formula for calculating forest aboveground biomass. Specifically, as follows:

[0036] S41: Taking the biomass-tree height-canopy area model as an example, its allometric growth equation is shown in equation (6-7):

[0037]

[0038] Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; cover area H represents the area covered by the crown; H represents the tree height.

[0039] S42: Convert the planar area of ​​the canopy into area multiplied by canopy coverage, as shown in formulas (6-8) and (6-9). Area multiplied by height can be approximated as an expression of the volume of a cylinder. After horizontal division, the volume of the entire cylinder can be considered as the sum of the volumes of individual division blocks, resulting in formula (6-10); and a new biomass density calculation formula (6-11) is obtained.

[0040]

[0041] Where S represents area; dh i AGBD represents the profile interval; aboveground biomass density.

[0042] Secondly, this invention provides a GEDIAGBD data correction and quality improvement system based on three-dimensional structural information; the system includes:

[0043] The data acquisition module is used to acquire L4A biomass data, L2A canopy height data, and L2B canopy cover information data of GEDI in the study area, and to perform data preprocessing.

[0044] The information extraction module is used to extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD, and perform preliminary data quality screening and geolocation error correction.

[0045] The data matching and quality assessment module is used to perform biomass inversion of the airborne area with the aboveground biomass data measured in the actual sample plot as a reference, and to perform spatial location matching with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform GEDI L4A data quality assessment.

[0046] The model training and optimization module is used to select high-quality data from the L4AAGBD dataset as training samples, train the model using an improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model.

[0047] The data correction and quality improvement module is used to correct all L4AAGBD data in the study area based on the optimized model in the model training and optimization module, so as to obtain corrected and improved L4AAGBD data.

[0048] In conjunction with the second aspect, the data acquisition module further includes a data preprocessing unit, which is used to perform format conversion and attribute information extraction on the acquired L4A biomass data, L2A canopy height data, and L2B canopy cover information data; and to perform location matching, deletion of empty data and non-forest area data on L2A canopy height data, L2B canopy cover information data, and L4A aboveground biomass data.

[0049] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0050] Fourthly, the present invention provides an apparatus comprising:

[0051] Memory, used to store instructions;

[0052] A processor for executing the instructions, causing the device to perform the steps of the method as described in the first aspect.

[0053] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0054] 1. This invention takes forests as the research object, and based on three-dimensional and volumetric perspectives, integrates two improved allometric growth equations to calibrate and improve the quality of GEDI AGBD data. It further verifies and supplements existing machine learning algorithms, breaking away from the old approach of two-dimensional estimation of forest aboveground biomass and shifting to a new approach of three-dimensional estimation. It clarifies the effective variables for estimating the light spot scale of forest aboveground biomass and has a clear theoretical basis. The method is simple to operate and yields accurate results. It not only effectively solves the problem of insufficient explanatory power of GEDI forest aboveground biomass but also promotes the further application of spaceborne lidar biomass products. Furthermore, it provides methodological and conceptual references for airborne lidar aboveground biomass estimation and data calibration for other vegetation types in GEDI. Attached Figure Description

[0055] Figure 1 This is a flowchart of the present invention.

[0056] Figure 2 This refers to the AGBD estimation accuracy under different parameter combinations in this invention.

[0057] Figure 3 This is a comparison chart showing the data quality improvement effect of GEDIAGBD in this invention. Detailed Implementation

[0058] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0059] Example 1

[0060] To verify the feasibility and effectiveness of the method provided by this invention, a sample area was selected in Northeast China for the experiment. The forests in this area have extensive forest coverage and rich tree species, making them suitable for experimental observation, calculation analysis and practical application.

[0061] refer to Figures 1-3 This invention provides a method for GEDIAGBD data correction and quality improvement based on three-dimensional structural information, the method comprising the following steps:

[0062] S1: Obtain aboveground biomass data of L4AAGBD, canopy height data of L2A, and canopy cover information data of GEDI in the study area, and perform data preprocessing.

[0063] S2: Extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD, and perform preliminary data quality screening and geolocation error correction.

[0064] S3: Based on the aboveground biomass data measured in the actual sample plots, biomass inversion of the airborne area is performed, and spatial location matching is performed with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform quality evaluation.

[0065] S4: Select high-quality data from the L4AAGBD dataset as training samples, train the model using the improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model.

[0066] S5: Based on the optimized model in step S4, correct all L4AAGBD data in the study area to obtain corrected and improved L4AAGBD data.

[0067] Specifically, in this embodiment, the data preprocessing in step S1 includes:

[0068] The acquired L4AAGBD aboveground biomass data, L2A canopy height data, and L2B canopy cover information data were converted in format and their attribute information was extracted. Furthermore, the L2A canopy height data, L2B canopy cover information data, and L4A biomass data were matched for location, and empty data and non-forest area data were deleted.

[0069] The specific operation is as follows: GEDI products are provided in HDF-5 format with a spatial resolution of 25m. MATLAB (2019 version) is used to convert all data to CSV format. Then, the required spatial data is cropped according to the geographical location of the experimental area, ensuring consistency in data acquisition time. Further filtering is performed during the three-category data matching process, removing data with empty 3D structural information, and deleting data from non-forest areas using land use and cover products.

[0070] Specifically, in this embodiment, the process in step S2 is as follows:

[0071] Based on the L2A canopy height data, the corresponding relative canopy height data, canopy top height information, and elevation information were extracted; simultaneously, based on the L2B canopy cover information data, the total coverage, vegetation area index data, vegetation area volume density data, and their vertical structure information for the corresponding geographical location were extracted; and AGBD information was extracted from the L4A biomass data.

[0072] Based on S1, preliminary screening of the quality of L2A canopy height data, L2B canopy cover information data and L4A biomass data and correction of geolocation errors were carried out.

[0073] Specifically, in this embodiment, step S2 further includes:

[0074] Various relative canopy height indicators, sensitivity, and quality flags were extracted, and poor-quality data were initially filtered out using built-in parameters. For example, the filtering criteria for L2A canopy height data included quality_flag=1, sensitivity>0.9, degrade_flag=0, and elev_lowestmode-digital_elevation_model_srtm<50. Similarly, L2B canopy cover information and L4A biomass data underwent preliminary quality screening based on quality flags in the attribute information. The geolocation error correction steps included: first, determining the window size to be 51*51; second, obtaining the elevation error matrix between the GEDI L4A elevation and the reference airborne lidar DEM; and finally, calculating the optimal offset by obtaining the position with the smallest elevation error within the window, thereby correcting the geolocation error and eliminating its impact.

[0075] Furthermore, in this embodiment, the specific process of step S3 is as follows:

[0076] Based on aboveground biomass data from sample plots, relative canopy height, coefficient of variation, standard deviation, variance, abundance, skewness, interquartile range, mean absolute deviation of height, maximum, minimum, median, mean, density, cover, and leaf area index were obtained from airborne lidar point cloud data. Using aboveground biomass data from the sample plots as a sample, a random forest model was established using the above parameters to invert the biomass of the airborne region. Based on the S2-processed data, AGBD data for the corresponding L4A locations were obtained from the airborne derived biomass data. The quality of the L4A data was then evaluated using assessment indicators, and low-quality L4A data was further removed.

[0077] Furthermore, in this embodiment, the specific process of step S4 is as follows:

[0078] S41: Based on the traditional allometric growth equation of "biomass-tree height-diameter at breast height" (6-1), calculate the aboveground biomass of the forest.

[0079]

[0080] Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; D represents diameter at breast height (DBH); and H represents canopy top height or tree height.

[0081] S42: From a three-dimensional perspective, the original biomass-tree height-diameter-thickness model is refined and updated to a three-dimensional structure model of biomass-tree height vertical structure-canopy cover, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area index, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area volume density.

[0082] S43: Taking the three-dimensional structure of biomass-tree height vertical structure-canopy cover as an example, the product between tree height and tree cover is proportional to the diameter at breast height, as shown in formula (6-2);

[0083]

[0084] Where D represents diameter at breast height (DBH); cover represents canopy coverage; h represents canopy height; H1 represents canopy base height; and H represents canopy top height.

[0085] S44: Substitute formula (6-2) into formula (6-1) to obtain formula (6-3);

[0086]

[0087] S45: The canopy cover is partitioned using GEDI L2B data partitioning intervals, and the partition at each i is defined using cover... i and h i The number of i is determined by the canopy top height and the profile interval. Therefore, formula (6-3) can be converted into the following formula (6-4), and the aboveground biomass AGB is divided by the area S to obtain the aboveground biomass density AGBD formula (6-5).

[0088]

[0089] Where dh represents the profile interval; h i Vertical subdivision information representing relative canopy height; S represents area;

[0090] S46: The parameters are standardized to obtain the final AGBD calculation model, as shown in formula (6-6);

[0091]

[0092] Where k0, k1, and k2 represent coefficients; H represents the height of the canopy top, and when i equals H / dh, h i =H.

[0093] It should be noted that in this embodiment, the three-dimensional structural information of biomass is related to canopy height and canopy coverage (or vegetation area index or vegetation area volume density). Therefore, step S4 can also be implemented using the following process: From a volumetric perspective, the original allometric growth equation of biomass-tree height-canopy area / diameter is updated to derive the three-dimensional structural information of biomass related to canopy height and canopy coverage (or vegetation area index or vegetation area volume density), thus obtaining the calculation formula for forest aboveground biomass. Specifically, as follows:

[0094] S41: Taking the biomass-tree height-canopy area model as an example, its allometric growth equation is shown in equation (6-7):

[0095]

[0096] Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; cover area H represents the area covered by the crown; H represents the tree height.

[0097] S42: Taking the biomass-tree height-canopy area model as an example, the planar area of ​​the canopy is transformed into area multiplied by canopy coverage, as shown in formulas (6-8) and (6-9). Area multiplied by height can be approximated as an expression of the volume of a cylinder. After horizontal division, the volume of the entire cylinder can be considered as the sum of the volumes of individual division blocks, resulting in formula (6-10); a new biomass density calculation formula (6-11) is obtained.

[0098]

[0099] Where S represents area; dh i AGBD represents the profile interval; aboveground biomass density.

[0100] In step S4, the AGBD derived from the airborne lidar is used as the reference ground truth data to perform an initial uncertainty evaluation of the GEDI L4A data. The evaluation index includes the coefficient of determination R. 2 RMSE, MAE, and rRMSE; these coefficients are shown in Table 1.

[0101] Based on Tables 1 and 2, R 2 The coefficient of determination represents the model; RMSE represents the root mean square error; MAE represents the mean absolute error; rRMSE represents the relative root mean square error.

[0102] Table 1: Preliminary Evaluation of GEDIAGBD Data Quality

[0103]

[0104] Table 2: Model accuracy under deformation of the allometric growth equation (logarithmic)

[0105]

[0106]

[0107] Table 2 above shows the model performance based on the improved allometric growth equation under different parameter combinations in this invention.

[0108] AGBD-H-PAI indicates that the model uses canopy top height (H) and total vegetation area index (PAI) as input variables to estimate aboveground biomass density (AGBD);

[0109] AGBD-H-PAI_i indicates that the model not only uses the canopy top height (H) and vegetation area index (PAI), but also considers the structural information of PAI in the vertical direction, that is, the PAI variation at different height layers;

[0110] AGBD-PAI_i indicates that the model uses only the plant area index (PAI) and its vertical structure information (PAI_i) to estimate the aboveground biomass density (AGBD), without considering the canopy top height (H);

[0111] AGBD-H-CC indicates that the model uses canopy top height (H) and total canopy cover (CC) as input variables to estimate aboveground biomass density (AGBD);

[0112] AGBD-H-CC_i indicates that the model uses the canopy top height (H) and canopy coverage (CC), and takes into account the structural information of canopy coverage in the vertical direction;

[0113] AGBD-CC_i indicates that the model uses only canopy cover (CC) and its vertical structure information (i) to estimate aboveground biomass density (AGBD), without considering the height of the top of the canopy (H);

[0114] AGBD-H-PAVD indicates that the model uses canopy top height (H) and total vegetation area volume density (PAVD) as input variables to estimate aboveground biomass density (AGBD);

[0115] AGBD-H-PAVD_i indicates that the model uses the canopy top height (H) and vegetation area volume density (PAVD), and takes into account the structural information of PAVD in the vertical direction;

[0116] AGBD-PAVD_i indicates that the model uses only plant area volume density (PAVD) and its vertical structure information (i) to estimate aboveground biomass density (AGBD), without considering the canopy top height (H);

[0117] AGBD-H indicates that the model uses only the canopy top height (H) as an input variable to estimate aboveground biomass density (AGBD), without considering other variables;

[0118] AGBD-Hi indicates that the model uses vertical structure information of the canopy height to estimate aboveground biomass density (AGBD);

[0119] AGBD-Hi-PAI_i indicates that the model uses canopy height (Hi) and vegetation area index (PAI), and takes into account the vertical structure information (i) of H and PAI to estimate aboveground biomass density (AGBD);

[0120] To verify the effectiveness of this scheme, horizontal and vertical comparison schemes were designed, including models that use only the overall information of canopy cover and models that use only the top height information of the canopy, as well as models that consider vertical structural information.

[0121] According to Table 2, this embodiment obtained the AGBD estimation accuracy under different parameter combinations in the present invention. It was found that the model considering the vertical structure information of canopy height and the vertical structure information of canopy coverage or vegetation area index has better accuracy.

[0122] Furthermore, nonparametric fitting was performed using machine learning, and the results are as follows: Figure 2 It can be concluded that, comparing horizontally (the first and second columns), the addition of vertical structural information i from PAI, CC, or PAVD, which represent canopy horizontal information, improves model performance. Comparing vertically, when only PAI, CC, and PAVD are used for modeling, the AGBD-H-PAVD model performs the worst, with an RMSE of 18.86 mg / ha; however, when the vertical structural information of all three is added, the performance of AGBD-H-PAVD_i is essentially equivalent to that of the other two models.

[0123] The first two lines of the last row Figure 2 For j and k, the addition of canopy height vertical structure information Hi results in better model performance than the addition of canopy top height H. Furthermore, the bidirectional addition of canopy height vertical structure information Hi and structural information representing canopy horizontal coverage (…) Figure 2 (l)AGBD-Hi-PAI_i) can achieve optimal model performance. That is... Figure 2 (l)AGBD-Hi-PAI_i has better accuracy than graphs j, k, a, b and c.

[0124] The third column, compared to the second column, can achieve a precision comparable to that of the second column. This can be explained by the formula mentioned above from a volumetric perspective.

[0125] In conclusion: (1) The models involving the three types of variables PAI_i, CC_i and PAVD_i have similar effects, but when only the overall horizontal information is combined with the canopy height, PAVD performs the worst (Figure 2(g)).

[0127] (2) Model accuracy using CC_i (or PAI_i or PAVD_i) vertical structure information Figure 2 (b), (e), and (h)) should have higher accuracy than simply using canopy top height and overall CC (or PAI or PAVD) level information. Figure 2 (a), (d), (g)).

[0128] (3) The model accuracy was achieved using the vertical structure information Hi of the canopy height. Figure 2 (k) is superior to the accuracy of models that only use the canopy top height H. Figure 2 (j)), while the AGBD-Hi-PAI_i model has the highest test accuracy (j) Figure 2 (l)).

[0129] Furthermore, this model (AGBD-Hi-PAI_i) was used for AGBD data correction and quality improvement at the GEDI spot scale, and the results are as follows: Figure 3 As shown.

[0130] based on Figure 3 , Figure 3 The horizontal axis of (a) and (b) represents the reference true value, mg / ha is the unit of aboveground biomass density, and the vertical axis represents the original GEDIAGBD value and the improved AGBD value, respectively.

[0131] from Figure 3 It is not difficult to see that the improved data R 2 Although the improvement is not significant, the RMSE is smaller, with the rRMSE decreasing by approximately 47.32%, resulting in higher accuracy than before.

[0132] Specifically, the improved rRMSE is 42.27%, a decrease of 37.97% compared to the original 80.24%, representing a reduction of approximately 47.32%; RMSE decreased from 61.63 Mg / ha to 32.46 Mg / ha, a reduction of 29.17 Mg / ha; bias decreased by 19.23 Mg / ha, a reduction of 58.03% compared to the original value; and MAE decreased by 20.06 Mg / ha.

[0133] In this experiment, only data with strong consistency with the reference data were selected. If the difference between the sample and the reference data is reduced, the accuracy of the final data will be further improved. However, this experiment did not blindly pursue the improvement of accuracy after correction. It mainly provides an idea and method. Even with the current sample data, rRMSE can still be reduced by about 47%.

[0134] Example 2

[0135] Based on Embodiment 1, this embodiment provides a GEDIAGBD data correction and quality improvement system based on three-dimensional structural information; the system includes:

[0136] The data acquisition module is used to acquire L4AAGBD aboveground biomass data, L2A canopy height data, and L2B canopy cover information data of GEDI in the study area, and to preprocess the data.

[0137] The information extraction module is used to extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD, and perform preliminary data quality screening and geolocation error correction.

[0138] The data matching and quality assessment module is used to perform biomass inversion of the airborne area based on the aboveground biomass data measured in the actual sample plots, and to perform spatial location matching with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform quality assessment.

[0139] The model training and optimization module is used to select high-quality data from the L4AAGBD dataset as training samples, train the model using an improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model.

[0140] The data correction and quality improvement module is used to correct all L4AAGBD data in the study area based on the optimized model in the model training and optimization module, so as to obtain corrected and improved L4AAGBD data.

[0141] Furthermore, in this embodiment, the data acquisition module also includes a data preprocessing unit, which is used to perform format conversion and attribute information extraction on the acquired L4A biomass data, L2A canopy height data, and L2B canopy cover information data; and to perform location matching, deletion of empty data and non-forest area data on the L2A canopy height data, L2B canopy cover information data, and L4A biomass data.

[0142] Example 3

[0143] Based on Embodiment 1, this embodiment provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0144] Example 4

[0145] Based on Embodiment 1, this embodiment provides a device, including:

[0146] Memory, used to store instructions;

[0147] A processor is configured to execute the instructions, causing the device to perform the steps of the method as described in Embodiment 1.

[0148] In summary, the method of this invention, based on a three-dimensional and volumetric perspective, integrates two improved allometric growth equations to calibrate and improve the quality of GEDI AGBD data, and then utilizes existing machine learning algorithms for further verification and supplementation. This invention can mitigate the significant uncertainty of GEDI AGBD data products and enhance the interpretability of GEDI AGBD data. Furthermore, in the quality improvement module, all feature variable data are derived from GEDI data products, making its feasibility and effectiveness superior to other related existing technologies.

[0149] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for GEDI AGBD data correction and quality improvement based on three-dimensional structural information, characterized in that, Includes the following steps: S1: Obtain L4AAGBD aboveground biomass data, L2A canopy height data, and L2B canopy cover information data of GEDI in the study area, and perform data preprocessing. S2: Extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD data, and perform preliminary data quality screening and geolocation error correction. S3: Based on the aboveground biomass data measured in the actual sample plots, biomass inversion of the airborne area is performed, and spatial location matching is performed with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform quality evaluation. S4: Select high-quality data from the L4AAGBD dataset as training samples, train the model using the improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model. S5: Based on the optimized model in step S4, correct all L4AAGBD data in the study area to obtain corrected and improved L4AAGBD data.

2. The GEDI AGBD data correction and quality improvement method based on three-dimensional structural information according to claim 1, characterized in that, The data preprocessing in step S1 includes: The acquired L4A aboveground biomass data, L2A canopy height data, and L2B canopy cover information data were format converted and attribute information extracted. Furthermore, the L2A canopy height data, L2B canopy cover information data, and L4A biomass data were matched for location, and empty data and non-forest area data were deleted.

3. The method for GEDI AGBD data correction and quality improvement based on three-dimensional structural information according to claim 1, characterized in that, The specific process in step S2 is as follows: Based on L2A canopy height data, corresponding relative canopy height data, canopy top height information, and elevation information were extracted; based on L2B canopy cover information data, horizontal information data such as total coverage, vegetation area index data, and vegetation area volume density, as well as their vertical structure information, were extracted for the corresponding geographical location; AGBD information was extracted from L4A biomass data. Based on S1, preliminary screening of the quality of L2A canopy height data, L2B canopy cover information data and L4A biomass data and correction of geolocation errors were carried out.

4. The method for GEDI AGBD data correction and quality improvement based on three-dimensional structural information according to claim 1, characterized in that, The specific process of step S4 is as follows: S41: Based on the traditional allometric growth equation of "biomass-tree height-diameter at breast height" (6-1), calculate the aboveground biomass of the forest. Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; D represents diameter at breast height (DBH); and H represents canopy top height or tree height. S42: From a three-dimensional perspective, the original biomass-tree height-diameter-thickness model is refined and updated to a three-dimensional structure model of biomass-tree height vertical structure-canopy cover, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area index, or a three-dimensional structure model of biomass-tree height vertical structure-vegetation area volume density. S43: Taking the three-dimensional structure of biomass-tree height vertical structure-canopy cover as an example, the product between tree height and tree cover is proportional to the diameter at breast height, as shown in formula (6-2); Where D represents diameter at breast height (DBH); cover represents canopy coverage; h represents height; H1 represents canopy base height; and H represents canopy top height. S44: Substitute formula (6-2) into formula (6-1) to obtain formula (6-3); S45: The canopy cover is partitioned using GEDI L2B data partitioning intervals, and the partition at each i is defined using cover... i and h i The number of i is determined by the canopy top height and the profile interval. Therefore, formula (6-3) can be converted into the following formula (6-4), and the aboveground biomass AGB is divided by the area S to obtain the aboveground biomass density AGBD formula (6-5). Where dh represents the profile interval; h i Vertical subdivision information representing relative canopy height; S represents area; S46: The parameters are standardized to obtain the final AGBD calculation model, as shown in formula (6-6); Where k0, k1, and k2 represent coefficients; H represents the height of the canopy top, and when i equals H / dh, h i =H.

5. The GEDIAGBD data correction and quality improvement method based on three-dimensional structural information according to claim 4, characterized in that, Step S4 can also employ the following process: From a volumetric perspective, the existing allometric growth equation based on biomass-tree height-canopy area / diameter is updated to derive the three-dimensional structural information relating biomass to canopy height and canopy cover (or vegetation area index or vegetation area volume density), thus obtaining a formula for calculating forest aboveground biomass. Specifically, as follows: S41: Taking the allometric growth equation of biomass-tree height-crown area as an example, as shown in equation (6-7): Where AGB represents aboveground biomass; a0, a1, and a2 represent coefficients; cover area H represents the area covered by the crown; H represents the tree height. S42: Convert the planar area of ​​the canopy into area multiplied by canopy coverage, as shown in formulas (6-8) and (6-9). Area multiplied by height can be approximated as an expression of the volume of a cylinder. After horizontal division, the volume of the entire cylinder can be considered as the sum of the volumes of individual division blocks, resulting in formula (6-10); and a new biomass density calculation formula (6-11) is obtained. Where S represents area; dh i AGBD represents the profile interval; aboveground biomass density.

6. The GEDI AGBD data correction and quality improvement method based on three-dimensional structural information according to claim 5, characterized in that, Based on formula (6-11), if equidistant partitioning is performed, dh i For fixed values, the model is directly related to the canopy top height and vertical structure of coverage.

7. A GEDI AGBD data correction and quality improvement system based on three-dimensional structural information, characterized in that, The system is used to implement the GEDI AGBD data correction and quality improvement method based on three-dimensional structural information as described in any one of claims 1-6, the system comprising: The data acquisition module is used to acquire L4AAGBD data, L2A canopy height data, and L2B canopy cover information data of GEDI in the study area, and to preprocess the data. The information extraction module is used to extract canopy height, total coverage, vegetation area index, vegetation area volume density and its vertical structure information respectively; extract attribute information that can represent data quality from each data, including AGBD, and perform preliminary data quality screening and geolocation error correction. The data matching and quality assessment module is used to perform biomass inversion of the airborne area based on the aboveground biomass data measured in the actual sample plots, and to perform spatial location matching with L4AAGBD data to obtain spatially overlapping reference AGBD data and perform quality assessment. The model training and optimization module is used to select high-quality data from the L4AAGBD dataset as training samples, train the model using an improved allometric growth equation, fit the nonlinear parameters using the GBM machine learning algorithm, compare the accuracy of different models, and obtain the optimal training model. The data correction and quality improvement module is used to correct all L4AAGBD data in the study area based on the optimized model in the model training and optimization module, so as to obtain corrected and improved L4AAGBD data.

8. The GEDI AGBD data correction and quality improvement method based on three-dimensional structural information according to claim 7, characterized in that, The data acquisition module also includes a data preprocessing unit, which is used to perform format conversion and attribute information extraction on the acquired L4A aboveground biomass data, L2A canopy height data, and L2B canopy cover information data; and to perform location matching, empty data deletion, and deletion of non-forest area data for L2A canopy height data, L2B canopy cover information data, and L4A biomass data.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the GEDI AGBD data correction and quality improvement method based on three-dimensional structural information as described in any one of claims 1-6.

10. A device, characterized in that, include: Memory, used to store instructions; A processor is configured to execute the instructions, causing the device to perform the steps of the GEDI AGBD data correction and quality improvement method based on three-dimensional structural information as described in any one of claims 1-6.

Images